Академический Документы
Профессиональный Документы
Культура Документы
Latin American
Remote Sensing
Revisión de
Week
trabajos 2010 Paper review - 2010
Con el propósito de dar a
documento que
esperamos contribuya a
difundir el conocimiento y
futuras investigaciones.
categorías.
www.lars.cl
Latin American Remote Sensing Week 2010
Página |1
Latin American Remote Sensing Week 2010
Productos y servicios
25.- CAMPAÑAS ALTIMETRICAS DE CALIBRACIÓN DEL TOPEX Y JASON-1 EN EL MEDITERRANEO
OCCIDENTAL .................................................................................................................................... 203
26.- ACURACY ASSESSMENT OF PROJECTS PERFORMED BY SAAPI SYSTEM ................................... 214
27.- EVALUACIÓN DE LA CALIDAD DE LOS DEM SRTM Y ASTER EN UNA CUENCA COSTERA DE LA
REGIÓN DE LA ARAUCANÍA, BAJO DISTINTAS CONDICIONES DE RELIEVE Y COBERTURA VEGETAL224
28.- REPRESENTACIÓN CARTOGRÁFICA DE LOS ECOSISTEMAS DEL ECUADOR CONTINENTAL ...... 231
29.- ADVANCES IN AIRBORNE LIDAR TECHNOLOGY........................................................................ 239
FOR FORESTRY AND OTHER 3D MAPPING APPLICATIONS .............................................................. 239
Página |2
Latin American Remote Sensing Week 2010
ABSTRACT:
Remote sensing technologies have seen unprecedented developments in the past
decade. New entirely digital and powerful multi-sensory systems have been
introduced regarding space, air, and ground platforms. Commercial satellite imaging
has become mainstream; state-of-the-art systems not only provide better spatial and
spectral resolution, but by using a constellation of satellites, temporal resolution has
significantly improved and, in some cases, even multiple daily coverage is available.
On airborne platforms, large-format digital aerial cameras and powerful LIDAR
systems, supported by GPS/IMU-based direct georeferencing, represent the state-of-
the-art. Also, terrestrial mobile mapping systems are rapidly spreading and are
increasingly used in a variety of applications. Powerful geospatial workstations,
offering a high level of automation, provide a common platform and a nearly identical
workflow for the data processing.
LIDAR technology, which is based on active sensing and provides a direct 3D
observation of the object space, is unquestionably the most important development
on airborne and terrestrial platforms in the past decade. Since their introduction,
LIDAR systems have seen phenomenal developments; the pulse rate increased by
two orders, multi-pulse techniques have been introduced, the use of multiple returns
with intensity has become the standard, and full waveform LIDAR data is about to
enter applications. This paper reviews the state-of-the-art in LIDAR technology, and
offers an outlook on expected future developments.
1. INTRODUCTION
LIDAR was probably the most significant remote sensing technology introduced in
mainstream mapping during this past decade (Shan and Toth, 2009). Although the
invention of laser goes back to the early 1960s, the lack of various supporting
technologies prevented its use in mapping until the mid 1990s. Most of the early
applications had nothing to do with mapping; e.g., fax machines, copiers, CD/DVD,
medical applications, etc. The first use of laser in remote sensing goes back to the
scientific mapping of ice-covered areas, primarily pursued by NASA since the late
70’s (Krabill et al., 1984). For a long while, the technology was not available to
support commercial mapping, including the high-cost of the sensor, the lack of
geodetic infrastructure and, in general, insufficient computer systems. The
introduction of direct georeferencing technologies in the mid-1990s, and general
advancements in computer technologies, enabled the development of the first
commercial LIDAR systems.
Airborne LIDAR technology, introduced to topographic mapping in the late 90s,
received wide acceptance in airborne surveying as a leading tool for obtaining high-
quality surface data in an unprecedentedly short turnaround time. The adoption of
Página |3
Latin American Remote Sensing Week 2010
the new technology was fairly smooth and quick, primarily due to the high-level of
automation of the data processing. With increasing numbers of the systems sold,
vendors were able to refine the technology rather quickly, delivering a continuously
improving performance, primarily measured in terms of the number of points per
second (PRR/PRF – Pulse Repetition Rate/Pulse Repetition Frequency). Not only has
the number of points per surface area increased, but the number of returns increased
and the intensity signal became available. Currently, we are in transition to exploit
the full waveform in post-processing, which will provide the possibility of further
analyzing the physical characteristics of the reflecting objects. Now that the LIDAR
technology has matured, the questions arise regarding what comes next and what can
be expected from the continuing improvements of technology. This paper reviews the
state-of-art and outlines some aspects of possible future directions in LIDAR.
2. STATE-OF-THE-ART TECHNOLOGY
The concept and system structure of airborne LIDAR systems have not changed over
the years, but the improvements in the performance of the components have been
significant. Figure 1 shows the data acquisition segment of LIDAR systems, including
the sensor platform (aircraft), georeferencing module, and laser sensor, which
consists of the laser source, opto-mechanical components, laser sensor, and data
recording system. Additional components of the LIDAR ground segment include
geodetic infrastructure, usually considered in the georeferencing system, and the
post-processing hardware and software.
pulses with large peak power at high repetition rate. The typical pulse width is in the
4- 10 ns range, equivalent to 1.2-3m distance, and its shape is typically described by a
Gaussian curve. The great majority of the airborne LIDAR systems work with the
1.064μ (or 1.550μ) wavelength. The LIDAR is based on line-of-sight measurements;
the sensor and the receiver are placed at the same location. When measuring the
travel time between the transmission and reception of the laser pulse, the range can
be computed as:
Where, r is the distance between the sensor and the object, c is the speed of light, and
tl is the time measured. Since time can only be measured with a given resolution, the
computed range varies in small steps. The accuracy of the range measurement
depends on the accuracy of two uncorrelated parameters:
Since c is known at high accuracy, the accuracy of time measurement determines the
ranging accuracy; in fact, the sampling of the return pulse is an extremely critical task
for the sensor system. Current airborne LIDAR systems can achieve 1-
accuracy under ideal conditions, such as perpendicular reflection from hard surfaces
with adequate reflectivity.
The ideal ranging accuracy is rarely realized in practice, as the object surface
characteristics vary over a large range due to the divergence of the laser beam. The
optical part of the LIDAR sensor module is built from conventional components,
including lenses, filters, etc. The aperture of the receiver optics is usually larger than
that of the transmitting one. The typical divergence of the laser pulse is in the 0.2-1.0
mrad range, which means that depending on the flying height, the footprint is
generally in the 15cm –
grow for larger angles (such as slanted surfaces). The shape of the return pulse
should consequently change, impacting the ranging accuracy for larger footprints. In
addition, the reflectivity of the surface defines the amount of energy reflected back to
the sensor. Simply stated, the incident angle, the reflectivity of the surface, and the
flying height together determine the shape and strength of the return signal, and thus
the achieved accuracy. Therefore, LIDAR vendors usually provide conservative
numbers in product specification for vertical accuracy; typically, an order larger than
the theoretical ranging accuracy. Figure 2 shows an accuracy specification for the
Leica ASL60 system.
Página |5
Latin American Remote Sensing Week 2010
The deformation of the return pulse could be more significant than an increased
footprint size for vegetated areas and for complex objects, which cannot be described
by a single elevation value. In these cases, instead of a single return pulse, there could
be more separate and/or overlapped pulses in the return signal, which is called
waveform or full waveform. Figure 3 shows three typical waveforms. The return
signal processing has improved a lot since the first LIDAR systems were introduced,
when a simple thresholding provided the measurement, the detection of a single
pulse. All state-of-the-art systems sample the return signal and store it temporarily
for pulse detection, which also includes an intensity signal determination from the
shape and strength of the detected pulse. Normally, up to four returns, first, second,
third, and last, are recorded along with intensity data. The separation of the returns is
practically defined by the emitted laser pulse width; say, for a 10 ns pulse, the object
surfaces should be at least 3m apart to produce separate pulses. Once the pulses are
extracted in real-time, the waveform is discarded, except if a waveform recorder is
used to store the full waveforms (as opposed to pulses). Processing the waveforms in
the office has several advantages. More sophisticated pulse detections can be applied,
and additional object information can be extracted that could be used, for example,
for classification. In particular, overlapping pulses can be better extracted or more
complex shape patterns can be identified. Full waveform is important for mapping
forested areas, where in addition to the terrain, biomass estimation and tree
structure/type extraction can be performed. From the shape of the waveform,
additional properties of the reflecting surface can be extracted, such as land
information, i.e., the classification of the reflecting objects into categories, such as
trees, other vegetation, roofs, pavement, etc.
Página |6
Latin American Remote Sensing Week 2010
To create a 2D range image from using a single laser beam, a scanning mechanism is
required, which usually provides scanning in one plane, and the other direction is
provided by the aircraft motion. There are a variety of scanning solutions, including
oscillating mirrors, rotating polygons, rotating prisms, and Palmer scanner. In
addition, there is a fiber scanning solution, allowing for very fast scanning rates. In
modern LIDAR systems, the scan rate falls into the 100-200Hz range, and can be
adjusted, as needed, to provide for nearly even LIDAR point distribution; the flying
height and PRF are the other parameters that affect the point distribution. The
Página |7
Latin American Remote Sensing Week 2010
scanning components, being the only mechanical parts of LIDAR systems, are subject
to wear and require frequent calibrations.
Based on Eqn. (3), the laser point accuracy in a mapping frame can be determined by
applying the law of error propagation; see earlier discussions on the errors and
model parameter recovery in (Schenk, 2001; Filin, 2003a and 2003b; Baltsavias,
1999) and a recent comprehensive discussion on the various terms of the error
budget, including analytical and simulation results can be found in (Csanyi May,
2007). Practically, the main error sources of the error budget can be grouped as:
-based navigation solution of the
sensor platform.
reflectance-based corrections.
-sensor calibration, such as boresight misalignment and lever arm errors.
beam divergence, time synchronization, atmospheric
effects, system stability, and terrain/object characteristics.
Página |8
Latin American Remote Sensing Week 2010
Since state-of-the-art systems provide excellent ranging accuracy, ~1cm (1ς). Since
system calibration can be done accurately and the typical object distance is not large,
the single most important component of the error budget is the platform navigation
accuracy. The GPS performance defines mainly the error budget of any integrated
GPS/IMU-based system, therefore, the GPS positioning method applied in the
georeferencing system is of prime importance in achieving the highest accuracy of
the LIDAR point cloud. Table 1 shows GPS technologies applicable to kinematic
positioning.
In current practice, the differential GPS with base station method is the most widely
used technique in airborne LIDAR data acquisition; it provides excellent accuracy
close to the base station, but over larger distances (> 50km) the accuracy
deteriorates. In contrast, the network-based differential solutions have no distance
dependency, and can provide a consistent and highly accurate solution, provided an
adequate network of CORS (Continuously Operating Reference System) is available;
these networks are rapidly developing and already available in the developed world.
The typical accuracy for a base station-based GPS solution falls, in general, into the 4-
7 cm range for shorter baselines. A network-based solution can reach 2-3 cm-level
accuracy. Note, assuming a medium grade IMU sensor, the attitude accuracy is
comparable to the positioning accuracy; the shift on the ground caused by attitude
error is comparable to the positioning error.
parameter, which primarily defines the average point density on the ground. The
current 400 kHz rate is by orders better than the 5-10 kHz PRF of the early systems
ten years ago. The difficulty of achieving high pulse rate comes from implementation
limitations. First, the pulse travel time imposes a limit on the maximum rate; say, for
a 1500m flying height, the pulse travels about 10μs, so the PRF should be less than
100 kHz. Figure 4 shows the relationship between the flying height and PRF for the
Leica ALS60 system.
There are two solutions available to further increase the pulse rate. First, if the
system does not have to wait for the arrival of the return pulse, several pulses can
simultaneously travel between the sensor and object surface; this solution is usually
called MPiA (Multiple Pulses in Air). Figure 4 shows the situation for two pulses.
Obviously, having multiple pulses in the air means that there is an ambiguity about
the measured range, which, given rather large distances between the pulses, can be
resolved in a robust way. Figure 5 shows the operational parameters for the Reigl
LMS-Q680i system, which boosts a 400 kHz PRF with three-pulse MPiA; note that the
In theory, the MPiA technology allows for a large number of pulses, but the
generation of laser pulses at such a high repetition rate is becoming an
implementation issue. A simple way to overcome this limitation is if multiple laser
sensors are used in the system, frequently called multi-channel LIDAR systems. The
Optech Pegazus system is a 400 kHz multi-channel system that was recently
introduced. To illustrate the point density with respect to the flying height, aircraft
speed and scan angle, Figure 6 shows these parameters for the Reigl LMS-Q680i
system.
P á g i n a | 10
Latin American Remote Sensing Week 2010
While the focus is mostly on the data acquisition capabilities of airborne LIDAR
systems, the ground processing aspects should be briefly mentioned too. In fact, there
was a significant gap between the data acquisition capabilities of LIDAR systems and
processing software at the beginning (ten years ago), as only vendor-specific basic
solutions were available. Fortunately, the LIDAR processing software market has
significantly improved recently, and there are several powerful packages available to
support the effective processing and visualization of the LIDAR point cloud. The
sophistication of the point cloud has also increased. In addition to the basic bare
earth extraction and surface modeling, there is an increasing need for objects feature
extraction, such as building extraction. Obviously, the number of applications is also
increasing. From the early applications in DEM/DSM extraction for topographic
mapping, corridor and power line mapping, and forestry use, now applications
include floodplain mapping, bathymetry, city modeling, biomass estimation, glacier
and snow mapping, agriculture, transportation, and, in general, object feature
extraction (Vosselman and Maas, 2009).
P á g i n a | 11
Latin American Remote Sensing Week 2010
3. FUTURE TRENDS
Reviewing the history of airborne LIDAR technology, it is clear that developments
were primarily driven by users’ needs (except for the early years). In other words,
system manufacturers have technologies already available to improve the
performance, and it is up to the users whether they can afford the higher price for
these improved capabilities. For example, waveform technology has been available
for about five years, yet, this capability has started to make inroads into mainstream
applications just recently. While it is rather difficult to make predictions in high-tech
system future developments, the likely short-term trends are:
channel system design, and it is likely to pass the one million PRF mark in a few
years. In terrestrial laserscanning, the Velodyne system, which uses 64 laser sensor
units, can already acquire one million points per second at shorter distances.
-time
processing to post-processing, as in most applications, there is no need for real-time
pulse detection, since accurate georeferencing is usually not available during the data
acquisition. The excessive volume of waveform data used to present a storage
capacity problem in the past, but as hard and solid state drive technologies continue
to improve, the space is not an issue anymore. In addition, various compression
technologies are also available.
also benefit in both better recovery of the object geometry and the classification of
objects.
4. SUMMARY
LIDAR systems and the laser scanning technology will continue to advance, following
similar trends in remote sensing imagery developments. The past gaps between
terrestrial and airborne point clouds and LIDAR images are not only disappearing,
but the overlap between the imagery acquired from various platforms will continue
to grow. In addition, airborne LIDAR technology, the primary source of surface data
at the local scale, is likely to extend to the feature extraction field, as point cloud
density continues to increase.
References:
Airborne LIDAR Sensors, Product Survey, GIM International, February, 2009.
Baltsavias, E.P., 1999. Airborne laser scanning: basic relations and formulas. ISPRS
Journal of Photogrammetry and Remote Sensing, Vol. 54: 199-214.
Csanyi, N. and Toth, C. (2007): Improvement of LIDAR Data Accuracy Using LIDAR-
Specific Ground Targets, Photogrammetric Engineering & Remote Sensing, Vol. 73,
No. 4, pp. 385-396.
Csanyi May, N. 2008. A Rigorous Approach to Comprehensive Performance Analysis
of State-of-the-Art Airborne Mobile Mapping Systems, PhD dissertation, The Ohio
State University.
P á g i n a | 13
Latin American Remote Sensing Week 2010
Filin, S., 2003a. Recovery of systematic biases in laser altimetry data using natural
surfaces, ISPRS Journal of Photogrammetric Engineering and Remote Sensing, 69
(11), 1235–1242.
Filin, S., 2003b. Analysis and implementation of a laser strip adjustment model.
International Archives of Photogrammetry and Remote Sensing, 34 (Part 3/W13):
65-70.
Krabill, W.B., Collins, J.B., Link, L.E., Swift, R.N., and Butler, M.L., 1984. Airborne laser
topographic mapping results, Photogrammetric Engineering and Remote Sensing, 50,
685-694.
Mallet, C., Bretar, F., 2009. Full-waveform topographic LIDAR: State-of-the-art, ISPRS
Journal of Photogrammetry and Remote Sensing, Volume 64, Issue 1, pp. 1-16.
Pfeifer, N., and Briese, C., 2007. Geometrical Aspects of Airborne Laser Scanning and
Terrestrial Laser Scanning, International Archives of Photogrammetry, Vol. 36 (Part
3/W52): 311-319.
Schenk, T., 2001. Modeling and Analyzing Systematic Errors in Airborne Laser
Scanners, Technical Notes in Photogrammetry, Vol. 19, The Ohio State University,
Columbus, USA, 46 pages.
Shan, J., Toth, Ch. 2009. Topographic Laser Ranging and Scanning: Principles and
Processing, CRC Press, 608 p.
Toth, C., 2004. Future Trends in LIDAR, Proc. ASPRS 2004 Annual Conference,
Denver, CO, May 23-28.
Vosselman, G., Maas, H-G. 2009. Airborne and Terrestrial Laser Scanning, Whittles
Publishing, 320 p.
P á g i n a | 14
Latin American Remote Sensing Week 2010
Abstract: Several processing techniques (or methods) for ASTER imagery are based on
spectral similarity between laboratory spectra and imagery spectra (pixel data). On the other
hand there are processing techniques based on basic or more advanced band math
operations, where bright pixels match certain spectral features. This process is also associated
with a certain level of thresholding defined by the user. The “Decision Tree Classification” is a
useful tool to select pixel spectra that matches exactly the feature you want to highlight, based
on the features that the user considers important to define a mineral or a mineral association.
INTRODUCCIÓN
Las curvas espectrales de cualquier tipo de material, son la base para el procesamiento de
imágenes en la percepción remota, cuando podemos obtener una imagen hiperespectral, las
comparaciones entre la curva de laboratorio y la curva del pixel entregan excelentes
resultados, con una identificación de los elementos de la superficie muy precisa, sin embargo,
cuando las imágenes son multiespectrales (Landsat/Aster) la curva espectral sólo muestra
parte de la información, por lo que el análisis es más complejo y limitado. Es aquí donde nacen
muchas metodologías para procesamiento donde se estudian características específicas de a
curva espectral en longitudes de onda específicas para generar algoritmos matemáticos,
análisis de componentes principales, razones de bandas, entre otras. La problemática cuando
se procesa una imagen, sin saber lo que hay realmente en terreno, es que estas metodologías
entregan como resultados, para cada índice, una banda en escala de grises, donde los pixeles
con mayores valores (o según la metodología, los menores) son los que representan el
material que se busca en la escena, sin embargo, aunque dicho material no se encuentre
presente en la zona de estudio, como el resultado es relativo, siempre tendremos un valor
mayor que otro que puede generarnos falsas anomalías en la interpretación. La “clasificación
por árboles de decisión” intenta eliminar este resultado relativo y solo localizar aquellos
pixeles que cumplan ciertos requisitos definidos por el usuario, para tener una mejor certeza
en nuestros resultados.
P á g i n a | 15
Latin American Remote Sensing Week 2010
ASTER
ESPECTROMETRIA DE IMAGENES
Las curvas de los pixeles se comparan con las curvas de laboratorio y de esta manera se
determinan en la imagen los pixeles que corresponden a cada material. Cuando tenemos
imágenes hiperespectrales podemos determinar con precisión los materiales, ya que la
complejidad de las curvas que entregan esas imágenes no difieren mayormente con las curvas
medidas en laboratorio. Sin embargo, cuando tenemos imágenes multiespectrales, se genera
una gran diferencia entre la curva de laboratorio y la curva que puede observarse desde el
sensor. Es por ello que las metodologías de procesamiento deben ser enfocadas a
P á g i n a | 16
Latin American Remote Sensing Week 2010
Figura 1: Curvas espectrales de laboratorio y su ajuste a las bandas VNIR + SWIR de ASTER
La vegetación, por ejemplo, tiene un gran salto de magnitud de la reflectancia entre la banda
roja (banda 2 -0.66 µm) y la banda infrarroja cercana (banda 3 – 0.82 µm). Si tenemos la curva
completa del pasto y de otro elemento vegetal (algún tipo de matorral o árbol) podremos
identificarlos y clasificarlos como distintos en la imagen, pero con la curva limitada que
tenemos en la imagen ASTER, no podremos hacer esa diferenciación y nos tendremos que
conformar con decir que el pixel corresponde a un elemento vegetal, lo mismo ocurre con
todos los elementos que se quieren mapear en una imagen, necesitamos saber cuál
característica es visible con la imagen y cómo vamos a manejar la información de las bandas
para resaltar su presencia.
P á g i n a | 17
Latin American Remote Sensing Week 2010
“El clasificador de árboles de decisión realiza clasificaciones de varias etapas utilizando una
serie de decisiones binarias para colocar píxeles en clases. Cada decisión divide los píxeles de
un conjunto de imágenes en dos clases basadas en una expresión. Puede dividir cada nueva
clase en dos clases más basada en otra expresión. Puede definir tantos nodos de decisión,
según sea necesario. Los resultados de las decisiones son clases. Usted puede utilizar los datos
de diferentes fuentes y archivos para hacer un clasificador de árbol de decisión individual.
Puede editar y "podar" los árboles de decisión de forma interactiva, y también puede salvar
los árboles y aplicarlos a otros conjuntos de datos”. (Ayuda ENVI – Decision Tree Classifier).
P á g i n a | 18
Latin American Remote Sensing Week 2010
Realizaremos una revisión de los minerales que queremos identificar, como es su curva
mirada desde el sensor ASTER y como podríamos construir el árbol de decisión para clasificar
dichos elementos. Enfocaremos este estudio para determinar presencia de Alunita, Kaolinita,
Muscovita y Calcita.
La figura 2 muestra las curvas espectrales de los minerales a estudiar, con el ajuste a la
información espectral que lee la imagen ASTER en el canal SWIR. Para realizar la clasificación
por árboles de decisión, enfocaremos las condiciones del árbol en las absorciones que
presentan los distintos minerales, estas absorciones pueden ser absolutas o relativas (figura
3).
P á g i n a | 19
Latin American Remote Sensing Week 2010
Las principales características que se observan en las curvas de los minerales se resumen en
la tabla 3:
Para poder detectar, dentro de la imagen, los pixeles cuyas curvas cumplen con esta
característica, debemos primero realizar una corrección atmosférica a la imagen para poder
tener los datos en valores de reflectancia.
Si consideramos la figura 3 como referente, tenemos 3 puntos (A,B y C) que reflejan ambos
tipos de absorciones, las condiciones que podemos ocupar en el árbol de decisión serían un
valle para la absorción absoluta y una inflexión en la absorción relativa, matemáticamente
podría escribirse de la siguiente forma:
Para tener una absorción relativa, necesitamos que haya un cambio de pendiente entre AB y
BC, que pase de una pendiente menor a una mayor.
De esta manera podremos construir nuestro árbol de decisión con las condiciones lógicas
antes mencionadas. Asignaremos las siguientes siglas para la representación del árbol:
• ABS-BX : Absorción absoluta en la Banda X
• RBD-X : Absorción relativa en la Banda X
P á g i n a | 20
Latin American Remote Sensing Week 2010
Los pixeles asignados a cada clase (Alunita, Calcita, Muscovita, Kaolinita), deberan entonces
cumplir con las absorciones estipuladas en el árbol de desición.
“El distrito Cuprita se encuentra en el extremo suroeste de la Gran Cuenca a unos 15 km al sur
de Goldfield, Nevada. Ocurrencias de cobre, plata, oro y plomo han sido reportados en calizas
del Cámbrico, y ocurrencias de azufre en el rocas sedimentáreas tobáceas y flujos de tobas de
ceniza soldadas del Terciario”. (Rowan, Hook y Mars, 2003).
Este distrito ha sido por excelencia la zona de prueba de muchos sensores multiespectrales e
hiperespectrales. Destacan en él distintos tipos de alteraciones hidrotermales, con extensas
apariciones de Alunita, Kaolinita, Muscovita, Calcita, Dickita, Buddingtonita, Ópalo, entre otros
(figura 5).
P á g i n a | 21
Latin American Remote Sensing Week 2010
P á g i n a | 22
Latin American Remote Sensing Week 2010
Figura 8: Clasificación final por Arboles de Decisión de imagen ASTER en el distrito Cuprita
La figura 9 muestra el mapeo realizado con el árbol de decisión final y el mapeo realizado con
la imagen hiperespectral AVIRIS, los colores de los minerales en el mapeo de la imagen ASTER
se modificaron para una mejor comparación entre los resultados de ambos procesos.
Figura 9: Mapeo en imagen ASTER (izquierda) v/s Mapeo en imagen AVIRIS (derecha)
Alunita Kaolinita Calcita Muscovita
P á g i n a | 23
Latin American Remote Sensing Week 2010
CONCLUSIONES
La clasificación por árboles de decisión es una herramienta muy útil para clasificar elementos
en imágenes multiespectrales, permite detectar pequeñas variaciones de las curvas
espectrales de los pixeles, que a su vez, marcan la diferencia entre un mineral y otro. Los
resultados obtenidos en Cuprita se asemejan con muy buena precisión al mapeo
hiperespectral realizado con imágenes AVIRIS, a pesar de solo tener 6 bandas en el infrarrojo
de onda corta (SWIR) mientras AVIRIS posee 50 bandas en dicho rango.
Esta metodología el aplicable para cualquier área de las geociencias, la construcción del árbol
de decisión va de la mano del conocimiento espectral que se tenga sobre los elementos que se
quieren mapear, y de las posibilidades que nos da la curva de ser diferenciable de otros
elementos.
REFERENCIAS
Abrams, M., Hook, S., Ramachandran, B. 2002. ASTER User Handbook – Version 2.
Rowan, L., Hook, S. y Mars, J., 2003. Mapping hydrothermally altered rocks at Cuprite, Nevada,
using the Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER), a new
satellite-imaging system. Economic Geology, Vol. 98.
Rowan, L., Mars, J. 2003. Geologic mapping and ASTER data analysis. USGS presentation.
ENVI Help
P á g i n a | 24
Latin American Remote Sensing Week 2010
Summary: The paper presents the first part of a project whose goal is to find indicators of
harmonic development of towns based on analysis of forty years development of fifty Czech
towns. The presented results show relation between urban land use classes and their
development of selected towns and road transport density. First thirty towns were processed
for the period between 1970 and 2009; first ten towns were mapped according to general
urban land cover classes and compared to the measured road density in large point network.
The town land use class areas were derived from combination of actual and historical city
plans, aerial photograph from various years, and Thematic Mapper data using GIS tools. The
road transport density measured repeatedly since 1973 to 2005 each 10 years was used for
all towns where the measurement was performed on pre-defined locations in towns and out
of towns. It was found that the traffic density within towns and to and from towns is more
depend on existence of close highways and by-pass roads unlike number of inhabitants, e.g.
Political change from the communist regime to the democratic one was also an important
breakpoint in town developments. Increase of the traffic density and enlarging of residential
areas are the characteristics proving the fact. The paper presents a methodology of spatial
mapping of land use classes utilized for determination of town development. The town
developments and their relation to road traffic is presented on maps and graphs.
1. INTRODUCTION
Urban development and its indicators are a matter of many projects, conferences and
meetings. An example of large set of indicators describing urban development was presented
in Global City Indicators Program. Many authors are discussing future development of urban
areas and definition on its indicators (Brugmann, 1997.)
The project is focused on detailed evaluation of relations between quality of life and present
behavior of the human society to create a model allowing improving the present state and less
demanding development in urban areas for their sustainable development. One fifth of the
Czech Republic population is living in three largest towns – Prague, Brno and Ostrava.
The spatial towns development was derived from city plans designed by local administration,
from aerial photographs in 1950, and 2008, and satellite (Thematic Mapper, MSS data) images
covering period between 1972 and 2000. The system of individual time level determination of
town maps was created from map vector data representing town land use maps and the
remote sensing data starting by the latest situation and heading the previous levels. The land
use maps are controlled by the Czech Statistical Office and Czech Office for Surveying,
Mapping and Cadastre data.
Large data base of road network development has been created by the Road and Motorway
Directorate. The data base comprises among others measurements of road transport density
in many points of roads of various road classes since 60-ies of the 20th century. The paper
presents relations between transport density and its development, and functional typology of
spatial changes in towns helping to find tools for the extrapolation of town developments and
their impact on transport density using remote sensing and GIS data.
P á g i n a | 26
Latin American Remote Sensing Week 2010
This spatial development is archived in Cadastre books in table systems and transfer of this
information can show towns development using cadastre boundaries. This transfer was
performed also into town city plans, whose processing intervals vary in individual towns. This
town size evolution cannot be derived from remote sensing data. If a town belongs to the
second or third group, there are large spatial changes.
The largest parts of these changes are in prevailing part represented by agricultural areas.
The main difference between the core area and associated area are separated urban parts
occurring in associated parts.
The first step of preparing urban land use classes is reclassification of detailed legend classes.
The final classes were residential, production, recreational, and facility areas, green ones,
forest, and traffic areas. These classes are called functional classes. Each class is formed by
higher number of city plan classes. The residential area is formed by mixed residential region,
general residential and rural ones and public areas. The reclassification means also including
of local roads classified as roads of low level in the state roads hierarchy into residential and
other areas. The reclassification is performed individually for each town according to its city
plan classes.
The next step was a verification of present city plans and real state of towns as the plans
comprise plans which may and really differ from the real state. This part of the processing was
done by visual interpretation combined with registration of changes of the city plan and the
result was a map of functional classes of the present state.
P á g i n a | 27
Latin American Remote Sensing Week 2010
The development for forty years was divided into determination of previous phase maps of
functional classes in 2000, 1990, 1980 and 1970. The process started in the latest period and
ended in 1970. These maps were processed using remote sensing data. Two sorts of remote
sensing data were used for change detection allowing mapping a given year. The latest map
(2008) was a result of the present city plan processing by implementing corrections found in
discrepancies between the plan and aerial orthophotographs. Satellite data - Thematic
Mapper data - were utilized for determination of land cover changes. They were derived from
subtraction of original bands and normalized vegetation index for satellite remote sensing
data. Found changes were pixels with high positive or negative values. To find changes
between 2009 and 2000, two TM were used for finding the differences. This approach yielded
areas with different land cover, however, there was an additional task to determine and
“translate” a land cover change into a land use change. Each functional class comprises a wide
range of land cover classes in aerial photographs spatial resolution; however, these detailed
classes are not in prevailing part detectable on Thematic Mapper data. The Thematic Mapper
resolution does not allow determining urban functional classes – agriculture area spectral
behavior can be for some plants similar to vegetated areas, etc. The areas with important
changes were verified using the aerial photograph taking into account also their shape and
texture. Satellite image pairs were always used for change detection. The oldest map showing
the 1970 year was also visually controlled using aerial orthophotomosaic created from aerial
orthophotographs mostly in 50-ies in the last century.
All functional classes were controlled by table data available at the Czech Office for Surveying,
Mapping and Cadastre for the town administrative area. The functional classes in individual
years were used for further evaluation between road transport density, town development
and investments into road network in the form of new by-pass, highways, etc. Following
indicators showing the relation were: development of functional class areas, development of
number of inhabitants, development of road transport density, and building of new
decongesting roads.
5. RESULTS
Kladno is an example of processed towns. The town consists both of core, and associated
parts. It is situated 30 km north-west from Prague. Fig. 1 shows spatial changes of four
functional classes between 1969 and 2009 mapped according to above mentioned method.
P á g i n a | 28
Latin American Remote Sensing Week 2010
Fig. 1. Development of Kladno functional classes for the 1969 – 2009 period
Town development will be further presented on five graphs. The first one (a) shows
administrative, core and associated areas with their built-up and non-built-up area. The other
represents built-up (b) core area Looking at the statistical evaluation presented in Fig. 2, we
can see that it was the area of production whose growth was the steepest in the core part.
Development of residential parts was higher than that of traffic areas within the core region
during last 20 years. However, there is a new highway passing the town in 5 km distance
enabling the town to be used less for a passing through by road traffic. The town has not yield
larger areas for recreation, leisure time, sport, etc. during last 40 years (Fig. 2).
P á g i n a | 29
Latin American Remote Sensing Week 2010
2000
non- built-up core
area
1000 associated area
0 built-up associated
1969 1982 1989 2009 area
forest area
300
recreational area
200
water
100
area of facilities
0
1969 1982 1989 2009 area of production
Fig. 2. Example of spatial development for 40 years shown in land use classes.
Administration area has not changed since 1982. The core area changed unlike most towns
between 1982 and 1989. Residential area and area of production cover similar part of the
built-up area, however the growth of production area growth is steeper. Non-built-up area is
in prevailing part forested. Associated areas are formed by agricultural and forest land,
however, their size declines after the political change in 1989.
Regular measurement of road traffic density proves intensive growth of road intensity
calculated as number of vehicles per 24 hours. Comparing the town development with traffic
density using intensity of traffic (number of vehicles per 24 hour/ha), we can see a following
trend. The traffic density has grown intensively since 1990. It is a year of economical change
due to the political change in 1989 from communist regime.
P á g i n a | 30
Latin American Remote Sensing Week 2010
Fig. 3. Sum of all measured segments on higher level roads used also for passing
transport and on local roads (left). Traffic intensity calculated as a ratio of all vehicles per 24
hours and size of functional areas (right).
Investments into highway and by-pass road construction can be easily recognized from
following two graphs. Ten towns with the highest number of vehicles per 24 hours entering
and leavig each town were selected and compared to number of their inhabitants.
P á g i n a | 31
Latin American Remote Sensing Week 2010
1800000
1600000
obyv_1965 obyv_1970 obyv_1980
1400000
1200000
1000000
800000
600000
400000
200000
0
Fig. 4. Comparing of traffic density since 1968 to 2005 and number of inhabitants in similar
periods
The Fig.4 shows the important influence of by-pass roads. Mělník has a very low number of
inhabitants in last 40 years compared to Ostrava; however, numbers of measured vehicles
leaving and coming to both towns are very similar. Mělník does not have any by-pass road and
lies on direction among Prague and other important Czech towns. If we analyse Kolin and
Hradec Králové their traffic density and number of inhabitants have analogic conditions.
6. CONCLUSIONS
Relations between town development and road traffic density showed interesting
dependences. The paper presents a very small analysis performed only for functional classes
and road traffic density. Changes in studied characteristics in 50 towns have already brought
a great deal of information. However, the chalenge is to determine a list and sequence of
indicators influencing the relations and their sustainable development.
P á g i n a | 32
Latin American Remote Sensing Week 2010
BIBLIOGRAPHY
Brugmann, J. Is there a method in our measurement? The use of indicators in local sustainable
development planning , Local Environment, Volume 2, Issue 1 February 1997 , pages 59 – 72
Vepřek, K. et al. Analysis of 100 years urban development of Hradec – Pardubice regional
agglomeration focused on detection of general tendencies and regularity, Terplan Praha,
1983, – research project VÚ P 16-521-503
Acknowledgement
The project is financed by Modeling of urban areas to lower negative influences of human
activities project of the Ministry of Education (OC1011), Czech Republic
P á g i n a | 33
Latin American Remote Sensing Week 2010
Samuel Martins da Costa Coura National Institute for Space Research – INPE
samuel@dsr.inpe.br
Gabriel Pereira – gabriel@dsr.inpe.br Yosio Edemir Shimabukuro – yosio@dsr.inpe.br
Flavio Jorge Ponzoni – flavio@dsr.inpe.br Andre Lima – andre@dsr.inpe.br
Daniel Capella Zanotta – zanotta@dsr.inpe.br
Summary: Los Árboles fuera del bosque (AFB o TOF, por sus siglas en ingles Trees Outside
Forest) se refieren a los árboles en el paisaje que no están definidos como bosques ni en
ninguna otra área boscosa. Estos pueden tambien incluir tierras agrícolas, comprendiendo
las áreas de praderas y pastizales, zonas edificadas y tierras áridas. La deforestación ha sido
cartografiada y cuantificada, pero muy poco se sabe sobre el destino de esas tierras que antes
estaban en bosque; la tala del bosque a menudo es seguida por el establecimiento de sistemas
de producción de la cual los árboles son una parte integrante. El objetivo de este trabajo es
evaluar el uso de diferentes imágenes de alta resolución (QuickBird, Ikonos, Eros y SPOT-5)
en el Bosque húmedo tropical atlántico Brasilero para precisa y clasificar los TOF, utilizando
diferentes métodos de tratamiento digital de imágenes, usando herramientas básicas de
procesamiento de imágenes (como la segmentación). La metodología utilizando
procesamiento de imágenes de alta resolución y SIG pueden ser efectivamente empleados
usando imágenes de satélite de alta resolución para identificar y cartografiar los recursos de
los TOF. Los resultados preliminares muestran que los árboles pueden ser cartografiados a
través de métodos de segmentación de imagen e interpretación visual, sin embargo, más
testes se están realizando para proporcionar algoritmos automatizados para extraer esos
datos. El aspecto del costo siempre debe contar a la hora de elegir la forma de identificar y
cartografiar los TOF, va depender de la escala y la precisión deseada, las opciones varían de
una amplia gama.
Palabras Importantes: árboles fuera del bosque, remote sensing,high resolution.
Abstract: Trees outside forests (TOF) refers to trees on land not defined as forest and other
wooded land. This may include agricultural land, including meadows and pasture, built-on
land and barren land. TOF as a separate entity, despite their ecological, economic and social
importance as (to estimate carbon sequestered in TOF, to evaluate the role of TOF in the
context of timber production, to estimate the volume of standing trees outside the forest area
etc.) has a great role in many environmental processes. Deforestation has been mapped and
quantified, but very little is known about the fate of land formerly under forest; forest clearing
is often followed by the establishment of production systems of which trees are an integral
part. The objective of this work is to evaluate the use of different high resolution imagery
(Quickbird, Ikonos, EROS and SPOT-5) in Brazilian Atlantic Rain Forest in order to indentify
and classify TOF using different methods of digital processing images, using basic tools of
image processing (such as segmentation). The methodology using high resolution imagery
processing and GIS can be effectively employed using high-resolution satellite imageries to
identify and map the TOF resources. The preliminary results depicts that the trees can be
mapped through image segmentation methods and visual interpretation, however more tests
are underway to provide automatized algorithms to extract such data. The costs aspects must
P á g i n a | 34
Latin American Remote Sensing Week 2010
always count when choosing how to identify and map TOF, depending on the scale and
accuracy desired the options vary from a wide range.
Keywords: tree outside forest, remote sensing, high resolution.
1-INTRODUCTION
The world has billions of trees that are not included in the FRA 2000 (Forest Resources
Assessment) definitions of “forests” and “other wooded land”. Trees outside forests (TOF) - i.e.
trees available on agricultural land, along road, railways, canals, ponds, orchards, parks,
gardens and homestead plays many role like forests. They make a critical contribution to
sustainable agriculture, food security and rural household economies; they supply many
products and services similar to forests; and they protect crops and the soil against water and
wind erosion, thus combating drought and desertification and protecting water resources
(FAO, 2010; Rawat et al., 2004).
Trees outside forests have been defined differently by different countries and international
agencies. In India, TOF is defined as all those trees, which have attained 10 cm or more
diameter at breast height, available on lands, which is not notified as forests. However, FAO
defines TOF as trees available on lands which is not defined as “forests” or “other wooded
land” (FAO, 2010; Kumar, 2006).
Deforestation has been mapped and quantified, but very little is known about the fate of land
formerly under forest; forest clearing is often followed by the establishment of production
systems of which trees are an integral part. Not much is known about the dynamics of trees on
farmlands and their corresponding contribution to the production of wood and other
products and services. Similarly, little is known about changes in tree cover in fields and
urban systems. (FAO, 2010).
The increasing forest cover loss and its fragmentation on one hand, the need to conserve
remnants of representative forest ecosystems and the increasing demand of forest products in
developing countries (Mateo, 1998; Rodríguez 1998; Salas, 1998) on the other hand, make the
development of innovative sustainable management tools imperative for other less studied
tropical forest resources such as trees outside forest (Herrera-Fernández, 2003).
Currently it is recognized that TOF embrace not only many ecological functions, such as
conservation of biodiversity, erosion control, and carbon sequestration, but also economic
functions, such as provision of firewood, fodder, fence posts, and living fence posts. In the
socioeconomic context, fuelwood is of particular interest, because it remains the first source
of energy in developing countries, representing almost 81% of the total wood harvest.
Although there is no precise data about the contribution of fuelwood from TOF as a
proportion of total fuel consumption, it is clear that agroforestry systems in developing
countries provide a large part of this resource (FAO,1999, 2010; Herrera-Fernández, 2003).
The objective of this paper was to identify and to evaluate the potential of 4 (four) subsets of
high spatial resolution imageries (Quickbird, Ikonos, SPOT-5 and Alos/AVNIR) to identify
trees outside forest (TOF) in the Brazilian Atlantic Rain Forest. This ecosystem has underwent
for a huge degradation since the early times of Brazil’s history, nowadays just remaining
around 7% of its initial area. The Atlantic Forest (mainly located along the Brazil’s shoreline)
P á g i n a | 35
Latin American Remote Sensing Week 2010
is now designated a World Biosphere Reserve, which contains a large number of highly
endangered species. It has been extensively cleared since colonial times, mainly for the
farming of sugar cane and for urban settlementsThis ecosystem is the richest in terms of
biodiversity in the whole world (SOS Mata Atlântica, 2010). Both visual and automatic
approaches will be used in this preliminary study.
The remote sensing data can provide stratification of the TOF resources, which can be utilized
to increase the precision level and may turn out time effective. Some time the objectives of
TOF resource assessment may require spatial distribution of resources on maps along with
several other features. This objective can be appropriately tackled by the use of aerial
photographs and satellite imageries in the assessment procedure (FAO, 2010). The choice of
tools and methods used to describe or assess trees outside the forest depends on the scale of
analysis, kind of data and degree of accuracy desired. The tools used are not generally specific
or new; rather, they are combined and implemented in original ways (Rawat et al. 2003).
The most commonly used remote sensing technology for TOF resources is aerial photography,
which can be used to describe spatial distribution and to distinguish TOF cover classifications,
providing the appropriate scale is chosen. However, high costs prohibit widespread use of
aerial photography for TOF assessments in most countries. The new 1-5m high resolution
satellite sensors represent a possible future alternative to aerial photography (FAO, 2010).
Some TOF field inventories are modeled on forest inventory methods and keep to biological
and physical criteria; others emphasize social aspects, choosing villages as the sampling units.
For measurements on the ground, sampling arrangements designed for forest stands may not
be the most effective arrangements for trees. Less traditional sampling plans which would
theoretically be better suited to this resource should be tested on various categories of TOF,
especially those covering fairly large areas, in Brazilian National Forest Inventory, which is
underway. TOF’s are supposed to be mapped using ALOS/AVNIR imagery with 2.5 m of spatial
resolution.
The imagery used in this research was previously processed, and then a data base was created
in order to apply the segmentation algorithms. The GIS SPRING® (free software) was the one
chosen to process and analyze the data.
All images subsets were submitted to the same process of segmentation with different
parameters of area and similarity, which define the size of the polygons delineated on the
images compositions and the variation the image’s gray level. Several thresholds were applied
in order to find out which one was more suitable to delineate the individual, vicinal or
aggregate of trees. In each situation (image subset) the thresholds varied due to the image
spatial resolution and many variables related to the landscape due to pattern and spatial
distribution of the TOF’s along the scene. Circumstantial and detailed visual interpretation
was done before applying any algorithm.
P á g i n a | 36
Latin American Remote Sensing Week 2010
In the Figure 1 from ALOS/AVNIR (06/20/2008) image over a stratified area in Sao Paulo
State (a) and respective image segmentation (b) with similarity of 4 and minimal area of 25
m² (corresponding to 4 pixels of ALOS/PRISM fusion through HIS method). In this area, trees
outside the forest (TOF) are located near water bodies; in land division and forest tree that
remnant in deforestation processes that leads to pastures and cattle areas. As shown in Figure
1b, the segmentation process, with the parameters described above, permitted a good
distinction between TOF and other land use and land cover (LULC) areas. The main factor for
separation is the difference between reflectance values in infrared region of electromagnetic
spectrum (EEM) of pasture and trees and key elements of image interpretation.
Figure 1 – (a) ALOS -2B3R4G image; (b) ALOS segmentation image (black lines) with minimal
area of 100 m² and similarity of 20.
Figure 2 shows the IKONOS (07/19/2008) true color image (1B2G3R) over a sample area in
Bahia State (northeast of Brazil) in (a) and respective image segmentation (b) with similarity
of 40 and minimal area of 5 m² (corresponding to 5 pixels). In this area, TOF are distributed in
superficial drainage channels and in pasture sites. Figure 2b shows that using only visible
region of EEM as a parameter estimator in segmentation method and small area detection,
several fragments caused by the variation of colors and shadows in the image appear, making
the visualization of TOFs in image not so clearly.
P á g i n a | 37
Latin American Remote Sensing Week 2010
Figure 2 – (a) IKONOS true color image; (b) IKONOS segmentation image (black lines) with
minimal area of 5 m² and similarity of 40.
Figure 3 shows the true color QUICKBIRD (04/09/2008) image (1B2G3R) in Bahia State
(northeast of Brazil) and image segmentation in (a) and (b), respectively. In this image, a
similarity of 40 and minimal area of 2.1 m² were applied to delimit TOF areas. As in IKONOS
image, the use of visible region of EEM to extract statistical parameters to delimited areas in
image led to the emergence of fragments in areas of strong variations between neighboring
regions. Otherwise, TOF area is also mapped, including trees shadows. The same results are
obtained in SPOT 5 (06/08/2006) true color segmentation with similarity of 40 and minimal
area of 5 m² as shown in Figure 4. Both areas correspond to pasture areas with cattle located
in the northeast of Brazil.
Figure 3 – (a) QUICKBIRD true color image; (b) QUICKBIRD segmentation image (black lines)
with minimal area of 2.1m² and similarity of 40.
Figure 4 – (a) SPOT true color image; (b) SPOT segmentation image (black lines) with minimal
area of 5 m² and similarity of 40.
The results of TOF delimitation using ALOS/AVNIR 2.5 m segmented image are shown in
Figure 5. In sample area located in Sao Paulo State, approximately 5% of area was classified as
TOF, mainly placed in superficial drainage channels according to Brazil environmental laws.
P á g i n a | 38
Latin American Remote Sensing Week 2010
To verify the assessment of TOF extraction, 60 random points are distributed in study area
(30 in TOF classification and 30 in unclassified areas), afterward, samples where compared
with image interpretation of high resolution images, obtaining a Kappa value of 97%.
However, some mistakes in TOF classification include areas of marsh and dense shrubs/tall
grass (less than 2%). Results indicate that ALOS/AVNIR presented a high capacity to detect
TOF, but, this study should be conducted in other biomes in order to verify the methodology
accuracy. Rawat et al. (2004) identified and classified TOF’s in India using IRS satellite data
with 5.8m. The TOF’s were classified based on its spatial distribution as (blocks, linear and
scattered) with a Kappa 0.92.
Figure 5 – (a) ALOS -2B3R4G image; (b) ALOS segmentation image (black lines) with TOF
classification.
4- CONCLUSIONS
Preliminary results present a positive perspective regarding the ability to identify the TOF’s
by visual interpretation as well by automatized techniques. On the next phase, it will be tested
principal components analysis, linear mixture model in order to improve the results and make
it faster and precise extraction of such data. Shadow effect is main obstacle to tackle in order
to achieve better results. As presumed, the better spatial resolution the major accuracy is
expected. In fact depending on the spatial resolution the TOF’s are mapped in form of groups,
not as an individual tree. What is called a mapped dossal, however through visual
interpretation the trees may be delineated by hand providing accurate results. The cost
effectiveness is always one of the main variables taken by policy makers, then, each situation a
particular decision. Simple, standardized, solutions, valid even for large-area coverage, are
hard to imagine. The composition, structure and distribution of trees outside forests are by
their very nature both variable and diverse. The methodology using orbital image processing
and GIS can be effectively employed using high-resolution satellite imageries to identify and
map the TOF resources. It is expected to provide better estimates of TOF resources than the
one generated through field survey only.
P á g i n a | 39
Latin American Remote Sensing Week 2010
5- BIBLIOGRAPHY
FAO- Food and Agriculture Organization – UN- United Nation –
http://www.fao.org/docrep/004/y1997e/y1997e09.htm
Kumar, O. Valuation and Evaluation of Trees Outside Forest (TOF) in India. Forest Survey of
India – Kaulagarh Road, PO: IPE Dehradun, India – 2006.
Rawat, J.K., Dasgupta, S.S., Kumar, R.S. Assessment of Tree Outside Forest based on Remote
Sensing Satellite Data. Forest Survey of India, Kaulagarh Road, PO. IPE, Dehradun – 248-195.
2004.
Rodríguez, J. Estado del Ambiente y los Recursos Naturales en Centroamérica 1998. Comisión
Centroamericana de Ambiente y Desarrollo. San José, Costa Rica.1998.
P á g i n a | 40
Latin American Remote Sensing Week 2010
Especialistas en Hidráulica
cervando@tlaloc.imta.mx
Especialista en Hidráulica
jbrena@tlaloc.imta.mx
Palabras clave: Modelación prospectiva del paisaje, uso del suelo y vegetación, análisis
espacio-temporal, escenario tendencial.
Abstract: This paper attempted to obtain a prospective landscape based upon space-
time modeling series of land use and vegetation. A set of algorithms, such as, Markov
chains, cellular automata, multi-criteria and multi-objective techniques were applied. A
small watershed located South of Mexico was used to exemplify the proposed
methodology (258206 ha). The procedure used land use maps developed from satellite
images taken during 1986 and 2006, these images served as hard data to predict future
land covers. For the multi-criteria assessment, three variables were considered: annual
P á g i n a | 41
Latin American Remote Sensing Week 2010
rainfall, soil fertility and ground slope. A weighting factor was allocated to each
variable by using an Analytical Hierarchy Process (AHP). The AHP methodology
involved the collaboration of a group of experts to come up with an appropriate set of
weighting factors. The final result is a soil suitability map for the year 2036. This map
will be helpful to decision makers to improve the natural resources management
throughout the watershed.
1. INTRODUCCIÓN
En esencia las cadenas de Markov lo que hacen es examinar el estado más recientes de
un proceso, considerando que los procesos anteriores son conocidos, entonces la
probabilidad de los estados futuros dependen únicamente del estado actual y no de los
anteriores. Que en nuestro caso podemos decir que el resultado prospectivo 2036 (T3)
depende de la comparación de los usos de suelos de t1 y t2 a partir de la generación de
una matriz de probabilidad de transición.
Para que la matriz sea aplicada es necesario llevar a cabo una evaluación multicriterio,
esta se aplica para obtener los mapas de aptitud del suelo que expresan los diferentes
niveles de uso potencial que pueden tener cada una de las clases de uso del suelo y
vegetación. El multicriterio también es la herramienta para obtener la aptitud de una
clase considerando la suma de más de un atributo para obtenerlo. Los factores de
ponderación o peso que se calculan para los atributos se obtuvieron utilizando el
Método de Análisis Jerárquico AHP creado por Thomas Saaty, es una herramienta que
P á g i n a | 42
Latin American Remote Sensing Week 2010
El algoritmo multiobjetivo define el cambio o destino de los pixeles que pasaran a ser
parte de otra clase de acuerdo con su probabilidad de cambio o de permanecer en la
misma si no hay condiciones para un cambio. Otra herramienta utilizada para
materializar el escenario futuro, son las operaciones lógicas entre imágenes, sin
embargo el peso sustancial en esta etapa corresponde a la discusión que se da entre
expertos para definir los criterios que se aplican para obtener los mapas de aptitud del
suelo para la recuperación de la vegetación, sin duda que esta etapa es la mas
enriquecedora del proceso porque abre la puerta para la discusión de ideas y
argumentos para lograr el consenso en los parámetros finales que se aplicarán los
mapas de aptitud. Podemos decir que esta etapa busca el consenso de ideas para
lograr o llegar a una meta.
3. METODOLOGÍA
Una vez definidas las superficies para el escenario futuro, la metodología requiere de
elementos para identificar las zonas dónde se llevarán a cabo los cambios, es decir, definir
las mejores áreas para aplicar los cambios.
Para cada una de las 20 clases se requiere de un mapa de aptitud del suelo, el objetivo es
evaluar la probabilidad que un área mínima exista en una clase determinada con base en
las características de esa área. La mínima área que se evalúa es un pixel. En estos mapas a
cada píxel se le asigna un valor de aptitud de 0 a 10, generalmente se utilizan valores de 0
a 1, pero se prefiere utilizar el intervalo de 0 a 10 para facilitar el cálculo y trabajar solo
con números enteros.
Un criterio aplicado para definir los mapas de aptitud de las clases es identificar la
tendencia de crecimiento o decrecimiento de la superficie de 1986 a 2006. Las clases que
tuvieron decrecimiento son las siguientes: Bosque templado, Selva mediana y bosque
mesófilo, Manglar Selva baja caducifolia, Popal y tular, Sabana, Vegetación de dunas
costeras, Vegetación riparia y Pastizal natural. En estos casos se considera que el área de
la vegetación original, la de 1986, es el área o mapa potencial porque la tendencia es
recuperar lo que se ha perdido o perturbado, a estas áreas se les asigna un valor de
aptitud de 10 y cero a las otras clases. Lo que se busca es revertir el escenario tendencial,
es decir, que no sigan disminuyendo las clases, lo que se plantea es recuperarlas.
P á g i n a | 44
Latin American Remote Sensing Week 2010
Ac = ∑PC j j (1)
J=1
Como ejemplo se muestra el cálculo del mapa de aptitud para la Agricultura de temporal.
De entrada se clasificó el mapa de pendientes en cuatro categorías a cada una se asignó un
valor de aptitud, los terrenos con pendientes menores de 8 grados se les asignó un valor
de 10 porque son los mejores para esta actividad. La precipitación media anual se dividió
en tres rangos y se asignó el valor de 10 al área donde el promedio es igual o mayor
1500mm. El mapa de fertilidad del suelo se clasifico en siete valores de aptitud porque no
alcanza la escala de diez, al suelo Luvisol ródico se le asignó el valor de 7 por ser el mejor
para la agricultura de temporal. En la Figura 2 se muestra la clasificación para cada uno de
los casos. Para establecer el peso ponderado de cada uno de los mapas se utilizó la técnica
de Saaty, denominada Proceso Jerárquico Analítico (AHP, por sus siglas en inglés). El AHP
“se trata de desmenuzar un problema y luego unir todas las soluciones de los
subproblemas en una conclusión. El AHP trata directamente con pares ordenados de
prioridades de importancia, preferencia o probabilidad de pares de elementos en función
de un atributo o criterio común representado en la jerarquía de decisiones”. El AHP
dispone de una escala (de 1 a 9) creada por Saaty que pondera los juicios emitidos por un
grupo de expertos que toma las decisiones valorativas en el proceso.
Figura 2. Aptitud de la Fertilidad del suelo, Figura 3. Pesos ponderados para la Agricultura
Precipitación y Rango de pendiente de temporal
P á g i n a | 45
Latin American Remote Sensing Week 2010
Steven Marquez (1), Hermogenes Suarez (2), Norah Gallucci (2), Hugo Rivas (2).
Universidad del Zulia, Escuela de Ingeniería Geodesica (1).
PDVSA, Departamento de Geofísica y Geodesia (2).
stevmarq@gmail.com, suarezhh@pdvsa.com, galluccin@pdvsa.com.
El objetivo del trabajo fue determinar la calidad de los diversos MDE Globales disponibles.
SRTM3, ACE2 y ASTER GDEM fueron evaluados. Enmarcando el estudio a las principales
regiones petroleras de Venezuela. El trabajo presenta una propuesta metodologica para
analizar y determinar la calidad altimetrica de los MDE, basandose en un estadístico de los
residuales altimetricos calculados a partir de los diferentes MDE (ASTER-GDEM, ACE2,
SRTM3) y diversas fuentes de datos altimetricos de referencia: Estaciones de redes
geodesicas, pozos petroleros medidos con GNSS y levantamientos sísmicos. Para el
tratamiento y analisis de los datos se aplicaron herramientas de GIS bajo OpenSource.
SRTM3 fue el MDE Global mas apropiado para las diversas aplicaciones de PDVSA en
Venezuela. El modelo SRTM3 reporto residuales altimetricos con un RMS de ±4,3m. El ACE2
arrojo residuales con un RMS de ±5,5m. El modelo ASTER GDEM reporto residuales
altimetricos con un RMS de ±9,8m.
Abstract: PDVSA in their studies and projects developed at the national and internationally
diverse uses of geospatial information, mainly based on satellite images of high spatial
resolution, which should be corrected, or orthorectified by Digital Designs Elevation (DEM), in
order to remove the distortions caused by topographic relief. Additionally, the MDE are
applied to several research and interpretations about surface geology, including applications
in oil industry.
The main objectives are to determine the quality and accuracy of several Global DEM available.
SRTM3, ACE2 and ASTER GDEM model were evaluated. Framing the study in main oil regions
in Venezuela. The paper presents a methodological proposal for analyze and determine the
DEM altimetry quality based In comparison an altimetric residuals calculated from several
altimetry geodetic data: geodetic network stations, GNSS measured oil wells and seismic
surveys. For processing and analysis of data were used GIS tools developed OpenSource.
SRTM3 was the most appropriate DEM for main oil regions in Venezuela. SRTM3 model
P á g i n a | 46
Latin American Remote Sensing Week 2010
reported altimetric residuals with ± 4.3 m RMS. ACE2 model reported altimetric residuals
with ± 5.5 m RMS. ASTER GDEM model reported altimetric residuals with ± 9.8 m RMS.
OBJETIVO
Evaluar la calidad altimétrica de los MDE: ASTER-GDEM, ACE2 y del SRTM3, mediante la
aplicación de una metodología basada en el análisis estadístico de residuales altimétricos
calculados a partir de diferentes fuentes de datos de referencia (redes geodésicas, pozos
petroleros, sísmica).
METODOLOGÍA
La metodología empleada para la evaluación de los MDT a través de datos altimétricos fue la
siguiente (Suárez H. et al., 2009):
DESCRIPCION DE DATOS
Para el estudio se contó con el modelo ASTER GDEM el cual fue generado a partir de datos
estereoscópicos, desde la época de 1999 y publicado en el 2009) (ERSDAC, 2009). El modelo
ACE2 resultante de la fusión del modelo SRTM3 y datos altimétricos provenientes de los
altímetros satelitales Jason, ERS-1, ERS-2, y Envisat (EAPRS, 2008). El SRTM3 (3” ≈ 90m) el
cual fue obtenido mediante la técnica InSAR (JPL, 2009). Se seleccionaron tres (3) zonas de
estudio donde se encuentran las principales áreas operacionales de PDVSA. La primera de
ellas es la zona nor-oriental de Venezuela correspondiente a la División Oriente (-65º a -63º λ
y 10º a 8,5º φ, WGS84). La segunda zona de estudio correspondiente a la División Centro Sur
(-70,5º a -69,5º λ y 9º a 8º φ, WGS84). La tercera zona fue la Costa Oriental del Lago de
Maracaibo (-71,5º a -70,5º λ y 11º a 10º φ, WGS84) (figura 1).
P á g i n a | 47
Latin American Remote Sensing Week 2010
b c
Figura 2: “DTM y datos altimétricos de referencia en la División Oriente. 2a: Datos sísmicos.
2b: Pozos petroleros. 2c: Redes Geodésicas.”
c
P á g i n a | 48 b
Latin American Remote Sensing Week 2010
Figura 3: “DTM y datos de referencia en la División Centro Sur. 3a: Datos sísmicos. 3b: Pozos
petroleros. 3c: Redes Geodésicas.”
b c
Figura 4: “DTM y datos de referencia en la División Occidente. 4a: Datos sísmicos. 4b: Pozos
petroleros. 4c: Redes Geodésicas.”
A continuación se presenta una tabla descriptiva de los datos utilizados en cada zona de
estudio en función de la cantidad de estaciones, calidad, espaciamiento promedio entre
estaciones y su densidad por área. En las tres zonas se pudo observar que los datos sísmicos
poseen la mayor densidad y con un espaciamiento promedio menor que el resto de los datos,
seguido por los pozos petroleros y las redes geodésicas.
P á g i n a | 49
Latin American Remote Sensing Week 2010
CÁLCULOS Y RESULTADOS
Se calcularon residuales altimétricos para cada MDT (ASTER GDEM, SRTM3 y ACE2) a partir
de los diferentes sets de datos geodésicos. La ecuación utilizada para el cálculo de los
residuales altimétricos se muestra a continuación:
En las 3 zonas se observó que los menores residuales altimétricos (redes geodésicos)
fueron reportados por el SRTM3 y eso se refleja en los diferentes indicadores estadísticos.
Los máximos residuales altimétricos (redes geodésicos) se detectaron en la zona División
Oriente, donde los RMS fueron ±10,7m, ±4,2m y ±4,6m para el ASTER-GDEM, SRTM3 y el
ACE2 respectivamente.
Para el 95% de los datos analizados se observó que los residuales se ubican en un rango
de 20m para el ASTER-GDEM y un rango de 7m para el SRTM3 y el ACE2.
En esta prueba se observó que el SRTM es el DTM de mejor calidad.
Resultados: Pozos petroleros
P á g i n a | 50
Latin American Remote Sensing Week 2010
Resultados generales:
En la tabla 5, se pueden observar las estadísticas generadas de las diferentes pruebas, los
cuales nos permiten comparar la calidad relativa que existen en los modelos, en las tres zonas
de evaluación. En esta tabla se muestra la media (X), desviación estándar (σ), RMS y el Ratio
calculado a partir del cociente de los RMS reportados para cada set de datos (2).
Ratio= RMSGDEM / RMSSRTM
(2)
P á g i n a | 51
Latin American Remote Sensing Week 2010
Tabla 5: “Resumen estadístico de residuales altimétricos MDT en las tres zonas de interés”
La comparación entre RMS de cada modelo con respecto al GDEM (Ratio) evaluado para cada
set de datos, muestra una tendencia de 2:1 a favor del SRTM3 y ACE2.
CONCLUSIONES Y RECOMENDACIONES
BIBLIOGRAFIA
EAPRS, Earth and Planetary Remote Sensing Laboratory (2008): “ACE2”. Web site:
http://tethys.eaprs.cse.dmu.ac.uk/ACE2/shared/overview
ERSDAC, Earth Remote Sensing Data Analysis Center (2009): “ASTER GDEM”. Web site:
http://www.ersdac.or.jp/GDEM/E/index.html
Suárez H., Méndez H., Villalobos M., Molero Y., Rivas H., Gallucci N. (2009): “DTMSis30, un DEM
mejorado a partir del SRTM3 y de data altimétrica de Proyectos Sísmicos 3D en el Distrito Norte
de PDVSA”. PDVSA, Puerto la Cruz, Venezuela.
P á g i n a | 52
Latin American Remote Sensing Week 2010
El uso del escáner láser como técnica del sensoriamento remoto es reciente, especialmente en
los estudios de afloramientos rocosos. Este dispositivo permite medir la posición en el espacio
y el color natural con una alta frecuencia de muestreo, detallando en tres dimensiones objetos,
superficies y estructuras. Se trata de una tecnología prometedora en términos de
productividad y eficiencia en varios ámbitos de aplicación. En el estudio de afloramientos,
dependiendo de la disposición y las característica de la roca, aparecen fallas en la imagen, es
decir algunas parte del afloramiento quedan sin información. En la búsqueda de la solución de
este problema se utilizarán las Redes Neuronales Artificiales. Se usará el algoritmo
backpropagation o una de sus variaciones y se probarán varias topologías, también varios
conjuntos de entrenamiento y de prueba como un intento de definir la red neuronal apropiada
para generalizar el problema en cuestión. El estudio de caso es la cuenca Camaquã situada en
el sur de Brasil en la región central del Escudo Sul-rio-grandense, que se encuentra ubicada a
250 km de Porto Alegre por la vía BR-290, en el afloramiento conocido como Pedra Pintada
del municipio de Caçapava do Sul – Rio Grande do Sul. Fue realizada la georreferenciación de
las imágenes utilizando una red geodésica con GPS modelo Leica 900, de doble frecuencia
(L1/L2), la cual fue ajustada con los puntos SMAR y POAL de la Red de Monitoreo Continuo
del IBGE (Instituto Brasileiro de Geografia e Estatística). El equipo usado fue el Laser Scanner
ILRIS - 3D (Intelligent Laser Ranging & Imaging System) de OPTECH Incorporated para el
procesamiento de las imágenes, el cual tiene capacidad de recoger hasta 2000 puntos por
segundo a una distancia máxima de 1500 m del albo. Los datos fueron procesados por el
programa Polyworks. Ambos, el equipo y el programa, pertenecen al Programa de Posgrado
en Geología de la Universidad Vale do Rio dos Sinos - UNISINOS. Las Redes Neuronales
Artificiales - RNAs han demostrado su uso en diversas áreas del conocimiento y se cree en el
éxito de su aplicación en esta investigación debido a su excelente capacidad de generalización,
clasificación, interpolación y extrapolación
P á g i n a | 53
Latin American Remote Sensing Week 2010
1-Introducción
Las rocas, laderas, bloqueos de carreteras o islotes, son afloramientos de rocas, con un 3% de
la superficie de los continentes. Las rocas minerales representan de gran importancia para la
humanidad y son utilizados de muchas maneras diferentes.
Afloramiento es la aparición de roca en la Tierra después de haber sido arrastrado todas las
materias objeto de facilitar el estudio y la cartografía geológica. Este arrastre de material
exponiendo las rocas puede ocurrir por procesos naturales como la erosión causada por los
ríos, las lluvias, los glaciares o corrimientos de tierras o por procesos artificiales, tales como
apertura de caminos y túneles.
Cada tipo de roca tiene un patrón de su topografía propia, la vegetación y el drenaje del
suelo. Esta pauta se puede diversificar sustancialmente con pequeños cambios en la
composición mineralógica de la roca, pero por otra parte, las rocas muy diferentes pueden
tener patrones similares.
El uso del escáner láser para realizar estudios sobre afloramientos de roca es relativamente
nuevo. Este dispositivo permite medir la posición en el espacio y el color natural de muchos
puntos por segundo, detallando los objetos en tres dimensiones, superficies y estructuras.
Cuenca Camaquã (Guaritas), tenemos el afloramiento conocido como Pedra Pintada que se
encuentra en la ciudad de Caçapava do Sul - RS, a unos 70 km del centro de la ciudad.
P á g i n a | 54
Latin American Remote Sensing Week 2010
La cuenca del Camaquã está sentado en las tierras de ígneas y metamórficas Escudo Sul-Río-
Grande, que incluye el cinturón de granito gnéisicas Dom Feliciano, Cinturón Tijucas, el
Cinturón de Vila Nova, rocas graníticas y Paleoproterozoica gnéisicas, incluyendo de Santa
María Chico Granulita Complejo rocas y facies de anfibolita gnéisicas granito. Por tanto NE y
SW Camaquã cuenca está cubierta por rocas sedimentarias del Pérmico al Cuenca de Paraná
(Teixeira et al, 2004).
Según Menegat y Fernandes (2003) la parte superior del Rift Guaritas incluye un paquete de
capas rojas con alrededor de 1,4 km de espesor, se originó durante un régimen extensional
que sucedieron a los eventos tectónicos y el cinturón colisional transcurrente de esta
cuenca ruptura fue llenado por sistemas eólicos y aluviales.
El relleno de la cuenca era controlado por fallas en la extensión NE. Estos fracasos, en gran
parte como resultado de la reactivación de las zonas de cizalla transcurrente Neoproterozoico
en el NE en condiciones frágiles, fueron reactivados para 470 Ma la edad de la secuencia
basal.
Guaritas cuenca contiene dos sistemas deposicionales, limitada por los desacuerdos entre sí
erosiva (PAIM et al, 2000). Las condiciones áridas registrado por los depósitos de esta cuenca
debe haber sido causado por su ubicación interfiere con la precipitación paleogeográficas
peculiar de agua del océano ciclo hidrológico-continente en lugar de la presencia de un centro
de alta presión estática.Posibles barreras naturales en el borde occidental de Gondwana son
las montañas de orogenia Cuyana y el arco magmático del Cinturón Dom Feliciano Occidental
impidió la penetración de masas de aire húmedo proveniente de paleoceano (Menegat
FERNANDES, 2003).
3. Marco Teórico
Las RNA son una técnica que no es nueva, en este estudio se ha iniciado en los años 40,
pasó por cierta negligencia en los años 70 y resurgió en los últimos años 80 como una
alternativa posible a la informática tradicional, (Haykin, 1999).
P á g i n a | 55
Latin American Remote Sensing Week 2010
Las redes neuronales en combinación con los sensores se han utilizado con éxito en los
estudios del suelo y de las asignaciones de YANG et all. (1998) y las copas para el perfil
y los estudios geoquímicos (Costa e Silva et all (2001). LAMMOGLIA y todo lo demás.
(2007)).
Brondino (1999) estudiaron el uso de redes neuronales para determinar el valor de las
parcelas de bienes raíces de la tierra. En 2004, TIMOSZCZUK realizó su tesis doctoral y
se analizaron mediante redes neuronales de impulsos automáticos para el
conocimiento del hablante (información temporal de la señal de voz) y obtuvo buenos
resultados.
Las RNA han visto su uso en esta investigación debido a su excelente capacidad para la
generalización, la clasificación, interpolación y extrapolación, tolerancia de errores y el
P á g i n a | 56
Latin American Remote Sensing Week 2010
Esto, que es el mismo utilizado por el equipo que se utiliza en este proyecto de
tesis (ILRIS Optech 3D) que tiene las siguientes especificaciones: Optech 3D) que
tiene las siguientes especificaciones: Precisión Alcance: hasta 7 mm;
Precisión Posición: hasta 8 mm; Alcance: 3m a más de 1 km; Campo de visión: 40°
x 40°; Frecuencia: 2 kHz; longitud de onda de láser (λ): 1500 nm; Clase de láser:
Clase 1 (inofensiva para los ojos); 6,6 megapíxeles (sensor CMOS);Se adjunta la
cámara digital. Peso del escáner: 13 kg; Duración de la batería - 3 horas.
Funcionamiento del sistema del equipo tiene una buena interfaz, donde puedes
tomar el control de la configuración a través de un Pocket PC o incluso un
cuaderno.No requiere un cable de conexión, donde se puede hacer de forma
inalámbrica a través de móvil e incluso tiendas de todos los datos adquiridos en la
memoria utilizando Pen Drive por tener una Controladores de USB (Universal
Serial Bus).
El Laser Scanner System es una tecnología que permite obtener datos de tres
dimensiones (X, Y y Z) de la nube de puntos, que puede ser georeferenciada con el
uso del GNSS (Global Navigation Satellite System), y agrega información de color y
textura como una función de una cámara conectada al equipo. Debido a que se ha
utilizado en varios ámbitos, sobre todo en la ingeniería, la cartografía y la
arquitectura.
Según Rocha (2002) y BARCHIK, y todo lo demás. (2007) El sistema láser es una
gran herramienta para el desarrollo de muchas aplicaciones en los procesos
mineros. Este sistema presenta claras ventajas sobre otros sistemas tradicionales,
con respecto a la aplicación, rendimiento, velocidad, precisión y seguridad, ya que
permite capturar información topográfica en lugares inaccesibles, inestable y de
riesgo en caso de elevada densidad de puntos fue adquirida y procesada en un
corto período de tiempo, con lo que por lo tanto capaz de considerar la situación
actual y futuro de la zona. Lo que también se puede observar en el estudio
realizado por Pérez (2008).
P á g i n a | 57
Latin American Remote Sensing Week 2010
4. Material y Métodos
Para alcanzar las metas y responder a las preguntas en el proyecto de tesis tendrá los
siguientes equipos y programas disponibles en el Laboratorio de Percepción Remota y
Cartografía Digital (LASERCA), Universidad de Vale do Rio dos Sinos (UNISINOS): láser
escáner 3D terrestres Optech - ILRIS 3D; Par de frecuencias del GPS dual, RTK (cinemática
en tiempo real) Leica, 900CS modelo; Procesamiento de los puntos adquiridos con el
programa de escáner láser 3D terrestre (Poliworks, de nube de puntos y analizador).
Programa de tratamiento de datos GPS e Programa de Matlab.
Con este trabajo queremos debatir y buscar respuestas a las cuestiones mencionadas. Los
datos se están organizando para comenzar las pruebas. É un gran desafio.
En busca de solución a estos fallos ("agujero negro") se pondrá a prueba utilizando Redes
Neuronales Artificiales. En la figura 2 se puede ver el afloramiento que fue fotografiada.
El uso de estas nuevas tecnologías integradas con otros datos y técnica de la visualización en
3D permite a los geólogos, la interpretación más precisa de los distintos sistemas de depósito,
desde la construcción de modelos de depósitos análogos a los depósitos de aceite.
Las Redes Neuronales Artificiales - RNAs han demostrado su uso en diversas áreas del
conocimiento y se cree en el éxito de su aplicación en esta investigación debido a su excelente
capacidad de generalización, clasificación, interpolación y extrapolación.
REFERENCIAS CONSULTADAS
P á g i n a | 59
Latin American Remote Sensing Week 2010
ARCHIK, E. ; MOSER, I.; SANTOS, D. S. dos. & MARTINS, B. D. Aplicação do Scanner Terrestre
ILRIS – 3D no Ramo da Mineração. IN: Anais XIII Simpósio Brasileiro de Sensoriamento
Remoto, Florianópolis, Brasil, 21-26 abril 2007, INPE, p. 3631-3635. 2007
CHAGAS, C.da Silva. Mapeamento digital de Solo por correção ambiental e Redes Neurasi
em uma Bacia Hidrográfica no domínio de Mar dos Morros. Viçosa, UFV. Tese de
doutorado.223p. 2006.
COSTA e SILVA, M. B.; SANTOS, R. V. Dos.; MARTINS, J.L.; FONTOURA, S.A.B. da. Predição de
propriedades de perfis com Redes Neurais: uma aplicação no campo Escola de
Namorado, Bacia de Campos, Brasil. In: 1º Congresso Brasileiro de P&D em Petróleo e Gaz.
UFRN – SBQ Regional RN. 2001.
HAYKIN, S.. Neural networks – A Comprehensive Foundation, second edition. Prentice Hall.
New Jersey. 1999.
HRUSCHKA, E. R. & EBECKEN, N. F.F. Extração de Regras de Redes Neurais por meio do
Algoritmo RX Modificado: Um Exemplo de Aplicação em Modelagem de Dados
Meteorológicos. In: IV Congresso Brasileiro de Redes Neurais. P.047-051. Jul.20-22. ITA. São
José dos Campos – SP. 1999.
LUK, K. C.; BALL, J. E.;SHARMA, A. A Study of Optimal Model Lag and Spatial Inputs to
Artificial Neural Network for Rainfall Forecasting, Journal of Hydrology (227) 56-65. 2000.
MEDEIROS, M. A. M.; FAVERA, J. C. D.; REIS, M. A. F. dos, VIEIRA e SILVA, T. L.; OLIVEIRA, E. B.;
BIASSUSI, R. A. & SILVEIRA, R. G. da O Laser Scanner e a Paleontologia em 3D. Anuário do
P á g i n a | 60
Latin American Remote Sensing Week 2010
NOBREGA, Jorge Emanuel dos Santos, SAWAKUCHI, André Oliveira e ALMEIDA, Renato Paes
de. Minerais pesados das porções média e superior do Grupo Guaritas (Eocambriano,
RS): considerações sobre a proveniência sedimentar. Rev. bras. geociênc., set. 2008, vol.38,
no.3, p.554-565. ISSN 0375-7536. 2008.
PAIM, P.S.G., CHEMALE, Jr F., LOPES, R.C. A Bacia do Camaquã. In: M. Holz & L.F. De Ros (eds)
Geologia do Rio Grande do Sul. CIGO/UFRGS, Porto Alegre, p. 231-274. 2000.
STEPHEN, T. W. Neural Network and Fuzzy Logic in C/C++. Wiley, USA. 1992.
P á g i n a | 61
Latin American Remote Sensing Week 2010
1. Introducción
El tiempo y los recursos que se invierten para evaluar el potencial de aprovechamiento de los
recursos forestal de cierta región, llega a ser considerable bajo las técnicas tradicionales de
muestreo. Por lo que deben buscarse alternativas que, en lo posible, sean más económicas,
pero sin menoscabo de la precisión (Flores et al., 2007). De esta forma se busca que, por lo
menos, se garantice la misma precisión que se obtiene por métodos tradicionales. En México,
la autorización de un aprovechamiento forestal está supeditada a que se cuente con
información recabada del campo, a través de un inventario forestal. El cual consiste, por lo
general, en un aserie de sitios de muestreo distribuidos sistemáticamente. Con una intensidad
P á g i n a | 62
Latin American Remote Sensing Week 2010
de muestreo que va de 1.5. a 3.0% (UCODEFO No. 2, 1997). Con base a la información
obtenida se calcula el ritmo de crecimiento del arbolado, ya sea definiendo el incremento
corriente anula, o el incremento medio anual. Posteriormente se calcula la posibilidad, que se
refiere al volumen aprovechable por hectárea, por año. Las cifras obtenidas en diferentes
sitios de muestreo, que se ubican dentro de un mismo rodal, se promedian. Dicho promedio se
usa para estimar variables dasométricas, como el potencial de madera que se puede extraer
por rodal de acuerdo a sus dimensiones. Es decir la posibilidad promedio se multiplica por e
número de hectáreas, resultando el volumen (m3/ha/año) que corresponde a dicho rodal. La
metodología anterior denota que poco se usan las estrategias y herramientas que ofrece la
geomática, principalmente en el sentido de apoyar la interpolación de la información
generada, para hacer estimaciones en aquellos lugares donde no se muestreo (Ek, 2004). De
esta forma, en lugar de estar trabajando con medias dentro de los rodales, se pueden hacer
estimaciones considerando las variaciones espaciales de la información de interés
(posibilidad, densidad, existencias, etc.). Esto ayudaría considerablemente al aumento de la
precisión en las estimaciones, y de esta forma se tendería a evitar tanto la sobreestimación,
como la subestimación.
Todas las implicaciones anteriormente señaladas se tienen que repetir en cada región de
bosque se va a ser aprovechada. Sin embargo, en ocasiones se tiene el caso de tener dos, o
más, regiones aledañas, las cuales tienen condiciones de bosque muy similares. Más aun,
alguna de estas regiones puede contar con un programa de manejo forestal, con lo que se
tendría información de sus existencias (m3/ha), posibilidad (m3/ha/año), etc. Dicha
información, podría usarse para el apoyo en la generación de los planes de manejo de las
regiones aledañas. Lo cual sería posible a través de un proceso de extrapolación. Sin embargo,
no se han desarrollado estrategias claras para llevar a cabo dicho proceso, principalmente por
las dificultades que implica el manejo de grandes cantidades de información georeferenciada
(Franco et al., 2001). No obstante, gracias la desarrollo de herramientas de la geomática, como
son los sensores remotos y la geoestadística, y que actualmente se cuenta con poderosos
equipos de cómputo, es posible probar diferentes alternativas de extrapolación (Flores y Omi,
2003). Para ejemplificar lo anterior se presenta un trabajo realizado en bosques templados
del estado de Chihuahua, México, en el que se extrapolo información sobre densidad,
posibilidad de aprovechamiento (m3/ha/año) y diámetro de copa. Se probaron varios
modelos basados en la tecnología de árboles de regresión, los cuales fueron definidos con base
a una serie de variables auxiliares.
2. Regiones vecinas
El área de estudio se ubica al oeste del estado de Chihuahua (Figura 1), el cual a su vez se localiza
en el límite norte de México. En general la vegetación arbórea que crece en estas regiones,
propias de las zonas montañosas del país, está por coníferas y latifoliadas (Allen, 1996). Las cuales
forman masas arboladas que por lo común poseen poca variación de especies; en Chihuahua.
Estas cubren las partes altas de las montañas, sobre todo de la Sierra Madre Occidental
P á g i n a | 63
Latin American Remote Sensing Week 2010
(Rzedowski, 1981). Respecto a su composición, existen variantes que van desde los bosques puros
de pino, de encino, bosques mixtos donde éstos se mezclan, además del bosque bajo-abierto, y
bosque de táscate; en conjunto, cubren 29% de la superficie estatal.
- Bosque de Pino
- Bosque de Pino-Encino
- Bosque de Encino-Pino
- Bosque de Encino
- Bosque Bajo-Abierto
Figura 1. Ubicación aproximada de dos regiones (circulo punteado), dentro del estado
de Chihuahua (México) con condiciones forestales similares.
3. MODELAJE ESPACIAL
Primeramente se trabajo con la región base (R. Chocachi), con la cual se desarrollaron una
serie de modelos que definen la distribución espacial de las variables de interés (densidad,
posibilidad de aprovechamiento [m3/ha/año] y diámetro de copa). Estos mismos modelos
fueron usados para generar las estimaciones de la región en extrapolación (R. Nopalera).
P á g i n a | 64
Latin American Remote Sensing Week 2010
Uno de los principales propósitos de este trabajo fue el de definir mapas de distribución de
cuatro variables dasométricas. Para esto existen varias técnicas, en este caso se uso la
denominada “árboles de regresión”, con base a la cual se asocia una serie de variables
conocidas con la variable de interés. En el primer caso se requiere que se tengan mapas de
cada una de las variables conocidas (independientes), como lo serian fotos aéreas (Figura 3),
imágenes de satélite, modelo de elevación digital, etc. Con base a las cuales se genera el mapa
de la variable de interés (dependiente). En total se analizaron 10 variables para cada una de
las variables dasométricas de Chocachi (Tabla 1).
P á g i n a | 65
Latin American Remote Sensing Week 2010
Para definir cuál de las variables auxiliares se asociaba espacialmente mejor con las variables
dasométricas, se generaron una serie de modelos, a través de la técnica de arboles de
regresión. A continuación se presentan un ejemplo (existencias [m3/ha]) de los estadísticos
resultantes de los árboles de regresión que se definieron. En estos se presenta el número de
variables que se usaron en la construcción de los árboles de regresión, así como el número de
nodos terminales resultantes en cada caso. Otro estadístico que se presenta es la desviación
media de los residuales (errores de estimación), con base a la cual se seleccionaron los
árboles de regresión más adecuados.
Figura 3. Orthofoto que se uso como una de las variables auxiliares para la generación
de arboles de regresión en la Región Chocachi,
VARIABLE DESCRIPCION
P á g i n a | 66
Latin American Remote Sensing Week 2010
Tabla 1. Variables con distribución espacial analizadas para estimar las variables
dasométricas de Región Chocachi.
El paso final en esta fase es la obtención de la representación grafica de los resultados de los
modelos de regresión (árboles). Esto se hace a través de un sistema de información
geográfica, a través del cual se puede manipular la información resultante. Es decir clasificar u
ordenar los resultados, de tal forma que el producto grafico obtenido sea de fácil
interpretación por el lector. Con esto se generan los mapas temáticos correspondientes, los
cuales están bajo el formato “raster”. Con lo que cada cuadricula (celda) tiene un valor
especifico de las variables dasométricas en cuestión. La Figura 5 muestra el resultado de la
implementación del árbol de regresión para la estimación de existencias en la Región
Chocachi. Observándose que en la mayor parte se estiman existencias superiores a los 15 m 3
por hectárea.
4. PROCESO DE EXTRAPOLACION
Con base a los modelos generados para la región base (R. Chocahi), se prepararon los mapas
de las variables auxiliares, que mejor estimaron cada una de las variables dasométricas, para
la región a extrapolar (R. Nopalera). Un ejemplo de estas variables se presenta en la Figura 6,
que corresponde a la banda 3 de la imagen Landsat TM.
P á g i n a | 68
Latin American Remote Sensing Week 2010
Figura 6. Banda 3 de una imagen de satélite usada como variable auxiliar en la Región
Nopalera.
Considerando que se uso, como base, el árbol de regresión generado para la Región Chocachi,
se prepararon las capas raster que correspondían a cada una de las variables auxiliares, Esto
de hizo para cada una de las variables dasométricas a estimar (densidad, posibilidad de
aprovechamiento (m3/ha/año) y diámetro de copa).
4.3. Validación
P á g i n a | 69
Latin American Remote Sensing Week 2010
Aprovechando la información del estudio que se hizo para desarrollar el programa de manejo
forestal de la Región Nopalera, se compararon los resultados obtenidos de la extrapolación,
con los obtenidos en terreno. El estudio para el programa se baso en un inventario de 202
sitios. La información recabada se estimo a nivel de hectárea, con el fin de hacerlo comparable
con las estimaciones definidas con la extrapolación. Con base a esto se hicieron algunos
análisis de correlación para ver cuál fue el grado de ajuste de los valores extrapolados con los
valores estimados con datos de terreno. La Figura 8 muestra un ejemplo en referencia a las
existencias. Donde se aprecia una tendencia definida, aunque en los valores bajos existe cierta
dispersión que propicia una baja correlación (R2= 0.3483). Esta correlación se ajusta a la
siguiente fórmula:
30
25
20
15
10
3
0 20 40 60 80 100 120 140 160 180 200 m /ha
220
N O P A L E R A
RODAL ha E D C
37 4.93 6.17 10.55 2.82
148 11.22 1.88 8.04 1.23
153 27.10 6.51 11.04 2.88
P á g i n a | 70
Latin American Remote Sensing Week 2010
5. CONCLUSIONES
6. REFERENCES
Ek, A. R. 2004. Mapping Forest Resources Using the k-Nearest Neighbor Method. Fact Sheet 6.
Remote Sensing and Goespatial Analysis Laboratory. University of Minnesota. College of
Natural Resources. Department of Forest Resources.
Flores G., J.G.; Mendoza B., M.A.; Celedonio A., B. 2007. Monitoreo de ecosistemas con
estrategias geoestadísticas, una aplicación de gran escala en Jalisco, México . Madera y
Bosques, 13 (2): pp. 97-104.
Franco L., H., Ek, A.R., y Bauer, M.E. 2001. Estimation and mapping of forest stand density,
volume, and cover type using the k-nearest neighbors method. Remote Sensing of
Environment. 77(3); pp. 251-274.
Allen, L.S. 1996. Ecological role of fire in the Madrean province. In: Proceedings: Effects of fire
on Madrean province ecosystems. December. Fort Collins, Colorado. USDA, For. Serv. Gen.
Tech. Rep. RM-GTR-289. pp 5-10.
Flores G., J. G. y P. N. Omi. 2003 Mapping forest fuels for spatial fire behavior simulations using
geomatic strategies. Agrociencia, 37(1): pp. 65-72.
UCODEFO No 2. 1997. Aplicación del inventario forestal continuo (I.F.C.) en los bosques del ejido
“El Largo”. Unidad de Conservación y Desarrollo Forestal No 2. Ciudad Madera, Chihuahua.
México. 34 p.
P á g i n a | 71
Latin American Remote Sensing Week 2010
The task of increasing the economic competitiveness of any country is not possible without
reinforcing the role of the higher professional education. Today, the traditional education, as
an access to learning, can not keep up with the current requirements of science and industry.
Russian company R&D Center SCANEX proves on a practice that setting up Remote Sensing
Centers allowing real-time imagery acquisition from Earth observing satellites within the
structure of Universities is one of the solutions of the above mentioned problems. This will
provide the proper environment for innovative education, to deliver the efficient training for
scientific and academic and teaching personnel, secure the role of the young professionals in
science, education and hi-tech, and maintain the continuity of generations in science and
education. SCANEX has delivered the land based UniScan™ centers to over 20 higher
education institutions in Russia, Kazakhstan, and Spain. These stations serve as the basis for
Earth monitoring from space centers providing the training and advanced training to produce
the specialists having the state-of-the-art knowledge in Earth Remote Sensing and GIS, as well
as the land-use monitoring and geo-data service for the economic operators in such diverse
areas as the nature resource management, agriculture, land property management, etc.
Since year 2009 satellite imagery service to Universities has extended worldwide outside the
Russian territory, providing EROS A satellite imagery at 1.8 m resolution, IRS-1D imagery at
5.8 and 23 m, SPOT 4 satellite imagery at 10 and 20 m resolution, and SAR RADARSAT-1
imagery with resolution ranging from 100 m to 8 m. Based on the Agreement between
SCANEX and the satellites Operators, and on unique flexible financial terms, the international
universities can purchase the universal compact UniScan™ stations including telemetry for the
first year of operation.
Creation of state-of-the-art remote sensing centers at universities will lead to a new quality
level for education and scientific studies and will enable to make education system in such
innovation institutions open to modern research work and economy.
Keywords: Earth remote sensing, space monitoring centers, UniScan™ ground station, geo-
portal
P á g i n a | 72
Latin American Remote Sensing Week 2010
Introduction
The task of increasing the economic competitiveness of any country is not possible without
reinforcing the role of the higher professional education, attracting youths to deliberate
choice of future profession. Today, the traditional education, as an access to learning, can not
keep up with the current requirements of science and industry.
Such an index as the education quality comes along with education results, currently
most often referred to as “competence”. Innovation of education should be more competence-
oriented than passing of knowledge that always gets out of date. Following this pattern
knowledge can be acquired on an individual bases. Such kind of education should be more
connected to practice than the traditional one.
One of the solutions to resolve these problems can be the introduction of Remote
Sensing Centers for Earth observation from Space for territories changes monitoring as
research and development links in the chain of universities. This will allow the students to
master practical skills having an ultra-modern laboratory base.
At the same time one who wishes to receive data with middle and high resolution
faces with such problems as high price for ground station and expensive telemetry fee.
Usually only national large remote sensing centers have a possibility to receive data with
middle and high resolution, both optical and radar. However, it was a dream for universities...
Since 2009 SCANEX provides the possibility of equipping universities all over the
world with technologies for receiving Earth remote sensing data of high resolution (up to 1.8
m) in real time. Participants of the world educational community can purchase the universal
UniScan™ ground station for receiving Earth observing data already completed with licenses
P á g i n a | 73
Latin American Remote Sensing Week 2010
for the right to receive 100 scenes of EROS A, 600 minutes of IRS-1D, unlimited access to SPOT
4 images and 50 scenes of RADARSAT-1 for the first year of operation. Initial inclusion of the
license in the functionality of UniScan™ ground station (pursuant to agreements with the
Operators of respective Earth remote sensing missions) leads to significant cost reduction of
receiving station and space imagery for universities. As a result the world universities will be
able to effectively utilize the state-of-the-art space technologies in their educational process
and scientific research by working with up-to-date satellite data received at their own stations
with the footprint of up to 2.5 thousand kilometers in radius.
Based on SCANEX technology Remote Sensing Centers have already been operating
with an excellent track record in such Russian Universities as Altay State University, Belgorod
State University, Ufa State Aviation Technical University, Tyumen State University, Moscow
State Technical University n.a. N.E. Bauman.
Samara Space Geoinformation Center has been operating in Samara State Aerospace
University since December 2006, and is currently the best equipped Earth Remote Sensing
Center in Russia and CIS countries, not only among the educational facilities, but also the state
and private space monitoring centers.
It has been currently the worldwide practice to set up Remote Sensing Centers at the
higher education entities and other educational organizations. Thus, UniScan™ based
laboratories and centers have been in operation at Kazakhstan-British Technical University of
Republic of Kazakhstan (Almaty) under the auspices of Kazakhstan System Modeling
Research Institute, at Satpayev Kazakhstan Research Science and Technology Institute
(Almaty), at two universities in Spain (Valladolid and Valencia).
The Remote Sensing Centers deliver the real-time training for executive decision
making support technology at the regional and municipal levels. The Centers can also provide
the commercial service as well, e.g., training at professional development programs,
contracted areal monitoring work management in a range of various applications
(management of natural resources, agriculture and forestry management, emergency
response, mapping updates, cadastral work, etc.) or research and development in various
fields of knowledge.
P á g i n a | 74
Latin American Remote Sensing Week 2010
Data received from the satellites in real-time are converted into ready-for-further-
analysis products within 0.5-1 hour after being received from space, which indicates a high
operational degree of the process (no data, available via Internet, can be obtained in such a
quick mode).
Remotely sensed data received by Remote Sensing Center of a university will allow
resolving following practical tasks concerning change detection using different types of
remotely sensed data both optical and radar:
- on/off shore oil spills detection within the oil production and transportation areas;
- ecological situation of water areas and new seaports, pipelines and oil terminals
construction sites;
P á g i n a | 75
Latin American Remote Sensing Week 2010
- agricultural monitoring for crop rotation rules observation and proper arable lands use;
- creation of up-to-date thematic maps of natural objects condition (vegetation, soil cover,
areas hazard rate, etc.);
- turn the university into one of the world leading education institutions equipped with
cutting-edge technology and firmware for Earth observation from space (footprint of ground
station is about 12 million square kilometers);
- carry out training and advance training of specialists having skills in remote sensing and GIS,
used for decision-making support;
- monitor territories and submit data in support of decision-making of regions and sub-
regions.
5. Summary
UniScan™ ground station technology developed by ScanEx and implemented for over
20 Remote Sensing Centers/Laboratories at universities enables new unique possibilities to
receive in real-time high resolution optical and radar data on standard personal computer of
users.
Since year 2009 satellite imagery service to Universities has extended worldwide
outside the Russian territory, providing EROS A satellite imagery at 1.8 m resolution, IRS-1D
imagery at 5.8 and 23 m, SPOT 4 satellite imagery at 10 and 20 m resolution, and SAR
RADARSAT-1 imagery with resolution ranging from 100 m to 8 m. Based on the Agreement
P á g i n a | 76
Latin American Remote Sensing Week 2010
between SCANEX and the satellites Operators, and on unique flexible financial terms, the
international universities can purchase the universal compact UniScan™ stations including
telemetry for the first year of operation. This is the first precedent in the world.
6. Conclusion
Creation of state-of-the-art Remote Sensing Centers/Laboratories at universities will
lead to a new quality level for education and scientific studies and will enable users to make
educational system in such innovation institutions open to modern research work and
economy.
P á g i n a | 77
Latin American Remote Sensing Week 2010
Ingeniero Forestal,
Erick Bollmann
Natania Núñez
Los proyectos del “Plan de Desarrollo Forestal Comunal (PDFC) en comunas rurales pobres”
se han iniciado a fines de la década pasada, en el contexto de las comunas rurales con
problemas ambientales y socioeconómicos3. La propuesta de acción es consensuada con
entidades publicas y privadas, para luego culminar con un convenio interinstitucional4
contando con el apoyo del programa Facility, la FAO, y el compromiso de un grupo creciente
de municipios rurales que forman parte de la Asociación Chilena de Municipalidades5.
El proyecto PDFC se refiere a una experiencia aplicada y sistematizada con los gobiernos
locales, en la que se ha logrado reducir la pobreza aplicando un enfoque de agroforestería
sostenible, acceso a tecnologías modernas (TI), “planificación – acción” participativas,
seguridad a los medios de vida, con apoyo a la asociatividad comunitaria a nivel de
microcuencas.
1
Ver www.pdfc.cl
2
Las comunas de Lolol y Litueche inicialmente, para luego integrar a las comunas de: Navidad, La Estrella,
Marchigue, Pichilemu, Paredones, Chepica, en la VI Región del Libertador Bernardo O´Higgins, y las comunas
de Empedrado y Chanco en la VII Región del Maule. En la actualidad se trabaja en la cuenca del río Toltén, IX
Región de la Araucanía, en la que existen 11 comunas rurales.
3
Ver www.munitel.cl
5
Las comunas de Lolol y Litueche inicialmente, para luego integrar a las comunas de: Navidad, La Estrella,
Marchigue, Pichilemu, Paredones, Lolol, Chepica, en la VI Región del Libertador Bernardo O´Higgins, y las
comunas de Empedrado y Chanco en la VII Región del Maule. En la actualidad se trabaja en la cuenca del río
Toltén, IX Región de la Araucanía, en la que existen 11 comunas rurales.
P á g i n a | 79
Latin American Remote Sensing Week 2010
P á g i n a | 80
Latin American Remote Sensing Week 2010
Contreras M.¹, Oyola N.12, Ramírez A2., Figueroa A.3, Novoa F.¹
Introducción
Los humedales son considerados dentro de los ecosistemas más importantes del planeta
(Mitsch & Gosselink, 2000). Éstos comparten una propiedad primordial, el agua, que juega un
rol fundamental en el ecosistema y en la determinación de la estructura y las funciones
ecológicas de éste (SAG, 2006).
Los humedales son hábitats para numerosas especies y sirven como áreas de refugio a
especies migratorias (Salinas et al., 2002). Además de lo anterior también funcionan como
filtro para los contaminantes. Otro factor a considerar es que los bienes y servicios que éstos
entregan son limitados y que su degradación puede significar reducción, fragmentación y, en
los casos más severos, su pérdida. Si se desea conservar sus múltiples beneficios, como
hábitats únicos y fuente de agua, no se debe sobrepasar el umbral crítico en el uso que
hacemos de ellos.
Lo anterior está siendo abordado por el proyecto “Diseño del Inventario Nacional de
Humedales y el Seguimiento Ambiental” de CONAMA. El trabajo que se presenta a
continuación se encuentra inmerso dentro de éste proyecto. En el documento se describen los
objetivos, la metodología y el estado de avance en que se encuentra la identificación de los
humedales (cuerpos de agua y vegetación hidrófila) del país, a escala regional, y como se
P á g i n a | 81
Latin American Remote Sensing Week 2010
Objetivos
General: generar una cobertura digital base para la realización del diseño del inventario
nacional de humedales y seguimiento ambiental utilizando imágenes multiespectrales (escala
nacional) e imágenes hiperespectrales (escala local).
Específicos: i) identificar humedales a escala regional para todo el país utilizando imágenes
de la serie Landsat y ii) identificar humedales indicadores, dentro de la cuenca hidrográfica, a
escala local, con alta resolución espacial y espectral, utilizando imágenes hiperespectrales
aerotransportadas del sensor AISA Eagle.
Metodología
Imágenes Landsat
El programa Landsat existe desde 1972 y es considerado uno de los programas más
fructíferos de teledetección espacial desarrollado hasta el momento. Ha sido utilizado con
éxito en diversos estudios (Villeneuve, 2005), debido a la buena resolución de sus sensores, su
carácter global, lo periódico de la observación que realiza y su gran difusión (Jensen 2007 y
Chuvieco, 2002). Además las características técnicas de las bandas de éste satélite permiten la
discriminación de superficies inundadas y de vegetación.
Hasta febrero de 2009 era necesario pagar por la adquisición de una imagen Landsat.
Actualmente las imágenes Landsat se pueden obtener gratuitamente en la United States
Geological Survey (USGS) y en el “Instituto Nacional de Pesquisas Espaciais” (INPE) de Brasil.
Lo anterior permite además de obtener las escenas más recientes disponibles de esta serie,
acceso a imágenes de archivo, con una data disponible de más de 30 años, lo que puede ser de
gran utilidad para estudios de líneas de base históricas de humedales y de vegetación.
Adquisición de imágenes
Procesamiento imágenes
Preproceso
En la etapa de preproceso se realizan las correcciones atmosféricas y geométricas. Para la
P á g i n a | 82
Latin American Remote Sensing Week 2010
primera de ellas se utiliza el modelo de Chávez, (Chávez, 1975 en Chuvieco, 2002) donde se
indica que los cuerpos u objetos de alta absortividad deberían tener valores de nivel digital
cero (Chuvieco, 2002), este procedimiento, conocido como Dark Object Subtraction Model
(DOSM), necesita datos de la variación solar, de la fecha del año e inclinación del ángulo de
elevación del sol, con ellos se puede eliminar el efecto de la niebla sobre la imagen. Esta
corrección a contribuido sustancialmente a mejorar la nitidez de la nubosidad en las imágenes
de la zona sur del país. En cuanto a la corrección geométrica se ha utilizado el modelo digital
de elevación de ASTER, obtenido en Internet. Además, para hacer más precisa esta corrección,
se han incorporado puntos de control obtenidos a partir de información vectorial de drenes, a
escala 1:50.000, lo que ha permitido comparar las imágenes con la línea de costa, quebradas
y ríos principales. Cabe señalar que las imágenes originales, en general, no presentan
mayores desplazamientos geométricos, con respecto a la cubierta de drenes y el DEM Aster,
lográndose un buen ajuste geométrico.
Procesamiento
El índice NDVI se relaciona con la biomasa o vigor de la vegetación (Jensen, 2000 en Polidorio,
2005). Para su cálculo se utiliza las bandas del rojo e infrarrojo cercano. Áreas de alto vigor
vegetacional poseen una mayor reflectividad en el infrarrojo cercano y una menor
reflectividad en el rojo. El índice se calcula utilizando la siguiente ecuación:
Los valores entregados por esta relación varían entre -1 y 1. Si el valor se acerca a 1 está
indicando una vegetación vigorosa y sana, los valores cercanos a cero se relacionan con suelo
fraccionado a desnudo, valores negativos generalmente corresponden a nubes o cuerpos de
agua. Mediante el empleo de este índice podemos identificar diferentes grados de cobertura
vegetal.
El cálculo del NDWI establece un índice que discrimina cuerpos de agua. Para ello se utilizan
las bandas del verde e infrarrojo cercano (Polidorio et al., 2005). La utilización de este índice
maximiza la reflectividad propia del agua realzando sus características. El índice se calcula
utilizando la siguiente ecuación:
P á g i n a | 83
Latin American Remote Sensing Week 2010
Los valores del NDWI varían entre -1 y 1, asociando valores positivos para las superficies con
agua o húmedas y cero o negativo para el suelo y la vegetación terrestre.
Imágenes hiperespectrales
Debido a la alta resolución espacial que se puede obtener con este sensor, la identificación y
delimitación de los cuerpos de agua y formaciones vegetacionales (u otras características del
terreno) alcanzan altos grados de precisión, en cuanto a ubicación y cálculo de superficie.
Vegetación
a) Contenido de agua en del dosel, WBI (Water Band Index) a medida que el contenido de
agua aumenta se incrementa la absorción alrededor de los 900 a 970 nm, las bandas límite de
este rango se utilizan para el cálculo del índice.
b) Vigorosidad, NDVI (The Normalized Difference Vegetation Index). Para su cálculo se utiliza
la banda de los 900 nm y la de 679, correspondientes al infrarrojo cercano y el rojo
respectivamente.
c) Eficiencia en cuanto al uso de la luz, PRI (Photochemical Reflectance Index) este índice
mide la sensibilidad a los cambios de pigmentos en el follaje, los que son indicativos de la
actividad fotosintética de la vegetación. Las bandas utilizadas en su cálculo son la 531 y la de
570 nm.
Sistema acuático
P á g i n a | 85
Latin American Remote Sensing Week 2010
en las longitudes de onda de 470, 550, 680 y 705 nm se producen los máximos contrastes
entre las variables mencionadas.
Figura 2. Respuesta espectral de agua clara y agua con microalgas (Han, 1997).
Productos y resultados
Como producto, dentro del proyecto, se contempla la generación de: i) cobertura digital de
humedales de todo el país a escala regional y base de datos descriptiva, se contempla cuerpos
de agua y vegetación azonal asociada si se presenta, ii) cobertura digital de humedales
indicadores piloto a escala local, y base de datos con información sobre rasgos bióticos,
abióticos.
Para el primer producto se tiene el 40% de avance a la fecha, lo que se aprecia en la Figura 3.
En tanto que en la cobertura para el nivel local aún no se ha realizado la toma de la imagen
hiperespectral, la que se contempla para mediados de octubre de 2010, según cronograma del
proyecto.
P á g i n a | 86
Latin American Remote Sensing Week 2010
Bibliografía
CONAMA. 2005. Estrategia nacional para la conservación y uso racional de los humedales en
Chile. http://www.sinia.cl/1292/articles-35208_recurso_1.pdf
P á g i n a | 87
Latin American Remote Sensing Week 2010
Jensen, J. R., 2007. Remote Sensing of the Environment: An Earth Resource Perspective. 2nd
Ed. Upper Saddle River. Prentice Hall, 592 pp.
Mitsch, W.J. & Gosselink J.G. Wetlands. 2000. John Wiley & Sons, Inc, New York,Third Edition.
Salinas, W., Treviño, E., Jaramillo, J., Campos, J. Identificación y Clasificación de Humedales
Interiores del estado de Tamaulipas por Percepción Remota y Sistemas de Información
Geográfica. Investigaciones Geográficas, número 049. Universidad Nacional Autónoma de
México, Distrito Federal, México, pp.74-91. Diciembre 2002.
SAG - BIOTA. Estudio de los sistemas vegetacionales azonales hídricos del del valle central y
precordillera del Bío-Bío. 2009.
Polidorio, A. M., Garcia, A.M., Nobuhiro N., Bueno M. 2005. Segmentação de corpos d’água em
imagens multiespectrais e temporais usando watershed com marcadores automaticamente
definidos. Anais XII Simpósio Brasileiro de Sensoriamento Remoto, Goiânia. INPE, p. 4249-
4257. Brasil, 16-21 abril 2005.
P á g i n a | 88
Latin American Remote Sensing Week 2010
Parviz Tarikhi
parviz_tarikhi@hotmail.com
Abstract: InSAR is a radar technique for combining synthetic aperture radar (SAR) single look
complex images to form interferogram and utilizing its phase contribution to land
topography, surface movement and target velocity. In recent years considerable applications
of Interferometric SAR technique are developed. It is an established technique for generating
high quality digital elevation models (DEM) from spaceborne and airborne data, and that it
has advantages over other methods for the generation of large area DEM. InSAR is capable of
producing DEMs with the precision of a couple of ten meters whereas its movement map
results have sub-centimeter precision.
DEMs are used in many applications in the context of earth sciences such as in topographic
mapping, environmental modelling, rainfall-runoff studies, landslide hazard zonation, and
seismic source modelling.
In this paper the experience gained on the subject in course of the continuous research work
since 1994 in the former Iranian Remote Sensing Center and the Iranian Space Agency is given
and the case studies on Izmit quake of August 1999, Bam quake of December 2003 and Haiti
Earthquake of 12 January 2010 using the related single look complex (SLC) images collected
by the European ERS-1&2 and Envisat remote sensing satellites is discussed. Using tandem
images of ERS-1&2 with the temporal baseline of one day and few seconds for the images of
Envisat leads to very good results for DEM generation while the spatial baselines smaller than
40m may not be suitable to this mean since slopes will be under-sampled.
________________________________________
Introduction: Digital Elevation Model (DEM) refers to the process of demonstrating terrain
elevation characteristics in 3-D space, but very often it specifically means the raster or regular
grid of spot heights. DEM is the simplest form of digital representation of topography, while
Digital Surface Model (DSM) describes the visible surface of the Earth (Fig. 1).
P á g i n a | 89
Latin American Remote Sensing Week 2010
Radar is a useful tool for land and planetary surface mapping. It is a good mean for obtaining a
general idea of the geological setting of the area before proceeding for field work. Time,
incidence angle, resolutions and coverage area all play important role at the outcome.
InSAR is a sophisticated processing of radar data for combining synthetic aperture radar
(SAR) single look complex (SLC) images to form interferogram and utilizing its phase
contribution to generate DEM, surface deformation and movement maps and target velocity.
The interferogram contains phase difference of two images to which the imaging geometry,
topography, surface displacement, atmospheric change and noise are the contributing factors.
Two coherent SAR images are required to produce an interferogram. The images are first co-
registered for finding the offset and difference in geometry between two amplitude images.
Normally the baselines of 80-300m are suitable for DEM generation. One SAR image is then
re-sampled to match the geometry of the other, meaning each pixel represents the same
ground area in both images. The interferogram is generated by multiplication of the first
image (master) to the complex conjugate of the second image (slave), and the interferometric
phase due to the reference ellipsoid is removed; the process is referred to as flattening.
In DEM generation by geodesic measurements, the planimetric coordinates and height values
of each point of the feature are summed point-by-point and using the acquired data the
topographic maps are generated with contour lines. The 1:25000-scale topographic maps are
common example. The method uses contour-grid transfer to turn the vector data from the
maps into digital data. For DEM generation by photogrammetry, the photographs are taken
from an aircraft or spacecraft and evaluated as stereo-pairs and consequently 3-D height
information is obtained.
P á g i n a | 91
Latin American Remote Sensing Week 2010
DEM generation by remote sensing can be made in some ways, including stereo-pairs, laser
scanning (LIDAR) and InSAR. There are three types of InSAR technique that is single-pass,
double-pass and three-pass (Fig.2). In double-pass InSAR, a single SAR instrument passes over
the same area two times while through the differences between these observations, height
can be extracted. In three-pass interferometry (or DInSAR) the obtained interferogram of a
double-pass InSAR for the commonly tandem image pairs is subtracted from the third image
with wider temporal baseline respective to the two other images. In single-pass InSAR, space-
craft has two SAR instrument aboard which acquire data for same area from different view
angles at the same time. With single-pass, third dimension can be extracted and the phase
difference between the first and second radar imaging instruments give the height value of
the point of interest with some mathematical method. SRTM used the single-pass
interferometry technique in C- and X-band. Earth’s height model generated by InSAR-SRTM
with 90-m horizontal resolution is available while the DEM with 4-to-4.5-m relative accuracy
is also available for restricted areas around the world.
Comparison
InSAR ability to generate topographic and displacement maps in wide applications like
earthquakes, mining, landslide, volcanoes has been proven. Although other facilities like GPS,
total stations, laser altimeters are also used, comparison between InSAR and these tools
reveals its reliability. Laser altimeters can generate high resolution DEM and low resolution
displacement maps in contrary to InSAR with the spatial resolution of 25m. However, most
laser altimeters record narrow swaths. Therefore, for constructing a DEM by laser altimeter,
more overlapping images are required. Displacement map precision obtained by terrestrial
surveying using GPS and total stations is similar or better than InSAR. GPS generally provides
better estimation of horizontal displacement and with permanent benchmarks slow
P á g i n a | 92
Latin American Remote Sensing Week 2010
deformations is monitored for years without being concerned about surface de-correlation.
The most important advantage of InSAR over GPS and total stations are wide continuous
coverage with no need for fieldwork. Therefore, wide and continuous coverage, high
precision, cost effectiveness and feasibility of recording data in all weather conditions are its
main privileges. However, it is important that the InSAR displacement result is in the line-of-
the-sight direction and to decompose this vector to parallel and normal components the
terrestrial data or extra interferograms with different imaging geometry are required. It is
shown that DEM generated by photogrammetric method is more accurate than the others. It
has approximately 5.5m accuracy for open and 6.5m for forest areas. SRTM X-band DSM is 4m
less accurate for open and 4.5m less accurate for forest areas.
Data availability and atmospheric effects limit using InSAR, however processing of its data is
challenging. For each selected image pair, several processing steps have to be performed. One
of the current challenges is to bring the techniques to a level where DEM generation can be
performed on an operational basis. This is important not only for commercial exploitation of
InSAR data, but also for many government and scientific applications. Multi pass
interferometry is affected by the atmospheric effects. Spatial and temporal changes due to the
20% of relative humidity produce an error of 10cm in deformation. Moreover, for the image
pairs with inappropriate baseline the error introduced to the topographic maps is almost
100m. In topographic mapping this error can be reduced by choosing interferometric pairs
with relatively long baselines, while in the displacement case the solution is to average
independent interferograms.
InSAR has some similarities to stereo-optical imaging in that two images of the common area,
viewed from different angles, are appropriately combined to extract the topographic
information. The main difference between interferometry and stereo imaging is the way to
obtain topography from stereo-optical images. Distance information is inherent in SAR data
that enables the automatic generation of topography through interferometry. In other words
DEMs can be generated by SAR interferometry with greater automation and less errors than
optical techniques. Moreover, using DInSAR surface deformations can be measured
accurately.
P á g i n a | 93
Latin American Remote Sensing Week 2010
Different DEM generation methods of Advanced Spaceborne Thermal Emission and Reflection
Radiometer (ASTER) stereoscopy, ERS tandem InSAR, and SRTM-InSAR are used. Both the
ERS-InSAR and SRTM DEMs are free of weather conditions, but ASTER DEM quality may be
affected by cloud coverage in some local areas. InSAR has the potential of providing DEMs
with 1-10cm accuracy, which can be improved to millimeter level by DInSAR. Its
developments are rapid however it is our requirements that say which one is better for use.
In course of the years of studies and verifications by our Microwave Remote Sensing Group
good and valuable achievements gained on InSAR technology applications. In this paper the
results and outcomes related to the generation of DEMs using Interferometric SAR are given
as example.
(I) Advent of the awful earthquake of December 26, 2003 in Bam, Iran has drawn the
attention of the many scientific and humanitarian organizations to study the phenomenon and
its causes and origins as well as its impacts and developments. Fig.3 shows the position of the
P á g i n a | 94
Latin American Remote Sensing Week 2010
area on a satellite image with the photos of the pre and post quake views of the Citadel Bam as
insets. In Fig.4, the left image is the topo-DInSAR product acquired from the Envisat-ASAR
data of 11 June and 3 December 2003, while the image at right is the topo-DInSAR product of
the 3 December 2003 and 7 January 2004. The team conducted by the author used the data
provided by ESRIN, and the DORIS and IDIOT softwares to generate the products. The middle
image obtained by NASA scientists is the 3-D perspective view of vertical displacement of the
land surface south of Bam during the 3.5 years after the 6.6 earthquake of December 26, 2003
that is derived from analysis of radar images. Blue and magenta tones show where the ground
surface moved downward; yellow and red tones show upward motion (particularly in south
of Bam). Displacements are superimposed on a false-color Landsat Thematic Mapper image
taken on October 1, 1999. In the right image that is obtained from the ASAR data pre and post
earthquake the curl-shape pattern south of Bam is distinguishable where such the torsion is
not visible in the left image that obtained from pre-earthquake data. For the left image the
normal baseline is 476.9 m and parallel baseline is 141.6 m, while for image at right the
normal baseline is 521.9 m and the parallel baseline is 268.3 m. The left image demonstrates
that the related interferogram includes four lobes. Since the displacement in the east is
greater than that in the west, the related lobes are larger. The displacements measured along
the radar line-of-sight direction are 30cm and 16cm at south-east and north-east lobes of the
interferogram, respectively. However, the displacement related to the western part of the area
is about 5cm along the radar line-of-sight direction.
P á g i n a | 95
Latin American Remote Sensing Week 2010
(II) In Fig.5, the top images at the left and right show the tandem amplitude data of 12 and 13
August 1999 (4 and 5 days pre-quake) of Izmit, Turkey as master and slave images
respectively. The research team conducted by the author used the ERS-1&2 data provided by
ESRIN, and the Earth-view and SAR Toolbox softwares to generate the variety of related
products. The normal baseline for the image pair is 224.2 m and parallel baseline is 91.1 m.
The image at the bottom left is the coherence image while the right image at the bottom
depicts the DEM image where the interferogram is overlaid on it. For each product the
relevant histogram is seen as inset. The similarity of the histograms of master and slave
images is considerable due to the high correlation of the images that is clearly seen in
coherence image. It is important to note that lowest coherence values (darkest values)
correspond both to steep slopes or vegetated areas (especially visible in the lower part of the
image) and to the lakes (image center and left). DEMs generated from the tandem images are
accurate because of the high correlation between master and slave images. Although both the
master and slave images are pre-earthquake data of the 7.8 earthquake of August 17, 1999,
the strain in the disaster area is visible a week before the quake. It could be a useful precursor
for the advent of a disaster like the earthquake in Izmit.
P á g i n a | 96
Latin American Remote Sensing Week 2010
(III) Fig.6 shows a sample of data products in the framework of the research and study work
on Haiti earthquake. On January 12, 2010, Tuesday a huge quake measuring 7.0 rocked the
Caribbean Haiti. It destroyed mainly the capital of Haiti, Port-au-Prince toppling buildings and
causing widespread damage and panic. Using the radar data provided by the European Space
Research Institute (ESRIN) affiliated to the European Space Agency (ESA) since early March
2010 an investigation and research work on the available data was carried out. The data
included 47 single look complex images (SLCI) of the C-band Advanced Synthetic Aperture
Radar (ASAR) image mode system of Envisat satellite in addition to other type of data. The
results sound good and below a sample of DEM generated for the area of study, Nord-Ouest
Department (North-West Province) is seen. The cities of Cap du Mole Saint-Nicolas and Bale-
de-Honne are situated in north-west and south respectively. The products are generated by
combining the Envisat ASAR images taken at the same day of 4 March 2010 with 8 seconds of
time interval and the virtual baseline of 13.23m while the parallel baseline amounts only
2.1cm that is too close and comparable to the radar wavelength of 56mm for Envisat.
P á g i n a | 97
Latin American Remote Sensing Week 2010
Acknowledgement
The SLC images for both the Izmit, Bam and Haiti areas for our study was generously provided
by the European Space Research Institute (ESRIN) of the European Space Agency (ESA) in
1999, 2004 and 2010 respectively.
Presenter’s CV
Parviz Tarikhi specializes in radar remote sensing since 1994 and heads the Microwave
Remote Sensing Research Core at the Mahdasht Satellite Receiving Station in Karaj, Iran. He
holds a PhD in physics focusing on microwave remote sensing. He has been involved with the
United Nations Committee on the Peaceful Uses of Outer Space (UN-COPUOS) since 2000,
including as Second Vice-Chair and Rapporteur in 2004-06 of the committee bureau. Since
2001 he has co-chaired Action Team number 1 of UNISPACE-III with the mission ‘to develop a
comprehensive worldwide environmental monitoring strategy’. From 2004-07 he led the
Office for Specialized International Cooperation of the Iranian Space Agency. He has made in
the mean time years of research and study on the developments and status of space science
and technology with a particular focus on Iran. For more information please see:
http://parviztarikhi.wordpress.com.
P á g i n a | 98
Latin American Remote Sensing Week 2010
Dr. B. S. Chaudhary
Associate Professor
Department of Geophysics,
bsgeokuk@yahoo.com
Haryana state may be considered as covered by three basins, namely the Yamuna basin, the
Ghaggar basin and internal basin which are parts of Ganges and Indus system. In northern
Haryana, the land area slopes from northeast to southwest, whereas in south, it slopes from
southwest to northeast. This makes almost a latitudinal depression along Sirsa-Fatehabad-
Hisar-Jind-Rohtak-Delhi axis. This saucer shaped physiography of the state is responsible for
many problems related to water resources along this central axis. These are flooding, water
logging, and salinity. The ground water conditions in northern and southern parts of state are
fresh but facing the problem of water level decline due to over exploitation for intensive
agriculture purposes. As a result, at present Haryana have problems of over drafting of fresh
quality water in northern and southern districts and excessive canal irrigation in the areas of
poor/ marginal ground water conditions due to highly water intensive & multiple cropping
systems as a result of green revolution. This is creating the two diverse problems that of
ground water table declining in the northern and southern part whereas the water logging
and salinization in the central portion. This call for integrated study of water resources in the
state for its sustained use.
Looking to the benefits of the RS and GIS technology a number of projects has been taken up
in the state by various organizations. The present paper discusses the preparation of ground
water prospects maps in various parts of Haryana state, India on one side and the initiatives
for artificial recharge of ground water in some other areas. First, Hydrogeomorphological
maps on 1:50,000 scale showing different ground water prospect zones are prepared for
different districts in Haryana State, India. These depict ground water worthy features. This
P á g i n a | 99
Latin American Remote Sensing Week 2010
information has been supplemented with the available inputs from existing sources about the
depth to water level and ground water quality etc. The other maps such as land use/ land
cover, geomorphology, drainage/ canal network and soils etc have also been consulted for
preparing water resources action plan. The maps thus prepared depict different units for
further ground water prospecting. It is to mention here that some of the Palaeo-channels have
been picked up first time. The surface water resources action plan maps have been prepared
by integrating geomorphology, slope, drainage, soils maps and various sites have been
suggested for site specific water resources conservation measures such check dams/ gully
plugging, earthen dams etc. The information thus developed has been submitted to various
departments involved in the planning and management of natural resources in the state for
further implementation of the activities suggested in different areas.
P á g i n a | 100
Latin American Remote Sensing Week 2010
ABSTRACT:
En Chile existe un marco legal que regula la actividad forestal, donde CONAF ha definido áreas
prioritarias para ser fiscalizadas, a partir de una serie de variables referenciales, cuyo
* Centro de Geomática, Universidad de Talca, casilla 721, Talca, Chile. Fono: 56-71-201580, Fax:
56-71-201708.
P á g i n a | 101
Latin American Remote Sensing Week 2010
En este contexto, el objetivo del presente estudio fue establecer una línea base para la
fiscalización forestal, a partir del uso de tecnologías geoinformáticas y la caracterización y
análisis, realizados por expertos, de la situación ambiental de zonas cubiertas con bosque
nativo y plantaciones en las comunas de Longaví y Parral, Región del Maule, determinando de
esta forma áreas prioritarias para una eficiente fiscalización. Mediante el uso de SIG se
elaboró una base de datos digital de la zona de estudio, conformada por variables tales como
vegetación, tipos forestales, pendientes, caminos, capacidad de uso, hidrografía, distancia a
centros urbanos y sitios poblados. A su vez, la utilización EMC, permitió la evaluación de las
variables ambientales consideradas, mediante la consulta a expertos relacionados con la
temática forestal. Finalmente, se generó un modelo de capacidad de acogida mediante el
método de la sumatoria lineal ponderada (SLP), que integró los criterios y alternativas
evaluadas por los expertos, estableciendo las zonas prioritarias para fiscalización forestal.
2. Cuerpo
Dadas las características climáticas, la zona de estudio presenta temperaturas que oscilan
entre los 3°C en invierno y 31°C en la época estival y sus precipitaciones alcanzan en un año
normal los 750 mm (Santibáñez y Uribe, 1993). Las clases de suelos predominantes de
acuerdo a la clasificación realizada por Ciren-Corfo (1983) son principalmente III (huertos
frutales y otros cultivos permanentes), IV (terrenos con cultivos extensivos) y VII (ganadería y
forestal).
Figura 1: Zona de estudio, comunas de Longaví y Parral, Región del Maule, Chile.
P á g i n a | 102
Latin American Remote Sensing Week 2010
2.2 Metodología
2.3 Fase 1
Determinación de las variables en estudio: las variables utilizadas deben cumplir con el
atributo de representatividad espacial, de acuerdo a ello se consideraron las siguientes
variables: tipo de vegetación, tipos forestales, pendiente del terreno, distancia a caminos,
capacidad de uso del suelo, red hidrográfica, cercanía a centros urbanos (puntos de
comercialización de productos forestales madereros). Cabe señalar, que la zona de estudio no
presenta superficies resguardadas por el Sistema Nacional de Áreas Silvestres Protegidas por
el Estado (SNASPE), motivo por lo cual esta temática no fue considerada.
Elaboración de la base de datos digital: Una vez definidas las variables a utilizar en el
estudio (incluyendo los factores y limitantes de la modelación), fue factible comenzar con la
construcción de la base de datos que se conformó por las coberturas digitales
P á g i n a | 103
Latin American Remote Sensing Week 2010
Definición de las alternativas para cada criterio: Las alternativas están representadas por
objetos o unidades espaciales, las cuales representan un sitio individual, constituyendo así el
conjunto global de alternativas de selección. A continuación, se exponen las clases temáticas
(alternativas de selección) utilizadas en cada criterio:
P á g i n a | 104
Latin American Remote Sensing Week 2010
indican los siguientes rangos de distancias: 0 a 50 m, 50 a 300 m, 300 a 400 m y 400 a 500
m.
- Clases de suelo: Esta variable considera la capacidad de uso del suelo y por ende, su
capacidad para acoger actividades productivas y agroeconómicas. Dada sus
características, no fue necesario realizar un mayor procesamiento a este tipo de cobertura,
puesto que los polígonos vectoriales de suelo que fueron rasterizados, poseen los
atributos de las clases de suelos que se encuentran en los mosaicos escala 1:20.000 de
CIREN. De esta forma, la cobertura estuvo conformada por las siguientes clases temáticas
de suelo: clase I, clase II, clase III, clase IV, clase V, clase VI, clase VII y clase VIII.
- Presencia de cursos de agua: todos aquellos sectores que forman parte de la cobertura
de hidrología, sectores no susceptibles de ser incluidos en la evaluación, fueron
codificados con el valor 0, mientras que los restantes sectores, apropiados para
fiscalización forestal, fueron codificados con el valor 1.
- Presencia de red caminera: Para esta situación, todo pixel catalogado como excluyente,
indica aquellos sectores con presencia de caminos, motivo por lo cual fue codificada con
valor 0. Sin embargo, los pixel codificados con valor 1, representaron todos los sectores
susceptibles de ser fiscalizados.
P á g i n a | 105
Latin American Remote Sensing Week 2010
fiscalizar, tales como: sitios poblados, áreas urbanas y cultivos agrícolas, codificándolos
con valor 0.
Fase 2
Aplicación del Método Sumatorial Lineal Ponderada (SLP): Obtenidas las matrices
evaluadas por los expertos, los pesos de los factores y un satisfactorio valor de consistencia en
la asignación de los juicios de valor (Gómez y Barredo, 2005), se optó por la utilización del
método de SLP, el cual se encuentra descrito por la ecuación 1:
n
ri (W
i 1
j * eij ) (1)
Donde: ri: capacidad del terreno para la ubicación de áreas prioritarias para fiscalización, Wj:
peso del criterio j, eij: valor normalizado de la alternativa i en el criterio j, n: número de
criterios involucrados en la investigación.
Esta ecuación fue aplicada a cada factor considerado en la evaluación realizada, obteniendo de
esta forma una cobertura que contiene los valores de capacidad de acogida (ri). Valores (ri)
más altos, indicaron sectores potencialmente adecuados para la determinación de áreas
prioritarias para fiscalización forestal en el área de estudio. Sin embargo, aún resta considerar
la exclusión de aquellas zonas limitantes, impuestas en las restricciones del modelamiento. La
aplicación de estas restricciones (proximidad a cursos de agua, proximidad a red caminera,
ocupación actual de suelos relacionados con áreas urbanas, sectores poblados y ocupación
actual de suelos relacionados con cultivos agrícolas), fue realizada a través de un proceso de
álgebra de mapas, en donde cada restricción o capa (1/0) fue multiplicada con la cobertura
que contiene los valores de capacidad de acogida (ri), extrayendo de esta forma las zonas
restrictivas.
Fase 3
P á g i n a | 106
Latin American Remote Sensing Week 2010
Verificación de los resultados: Los resultados obtenidos fueron visitados en terreno, con la
finalidad de validar y corregir el modelo obtenido a partir de la aplicación de las Técnicas de
Evaluación Multicriterio. Esta validación fue realizada a partir de la inspección aleatoria de
algunos puntos pertenecientes a las comunas de Longaví y Parral.
Resultados y discusión
Superficie Superficie
Clase Puntaje Aptitud
(ha) (%)
80 –
1 Muy Alta - -
100
Tabla 1: Categorías, puntaje y superficie considerada para las zonas del Modelo de Capacidad
de Acogida.
P á g i n a | 107
Latin American Remote Sensing Week 2010
Respecto a las superficies alcanzadas de cada categoría, al interior de las Comunas de Longaví
y Parral, no existen sectores con una “Muy Alta Prioridad” para efectuar actividades de
fiscalización forestal; la clase “Alta” cuenta con 1.241 ha, lo que representa un 1,20% del total
de la superficie potencialmente fiscalizable, la clase “Media” cuenta con 39.524 ha, que
corresponden a un 38,26% del total, la clase “Baja” cuenta con 30.710 ha, que corresponden a
un 29,73% del total; la clase “Muy Baja” cuenta con 23.240 ha, que corresponden a 22,50%; y
finalmente la clase “Excluyente” cuenta con alrededor de 8.570 ha, correspondiendo a un
8,3% de suelos con presencia de cultivos agrícolas, sectores urbanos y sitios poblados,
caminos e hidrografía. Estos resultados demuestran que la zona de estudio se encuentra
representada mayoritariamente por sectores con “Media Prioridad” para fiscalización forestal,
ubicados principalmente en la pre-cordillera andina del área de estudio.
CONCLUSIONES
P á g i n a | 108
Latin American Remote Sensing Week 2010
Respecto al recurso forestal nativo, los tipos forestales de mayor preocupación en cuanto a su
conservación y preservación, corresponden a la categoría de vulnerabilidad, tal es el caso del
Ciprés de la cordillera. En este marco, cabe señalar que los tipos forestales Esclerófilo y Roble
– Hualo, a pesar de ser los más intervenidos para su aprovechamiento productivo, se
encuentran siendo explotados sustentablemente en la zona de estudio, gracias al trabajo y
presencia permanente en terreno de profesionales de CONAF. En este contexto, existe una
mayor preocupación por manejo y uso sustentable del recurso forestal nativo, debido a que
éste presenta una mayor fragilidad y susceptibilidad ecosistémica. De igual forma, las
plantaciones forestales son foco de fiscalización constante, principalmente en lo referido al
cumplimiento en la obligación de reforestación después de efectuada la cosecha, tal como lo
señala el Art. 22º del D.L. 701/1974.
La integración de técnicas de EMC y SIG constituye una adecuada herramienta para tareas de
asignación – localización de actividades. Sin embargo, debido a la estructuración de la EMC
con panel de expertos, en ocasiones es posible la obtención de resultados con cierto grado de
incertidumbre y subjetividad, debido a que los resultados dependen del conocimiento de los
profesionales y de la opinión que ellos entreguen respecto a la problemática en análisis. En
este sentido, la mantención y actualización de la base de datos geoespacial es relevante para la
obtención de resultados precisos.
La metodología propuesta, puede ser considerada como un apoyo a los procesos de toma de
decisiones tradicionalmente utilizados para la determinación de áreas para fiscalización
forestal por parte de CONAF, pues además de reducir la superficie en análisis a través de la
aplicación de las restricciones, permite la simulación de diversos escenarios y la obtención de
múltiples resultados dependiendo de los valores que adopten los diversos criterios evaluados
por los profesionales.
REFERENCIAS
Barredo, J., 1999. Los SIG en la ordenación del territorio: Posibilidades y desarrollo utilizando
evaluación multicriterio. En: Sistemas de Información Geográfica y Teledetección espacial
Aplicadas a la ordenación del territorio y el medio ambiente. Talca: Universidad de Talca, Chile,
pp. 105-115.
P á g i n a | 109
Latin American Remote Sensing Week 2010
Séptima Región. Proyecto CONAF – CONAMA – BIRF. Universidad Austral de Chile, Pontificia
Universidad Católica de Chile, Universidad de Temuco. Santiago, Chile. 116 p.
Corporación Nacional Forestal (Conaf), 2004. Decreto Ley Nº 701 de 1974, sobre fomento
forestal y otras disposiciones legales. Gobierno de Chile, Corporación Nacional Forestal.
Santiago, Chile. 96 p.
Oyarzo, N., 2005. Establecimiento de áreas de riesgo para la fiscalización de las cortas no
autorizadas en la Región del Maule. Memoria para optar al título de Ingeniero Forestal.
Universidad Católica del Maule, Escuela de Ingeniería Forestal, pp. 3-32.
Saaty, T. 1982. The analytical hierarchy process. New York: Mc Graw Hill. Primera edición. 281
p.
Santibáñez F, Uribe J., 1993. Atlas agroclimático de Chile, Regiones VI, VII, VIII, y IX.
Universidad de Chile. Santiago, Chile. 99 p.
P á g i n a | 110
Latin American Remote Sensing Week 2010
Abstract: Aboveground objects are easily detected from height data. Buildings and
trees can be distinguished using texture information, since these objects have
different textural characteristics. This paper aims to detect and remove trees along
the streets from height data obtained by Laser Scanning Systems. If the trees are
previously removed from original laser height data, street ribbons become more
homogeneous and, as such, the subsequent process of urban road network extraction
usually becomes less complex. In a normalized Digital Surface Model, which is
obtained by subtracting a Digital Terrain Model from the corresponding Digital
Surface Model, the streets have height values around zero, while the aboveground
objects have height values around their corresponding heights. Based on these
characteristics, it is possible to separate the aboveground objects through a
histogram thresholding. Trees detection is performed by using a texture filter, which
calculates the standard deviation of gray levels, enhancing the most heterogeneous
regions. Using a thresholding procedure, the candidates regions corresponding to
trees are isolated and then subtracted from the elevated regions, remaining only
buildings. The experimental results showed that most trees were successfully
detected and removed, mainly large ones, which are just more difficult to handle in
urban road network extraction from laser height data.
Resumen:Los objetos altos pueden ser fácilmente detectados a partir de los datos de
altura. La distinción entre las edificaciones y los árboles se puede realizar con
información de textura, ya que estos objetos tienen diferentes características. El
objetivo de este trabajo es detectar y eliminar los árboles de las calles a partir de los
datos altimétricos obtenidos a través de los Sistemas de Láser Escáner. Si los árboles
son previamente suprimidos de los datos altimétricos originales, las calles se tornan
más homogéneas y así el posterior proceso de extracción de la malla viaria se hace
menos complejo. En un Modelo Digital de Superficies normalizado, que se obtiene a
través de la diferencia entre el Modelo Digital de Superficies y el Modelo Digital del
Terreno, las calles tienen valores de altura cerca de cero, mientras que los objetos
altos tienen valores de altura cerca de sus correspondientes alturas. Con base en
estas características, es posible separar los objetos altos a través de una
umbralización del histograma. La detección de los árboles se realiza usando un filtro
de textura que calcula la desviación estándar de niveles de gris, realzando las
regiones más heterogéneas. Utilizando un proceso de umbralización, estas regiones,
que supuestamente corresponden a los árboles son aisladas y luego, son sustraídas
de las regiones elevadas, quedando sólo las edificaciones. Los resultados
experimentales mostraron que la mayoría de los árboles fueron detectados y
eliminados con éxito, especialmente los grandes, que causan mayores dificultades en
P á g i n a | 111
Latin American Remote Sensing Week 2010
Keywords: laser scanner data, normalized Digital Surface Model, feature detection.
1. INTRODUCTION
The interest in using laser scanner data for cartographic feature extraction has
emerged due to maturity of the system and the increasing availability of data.
Developed methods for individual trees detection in forest environments using a
Canopy Height Model (CHM) generated from laser scanner data (Persson et al.,
2002; Pitkänen et al., 2004; Yu et al., 2004; Solberg et al., 2006) have produced
successfully results not only tree location, but also in measure single tree
parameters (height, crown width and even tree species).
In urban environments, many methods for the tree detection are related to the
building detection or reconstruction. The methods normally begin by extracting
the ground surface using a filtering algorithm. After this, the most important task
is to distinguish buildings from trees (Matikainen et al., 2007). Even though
aboveground objects can be easily detected from height data (e.g. normalized
Digital Surface Model - NDSM), the discrimination between buildings and trees
can be difficult, if only simple criteria like size or shape of region are considered.
Texture information can greatly contribute to distinction between these objects,
since they have different textural characteristics (Elberink & Mass, 2000). The
authors used anisotropic texture measures in laser scanner data. The difference
between the first and last pulse laser scanner data is used in Elberink & Maas
(2000) and Alharthy & Bethel (2002). Matikainen et al. (2007) used geometric
and textural attributes calculated from first and last pulse laser and color aerial
image. Other works used a segmentation process followed by classification
(Secord & Zakhor, 2007; Chen & Zakhor, 2009).
This paper aims tree detection and removal along the streets from NDSM
generated by height data of the Laser Scanner System. The Section 2 describes
the proposed methodology for trees detection and removal. The experimental
results are shown in Section 3. Some conclusions are present on Section 4.
2. METHODOLOGY
Figure 1 shows the flowchart of the proposed methodology. A point cloud data
obtained by Laser Scanner Systems is used as input data. From this data, a NDSM
is generated. Two steps occur in parallel. A step is aboveground objects
detection that results a binary image, representing the elevated objects and
ground surface. Other step is trees detection using a texture filter that enhances
the most heterogeneous regions. Trees detected are removed from aboveground
objects by difference operation, remaining only buildings. To obtain elongated
streets ribbons, without occlusions of trees and blocks as homogeneous regions,
morphological operators are applied.
P á g i n a | 112
Latin American Remote Sensing Week 2010
NDSM generation
Streets ribbons
The laser scanner data are initially interpolated to generate a regular grid
(DSM) using the nearest neighbor interpolator with resolution of 0.3 meters.
From this grid, points representing the ground surface are collected and is
generated another grid, the DTM (Digital Terrain Model). For a
representation of elevated objects on a flat surface, a NMDS is obtained by
difference between the DSM and DTM. The Figure 2 shows the image
representing the normalized Digital Surface Model obtained.
In a NDSM, the streets have height values around zero, while the
aboveground objects have height values around their corresponding
heights, as shown in Figure 2. Based on these characteristics it is possible to
separate the aboveground objects through a histogram thresholding,
resulting in the binary image, where the aboveground objects are
represented in white, while the ground is represented in black.
P á g i n a | 113
Latin American Remote Sensing Week 2010
Texture filter: Trees are more heterogeneous regions in NMDS and present
a texture that differentiates them from buildings (see Figure 2). Using a
texture filter know as standard deviation, these heterogeneous regions are
enhanced. This filter calculates the standard deviation of gray levels within a
rectangular mask, whose dimensions (height and width) are parameters
that must be provided. This operation results in a standard deviation-valued
image.
3. EXPERIMENTAL RESULTS
P á g i n a | 114
Latin American Remote Sensing Week 2010
Figure 3 - Results for the image patch 1: (a) NDSM image. (b) Binary image
representing the aboveground objects. (c) Standard deviation image from NDSM. (d)
Contours of trees detected. (e) Trees removed from elevated regions. (f) Elongated
and homogeneous street ribbons.
Figure 3 presents the results for the first image patch. Figure 3(a) shows the
NDSM image generated from the point cloud data. Note trees along the streets.
The binary image that represents the aboveground objects detected by histogram
thresholding is shown in Figure 3(b). Figure 3(c) shows the standard deviation
image from NDSM. In this image it is observed that trees and buildings edges are
enhanced. Figure 3(d) shows the contours of trees, which are successfully
detected, although some regions that are not trees are also detected. The
difference between elevated regions and tree regions is shown in Figure 3(e).
Please, note in the background image the trees removed. Figure 3(f) shows the
streets as homogeneous and elongated ribbons, without occlusions of trees. The
holes inside the blocks are filled and street edges become more regular.
P á g i n a | 115
Latin American Remote Sensing Week 2010
Figure 4 - Results for the image patch 2: (a) NDSM image. (b) Binary image
representing the aboveground objects. (c) Standard deviation image from NDSM. (d)
Contours of tree detected. (e) Trees removed from elevated regions. (f) Elongated
and homogeneous street ribbons.
Figure 4 presents the results for the second patch. Figure 4(a) shows the NDSM
image generated from the point cloud data. Note in the right side of image a lot of
trees occluding almost completely the street. The aboveground objects detected
by histogram thresholding are shown in Figure 4(b). Figure 4(c) shows the
standard deviation image derived from NDSM, with enhanced trees along the
streets and inside the blocks. The building edges also are enhanced. Figure 4(d)
presents the contours of trees detected, showing that trees are successfully
detected, including those within blocks. The difference between aboveground
regions and tree regions is shown in Figure 4(e), with trees removed shown in the
background image. The streets as homogeneous and elongated ribbons are
presented in Figure 4(f), which also shows that the holes inside de blocks are filled
and street contours become more regular.
4. CONCLUSIONS
This paper presented a methodology for trees detection and removal along the
street in NDSM image generated from laser scanner data. The steps of the
methodology were described and two experiments were presented. From a
visual analysis of the obtained results it is possible to conclude that trees were
successfully detected and removed. Moreover, the regularization strategy of
block contours produced streets closer to elongated and homogeneous ribbons.
P á g i n a | 116
Latin American Remote Sensing Week 2010
improve block contours. This will require other regularization strategies to make
block contours as straight as possible, filling the gulfs caused by regions without
buildings (e.g. wastelands, backyards, squares and parks) and removed trees.
ACKNOWLEDGEMENTS
This paper is part of a Ph. D. research supported by Brazilian National Agency for
Science and Technology (CNPq) which is developed in the Cartographic Sciences
Postgraduate Program at UNESP – Universidade Estadual Paulista – Brazil. The
authors thank the LACTEC for furnish the data for this research.
REFERENCES
Alharthy, A. & Bethel, J., 2002. Heuristic filtering and 3D feature extraction from
LiDAR data. In: The International Archives of the Photogrammetry, Remote Sensing
and Spatial Information Sciences, Austria, Vol. XXXIV, Part 3A, pp. 29-34.
Chen, G. & Zakhor, A., 2009. 2D Tree detection in large urban landscapes using aerial
LiDAR data. IEEE International Conference on Image Processing 2009, pp. 1693-
1696.
Elberink, S. O. & Mass, H. G., 2000. The use of anisotropic height textures for the
segmentation of airborne laser scanner data. International Archives of
Photogrammetry and Remote Sensing, Amsterdam, vol. 33, part B3, pp. 616-623.
Matikainen, L., Kaartinen, H. & Hyyppä, J., 2007. Classification tree based building
detection from laser scanner and aerial image data. In: ISPRS Workshop on Laser
Scanning 2007 and SilviLaser 2007, Finland, pp. 280-287.
Persson, Å., Holmgren, J. & Söderman, U., 2002. Detecting and measuring individual
trees using an airborne laser scanner. Photogrammetric Engineering & Remote
Sensing, 68, pp. 925-932.
Pitkänen, J., Maltamo, M., Hyyppä, J. & Yu, X., 2004. Adaptive methods for individual
tree detection on airborne laser based canopy height model. International Archives of
Photogrammetry, Remote Sensing and Spatial Information Sciences, 36, pp. 187-191.
Secord, J. & Zakhor, A., 2007. Tree Detection in Urban Regions Using Aerial Lidar and
Image Data. IEEE Geosciences and Remote Sensing Letters, vol. 4, issue 2, pp. 196-
200.
P á g i n a | 117
Latin American Remote Sensing Week 2010
Solberg, S., Naesset, E. & Bollandsas, O. M., 2006. Single Tree Segmentation Using
Airborne Laser Scanner Data in a Structurally Heterogeneous Spruce Forest.
Photogrammetric Engineering & Remote Sensing, Vol. 72, No. 12, pp. 1369-1378.
Yu, X., Hyyppä, J., Kaartinen, H. & Maltamo, M., 2004. Automatic detection of
harvested trees and determination of forest growth using airborne laser scanning.
Remote Sensing of Environment, 90, pp. 451-462.
P á g i n a | 118
Latin American Remote Sensing Week 2010
Summary: The forest gaps naturally formed are essential in the process of forest
regeneration, particularly by allowing greater penetration of light inside the canopy. This
study presents a methodology for mapping forest gaps in Atlantic Forest using LIDAR data.
LIDAR remote sensing has great potential for application in forest studies, because this
technique produces a large set of georeferenced points on Earth's surface with height
information, in addition, the pulse laser has the ability to penetrate the canopy and create
points in the ground beneath the trees. With information from the treetops and the ground
beneath the canopy, the vertical and horizontal structure of forest can be analyzed. The
techniques of digital image processing can be used to analyze the large volume of data
available through the LIDAR. In this study we applied filtering procedures in the spatial
domain, segmentation based on region growing and classification on the images generated
from the interpolation of points obtained by LIDAR, aiming mapping of forest gaps. We used
two algorithms for image classification, Maxlike and Bhattacharya. To evaluate the accuracy of
the mapping the Kappa index was generated. The LIDAR data used in the study were acquired
using an airborne laserscanner Leica ALS-50 with a frequency of 150 kHz.
Resumen: Los claros del bosque forman de manera natural son esenciales en el proceso de
regeneración de los bosques, en particular, al permitir una mayor penetración de la luz en el
interior del dosel. Este estudio presenta una metodología para el mapeo de los claros
forestales en el Bosque Atlántico a partir de datos LIDAR. LIDAR teledetección tiene un gran
potencial para su aplicación en estudios forestales, ya que esta técnica produce un gran
conjunto de puntos georreferenciados en la superficie de la Tierra con información sobre la
altura, además, el láser de pulso tiene la capacidad de penetrar en la bóveda y crear puntos de
la tierra debajo de las árboles. Con información de la copa de los árboles y el suelo bajo la
copa, la estructura vertical y horizontal de los bosques pueden ser analizados. Las técnicas de
procesamiento digital de imágenes se pueden utilizar para analizar el gran volumen de datos
disponible a través del LIDAR. En este estudio hemos aplicado los procedimientos de filtrado
en el dominio espacial, la segmentación basada en región de cultivo y clasificación de las
imágenes generadas a partir de la interpolación de puntos obtenidos por LIDAR, con el
objetivo de mapeo de claros del bosque. Se utilizaron dos algoritmos para la clasificación de
imágenes, Maxima Verosimilitud y Bhattacharya. Para evaluar la exactitud de los mapas el
índice de Kappa fue generado. Los datos LIDAR utilizado en el estudio fueron obtenidas
utilizando un aire Laserscanner Leica ALS-50 con una frecuencia de 150 kHz.
P á g i n a | 119
Latin American Remote Sensing Week 2010
1. INTRODUCTION
The LIDAR technology can be used to obtain elevation data with speed and accuracy even in
difficult to access land. Equipment with LIDAR technology can be static or dynamic (Dalmolin,
2004). Dynamical systems are installed on airborne platforms, usually small planes or
helicopters, and for this reason cited in literature as Airborne Laserscanner (ALS) and it was
from this system that obtained the data for this research. ALS are composed of three main
components: (1) laserscanner, equipment that emits a pulse laser, (2) the Differential Global
Positioning System (dGPS) which is technology that allows the exact definition of the location
of the point generated by the laser pulse surface and (3) the Inertial Navigation System (INS)
equipment that determines the attitude of the aircraft during flight (Wehr and Lohr, 1999).
Basically, the principle of ALS is based on the emission of laser pulses toward the surface, each
return of these signs has its measured distance, thus providing information on surface height
and together with the positioning provided by dGPS it also has the coordinates X and Y. The
equipment with LIDAR technology can emit pulses at rates of over 100 000 pulses per second,
creating a cloud of points with information X, Y and Z. From this point cloud processing,
products such as Digital Terrain Model (DTM) and Digital Surface Model (DSM) are derived.
LIDAR instruments with small footprint with a laser diameter less than 1 meter has the ability
to penetrate dense forest canopies and create georeferenced points below these sites allowing
the construction of a DTM of the area, this in conjunction with the DSM enables forest studies
since data will be available from the top of the trees and the ground below it.
Knowing the location and size of forest gaps are required to understanding the dynamics of
forest systems, are factors that determine the amount of light received by the clearing, thus
identifying places prone to forest succession. The formation of gaps is of fundamental
importance in the process of germination and development of species (Alvarez-Buylla and
Martinez-Ramos, 1999), ecological succession (Whitmore, 1989), distribution and species
richness (Denslow, 1995). The aim of this study is to assess the ability of the LIDAR
technology to mapping forest gaps, testing the ability of this remote sensing technique to
assist in these types of forest studies.
2. METHODS
The study was conducted at the Ecological Station of Universidade Federal de Minas Gerais,
located on campus Pampulha, in Belo Horizonte, state capital of Minas Gerais, Brazil, taking
the seat with UTM coordinates 607,612 E and 7,802,154 N, Zone 23 K. The vegetation of
station consists of typical semideciduous forest, a subtype of Atlantic Forest. Was made a cut
of about 108 Km2 in an area called the butterfly forest (Figure 1), which is a more advanced
stage of regeneration.
P á g i n a | 120
Latin American Remote Sensing Week 2010
Data were acquired using an system Leica ALS-50, with up to 150 kHz frequency, assigned by
Enterprise Computing and Information Belo Horizonte (PRODABEL) for purposes of academic
research, with parameters described in (Table 1).
The cloud of points generated by the ALS system available for this study was filtered using
software Terrascan. A georeferenced point cloud created by the LIDAR technology as the DSM
and from this point cloud filtering program in Terrascan is possible to create the DTM. The
Terrascan has a tool (classify - routine - ground) that performs an automatic filtering the point
cloud generated by LIDAR. The principle that guides the operation of the tool is the slope
(slope of the ground surface in the horizontal plane), and considering the altimetric point
cloud measured by LIDAR, the tool (classify - routine - ground) filters the items by drawing a
tangent to the surface from a point P. The models created from the processing in Terrascan
are produced in vector format of points and arranged irregularly.
Digital image interpretation and analysis involves the manipulation and interpretation of
digital images with the aid of a computer (Lillesand and Kiefer, 2000). Several techniques of
digital image processing have been developed for analyzing the large number of remote
sensing images currently available for many platforms orbital and ground-based imaging of
land surface.
P á g i n a | 121
Latin American Remote Sensing Week 2010
The creation of digital models consists of convert models of vector to raster format and layout
of the points regularly. For this purpose the vector files were imported into the software ENVI
4.4, which has a module for processing and displaying LIDAR data. The initial procedure for
the generation of models is the interpolation of the height values contained in the DSM and
DTM for the purpose of creating a two-dimensional array or grid of regular shape in each grid
cell or pixel formed by a pair of coordinates (x, y) describe the height of the laser pulse at this
point. The interpolation method used was the nearest neighbor, this method was chosen for
its characteristic to ensure that the interpolated value is one of the original values, does not
generate new values. The resolution of the pixel grid in all models was 1 meter. In the
interpolation process were considered the highest elevations contained in the DSM plots
demarcated as this information corresponds to the top of the canopy, not including the laser
pulses that penetrated the canopy reaching the ground or the trees, and this model is a Digital
Crown Model (DCM). In this same logic, the same procedure was performed for the DTM, but
using the minimum elevation, minimizing the interference of small obstacles along the ground
in the modeling of the terrain. From the regular grids of DTM and DCM, another processing
step is to obtain a grid with the heights of trees in the plots studied, namely the establishment
Digital Tree Height Model (DTHM) obtained from the subtraction of the DCM - DTM. After
obtaining DTHM will be on this model that will provide applications and routines of the
algorithms for mapping the forest gaps.
Image enhancement involves techniques for increasing the visual distinctions between
features in a scene, with the objective to create new images from the original image data in
order to increase the amount of information that can be visually interpreted from the data
(Lillesand et al 2008). The Spatial Filtering is an operation that modifies each pixel based on
values of neighboring pixels. Low pass filters is a type of spatial filter developed to emphasize
low frequency features and deemphasize the high frequency components of an image. To
smooth MDAA was applied a gaussian low pass filter in a convolution with a mask of 3 X 3
(Figure 2).
2.4.3. Segmentation
P á g i n a | 122
Latin American Remote Sensing Week 2010
are considered similar, they are grouped. The area is the threshold value of minimum area for
a region to be individualized, is the minimum number of pixels required for a cluster region of
the image. The definition of similarity and area thresholds occurred systematically. Similarity
began with value 2 and made up an increment of 2 digital levels up to the value 8. Threshold
for the area started with the value 15, performing an increment of 5 pixels for the following
tests.
2.4.4. Classification
The first step in the process of mapping the forest gaps was the creation of the DCM, DTM and
DTHM (Figure 3). The gray scale of the models is distributed according to the variation in
height. DTHM in the gray scale allows viewing of the canopy, with the highest levels, the
lightest correspond to greater heights in the forest canopy and the darker areas represent
lower altitudes or absence of vegetation. The next step was the application of a smoothing
filter on the DTHM (Figure 4a), optimizing the segmentation process. The combination of
thresholds for the segmentation was 6 for similarity and 25 for area (Figure 4b).
Figure 4. (a) Image after applying the smoothing filter and (b) Threshold of similarity (6) and
area (25)
P á g i n a | 123
Latin American Remote Sensing Week 2010
(a) (b)
Kappa coefficient was calculated with the results obtained by the two classification methods.
The Kappa coefficient obtained by Maxlike was 0.92 and by Bhattacharya was 0.81,
considered excellent concordance rates. By analyzing the quantification of mapped classes, it
was found that the study area showed 15.13% of forest gaps by Maxlike classifier (Figure 5a)
and 20.12% by the Bhattacharya (Figure 5b). In the experiments with the classifiers, it is
noted that both tend to confuse the class areas canopy with sub-canopy and also show
confusion between the classes forest gap with sub-canopy. It is believed that this confusion is
due to the fact that the values of gray levels of pixels of these two classes are very close.
The Maxlike classifier identified as forest gaps the areas devoid of vegetation and the
Bhattacharya classifier considered as forest gaps the surrounding areas containing low
vegetation, thus identifying forest gaps bigger. This same logic can explain what happened
between the classes sub-canopy and canopy, where the algorithm Bhattacharya identified
areas with more high as sub-canopy, and then more areas were classified as being in this
class.
P á g i n a | 124
Latin American Remote Sensing Week 2010
(a) (b)
4. CONCLUSIONS
The use of LIDAR data combined with the techniques of digital image processing proved to be
good tools for mapping forest gaps. The 108 km2 mapped, it was found that between 15-20%
of the total area are formed by forest gaps. This information coupled with the location and
form of distribution of forest gaps become useful in research studies, preservation and
management of these ecosystems. The two classification methods had good results in the
mapping of gaps especially the classifier Maxlike.
BIBLIOGRAPHY
Denslow, J.S. 1995. Disturbance and diversity in tropical rain forests: the density effect.
Ecological Application 5:962-968.
Lillesand, T.M., Kiefer, R.W. 2000. Remote Sensing and Image Interpretation. 2a Ed. New York.
John Wiley& Sons.
SPRING: Integrating remote sensing and GIS by object-oriented data modelling. Camara G,
Souza RCM, Freitas UM, Garrido J . Computers & Graphics, 20: (3) 395-403, May-Jun 1996.
Wehr A., Lohr U. 1999.Airborne Laser Scanning – An Introduction and Overview. ISPRS
Journal of Photogrammetry & Remote Sensing, pp. 68-82.
P á g i n a | 125
Latin American Remote Sensing Week 2010
17.- LINKING REMOTE SENSING AND SOCIAL DATA FOR THE INVESTIGATION
ON DRIVING FACTORS OF LAND USE AND COVER CHANGE (LUCC) IN BRAZILIAN
SOUTH REGION
Summary: In this work we integrated multitemporal (2002-2008) land use and cover
classification of Landsat 5 – TM images with census social data and SRTM derived topographic
data. The objective of this data integration was the application of linear and spatial regression
models for the analysis of LUCC driving factors. The LUCC processes were defined as land
cover degradation, land cover regeneration, agricultural expansion, agricultural retraction,
forestry expansion and urbanization. The driving factors were represented by 17 social,
economical, infrastructural and topographical variables. The methodological procedure was
based on the following steps: a) multi-temporal classification of remote sensing data based on
object-oriented image analysis; b) data integration in 206 rural census tracts; c) analysis on
the spatial autocorrelation of LUCC processes; d) backward stepwise linear regression models
(OLS) for a exploratory data analysis; e) Geographically Weighted Regression (GWR) spatial
regression models to incorporate the issue of spatial autocorrelation and non-stationarity.
The OLS allowed the selection of variables to be used in the GWR models based on
significance level (α = 0.05) and presented low to medium adjusted R² (0.17 to 0.35) and AICc
(-264 to -500). The GWR presented a best fit compared with OLS (lower AICc values between -
541 and -347 and adjusted R² from 0.42 to 0.62). The LUCC processes driving factors analysis
revealed a strong interaction among biophysical and social variables highlighting social
(population density, dependence ratio, literacy rate and age), economical (income),
infrastructural (distance to roads) and topographical (slope and elevation amplitude) factors.
Abstract: En este trabajo hemos integrado la clasificación multitemporal (2002-2008) del uso
y de cubierta de la tierra de imágenes Landsat 5–TM con datos del censo social y de datos
topográficos derivados del SRTM. El objetivo de esta integración de datos fue la aplicación de
modelos de regresión lineal y espacial para el análisis de factores de conducción de cambios
de uso y cubierta de la tierra (LUCC). El análisis de los factores de conducción de procesos de
LUCC revelaron una fuerte interacción entre variables biofísicas y sociales destacando
factores sociales (densidad de población, relación de dependencia, tasa de alfabetización y
edad), económicos (ingreso), infraestructurales (distancia a los camiños) y topográficos
(amplitud de elevación y pendiente).
Keywords: Geographically Weighted Regression, Land Use and Cover Change, Remote
Sensing.
1. INTRODUCTION
The land use and cover change dynamics, known as LUCC (Land Use and Cover Change), is
currently one of the main factors of global environmental changes underway since the mid-
20th century (Turner II et al., 2007). LUCC studies are based on the integration of remote
sensing, biophysical and social data for the analysis on the driving factors of landscape
P á g i n a | 126
Latin American Remote Sensing Week 2010
processes (Geoghegan et al., 1998). In this work was carried out an analysis of variables
related with LUCC in an area of Uruguay River Basin (border of the States of Rio Grande do Sul
and Santa Catarina, Brazil).
The use of spatial regression models incorporating the issue of spatial dependence on LUCC
factors is recent and allows the analysis on the spatial autocorrelation of these processes
(Overmars et al., 2003; Lesschen et al., 2005; Soares-Filho et al., 2008), e.g., the Spatial Lag
Model (SAR) which has been adopted in different knowledge areas and incorporates the
spatial issue globally, based on global spatial autocorrelation methods (Moran's I) and spatial
homogeneity and stationarity assumptions (Anselin, 2002). In this study we used backward
stepwise linear regression models (OLS) for the first exploratory analysis of the variables, and
local spatial regression models (Geographically Weighted Regression - GWR) incorporating
the issue of spatial autocorrelation, heterogeneity and non-stationarity (Fotheringham et al.,
2002) on the driving factors of LUCC processes in the study area with the aggregation unit of
census tract.
2. METHODOLOGY
The study area is composed by 18 municipalities (the main are Lages and Campos Novos, in
Santa Catarina State) on the border of the States of Rio Grande do Sul and Santa Catarina, with
an approximated area of 12,000 km², representing the upper Uruguay River Basin (Alto
Uruguay) which exposes traditional agricultural land uses (soybeans, wheat, corn and others),
livestock farming and forestry.
The land use and cover mapping was based on Landsat-5 TM multitemporal data, scenes
221/79 (12/11/2002 and 03/10/2008), 221/80 (25/09/1999, 03/10/2008) and 222/79
(24/09/1999 and 29/10/2009). The cartographic base was comprised by topographic maps
in 1:100,000 scale, limits of municipalities (total of 17) and rural census tracts (total of 206)
of the 2000 year. Geomorphometric data of TOPODATA Project were used for the
characterization of physical variables (such as Digital Elevation Model and slope). The
softwares used were Definiens Developer (object-oriented images analysis), ArcGIS
(geoprocessing tools), Geoda (exploratory spatial analysis), SAM (GWR) and XLStat (statistical
analysis).
The definition of the dependent variables considered the following land use and cover classes
(FIGURE 1): forest and advanced secondary vegetation, native grasslands, forestry
(reforestation activities), mixed agriculture (small land use extensions with heterogeneous
patterns of pasture, crops and secondary vegetation), large-scale agriculture (extensive
cultures of soy, wheat and others), urban and water.
P á g i n a | 127
Latin American Remote Sensing Week 2010
FIGURE 1: Land use and cover maps of 2002 (left) and 2008 (right).
Initially, we defined the LUCC processes in the study area based in the land use and cover
conversions between 2002 and 2008 period used as dependent variables (areal proportion of
each census tract) in this work as:
P á g i n a | 128
Latin American Remote Sensing Week 2010
Education: mean of study years by person responsible for home (study_years) and
literacy rate (lit_rate).
Population: population density (dem_dens), dependence ratio (dep_ratio), percentage
of population lesser than 15 years (<15_years_pop), between 15 and 64 years
(15_64_pop), and with more than 64 years (pop+64)
Infrastructure: mean euclidian distance to roads (mean_dist) and standard deviation
of the Euclidian distance to roads (std_dist).
Geomorphometric: elevation amplitude (elev_amp), mean elevation (mean_elev),
mean slope (mean_slp), median slope (med_slp) and slope standard elevation
(slp_std).
With the definition of dependent and independent variables, it was possible to carry out an
exploratory spatial analysis for checking of spatial autocorrelation problems related to the
dependent variables by the method of global Moran's I. The linear regression models (OLS)
were used as an exploratory analysis tool, being generated by the backward stepwise
technique, where a model was created for each dependent variable, with variable selection
based on significance of F-test from the dataset starting with all original and square root
transformed independent variables. From the result of the previous selection, were applied
GWR models that performs individual regressions for each observation (census tract), with
the estimation of individual parameters. This allows us to evaluate how these regressive
parameters behave spatially. The technique used in the models generation was a Gaussian
adaptive kernel with an AIC optimization (minimum of 10% and maximum of 15% of
neighbors). (Rangel et al., 2006)
The parameters used in the diagnosis of the linear regression models were: a) the selected
variables, b) measures of model fitting (R²), c) significance of the model (F-statistic), d)
multicolinearity (Condition Number - CN), d) normal residual distribution (Jarque-Bera test),
e) heteroscedasticity or non-constant variance of the residuals (Koenker-Bassett, Breusch-
Pagan and White tests).
Based on the previous OLS model variables selection, we generated GWR models which
demonstrated efficacy in treating problems of spatial autocorrelation (measured by the
Moran's I tests of residuals), permanence of significance levels, higher coefficients of
determination (R²) and improving the fit of the models (measured by lower values of the
Akaike Information Criteria - AICc) (TABLE 1).
P á g i n a | 129
Latin American Remote Sensing Week 2010
TABLE 1: Results of OLS and GWR regression models, all models were significant (F-statistic
p-value <0.001).
LUCC OLS* GWR*
Variables
Process R² AICc Moran’s I R² AICc Moran’s I
pop+64, income,
Land Cover pop_dens, elev_amp,
0.20 -374.109 0.33 0.636 -441.891 0.01
Degradation median_slp, mean_slp,
slp_std
Land cover income, lit_rate,
0.33 -409.001 0.18 0.616 -470.895 -0.004
regeneration mean_elev, mean_slp
pop_dens, elev_amp,
Agricultural
mean_elev, 0.35 -264.256 0.28 0.673 -347.653 0.01
retraction
median_slp, mean_dist
Forestry pop+64, pop_dens,
0.18 -497.957 0.36 0.474 -541.358 0.15
expansion mean_elev
dep_ratio ,
Agricultural
avg_income, pop_dens, 0.27 -309.394 0.23 0.514 -327.165 0.18
expansion
median_slp, std_dist
Another result of the GWR models is the census tract mapping of the coefficients of
determination (local R²) and local regressive parameters, with a mask in the sectors where
the significance of the models (p-value) was below 90% (FIGURE 2). This mapping technique
allows a more refined analysis on the spatial regression patterns, with a more precise
identification of where the model was significant and of the positive or negative sense of the
estimated parameters (Mennis, 2006).
P á g i n a | 130
Latin American Remote Sensing Week 2010
FIGURE 2: Mapping of regressive parameters (A) and the local coefficient of determination
(R²).
4. CONCLUSION
The integrated use of linear regression (OLS) and local spatial regression (GWR) models
based in remote sensing and social data allowed an exploratory analysis of factors affecting
the processes of land use and cover change. The main potential of GWR models is related to
local generation of coefficients of determination and significance levels for each aggregation
unit, which allows a better interpretation of the relationships between variables and a better
definition of spatial regimes to be used in later LUCC modeling stages. This exploratory spatial
analysis revealed a large interaction between the physical, ecological and social LUCC driving
factors in all models, emphasizing social (population density, dependency ratio, literacy rate,
age and income), infrastructural (distance to roads) and topographic (slope and elevation
amplitude) variables.
P á g i n a | 131
Latin American Remote Sensing Week 2010
BIBLIOGRAPHY
Anselin, L. Under the hood: issues in the specification and interpretation of spatial regression
models. Agricultural Economics, n.27, 2002, p.247-267
Geoghegan, J. et al. “Socializing the pixel” and “pixelizing the social” in land-use and land-cover
change. In: Liverman, D. et al. People and pixels: linking Remote Sensing and Social Science.
Washington, National Academy Press, 1998, p.51-69.
Lesschen, J. P. et al. Statistical methods for analysing the spatial dimension of changes in land
use and farming systems. LUCC Report Series No. 7, Nairobi/Wageningen, The International
Livestock Research Institute/LUCC Focus 3 Office, 2005, 80p.
Overmars, K.P. et al. Spatial autocorrelation in multi-scale land use models. Ecological
Modelling, n.164, 2003, p.257–270.
Rangel, T. et al. Towards an integrated computational tool for spatial analysis in macroecology
and biogeography. Global Ecology and Biogeography, n.15, 2006, p.321-327.
Turner II, B. L. et al. The emergence of land change science for global environment change and
sustainability.. Proceedings of the National Academy of Sciences, v.104, n.52, 2007, p.20666-
20671.
Verburg, P. H. et al. Modeling the Spatial Dynamics of Regional Land Use: The CLUE-S Model.
Environmental Management, v. 30, n. 3, 2002, p. 391– 405.
P á g i n a | 132
Latin American Remote Sensing Week 2010
ABSTRACT:
The objective of this work was to evaluate the potential use of the Multiple Endmember
Spectral Mixture Analysis (MESMA) when applied to EO-1 Hyperion hyperspectral data to
discriminate land covers in the southern state of Rio Grande do Sul, Brazil. The methodology
involved: (a) pre-processing and atmospheric correction of Hyperion data; (b) sequential use
of the Minimum Noise Fraction (MNF), Pixel Purity Index (PPI) and n-Dimensional Visualizer
techniques in the 454-2334 nm range for the initial selection of a general group of
endmember candidates (first spectral library) and of another group of pixels to be used for
model validation; (c) use of the Visualization and Image Processing for Environmental
Research Tools (VIPER Tools) to perform the final selection of end members based on the
first spectral library and to obtain MESMA models; and (d) evaluation of resultant fraction
images and root mean square error (RMSE) values to determine the optimal number of
components of the MESMA model. Results showed that a four-endmember MESMA model
(soil = dunes and dry fields; green vegetation = pinus, eucalyptus and grasslands; water =
without sediments, with sediments, and with chlorophyll; and shade) adequately described
the diversity of the scene components, including
that of materials within the same class (e.g., pinus and eucalyptus) and produced the largest
fractions and the lowest RMSE values on a per-pixel basis. Results demonstrated the potential
use of the MESMA with EO-1 Hyperion hyperspectral data to discriminate land covers in the
coastal plains of Rio Grande do Sul, even considering the low signal-to-noise ratio of the
instrument, especially in the shortwave infrared range.
1. INTRODUCTION
Numerous techniques for the classification of orbital imageshave been used in the extraction
of land surface data (land covers and land use, for instance). Multispectral sensors such as the
Enhanced Thematic Mapper (ETM+, Landsat 7) with 8 spectral bands are generally used in
such studies. However, these sensors do not allow for a more detailed study of the spectral
behavior of the targets. On November 21st, 2000, the Hyperion sensor was launched aboard
the National Aeronautics and Space Administration’s (NASA) Earth Observing-1 satellite (EO-
1), becoming the first hyperspectral sensor to operate from the Earth’s orbit. With 242
spectral bands laid out between 0.4 and 2.5 μm, the sensor enables the extraction of a
practically seamless detailed reflectance spectrum for each scene element (pixel) in the image
(Goodenough et al., 2002; Galvão et al., 2005).
P á g i n a | 133
Latin American Remote Sensing Week 2010
On the other hand, the spectral response of a pixel is, in fact, the integrated sum of the
spectral response of the scene components as registered in the field of view of the sensor
(Instantaneous Field of View - IFOV) such as those of the soil, vegetation, water and of the
shade projected on the land, as well as atmospheric contributions Shimabukuro and Smith,
1991).
In order to identify the ratio of the different materials that comprise a pixel, one can use
models which decompose the pixel into its pure components or endmembers. The Spectral
Mixture Analysis (SMA) (Smith et al., 1985) is a digital imageprocessing tool that can separate
the relative ratio of each material within a pixel by using a set of pure components of the
image (Pereira et al., 1998). Nevertheless, the SMA fails in that it considers all pixels to be a
mixture of a single initial set of endmembers, thereby possibly allowing a pixel to be modeled
by endmembers which might not actually be present therein.
In order to correct such error, the Multiple Endmember SpectralMixture Analysis model
(MESMA) (Roberts et al. 1998a) was proposed as it allows the number and type of
endmembers, as well as their prevalence, to vary from pixel to pixel and establishes the best-
fit mixture model for each pixel individually.
In this context, the objective of this research was to assess the potential of the data deriving
from the Hyperion orbital sensor and from the MESMA model for the discrimination of land
cover classes in the Coastal Plains of Rio Grande do Sul.
P á g i n a | 134
Latin American Remote Sensing Week 2010
In order to identify the type of materials present in the scene and characterize land cover
classes and subclasses, two ground truth measurements took place in 2007. The stages of the
preprocessing of Hyperion data involved:
correction of anomalous pixels (stripes) by interpolation; exclusion of the bands which had
no data of the L1R product and which were positioned around the atmospheric absorption
bands (1,400 nm and 1,900 nm); onversion of data into surface reflectance values through
the use of a radiative transfer model
based on the MODTRAN-4; and image georeferral using a GeoCover image of the study area as
the reference.
The Minimum Noise Fraction (MNF), the Pixel Purity Index (PPI) and the n-Dimensional
viewer processing sequence was applied over the 143 bands selected from the 454 to 2334
nm interval in order to compose the first spectral library and generate a group of samples for
the validation of the results obtained by MESMA. For the selection of the most representative
spectra of the first spectral library, the following metrics were used: EAR
(Endmember Average RMSE), MASA (Minimum Average Spectral Angle) and CoB (Count
Based Endmember Selection), all included in the VIPER (Visualization and Image Processing
for Environmental Research) Tools program (Roberts et al., 2007). The spectra that had the
best performance in each metric were selected to generate the second spectral library. From
this second spectral library, the two, three and four endmember MESMA models were applied
to the Hyperion data using the “Run SMA/MESMA” module of VIPER Tools. Restriction
parameters required by the program regarding the type of shade (photometric or non-
photometric), minimum and
P á g i n a | 135
Latin American Remote Sensing Week 2010
maximum fractions of the non-shade component, maximum shade fraction, and maximum
RMSE allowed are all shown on Table 1. Shade Photometric Minimum Allowable Fraction -6%
Maximum Allowable Fraction 106%
Maximum Allowable Shade Fraction 80% Maximum Allowable RMSE 0,025 Residual
Threshold 0,025 Number of Contiguous Bands 7 Table 1. Parameters used in the “Run
SMA/MESMA” module of the VIPER Tools for the application of the MESMA models.
In order to generate the fraction image for each subclass represented in this study as well as a
classified image, a MatLab routine was developed, in which the subclass with the greatest
ratio was attributed to the pixel.
The accuracy assessment was obtained in relation to the “pure” pixel demixing (Validation
Sample) resulting from the PPI process. Thus, the fractions produced by each model for each
subclass were compared in relation to the RMSE produced so as to assess the performance of
the MESMA mixing models.
P á g i n a | 136
Latin American Remote Sensing Week 2010
P á g i n a | 137
Latin American Remote Sensing Week 2010
The number of “pure” pixels selected manually from the Hyperion image, with the aid of the
n-Dimensional viewer to compose the first spectral library and the validation sample is
shown on Table 2.
P á g i n a | 138
Latin American Remote Sensing Week 2010
Approximately 85% of the Hyperion image was modeled by the 4 endmember MESMA, which
corresponds to 195,725 pixels (a total of 230,400 pixels). Most pixels of the water, dry field,
pinus, grasslands and eucalyptus subclasses were properly modeled and classified. The dune
subclass only had 46.25 % of its sample modeled (Table 5).
The RMSE image indicated that the most significant mistakes were produced by MESMA
when modeling grasslands and dunes due to the high spectral variation of those targets. With
P á g i n a | 139
Latin American Remote Sensing Week 2010
the aid of histograms of the fraction images of each subclass, it was found that most of the
pixels of the Water Body Class were modeled with fractions superior to 90%. In the
Vegetation Class, the average fractions produced varied between 60% (grasslands) and 90%
(pinus and eucalyptus). The Soil Class presented the smallest fraction values, with
approximately 50% for the dry field subclass. The dune subclass had a practically linear
distribution, with intervals between 20% and 104%.
The final classification (Figure 4) obtained from the land vegetation and water fraction
images depicts the spatial
distribution of the eight subclasses presented in this study. It was observed that some areas
that actually corresponded to the cultivation of eucalyptus were erroneously classified as
pinus, yet the opposite did not occur. For the water classification, the result seems to be
consistent with the normal color image in Figure 2. Nevertheless, in shallow waters
(margins), there was a slight confusion between the subclasses Water with Sediments and
Water with Chlorophyll, mainly due to the background effect. Similarly to what happened in
the validation samples, dune-covered areas were mostly not modeled. The urban area, which
was not represented in the mixing models, was partially modeled as dry field (soil).
4. CONCLUSIONS
When applied over a Hyperion image and over the validation sample, the 4 endmember
MESMA model (soil = dunes and dry field; green vegetation = pinus, eucalyptus and
P á g i n a | 140
Latin American Remote Sensing Week 2010
grasslands; water = with sediments, without sediments and with chlorophyll; shade) properly
described the diversity of the scene components, including that of materials within the same
class (e.g.: pinus and eucalyptus).
The Fractions produced by the MESMA models when modeling “pure” pixels corresponded to
the expected values, mostly reaching fractions above 90% with low RMSE.
The results, as a whole, demonstrate the potential of the application of the MESMA model
with EO-1 Hyperion sensor hyperspectral data, even considering the low signal-to-noise
relationship of the instrument, especially in the SWIR.
5. REFERENCES
Galvão, L. S.; Formaggio, A. R.; Tisot, D. A. Discrimination of sugarcane varieties in
southeastern Brazil with EO-1 Hyperion data.
Remote Sensing of Environment, v. 65, n. 3, p. 267-279, 1998a. Roberts, D. A.; Halligan, K.;
Dennison, P. VIPER Tools User Manual. Versão 1.2. Santa Barbara: University of Califórnia
Santa Barbara, 2007 p. 95. Disponível em: <http://www.vipertools.org> accessed: 18 abr.
2007.
Shimabukuro, Y.E. & Smith, J.A. The least-squares mixing models to generate fraction images
derived from remote sensing multispectral data.
IEEE Transactions on Geoscience and Remote Sensing, v. 29, n.1, p. 16-20, 1991.
Smith, P. S.; Balonek, T. J.; Heckert, P. A.; Elston, R.; Schmidt,
G. D. UBVRI field comparison stars for selected active quasars
and BL Lacertae objects. Astronomical Journal, v. 90, p.
1184-1187, 1985.
P á g i n a | 141
Latin American Remote Sensing Week 2010
Peña M.A.*
marco.pena@umayor.cl
Cruz P.C.*
pablo.cruzj@umayor.cl
Castro B.P.*
benjamin.castro@umayor.cl
Abstract: In this work we explore the capacities that airborne hyperspectral imaging offers to
estimate water's turbidity and depth, and to map riparian vegetation species in the Negro
river, Spain. Ground-based data of both state variables, and representative locations of the
main tree vegetation species were collected at the same time of the remotely-sensed data
acquisition. Correlations were carried out between both data sets in order to find the
normalized difference index (NDI) that best relates each state variable. A supervised
hyperspectral classifier, trained with field data, was used to map the distribution of the tree
species. Results show that water's depth changes are best tracked by a NDI constructed with
the near infrared (NIR) band at 705.5 nm and a green visible band at 531.1 nm (r = -0.69).
Meanwhile, water's turbidity changes are best tracked by a NDI constructed with the near
P á g i n a | 142
Latin American Remote Sensing Week 2010
infrared bands at 905.34 nm, and at 760 or 774.54 nm (r = 0.82). Both results were found to
be consistent with the theoretical background. The visual inspection of the output image
rendered by the classification algorithm suggests that hyperspectral imaging is encouraging
for the riparian vegetation species mapping. An improvement of the methodology employed
in this work is expected in the future, in order to provide a solid validation that supports the
use of this technology in other water bodies of the country.
Keywords: hyperspectral imaging, spectral indices, riparian vegetation, water turbidity, water
depth
1. Introduction
Optical hyperspectral remote sensing is a scientific discipline that retrieves information from
an element by detecting and analyzing its radiated energy throughout hundreds of spectral
samplings (from now bands) continuously located in the visible and infrared dimensions of
the electromagnetic spectrum (i.e., optical spectral range). In specific spectral wavelengths of
that range most of the biotic and abiotic elements of the Earth surface produce absorption and
reflection features that allow their physical and chemical characterization. Nevertheless, only
remote sensors with hyperspectral sensitivity can properly measure the wide range of those
signals (Lucas et al., 2004; Borengasser et al., 2008; Goetz, 2009).
Optical remote sensing has largely assisted the study of water resources, mainly by estimating
and predicting water quality parameters such as turbidity, dissolved organic matter and algae
concentration (Govender et al., 1999; Ritchie et al., 2003). In the optical spectral range
sediments and microorganisms contained in a water body produce distinctive signals that can
be isolated by the design of spectral indices, or arithmetic combinations of bands sensitive to
the presence of a biogeochemical compound. The strength of a spectral index relies on its
simple formulation (usually it combines a band sensitive to the compound variations and
another one insensitive to it, that serves as baseline) and its capacity of minimize factors that
may interferes in the measured compound (e.g., solar illumination and sensor viewing
geometry, scattered path radiance, signals from other elements contained within the pixel)
(Treitz and Howarth, 1999; Asner et al., 2003; Liang, 2004; Blackburn, 2007).
supervised classification procedures were applied on the images in an attempt to map the
vegetation species of interest.
2. Methods
Image pre-processing: Original digital numbers (DNs) recorded on the image’s pixels were
automatically converted into absolute radiances by using the scanner’s radiometric-
exportation software. Though ideally at-sensor radiances should be converted into apparent
surface reflectances (i.e., the ratio of irradiance to radiance, both free of atmospheric effects),
the image’s atmospheric correction cannot be performed because during the period of this
study the software considered for this purpose (ATCOR©, Atmospheric Correction) was being
tested. Nonetheless, calibrated radiances were considered suitable for the achievement of the
study goals (RSI, 2009; Liang, 2004).
P á g i n a | 144
Latin American Remote Sensing Week 2010
radiance values of their four nearest uncorrected pixels. Nonetheless, this radiometric
smoothness was considered as positive, since contributed to minimize spatial mismatches
that subsequently could arise between field sampled plots and their corresponding image
pixels, as a result of geographic coordinate inaccuracies on each data set. The geographic
coordinates system assigned to the output images was UTM (Universe Transverse of
Mercator), datum WGS84 (World Geodetic System 1984), zone 29, North. Finally, an images'
mosaic was constructed, which subsequently was trimmed according to a distance of 250 m
from the river shoreline with the sense of only submitting to later processing the interest
area.
Field data collection: Field campaign was carried out at the same time of the flight, with the
sense of minimizing the negative effect that any temporal variability in the compounds of the
measured variables may have in the later statistical analysis accomplished between ground-
based data (physical units of the variable) and remotely-sensed data (pixel radiances). The
first task consisted in the identification of tree species within three spatial windows that were
defined at regular intervals along the studied river's transect. The species identified were:
Chopo blanco/Chopo negro (Populos alba/Populus nigra), Aliso (Alnus glutinosa), Encino
(Quercus ilex), Roble (Quercus pyrenaica), and Pino (Pinus pinaster). All these species were
marked as plots on a high spatial resolution cartography of the study area by field personnel.
Afterward, inside each window water turbidity (nephelometric turbidity units, NTU) and
water depth (m) were measured on 23 plots each one, in order to statistically relate these
data with their corresponding pixel radiances. Plots were selected according to an ad-hoc
sampling method and the measurements were accomplished by using a zodiac boat,
preferably moved by oars in order to not disturb the sampled waters. To minimize the
confusion that water depth may introduce in the water turbidity analysis, all the turbidity
plots were sampled on areas with a similar depth. By the same token, water depth was
sampled on areas with similar turbidity. Finally, with the purpose of minimizing spatial
coregister inconsistencies between field sampled plots and their corresponding image pixels,
all the measurements were carried out by considering a plot's support region of about 2 m of
diameter. Geographic coordinates of the field data were collected by a geodetic GPS (Global
Positioning System), set to a positional accuracy below 0.5 m.
Supervised classification: Field sampled plots of the tree species were superimposed on the
image (i.e., trimmed images mosaic) to define training areas corresponding to each species.
Additionally, training areas corresponding to other elements such as water, bare soils,
impervious surfaces, rooftops and grasslands were also defined. The statistical separability
between the spectral curves of all pairs of the interest elements (from now classes) was
determined through the Divergence Transformed measure. In all cases, a high separability (≅
2,000, the maximum possible divergence value) was retrieved. For the mapping purpose, the
Spectral Angle Mapper (SAM) was used. This is a hyperspectral classification algorithm that
assigns each image pixel to a given class based on their spectral similarities. Specifically. the
algorithm calculates the angular distance (in radians) between the spectral curve of each pixel
and a given class, assigning the pixel to the class which more resembles. Though the
P á g i n a | 145
Latin American Remote Sensing Week 2010
algorithm can operate with a threshold from which a pixel will not be classified, we decided to
force the classification of all pixels, mainly because of our uncertainty in the spectral
variability that each class may comprise. Ideally, an image classification process should ends
with an accuracy assessment based on a contingency table (i.e., the comparison of each class
with their corresponding ground truth data). Nonetheless, this procedure was beyond the
study scope. So, we recognize that no quantitative approach was attempted for such an
evaluation, but just a simple visual inspection based on the knowledge of the study area.
Empirical spectral indices retrieval: Field sampled plots of water turbidity and depth were
superimposed on the image for extracting their corresponding pixel radiance values.
Afterward, each parameter was correlated to the radiances resulting from a normalized
difference index (NDI).
Where Lλ1 and Lλ2 are the absolute radiances of two image bands. We first found the band that
best correlates their corresponding field data, and that was the Lλ1 formula term. Then, each
one of the remaining bands were entered in the formula as the Lλ2 term, and new correlations
were retrieved. By doing this step, we obtained the NDI that best tracks a given parameter.
Pearson correlations were used for water depth field data. Since the distribution of water
turbidity field data was not normal, Spearman correlations were used.
grassland class, which evidences the high cover and health status of that physiognomic type.
In general, spectral curves of the sampled tree species exhibit their larger separation in the
NIR region, that usually has been referred to as the spectral region that most contributes to
the species discrimination, because of their sensitivity to the vegetation canopy structure
attributes, which can reach notorious differences among species (Asner, 1998; Liang, 2004;
McCoy, 2005). Figure 1b shows the typical spectral behaviour of clear water, which is
characterized by a very low reflection in the whole visible region and an almost complete
absorption in the NIR region (Lillesand et al., 2004; McCoy, 2005). Meanwhile, bare soil
classes present a regular and relatively high reflection on those spectral regions. Spectral
curves from vegetation as well as from the remaining elements show strong absorption
features between 760 and 940 nm of the spectrum, which are due to oxygen and water vapour
content in the atmosphere, respectively (Schläpfer et al., 1998, 2005; Teillet et al., 1998;
Richter, 2010). These spectral signals are specially notorious when at-sensor radiances are
used, since the scattered path radiance has not been compensated and spectral irradiance has
not been considered.
Mapping of riparian vegetation species: Figure 2 shows a representative spatial window of the
vegetation species classified across the riparian zone of the Negro river. During the training
area stage, no attempt was made to identify shadowed features, since they could comprise a
wide spectral variability (according to their intensity and the land cover type that they cover)
that may confuse the algorithm's class assignment, thus promoting the image
misclassification. In other words, it was the algorithm who assigned shadowed features to the
class that more resembles. As a result, most of them were classified as bare soils. Due to their
spectral similarity, some shallow waters were confounded with bare soils. Regarding
vegetation classification performance, Aliso class tends to be identified across the river
shoreline, though in some cases it is confounded by Roble and Encina classes. Chopo and
Roble canopies exhibit quite distinguishable visual appearances, so it was easy to check if they
were correctly classified. A visual inspection suggests that both classes were classified with
high accuracy. There are many cases in which a single tree canopy shows more than one class.
This evidences the spectral complexity that vegetation species may achieve, and highlights the
need of a more representative and accurate selection of the training areas, in such a way that
these misclassifications be minimized. At this point, it is important bearing in mind that the
spectral behaviour of vegetation results from a combination of phenological, physiological,
physiognomic and site-specific factors. According to this, a given vegetation species class
might be divided into several subclasses, even within a relatively small and apparently
homogenous study area.
P á g i n a | 147
Latin American Remote Sensing Week 2010
Figure 1. Spectral curves of vegetation species (A) and other elements (B) of interest extracted
from the hyperspectral data.
Figure 2. Vegetation species mapping for a spatial window of the study area.
Water's depth and turbidity indices: Figure 3a allows us to observe the bands significantly
correlated to water depth. The highest correlations were found for NIR bands with central
wavelengths at 705.5 nm (band 82), 709.14 nm (band 83), 701.87 (band 81), 712.77 nm
(band 84) and 720.03 nm (band 86) (r = -0.69; -0.69; -0.68; -0.67; -0.67, respectively). It is
likely that these spectral wavelengths are sensitive to changes in the amount of suspended
sediment contained throughout the water column, which indirectly are related to water depth
(if water depth increases then suspended sediments should also increase). According to figure
3b the NDI that best correlates water depth (r = -0.75) results from the arithmetic
combination of band 82 (as the first term of the formula) and band 34 (with central
wavelength at 531.1 nm). In general, a large number of bands entered as the second term of
the NDI formula renders NDIs with good correlation coefficients (r ≥ -0.5), but only a few blue
P á g i n a | 148
Latin American Remote Sensing Week 2010
visible bands render NDIs with correlation coefficients higher than the band 82 by itself (r ≥ -
0.69). The improvement that these bands introduce in the NDI can be explained by the well-
known capacity of the blue spectral region to penetrate a water body, which has been
successfully used in bathymetry remote sensing systems (Aronoff, 2005; McCoy, 2005).
Figure 3. Correlations for hyperspectral bands and water's depth (A), and for NDI and water's
turbidity (B).
Figure 4a shows that NIR bands with central wavelengths at 905.34 nm (band 137) and
894.44 nm (band 134) present the highest correlations with water turbidity (r = 0.77 and
0.76, respectively), which is due to the sensitivity of that spectral region to changes in the
amount of suspended sediments (Bhargava and Mariam, 1991; Doxaran et al., 2002a, 2002b).
Figure 4b shows that the NDI that best correlates water turbidity (r = 0.82) results from the
arithmetic combination of band 137 (as the first term of the formula) and bands with central
wavelength at 760 nm or at 774.54 nm (bands 97 and 101, respectively). It should be noted
that the NDI also renders high correlation coefficients (r ≥ 0.75) with many other blue visible
bands used as the second term of the formula. The improvement that these bands introduce in
the NDI performance should be related to the capacity of that spectral region to penetrate the
water body and therefore, to their relative insensitive to changes in water turbidity, thus
acting as baseline of the index. Nonetheless, it was an arithmetic combination of bands located
on the NIR region which rendered the highest NDI correlation coefficient. This is because of
the insensitive that band 97 presents to turbidity changes, whose absorption feature is caused
by atmospheric oxygen. The above-mentioned stands out the sensitivity of a spectral index to
atmosphere-induced effects. So, in cases where it wants to be temporally or spatially
extrapolated, equivalent radiometry units of the images should be employed, or even more
important, both images should be previously subject to an atmospheric correction procedure
(in order to retrieve surface reflectance units).
P á g i n a | 149
Latin American Remote Sensing Week 2010
Figure 4. Correlations for hyperspectral bands and water's turbidity (A), and for NDI and
water's turbidity (B).
4. Conclusions
This work explored the capacities that optical hyperspectral remote sensing offers for
estimating water state variables and for mapping riparian vegetation species in the Negro
River, Spain. The results support the potential of this technology to accomplish aims as the
pursued in this work. Empirical statistical analysis show that it is possible to design spectral
indices closely related to water's turbidity and depth changes. Meanwhile, an encouraging
discrimination of vegetation species may be derived from this type of images if appropriately
representative spectral curves may be extracted from the targets of interest.
From the experience gained in this work, some methodological enhancements are
recommended for similar future applications of this technology. First, it would be desirable to
minimize the atmosphere-induced effects on the image data. Even when scattered path
radiance should be low for images acquired near the ground (mainly because of to the low
optical thickness of the atmosphere), the use of atmospherically corrected radiometric units
(i.e., surface reflectances) would provide more reliability and validity to the results obtained.
Second, in spite of the expensive that it can be to increase the number of field sampled plots, it
would be recommendable to procure the highest number possible in order to minimize
potential overfitting problems in the subsequent statistical analysis. Third, a classification's
accuracy assessment should be accomplished at least over sensible or representative areas, in
order to provide a quantitative measure of the degree of success achieved by the supervised
classification procedure.
References
Aronoff S. 2005. Remote Sensing for GIS Managers. ESRI Press. New York, USA. 487 p.
Asner G.P. 1998. Biophysical and biochemical sources of variability in canopy reflectance.
Remote Sensing of Environment, 64: 234–253.
P á g i n a | 150
Latin American Remote Sensing Week 2010
Asner G., Hicke J., Lobell D. 2003. Per-pixel analysis of forest structure: Vegetation indices,
spectral mixture analysis and canopy reflectance modeling. In Wulder M.A., Franklin S.E. eds.
Remote Sensing of Forest Environments: Concepts and Case Studies. Kluwer Academic
Publishers. Dordrecht, Holland. p. 209-254.
Bhargava D.S., Mariam D.W. 1991. Light penetration depth, turbidity and reflectance related
relationships and models. ISPRS Journal of Photogrammetry and Remote Sensing, 46: 217-230.
Blackburn G.A. 2007. Hyperspectral remote sensing of plant pigments. Journal of Experimental
Botany, 58(4): 855-867.
Borengasser M., Hungate W.S., Watkins R. 2008. Hyperspectral Remote Sensing, Principles and
Applications. CRC Press. Florida, USA. 119 p.
Doxaran D., Froidefond J.M., Lavender S., Castaing P. 2002a. Spectral signature of highly turbid
water. Application with SPOT data quantify suspended particulate matter concentrations.
Remote Sensing of Environment, 81: 149-161.
Doxaran D., Froidefond J.M., Lavender S., Castaing P. 2002b. A reflectance band ratio used to
estimate suspended matter concentrations in sediment-dominated coastal waters.
International Journal of Remote Sensing, 23(23): 5079-5085.
Goetz A.F.H. 2009. Three decades of hyperspectral remote sensing of the Earth: A personal
view. Remote Sensing of Environment, 113: S5-S16.
Govender M., Chetty K., Bulcock H. 2007. A review of hyperspectral remote sensing and its
application in vegetation and water resource studies. Water SA, 33(2): 145-152.
Liang S. 2004. Quantitative Remote Sensing of Land Surfaces. Wiley. New Jersey, USA. 560 p.
Lillesand T.M., Kiefer R.W., Chipman J.W. 2004. Remote Sensing and Image Interpretation.
Wiley. New York, USA. 763 p.
Lucas R., Rowlands A., Niemann O., Merton R. 2004. Hyperspectral sensors and applications. In
Varshney P.K., Arora M.K. eds. Advanced Image Processing Techniques for Remotely Sensed
Hyperspectral Data. Berlin, Germany. Springer. p. 11-49.
McCoy R.M. 2005. Field Methods in Remote Sensing. The Guilford Press. New York, USA. 159 p.
Peñuelas J., Fillela I., Biel C., Serrano L., Savé R. 1993. The reflectance at the 950-970 nm
region as an indicator of plant water status. International Journal of Remote Sensing, 14(10):
1887-1905.
Peñuelas J., Peñol J., Ogaya R., Fillela C. 1997. Estimation of plant water concentration by the
reflectance Water Index WI (R900/R970). International Journal of Remote Sensing, 18(13):
2869-2875.
P á g i n a | 151
Latin American Remote Sensing Week 2010
Ritchie J.C., Zimba P.V., Everitt J. H. 2003. Remote sensing techniques to assess water quality.
Photogrammetric Engineering and Remote Sensing, 69(6): 695-704.
RSI (Research Systems Incorporation). 2009. ENVI (Environment for Visualizing Images)
Online Help. ITT Visual Information Solutions. Boulder, USA. (unpaginated).
Schläpfer D., Borel C.C., Keller J., Itten K.I. 1998. Atmospheric pre-corrected differential
absortion techniques to retrieve columnar water vapour. Remote Sensing of Environment,
65(3): 353-366.
Schläpfer D., McCubbin I.B., Kindel B., Kaiser J.W., Ben-Dor E. 2005. Wildfire smoke analysis
using the 760 nm oxygen absorption feature. In proceedings of the 4th EARSeL Workshop on
Imaging Spectroscopy. April 27 – 29. Warsaw University. Warsaw, Poland. 10 p.
Teillet P.M., Staenz K., Williams D.J. 1997. Effects of spectral, spatial, and radiometric
characteristics on remote sensing vegetation indices of forested regions. Remote Sensing of
Environment, 61: 139-149.
Treitz P.M., Howarth P.J. 1999. Hyperspectral remote sensing for estimating biophysical
parameters of forest ecosystems. Progress in Physical Geography, 23(3): 359–390.
P á g i n a | 152
Latin American Remote Sensing Week 2010
ABSTRACT:
In this paper we present the displacement of the TIGO geodetic station, located in
Concepci´on, Chile, after the major 2010 earthquake. The displacement has been obtained by
using the Satellite Laser Ranging (SLR) technique, which allows to get the distance between
the station and the satellite by measuring the round-trip time of high power laser pulse with
picosecond precision clocks. By processing the data of the global SLR network, it is possible to
determine the local geographic coordinate and the relative displacement among stations over
the time, achieving millimetric accuracy. A total post-seismic displacement of 3.19 m mainly,
was measured by using the SLR data obtained from the TIGO-SLR station. The motion
detected by SLR confirms the results which have been obtained by the colocated GPS/Glonass
site at TIGO.
1 INTRODUCTION
On February 27, 2010 at 3.34 local time, Concepci´on was struck by one of the strongest
earthquakes ever recorded, reaching a
magnitude of 8.8 in the Richter seismic scale. This earthquake has awaken considerable
scientific interest given its strength, and
the fact that during the last decade the area has been frequently monitored by several state of
the art scientifics instruments, so
there is an extraordinary opportunity to follow up the post-earthquake evolution of crustal
motion.
The Transportable Integrated Geodetic Observatory (TIGO), located in Concepci´on, Chile, is a
fundamental station with a number of independent measurements systems: Satellite Laser
Ranging (SLR), Very Long Baseline Interferometry (VLBI), Global Positioning System (GPS)
and gravimetry. TIGO contributes significatively to the volume and accuracy of SLR data in
the world.
Moreover, its privileged location in a highly seismic zone, turns Figure 1: The SLR station at
TIGO performing a satellite tracking with the blue laser beam through its Coud´e Telescope.
The optical system, electronics and control room are inside the container it into a suitable
station for the study of tectonic movements occurring there.
P á g i n a | 153
Latin American Remote Sensing Week 2010
Since the first pulsed laser satellite tracking carried out by a NASA team in 1964 (Plotkin et
al., 1965), the development of the SLR
technique has improved its accuracy achieving millimeter resolution in satellite orbit
measurements, with respect to a reference point (Degnan, 1993). The SLR technique is based
on the pulseecho principle where a high power pulsed laser is directed to satellites that carry
retroreflectors, reflecting back the pulses to the SLR station, where photons are detected with
Single Photon Avalanche Detectors (SPAD), and the round-trip time is measured with
picosecond accuracy instruments. Then, given the speed of light and an atmospheric
propagation model (Mendes and Pavlis, 2004), it is possible to determine the distance to the
satellite.
Worldwide processing centers provide new satellite orbit predictions on a daily basis by each
station. Thus, the SLR technique allows to directly measure the error between the prediction
and the actual satellite orbit. Henceforth, the use of SLR also allows carrying out scientific
studies of tectonic-plate motion, general relativity principles, gravitational fields,
oceanography, topography, lunar physics and fundamental physics (Smith et al., 1990;
Pearlman and Bosworth, 2002).
SLR data is collected, merged, analyzed, archived and distributed by the International Laser
Ranging Service (ILRS). These data sets are used by the ILRS to generate a number of
fundamental data products such as: centimeter accuracy satellite ephemerides, earth
orientation parameters (EOP), three-dimensional coordinates and velocities of the ILRS
tracking stations, for instance (Pearlman and Bosworth, 2002).
By using the TIGO SLR data, we have been able to measure the crustal displacements
produced by the earthquake with high accuracy, and the results are reported herein.
P á g i n a | 154
Latin American Remote Sensing Week 2010
_ The Optical Bench (fig. 3) consists of a pumping Nd:YAG pulsed laser at 532 nm (diode
pumped) called JEDI manufactured
by THALES Laser; the oscilator and the regenerative amplifier operating at 847 nm from
High-Q Laser; a linear amplifier implemented at TIGO with a Ti:Sapphire crystal pumped by
the JEDI laser; a fast photodiode which triggers the start signal; a security shutter and a
Second Harmonic Generator (SHG) crystal which generates the blue laser beam at 423.5 nm.
_ The Transmitter/Receiver Unit (not shown here), is made up of opto-mechanical
components used to focus the returned laser beam to the SPAD detectors, a CCD camera and
an eyepiece to see through the telescope, two small lasers and a spectrometer for avelength
measurement.
_ The Coud´e Telescope (fig. 1) consists of a main lens of 50 cm of diameter, two secondary
mirrors, a triple lens array to achieve beam collimation in the two wavelengths, and four
prisms that fold the optical path into the axes of the Coud´e mount. This telescope operates in
dual mode, i.e. transmission and reception of laser pulses pass through the same optical
channel.
Several factors affect the measured apparent distance to the satellite d so it is given by:
P á g i n a | 155
Latin American Remote Sensing Week 2010
South, 3.13 m to the West and 0.04 m down in Altitude, as seen on the plots. The horizontal
direction of this displacement corresponds to an azimuth angle of 259_ w.r.t. North, which is
contrary to the preseismic velocity vector, plus an additional component of 25_ westward.
The displacement measured by SLR confirms the results which have been obtained by the
independent but colocated GPS/Glonass and VLBI techniques at the Geodetic Observatory
TIGO, as is shown in figure 6 where the co- and post-seismic isplacement SLR vector is
compared to the GPS “CONZ” station at the observatory.
Compared to the microwave based techniques, the strength of the optical SLR is its ability to
determine the altitude component of the displacement with higher accuracy.
The seismic activity is still present in the southern central part of Chile, so the site
displacement will continue in function of the tectonic movements in the area. The ongoing
monitoring by TIGO-SLR contributes to investigations of the pre- and postseismic behaviours
of crustal motion in a subduction zone, stressing the importance to continue monitoring of
the area by a fundamental geodetic station.
P á g i n a | 157
Latin American Remote Sensing Week 2010
ACKNOWLEDGEMENTS
The authors appreciate the tremendous effort made by Alejandro Fernandez, Victor Mora and
Cesar Guaitiao from the TIGO SLR team, who quickly recovered the station after the
devastating earthquake. We thank to the International Laser Ranging Service ILRS for
upporting the SLR data processing and provide the geocentric coordinate for the tracking
stations. We also thank Esteban Vera, ebasti´an Godoy and Sergio Sobarzo for suggestions
and proofreading this paper.
References
Degnan, J., 1993. Millimeter Accuracy Satellite Laser Ranging: A Review. AGU Geodynamics
Series 25, pp. 133–162.
Mendes, V. and Pavlis, E., 2004. High-accuracy zenith delay prediction at optical wavelengths.
Geophysical Research Letters 31, pp. L14602.
Pearlman, M.R., D. J. and Bosworth, J., 2002. The International Laser Ranging Service.
Advances in Space Research 30(2), pp. 135–143.
Plotkin, H. H., Johnson, T. S., Spadin, P. L. and Moye, M., 1965. Reflection of ruby laser
radiation from Explorer XXII. Proceedings IEEE 53(3), pp. 301–302.
Smith, D. E., Kolenkiewicz, R., Dunn, P. J., Robbins, J. W., Torrence, M. H., Klosko, S. M.,
Williamson, R. G., Pavlis, E. C.,
Douglas, N. B., and Fricke, S. K., 1990. Tectonic Motion From LAGEOS. Journal of Geophysical
Research 95(B13), pp. 22013–22041.
P á g i n a | 158
Latin American Remote Sensing Week 2010
m-luza@upla.cl
roberto.richardson@gmail.com
flavia.landeros@gmail.com
of the aerial photograph from Hycon IGM flight (1954-1955), to 1:70.000 scale, was made, in
order to obtain a DEM and compare it to a newer Aster satellite Digital Elevation Model. As a
result, differences were found in regard to sediment accumulations in floods bed positioned
under the “Fiscal” bridge that is located upstream in Ovalle city. 1976 (MSS Landsat 2), 1986-
1996 (TM Landsat 5), and 2002 (ETM Landsat 7) Landsat satellite images were used to
support our research. A false-color combination 754 (near-infrared band, visible bands, red
and green) was applied to 1976 image, and a combination of 453 was applied to the other
images, using both ENVI and PCI Geomatics software, in order to stress changes in bed
pattern and riverbed through time. Cross sections were produced by using DEMs to
understand the differences between sediment loads, giving notable results. Riverbank
stability was associated to the presence or absence of riparian vegetation, by applying a false-
color band combination of 456 to 1976 images, and a combination 543 to the others. A true
color band combination was produced, and processed using the cluster module of Idrisi
Kilimanjaro software. The classification result evidenced urban expansion through time.
Key words: Teledetection, urban expansion, fluvial dynamics, runoff pattern, riparian
vegetation.
1. INTRODUCCIÓN
2. METODOLOGÍA
Los resultados obtenidos han sido verificados en terreno, no obstante la generación de cluster
para visualizar la expansión de la urbe e identificación de vegetación riparia, no han sido
capturados punto GPS específicos, debido a la gran extensión de estos. No obstante, la
P á g i n a | 160
Latin American Remote Sensing Week 2010
Se utilizaron las curvas de nivel para corregir geométricamente, a Datum PSAD 1956, el
scanner (600 dpi) de la fotografía aérea del vuelo Hycon IGM año 1954 -1955 a escala
1:70.000, en el programa Arc Gis 9.2 obteniendo un error de 1.5 con 110 puntos de control.
Además se extrae DEM Aster el cual corresponde al año 2000, en software ENVI, para
posterior comparación a través de perfiles longitudinales (tomado desde confluencia río
Grande y Hurtado hasta bajo Puente la Chimba) y transversales (trazados de terraza 3 a
terraza 4 descritas por Paskoff, 1993), respecto a DEM año 1955.
Para realizar NDVI (Normalized Difference Vegetation Index), se efectuó en primera instancia
combinaciones de falso color (RGB) 543 en el programa Geomática PCI, para las imágenes TM
de los años 1986 y 2007; y para imagen MSS año 1975, se realizó una combinación de falso
color (RGB) 456, que son las bandas con que cuenta el Sensor.
Luego de esto, en el programa ERDAS, se lleva a cabo el procesamiento de valores que van de
0-1, los cuales son resultantes de la mezcla algorítmica automática reconocida por el software,
según sensor. De los cuales se consideran los valores máximos y mínimos cercanos al cauce,
en los distintos años, debido a que se buscó identificar la vegetación riparia, para interpretar
estabilidad del cauce, en torno a esta.
3. RESULTADOS
P á g i n a | 161
Latin American Remote Sensing Week 2010
de servicios para la población del amplio territorio de la cuenca del río Limarí, este se funda
adyacente a la caja del río, con un trazado ortogonal, que definió manzanas, el cual se fue
extendiendo en dirección Sur, poniendo presión sobre ribera derecha del cauce. Este aumento
es constatado, a partir de los siguientes cluster, los cuales corresponden a cuatro décadas.
Figura 1. Expansión Urbana, año 1976. Figura 2. Expansión Urbana, año 1986.
Figura 3. Expansión Urbana, año 1996. Figura 4. Expansión Urbana, año 2009.
1
2
P á g i n a | 162
Latin American Remote Sensing Week 2010
Figura 5. Fotografía Aérea, año 1955. Figura 6. Imagen Landsat MSS, año 1976.
3
4
Figura 7. Imagen Landsat TM, año 1986. Figura 8. Imagen Landsat TM, año 1996.
5
6
Figura 9. Ortofoto, año 2002. Figura 10. Imagen Landsat TM, año 2009.
P á g i n a | 163
Latin American Remote Sensing Week 2010
1 1
2 2
Figura 11. DEM 1955 Figura 12. DEM Aster
P á g i n a | 164
Latin American Remote Sensing Week 2010
Es a partir de la rectificación del río hacia la ribera izquierda (Véase Figura 20), que este
índice se encuentra mayormente reflejado en las imágenes, puesto que el cauce, mantiene su
caudal en la actual dirección, sin variaciones.
La estabilidad del cauce, en el tramo del meandro cercano a puente Fiscal, que se mantiene
hasta hoy, no correspondería a la “firmeza” que le da la vegetación, puesto que índice NDVI en
esta zona, arroja los valores más bajos, sino que lo constituiría un depósito elongado de
material sedimentario adosado a la ribera derecha, que tampoco ante un evento extremo,
resultaría de soporte.
Figura 18. “NDVI, Landsat MSS, 1975”. Figura 19. “NDVI, Landsat TM
1986”.
P á g i n a | 165
Latin American Remote Sensing Week 2010
4. CONCLUSIONES
El manejo del flujo para almacenar agua ha causado inadvertidamente una degradación
considerable de la biodiversidad asociada a ecosistemas riparios (planicies de inundación),
aguas arriba y abajo, a lo largo del tramo sinuoso estudiado.
5. REFERENCIAS
Arche, A., 1992. Sedimentología. Vol 1. España, Edit. Consejo Superior de Investigaciones
Científicas.
Chuvieco, E., 1995. Fundamentos de Teledetección Espacial. (2. ed.). España: Edit: Rialp, S.A.
P á g i n a | 166
Latin American Remote Sensing Week 2010
Leopold, L., Gordon M., Millar J., 1964. Fluvial Processes in Geomorphology. San Francisco: W.
H. Freeman and Company.
Naiman, R.J.; Lonzarich, D.G.; Beechie, T.J. y Ralph, S.C. (1992) General principles of
classification and the assessment of conservation potential in rivers. In: Boon, P.J.; Carlow, P. y
Petts, G.E. (Eds.) River Conservation and Management, John Wiley & Sons,Chichester, 93-123.
Paskoff, R., 1993. Geomorfología de Chile Semiárido. (Le Chile semi aride: recherches
géomorphologiques). Trad. Español E. Novoa, La Serena, Chile: Editorial: Universidad de la
Serena.
AGRADECIMIENTOS
Se agradece, el aporte de los distintos organismos públicos que nos entregaron las
herramientas técnicas y estadísticas, que sin las cuales hubiese sido imposible la realización
de esta investigación, nos referimos a la Centro de información de Recursos Naturales
(CIREN) e Instituto Geográfico Militar (IGM), por el material visual referido a la zona de
estudio y a la Dirección General de Aguas (DGA), quien nos entregó el material estadístico
referido a los datos de precipitación y caudales de la localidad de Ovalle.
P á g i n a | 167
Latin American Remote Sensing Week 2010
fpino@uchilefau.cl
Pamela Smith
pamelasmit@gmail.com
Abstract: From digitally processed satellite images, an analysis of land cover and land use a
modernized agricultural valley in Central Chile. We seek to determine the evolution of land
use in the next ten years, based on current behavior and past land use, for which the method
is complemented by the incorporation of stochastic models of type Markov chains of first
order and then proceed to map the spatial distribution projected by cellular automata. The
information generated will estimate future labor demands in terms of the main crops and use
this background to help in the preparation of local development plans in the area.
Palabras Claves: Desarrollo local, uso y cobertura del suelo, cadenas de markov, detección
de cambios.
1. Introducción
P á g i n a | 168
Latin American Remote Sensing Week 2010
2. Área de Estudio
Con una superficie de 952,5 Km2 y una población estimada al año 2008 de 27.751
habitantes, la comuna de Casablanca representa una zona importante desde el punto de vista
productivo en el contexto de la Región de Valparaíso. Actualmente es considerada como una
de las comunas con mayor vocación vitivinícola en la Provincia, alcanzando una gran
dinámica en la última década, debido a la incidencia directa de variables endógenas como la
dotación de los recursos suelo, agua y clima, así como por variables externas de demanda
constante y precios internacionales que garantizan niveles de rentabilidad superiores a los de
la agricultura tradicional, provocando con esto un incremento significativo en su superficie
con viñedos para vinificación, que junto con modernizar el espacio rural ha estimulado la
valorización de la tierra, favoreciendo dicha reconversión (Pino, 2008).
4. Objetivos
5. Método y Materiales
P á g i n a | 169
Latin American Remote Sensing Week 2010
Universidad de Maryland. Las fechas de las imágenes se extiende entre enero febrero y marzo
respectivamente. Dichas imágenes fueron preprocesadas (correciones radiométricas) con la
finalidad de facilitar la comparación entre las tres fechas y luego procesadas con el software
IDRISI Kilimanjaro, los modelos cartográficos derivados de las imágenes se terminaron con
ArcGis 9.3.
Se contó además con la cartografía regular 1: 50.000 del IGM en formato digital, la que fue
utilizada para realizar la corrección geométrica de cada una de las imágenes, todas las bandas
fueron georeferenciadas en el programa Arcgis 9.3, usándose para dicho efecto un polinomio
de primer orden. El sistema de proyección seleccionado corresponde a la UTM (Universal
Tranversal de Mercator) y el elipsoide de referencia elegido para todas las imágenes y los
modelos finales es el WGS 84.
P á g i n a | 170
Latin American Remote Sensing Week 2010
proceso Markov se lleva a efecto mediante operaciones entre matrices, pudiendo hacerse de
forma automatizada, con el software apropiado obteniéndose de este modo una matriz de
probabilidades de transición a partir de la matriz de conteo de Scherer (1972), generada de la
tabulación cruzada de los datos de dos fechas que tienen las mismas categorías de uso en la
clasificación. A partir de la matriz de probabilidades de transición o de tabulación cruzada
proporcional, se genera la matriz de probabilidades condicionales que permite hacer el
pronóstico de los cambios en función de las probabilidades de cambio.
De esta forma, cada una de las celdas de la diagonal principal de la matriz representa la
superficie (en hectáreas) de cada clase de cobertura vegetal y uso del suelo que permaneció
en la misma categoría en el período de tiempo considerado, mientras que el resto de las
celdas estiman la superficie de una determinada cobertura o tipo de uso de suelo que pasó a
otra categoría (Aaviksoo, 1995; Bocco et al., 2001). Por otra parte, los autómatas celulares
(AC), proporcionan una poderosa herramienta para la modelización dinámica de los cambios
de uso del suelo, consisten en un método común para tomar en cuenta las interacciones
espaciales. Se trata de sistemas dinámicos discretos cuyos elementos tienen una interacción
constante entre sí, tanto en el espacio como en el tiempo y tienen por tanto la capacidad de
representar comportamientos complejos a partir de una dinámica sencilla. De acuerdo con
Kumar (2003), los autómatas celulares son modelos dinámicos que integran inherentemente
una dimensión espacial y una temporal, reciben su nombre debido a que se componen de
células, como las celdas de un tablero de ajedrez, y los estados de cada célula pueden
evolucionar en función de unas reglas de transición simple. La fase descrita culmina con la
aplicación del modulo de análisis multicriterio que facilita la aplicación de las reglas, modulo
que también se encuentra disponible en el software IDRISI.
6. Resultados y Discusión
P á g i n a | 171
Latin American Remote Sensing Week 2010
Tal como se mencionó anteriormente, los fondos de valle se caracterizan por concentrar la
actividad silvoagropecuaria del área de estudio, destacando por una parte el cambio de uso
desde actividades típicas de una agricultura tradicional en los años 80, hacia unos usos en los
que predomina el viñedo orientado a la producción de vino para exportación en el presente y
que en algunos sectores se presenta casi como un monocultivo (Foto 2).
A partir de la clasificación se definieron desde las imágenes las siguientes clases de uso del
suelo:
Realizado el análisis y clasificación de las imágenes se construye la tabla 1, que muestra los
cambios más significativos entre los diferentes años y también el porcentaje de cambios
entre el año inicial del período (1989) y el año final (2008).
Las coberturas han presentado cambios muy significativos durante el período, destacando las
disminuciones de superficie que afectan a la agricultura tradicional, lo que se explica por un
proceso de reconversión hacia cultivos de exportación, como las viñas en el caso del valle, el
cambio va desde un 0.9 % de la superficie productiva (35.000 hás. aprox.) comunal dedicada
a viñas, hasta un 23.9 % en el año 2008. También es interesante observar como la superficie
destinada a la creación de tranques destinados a almacenar agua para riego (0.6 %) aumenta
al doble (1.3 %), lo que se explica por el aumento de la superficie plantada con viñas y por
ende, un aumento de las demandas de agua particularmente durante el período estival.
Una de las dificultades que debe enfrentar el valle para hacer sostenible el enorme
crecimiento de la actividad vitivinícola, es la disponibilidad del recurso agua, situación que se
ha visto agravada con la sequía de los últimos años. De acuerdo con la tabla 1, el incremento
de superficie con viñas para todo el período supera el 2000 %, lo cual explica el notable
aumento de las demandas hídricas, provenientes fundamentalmente de la captación del agua
subsuperficial a partir de pozos, situación que durante los últimos años se ha traducido en
una disminución del nivel freático, generándose un nivel de incertidumbre respecto del riego,
que ni siquiera el riego por goteo, modalidad de predominante, ha sido capaz de mitigar. En
este contexto la construcción de tranques de almacenamiento de agua proveniente de la
precipitaciones viene a subsanar en parte las demandas estivales del recurso agua.
P á g i n a | 173
Latin American Remote Sensing Week 2010
Con la finalidad de validar la aplicación del modelo de Markov se realizó en una primera
etapa una simulación para el año 2008 (tabla 2) a partir de los años 1989 y 1998 y luego
después de observar una buena coincidencia con los datos obtenidos de la clasificación,
destacando un tendencia a sobrestimar algunas superficies, se procedió a realizar la
simulación para el año 2020, obteniéndose la matriz de probabilidades de cambio que se
muestra en la tabla 3.
Tabla 2.
Matriz Probabilidades de Cambio (1989 – 1998) para 2008
Usos Agr. Tradic. Asent.Pob. Bosq. Plant. Veg.Natural Tranque Viñas
En esta matriz destaca por una parte la probabilidad hacia una estabilidad o menor velocidad
de cambio estimada por el modelo para las categorías asentamientos poblados (p = 0.96),
vegetación natural (p = 0.74), tranque (p = 0,80) y viñas (p = 0.93). Por otra parte se refleja
extraordinariamente bien el cambio desde uso agrícola a viñas (p = 0.23). La estimación
markoviana para el año 2020 (tabla 3) es quizás más impactante, destacando la probabilidad
de cambio desde agricultura tradicional hacia viñas (p = 0.52) y desde bosque plantado a
vegetación natural (p = 0.51) lo que podría explicarse por regeneración del matorral
esclerófilo, en zonas cuya cubierta de bosque plantado es explotada y no se aplica plan de
reforestación.
Probablemente lo más interesante es que los aportes al incremento de las viñas en los
próximos 10 años, deberían venir por un lado, desde un aumento de la reconversión de la
agricultura tradicional (p = 0.52) y desde la vegetación natural (p = 0.09), la que está siendo
reemplazada en laderas de solana por viñedos, proceso que se ha incrementado
notablemente durante los últimos tres años en los sectores de la vinilla y otros cercanos.
P á g i n a | 174
Latin American Remote Sensing Week 2010
Tabla 3.
Matriz Probabilidades de Cambio (1989 – 2008) para 2020
Usos Agr. Tradic. Asent.Pob. Bosq. Plant. Veg.Natural Tranque Viñas
Agric. Tradic 7898,8 22,5 8481,4 24,1 3611,3 10,3 2585 7,4 -67,27
Asent. Poblad. 245,5 0,7 478,3 1,4 771,26 2,2 1081 3,1 340,32
Bosque plant. 1758,3 5,0 1529,5 4,4 1657,9 4,7 1527 4,3 -13,15
Cober.natu. 24708,1 70,3 20367,5 57,9 20249 57,6 17791 50,6 -27,99
Tranque 209,8 0,6 216,9 0,6 449,2 1,3 536,1 1,5 155,53
Viñas 329,2 0,9 4076,0 11,6 8411,4 23,9 11630 33,1 3432
Totales 35150 100 35149,7 100 35150 100 35150 100 1989 -
2020
P á g i n a | 175
Latin American Remote Sensing Week 2010
mapas, el aumento de la superficie con viñas entre 1989 y 2008 es notable, importantes
superficies en los fondos de valle han pasado de agrícola a viñedos.
Mapa 1.
Mapa 2.
7. Conclusiones
De acuerdo a la información proporcionada por los mapas elaborados desde las imágenes
satelitales clasificadas y considerando la proyección de los cambios a partir de la aplicación
de las cadenas de markov y autómatas celulares, es posible advertir una disminución
considerable de las tierras dedicadas a la agricultura tradicional, la que se da con mayor
intensidad hacia el norte de la ruta que conecta Santiago con Valparaíso.
Una de las coberturas que muestra para todo el período 1989 – 1998 – 2008 y luego en el
pronóstico hacia 2020 una tendencia a la disminución de superficie es la vegetación natural,
cobertura que de acuerdo a los antecedentes de terreno ha sido reemplazada por
plantaciones de viñas en laderas de solana e inclusive de umbría, lo que en la práctica
representa una expansión de la frontera agrícola en el valle, incorporándose al sistema
productivo tierras con fuertes pendientes, sin vocación agrícola, pero con grandes
expectativas para la producción de vides. De acuerdo con lo anterior, y si se relaciona este
aumento de la superficie plantada con vides y los requerimientos de mano de obra de esta
actividad, entonces debería esperarse incrementos en las demandas de mano de obra, en el
mediano plazo.
P á g i n a | 177
Latin American Remote Sensing Week 2010
Si consideramos algunas referencias según las cuales los requerimientos de mano de obra en
el cultivo de la vid pueden estar cerca de 548.5 horas/hombre/Hás/año (Perlbach, Calderón y
Ríos, 2005) y por otra parte, otro estudio (Neiman, 2007) que nos presenta una estimación de
demanda de trabajo en un establecimiento típico dedicado a la producción de uvas de alta
calidad enológica según el cual, una hectárea requiere 317 horas/año de trabajo (equivalente
a 40 jornales/año de 8 horas cada uno) para completar todas las tareas requeridas por el
ciclo de cultivo, es posible hacer algunas estimaciones más bien optimistas, pero con
precaución debido a las debilidades del modelo, en cuanto al impacto sobre las demandas de
mano de obra que puede significar el aumento de la superficie con vides en los próximos 10
años.
8. Bibliografía
Aaviksoo, K.1995. Simulating vegetation dynamics and land use in a mire landscape using a
Markov Model, Landscape and Urban Planning 31 129 – 142.
Hang, X. y Andersson, C. 2004. Assessing the impact of temporal dynamics on land-use change
modeling Computers, Environment and Urban Systems, 28 (2004) 107–124
Jensen, J.R., 1996. Introductory Digital Image Processing, a Remote Sensing Perspective,
Prentice Hall Series in Geographic Information Science, Keith C. Clarke, Series Editor, Second
Edition.
Neiman, G., 2007. Aspectos Sociales del Desarrollo Empresarial en la agricultura intensiva de
Argentina. Primer Seminario de Cooperación y Desarrollo en Espacios Rurales
Iberoamericanos. Sostenibilidad e Indicadores.
Kumar, 2003. Modelling Land Use cover changes using cellular Automata in a Geo Spatial.
Thesis submitted to the International Institute for Geo-information Science and Earth
Observation in partial fulfilment of the requirements for the degree of Master of Science in
Geoinformatics.
Pino, F., 2008. Evaluación del cambio de uso del suelo en el Valle de Casablanca, Anales de la
sociedad Chilena de Ciencias Geográficas, ISSN0717 – 3946, Santiago, Chile.
Pino, F., 2007. Use Geographical Information Systems (GIS) in the evaluation dynamics and
changes in the rural landscape Central Chile, Publicado en Proceedings International
Cartographic Conference, Moscú, Rusia. Pags. 159 – 169.
P á g i n a | 178
Latin American Remote Sensing Week 2010
http://upload.wikimedia.org/wikipedia/commons/6/6f/Santiagocontaminado.jpg
Enrique Blondel
Especialista en Teledetección
Ottawa Ontario Canadá
P á g i n a | 179
Latin American Remote Sensing Week 2010
Introducción
Intensas y prolongadas olas de calor se registraron en las ciudades de Paris, Francia en Agosto
2003 y Moscu, Rusia en Julio del 2010 registrándose 14.802 y 700 fallecimientos
respectivamente.
Figura 1. http://www.washingtontimes.com/news/2010/aug/9/amid-smog-moscow-
deaths-double-700-day/
La generación de Islas Térmicas en medios Urbanos es un problema crónico que tiene origen
primordialmente en la implementación y practica de regulaciones urbanísticas no cónsonas
con la realidad que han introducido los recientes patrones de interacción demográfica y el
cambio climático en las ciudades.
Así el problema tiene raíces en el plan regulador del Uso del Suelo, por lo tanto modificar,
adaptar y hacer aplicar de forma constructiva e innovadora las actuales regulaciones
normativas Urbanas con la ayuda de recursos como la Teledetección, los Sistemas de
Información Geográfica y otras disciplinas, potenciarían el incremento en la calidad de vida de
los habitantes de las ciudades, sobre todo las capas de población mas vulnerables.
El presente trabajo investigativo propone una orientación práctica sobre el uso de las Bandas
Térmicas de las Imágenes Satelitales Aster y Landsat 7 en la caracterización espacio temporal
del fenómeno de generación de Islas Térmicas a través de la estimación de Temperaturas de
Superficie.
P á g i n a | 181
Latin American Remote Sensing Week 2010
P á g i n a | 182
Latin American Remote Sensing Week 2010
Son Isotermas que indican áreas de calor relativo por encima de la capa superficial en
regiones urbanas y que están directamente asociadas a áreas perturbadas por intervención
humana a causa del proceso continuo de urbanización.
Las Islas Térmicas por si mismas pueden contribuir nocivamente en deteriorar la calidad del
aire y el agua potable por el incremento de los índices de contaminación, así como en
aumentar las emisiones de gas de efecto invernadero, además de conllevar a un exceso del
consumo energético y que, en consecuencia de todo lo anterior, pueden ocasionar graves
lesiones respiratorias y cutáneas en los seres vivos y hasta la muerte.
La magnitud en la generación de las Islas Térmicas puede ser amplificada en eventos de Olas
de Calor intensas y prolongadas. La disposición actual de los patrones urbanísticos igualmente
evita que las corrientes de aire fluyan libremente dentro de las ciudades impidiendo el
proceso de Convección Térmica o el flujo normal de transferencia de calor.
Evitar, Minimizar y en último caso Mitigar las causas que dan origen a este fenómeno es
responsabilidad de todas las instancias decisorias y reguladoras de la Planificación
Urbanística en las ciudades, en conjunción con sus habitantes.
Las bandas utilizadas para el procesamiento son la 6 (Low Gain) en Landsat 7 y la banda 13 en
Aster, ambas tienen en común similares rangos en sus longitudes de onda. El algoritmo
monocanal, usado en la estimación de temperaturas a través de la banda térmica infrarroja 13
del sensor Aster, tiene como concepto el inverso de la ecuación de la ley de Planck y el
concepto de efectividad de longitud de onda. (A Single-Channel Algorithm for Land-Surface
Temperature Retrieval from ASTER Data. Juan C. Jiménez-Muñoz y José A. Sobrino)
P á g i n a | 184
Latin American Remote Sensing Week 2010
seleccionado por presentar este día una temperatura ambiental máxima de 260C a las 11:00
AM horal local (www.wunderground.com), representando condiciones ideales para el
presente estudio.
P á g i n a | 185
Latin American Remote Sensing Week 2010
Sin embargo Aster sobreestima ligeramente las temperaturas con respecto a Landsat 7 sobre
todo en sitios desprovistos de vegetación como se evidencia al verificar con respecto al raster
obtenido en el procesamiento de Índice de Vegetación de Diferencia Normalizada (NDVI).
Landsat 7 estima las temperaturas de forma más precisa por su mejor resolución espacial de
60 metros con respecto a Aster (90 metros), también ofrece una gran exactitud y resolución
radiométrica. Es importante mencionar que debido a una falla mecánica que presenta el
corrector de escaneamiento lineal desde el 31/05/2003, y cuya función de diseño es la de
compensación por movimiento de avance, las imágenes presentan un patrón de zig-zag, lo que
conlleva a que presenten zonas que han sido doblemente captadas mientras que otras no lo
han sido durante el proceso de adquisición, conllevando a perdida de información.
P á g i n a | 186
Latin American Remote Sensing Week 2010
Se puede observar en la tabla anterior y en los gráficos que se muestran a continuación, que
los valores estimados de temperatura calculados a partir de la imagen Landsat 7 se
correlacionan mejor con respecto a los valores de NDVI en relación a su par de Aster.
P á g i n a | 187
Latin American Remote Sensing Week 2010
P á g i n a | 188
Latin American Remote Sensing Week 2010
44oC 42.2 ºC
o 18oC
33 C
39oC
Quinta Normal
Zonas cercanas a Ave. Libertador O’Higgins con Gral. Velásquez presentan aumentos de
temperatura ambiental de hasta 39 ºC debido a la aparición de Islas Térmicas adyacentes a
edificaciones con techos de bajo albedo que exhibe en este caso una temperatura superficial
de 44 ºC. Caso contrario a las adyacencias de la Quinta Normal y el Centro de Eventos
MoviStar Arena, en donde las temperaturas superficiales alcanzan valores de la temperatura
ambiente promedio de la hora de adquisición de la imagen o ligeramente superiores. Véase
P á g i n a | 189
Latin American Remote Sensing Week 2010
también como estas Temperaturas Superficiales se hacen ligeramente más frías en las
adyacencias de los Cerros San Cristóbal y Santa Lucia.
El centro de eventos MoviStar Arena cuya cubierta de techo registró una estimación de
temperatura superficial de 18 ºC, valor que contrasta notablemente con el registrado en la
cubierta de techo de un conocido establecimiento comercial situado en la esquina de la
Avenida Irarrázabal con Suecia en el sector Ñuñoa, de 42,2 ºC, demuestra como el uso de
materiales alternativos en edificaciones y en redes viales, puede lograr reducir o evitar la
formación de Islas Térmicas en sus entornos inmediatos.
Temperaturas Superficiales clasificadas por Zona de Uso del Suelo en la Región Metropolitana
de Santiago de Chile
P á g i n a | 191
Latin American Remote Sensing Week 2010
60 500,000
450,000
50
400,000
350,000
40
300,000
LST Landsat 7 (ºC)
Población
30 250,000
200,000
20
150,000
100,000
10
50,000
0 0
SAN JOSE DE MAIPO
CALERA DE TANGO
LAMPA
VITACURA
LA GRANJA
LA PINTANA
CERRO NAVIA
SAN RAMON
QUINTA NORMAL
SANTIAGO
PADRE HURTADO
SAN BERNARDO
LA REINA
PROVIDENCIA
LA CISTERNA
LA FLORIDA
HUECHURABA
LO BARNECHEA
MACUL
MAIPU
SAN MIGUEL
SAN JOAQUIN
ESTACION CENTRAL
LO PRADO
PUENTE ALTO
CONCHALI
NUNOA
COLINA
RENCA
RECOLETA
LAS CONDES
QUILICURA
INDEPENDENCIA
PAC
PENALOLEN
PUDAHUEL
LO ESPEJO
PIRQUE
EL BOSQUE
CERRILLOS
Comunas
Conclusiones
Es interesante destacar hoy en día como en la práctica y debido al Cambio Climático, el uso
extendido de ciertos tipos de tecnologías diseñadas para mejorar la disposición y las
características estructurales, estéticas y, sobre todo, funcionales de cada uno de los
P á g i n a | 193
Latin American Remote Sensing Week 2010
componentes de las edificaciones modernas; así como de las infraestructuras que conectan la
intricada red urbanística de las ciudades; teniendo, como base, la rentabilidad energética y el
incremento en la calidad del aire y del agua potable, pueden tener un impacto favorable sobre
el medio ambiente que las rodea.
Una nueva iniciativa ecológica introducida en los medios urbanos como la incorporación de
revestimientos de vegetación en las cubiertas de techos de las edificaciones así como la de
implementar el uso alternativo de nuevos materiales con el fin de incrementar el grado de
albedo en las superficies expuestas a la radiación solar tanto de edificaciones como de las
redes de infraestructura urbana, pueden influir rigurosamente en la reducción de las Islas
Térmicas, si su empleo se intensifica por medio de la imposición de regulaciones y ordenanzas
urbanísticas firmes y consensuadas.
Los resultados obtenidos en esta investigación indican que diferentes zonificaciones de uso de
suelo así como la de los espacios verdes y las edificaciones que la conforman, pueden influir
adversamente o favorablemente en la formación de micro-climas en el denso patrón
urbanístico de la ciudad.
El uso de las tecnologías Geo-espaciales y más aun la Teledetección puede ayudar, no solo a
identificar y caracterizar el problema en estudio, sino también a monitorear el grado de
evolución de la calidad del aire y el agua debido a la formación de Islas Térmicas, en función
de la implementación de las medidas pertinentes que mitiguen de forma efectiva este tipo de
desajuste climático.
Referencias.
IEEE GEOSCIENCE AND REMOTE SENSING LETTERS, VOL. 7, NO. 1, JANUARY 2010
A Single-Channel Algorithm for Land-Surface Temperature Retrieval From ASTER Data
Juan C. Jiménez-Muñoz and José A. Sobrino
P á g i n a | 194
Latin American Remote Sensing Week 2010
Agradecimientos
GeoEye Foundation
21700 Atlantic Blvd. Dulles, VA 20166
NASA Land Processes Distributed Active Archive Center User Services USGS Earth Resources
Observation and Science (EROS) Center
47914 252nd Street
Sioux Falls, SD 57198-0001
P á g i n a | 195
Latin American Remote Sensing Week 2010
Los resultados obtenidos indican que en la cuenca del Río Papaloapan existe una superficie de
479,258 ha distribuidas en 20 humedales, destacando por su importancia el humedal Laguna
de Alvarado (379,900 ha equivalentes al 79% de la superficie evaluada). La
comparación entre coberturas mostró cambios negativos en la clases hidrófila, cuerpos de
agua y forestal. Lo que constituye una seria amenaza a la sustentabilidad de los humedales.
Para revertir el deterioro de estos ecosistemas, se proponen acciones de recuperación forestal
en zonas estratégicas de la cuenca.
Abstract: México, due to its physical and biological conditions arising from its geographical
location, has a large area covered by wetlands, which are considered as a strategic reserve of
water under the responsibility of the National Water Commission.
The main objective of this study was to inventory and characterize wetlands in the
Papaloapan River Basin into two levels of analyses: the first one implemented at 1:250,000
scale at river basin rank and the second one at 1:40,000 scale, at wetland category. The first
level involved the mapping of land use and vegetation based on satellite images taken at 1979
and 2003 to obtain the latest features and detect changes. This was complemented by map
analyses evaluation of indicators to define the wetlands specific areas. The second level
included the development of detailed maps of land use and vegetation with 2006 images, for
the identified wetlands.
P á g i n a | 196
Latin American Remote Sensing Week 2010
The results indicated that in the Papaloapan River Basin there is a surface of 479.258 ha
distributed in 20 wetlands. The Laguna de Alvarado wetland highlights due to its area size
(379.900 ha is equivalent to 79% of the assessed area). The comparison showed negative
changes in the hydrophilic, water bodies, and forest coverage. This condition constitutes a
serious threat to the sustainability of the wetlands. To reverse the decline of these
ecosystems, we propose actions of forest recovery in strategic areas of the basin. One
conclusion is that the proposal of projects and programs for the conservation and
sustainability of the wetlands will be more effective if they are conceived at the basin scale.
1. INTRODUCCIÓN
2. OBJETIVOS
P á g i n a | 197
Latin American Remote Sensing Week 2010
3. METODOLOGÍA y RESULTADOS
Figura 1. Mapas de distribución de uso de suelo y vegetación de los años 1979 y 2003.
El mapa de uso del suelo y vegetación de la cuenca del río Papaloapan se presenta en escala
1:250,000. Con un área mínima cartografiable de 1 ha, como se menciona en el párrafo
anterior se hace una análisis retrospectivo con base en imágenes tomadas en 1979, un análisis
actual con imágenes tomadas en el año 2003 y un análisis prospectivo para el año 2033,
P á g i n a | 198
Latin American Remote Sensing Week 2010
empleando para ello el método Markov alimentado con los datos obtenidos con las imágenes
de satélite. El mapa prospectivo de uso de suelo y vegetación, apoyará la integración de
escenarios futuros de planeación para la conservación y uso sustentable de los humedales. La
figura 2 presenta el diagrama metodológico aplicado para la generación del mapa prospectivo.
P á g i n a | 199
Latin American Remote Sensing Week 2010
P á g i n a | 200
Latin American Remote Sensing Week 2010
Figura 5. Diagrama metodológico empleado para delimitar las áreas de recuperación forestal
Respecto a las necesidades de recuperación forestal (RF) por uso de suelo, se identificó que la
mayor superficie con necesidades de atención corresponde al bosque fragmentado (52.7%),
siguiéndole el matorral fragmentado (13.1%), selva baja fragmentada (11.3%), selva media
fragmentada (10.3%) y la agricultura de temporal (5.8%), siguiéndole el resto de usos en
proporciones bajas. Esta distribución de necesidades puede verse gráficamente en la Figura 6.
Figura 6. Necesidades de recuperación forestal por uso de suelo en la cuenca del Papaloapan.
P á g i n a | 201
Latin American Remote Sensing Week 2010
4. CONCLUSIONES
5. BIBLIOGRAFÍA
Comisión Nacional del Agua. 2005. Estadísticas del Agua en México 2005. 104p. México, D.F.
Comisión Nacional del Agua. 2001. Programa Hidráulico de Gran Visión, Región X Golfo
Centro.
CONAFOR, Mayo 2006. Gerencia Estatal XII Península de Yucatán. Quintana Roo, Informe de
Gestión 2000 – 2006.
IMTA, 2004. Formulación del Programa Regional Hidrológico Forestal para La Región XI
Frontera Sur, 324p. Jiutepec, Mor.
INEGI. 1996. Modelos digitales del terreno escala 1:250000, Aguascalientes, Ags.
INEGI. 1999. Mapas topográficos digitales escalas 1:50000 y 1;250000. Aguascalientes, Ags.
INEGI. 1999. Modelos digitales del terreno escala 1:50000, Aguascalientes, Ags.
P á g i n a | 202
Latin American Remote Sensing Week 2010
Juan Jose Martinez Benjamin, Marina Martinez Garcia, Miquel Angel Ortiz Castellon, Jose
Martín Davila, Jorge Garate Pasquin, Begoña Perez Gomez , Pascal Bonnefond
PALABRAS CLAVE: calibración altimétrica, boyas GPS, mareógrafos, nivel del mar, gradiente
del geoide
ABSTRACT: Three Begur Cape experiences on radar altimeter calibration and marine geoid
mapping made on 1999, 2000 and 2002 are overviewed. One campaign has also been made in
June 2003 at the Ibiza island area. Direct absolute calibration estimating the Topex Alt-B bias
was performed during the satellite overflight by using GPS buoys. The advantage of that
method is that neither geoid modelling nor tidal error is needed. Other main objective was to
map the profile of the Mean Sea Surface (mss) along the closest T/P and Jason-1 groundtrack.
Mapping the marine surface for indirect altimeter calibration has de advantage of allowing the
calibration of any radar sensor that crosses the studied area but, in turn, the disadvantage is
P á g i n a | 203
Latin American Remote Sensing Week 2010
that the method requires ocean tide and geoid knowledge, which reduces the accuracy of the
bias estimate by a factor of 2.
A technical Spanish contribution to the calibration experience has been the design of GPS
buoys and GPS catamaran taking in account the University of Colorado at Boulder and
Senetosa/Capraia. For the mapping of the extended calibration areas centered on satellite
ground tracks, the catamaran was tracked by the Patrol Deva, from the Spanish Navy. An
additional absolute altimeter direct calibration was performed on June 14. Complementary
data came from five GPS reference stations deployed at Ibiza , San Antonio and Portinatx, and
from vertically-referenced tide gauges located at Ibiza and San Antonio. We present first
results on Jason-1 altimeter calibration using the marine geoid derived from data collected
during the campaign. Moreover, the geodetic activities (e.g., GPS, leveling) has permitted to
build a very accurate (few mm) local network linked to the european one, with a reference
frame compatible with the satellite altimetry missions (ITRF2000).
KEY WORDS: altimeter calibration, GPS Buoys, tidegauges, sea level, geoid gradient
1. INTRODUCCION
1.1 Altimetria radar
El radar altímetro mide la distancia relativa entre su antena y la superficie marina instantánea
a lo largo de la traza del satélite. Ello permite establecer medidas de la superficie topográfica
marina con precisiones de pocos centímetros, medidas de la altura significativa de las olas con
precisiones de 1 cm o sub-centrimétricas y medidas de la velocidad del viento en superficie
con precisiones de unos 2-3 m/s. Todo ello con una visión global y en un corto intervalo de
tiempo (repetibilidad de 10 días, aprox). Todas estas medidas han supuesto una contribución
excepcional a la Observación de la Tierra y un reto para la tecnología espacial y el desarrollo
de sensores. En la actualidad algunos de los satélites altimétricos en funcionamiento son:
La investigación en altimetria por satélite en zonas costeras aún está en sus pasos iniciales.
Esto es debido a las limitaciones intrínsecas de la tecnologia y las dificultades en el procesado,
como la proximidad a tierra o las variaciones rápidas debidas a las mareas y efectos
atmosféricos. Sin embargo aún hay posibilidades a explorar, como el retracking del eco del
radar, más datos a lolargo de la traza (along-track) y los escenarios multialtimétricos. Los
datos sobre las regiones costeras necesitan ser reanalizados, mejorados y más eficientemente
procesados. Esto resulta extremadamente importante para la Oceanografia operacional, para
investigar la dinámica compleja de las areas costeras y la variación del nivel del mar con su
impacto en la línea de costa.
La fuente principal de datos de nivel del mar son las redes nacionales de mareógrafos en la
costa, en España perteneciendo a diferentes instituciones como el Instituto Geográfico
Nacional (IGN), Puertos del Estado (PE), Instituto Hidrográfico de la Marina, etc.Los
mareógrafos miden el nivel del mar relativo a la tierra. Buenos modelos de mareas existen en
el Mediterráneo Occidental. Críticamente dependen de la información batimétrica precisa que
ha mejorado recientemente. La variación del nivel del mar para períodos largos ha sido
cuantificada recientemente usando datos del Topex/Poseidón y Jason-1. (fig.1)
Figura 1. Variación del nivel medio marino a partir de datos de los satélites altimétricos del
Topex y Jason-1.
-Calibración directa:
P á g i n a | 205
Latin American Remote Sensing Week 2010
donde SSH_GPS es la altura de la superficie instantánea del mar medida por la boya GPS y
Delta_Tide es la medida del mareógrafo menos el valor medio obtenido después de años de
observaciones. Se procura que las observaciones se realicen en diferentes horas del dia y en
diferentes estaciones del año para tener un buen Delta_Tide (sugiriéndose el mareógrafo no
se encuentre muy distante de la zona de observación con boya GPS).
-Calibración indirecta
donde SSH_ALT(Xs,Ys) es la altura instantánea del nivel del obtenida de la medida altimétrica
realizada por el satélite cruzando la banda.
La expresión calibración indirecta se refiere al hecho de que no hay boya GPS en el momento
del sobrevuelo del satélite pero la solución encontrada antes permitió ajustar la forma de la
MSS para tener una referencia absoluta a lo largo de la banda (superficie), que sirve para
calcular el bias del altímetro de cualquier satélite que atraviese la misma. Por esta razón la
calibración indirecta es menos precisa que la calibración directa donde la boya se encuentra
exactamente en la misma vertical del sobrevuelo del satélite. En definitiva la MSS encontrada
por boyas/catamarán GPS suministra la referencia para calcular el bias de los diferentes
satélites altimétricos que cruzan la banda.
La calibración del altímetro es un requisito esencial para la medida absoluta del nivel del mar
mediante el conocimiento preciso de su bias (desviación) y de su drift (deriva). Debido a
diferencias sistemáticas entre los instrumentos de los diferentes satélites es esencial
relacionar sus medidas usando instrumentos in-situ. La calibración es el objetivo final de
todos los procesos de validación (órbitas, correcciones, modelos geofísicos, EM bias, …).
La altura real del nivel del mar en el punto de calibración se determina, independientemente a
la medida altimétrica, a partir de sensores locales terrestres como p.e.: mareógrafos, boyas
GPS, etc. No se descarta el uso de otras tecnologías espaciales o aerotransportadas que
permiten también la determinación de la altura del nivel del mar de forma independiente.
Estas son p.e. la altimetría láser espacial –ICESAT- o la altimetría láser aerotransportada –
LIDAR- [Limpach er al., 2006]. Sin embargo, estas otras tecnologías aplicadas en el medio
marino se hallan en un estadio incipiente, en relación a las tecnologías antes mencionadas.
P á g i n a | 206
Latin American Remote Sensing Week 2010
Para los lugares geográficos de calibración, una configuración clásica sería el uso de una isla
muy pequeña, como el caso de las primeras calibraciones realizadas en el Mediterráneo en la
isla de Lampedusa (Italia) para el T/P (Ménard et al., 1994); posteriomente la isla de Córcega
(Bonnefond et al. 2003a, 2003b), que ha sido el site oficial de calibración de CNES para Jason-
1; la isla de Gavdos, también para T/P y Jason-1 (Pavlis 2002), la plataforma pretrolífera de
Harvest (California), utilizada como site oficial de NASA para la monitorización y calibración
continua de T/P y Jason 1 (Haines et al., 2003), asi como la isla de Ibiza (Martinez Benjamín et
al., 2004, 2005) que constituyó una contribución española a la calibración de Jason-1. (fig. 2).
El objetivo Principal era Integrar Ibiza y de forma complentaria el Cabo de Begur en las áreas
de Calibración Altimétrica Permanente en el Mar Mediterráneo Occidental, complementando
el área de Córcega/Senetosa (CNES, Francia), área principal de Calibración Altimétrica. En
oposición a las áreas principales de calibración, la isla de Ibiza es demasiado grande para no
afectar las medidas altimétricas en términos de ecos de radar y correcciones por la troposfera
húmeda (debido a la contaminación terrestre). Esto implica preferiblemente a utilizar datos
altimétricos medidos lejos de la costa (>10 km).
Existen configuraciones de areas costeras que monitorizan las trazas provenientes del mar
como son Bass Strait (en Tasmania, Australia), que es el único site en el hemisferio sur para
T/P y Jason-1 (Watson et al. 2003) y el Cabo de Begur (Costa Brava, Gerona) para T/P y Jason-
1, que además correspondió a la primera experiencia de calibración de un altímetro por un
grupo español en Marzo de 1999 (Alt-B del T/P). Finalmente, otro site remarcable es el del
Lago Erie (USA), se supone el único ejemplo de calibración altimétrica (ERS, GFO) en aguas
someras (Shum et al. 2004). Las áreas de calibración ayudan a controlar los errores
correlacionados geográficamente que son significativos en cada área aislada. Se espera en el
futuro realizar otra campaña incluyendo Jason-2 y Altika permitiendo datos de altimetria
costera, que deberia ser lanzado en 2011.
P á g i n a | 207
Latin American Remote Sensing Week 2010
Figura 3 Boya GPS Este tipo de boya /fig.3) consiste básicamente en una antena GPS,
albergada directamente sobre un aro flotador mediante una pieza cilíndrica por su base y
protegida por una cúpula en su parte superior, ambas piezas de metacrilato. La antena recibe
las señales de los satélites GPS y las envía mediante un cable de conexión, debidamente
preparado para no interferir en el movimiento libre de la boya sobre el agua, al receptor GPS
que se encuentra situado en una embarcación auxiliar.
P á g i n a | 208
Latin American Remote Sensing Week 2010
c)Las características de la cúpula, o sea el efecto que produce en la trayectoria de las señales
GPS que llegan y su disposición respecto a la antena GPS.
d)La correcta determinación de la geometría de la boya y en particular el valor de la distancia
entre el centro de fase de la antena GPS y la linea de flotación del mar en calma.
e)Estructura de la boya manejable, de fácil transporte y reciclable para ser utilizada en futuras
campañas.
La instrumentación de las medidas del nivel del mar será mejorada con la instalación de un
nuevo mareógrafo radar de pulsos DATAMAR 3000C, de Geónica S.L., en el Puerto de
Barcelona.
La Autoridad Portuaria de Barcelona (APB) ya dispone de una estación GPS, formada por un
sensor de Leica Geosystems GRX1200 GG Pro y una antena AX 1202 GG capaz de realizar un
seguimiento constante de las constelaciones NAVSTAR y GLONASS asi como de la futura
constelación GALILEO, en la nueva torre de control situada en el muelle de inflamables
constituyendo el conjunto una estación CGPS.
Se espera constituya una CGPS de las redes ESEAS (European Sea Level) y TIGA (GPS Tide
Gauge Benchmark Monitoring).
Puertos del Estado también dispone de un mareógrafo radar MIROS en el Puerto de Barcelona
además de un CGPS en el Puerto de Ibiza básico en la campaña IBIZA2003 de calibración
altimétrica del Jason-1 (fig.4).
P á g i n a | 209
Latin American Remote Sensing Week 2010
En la campaña de 1999 se siguió la traza ascendente (187) del ciclo 239 el 18 de Marzo de las
8:15 UTC a las 12:00 UTC. La calibración directa se realizó a las 8:45 UTC durante el
sobrevuelo del T/P sobre las dos boyas GPS. En la campaña de 2000 se recorrió la misma
traza ascendente del ciclo 287 el 6 de Julio. La calibración directa se efectuó a las 7:34 UTC
durante el sobrevuelo del T/P sobre la boya GPS. Se realizó además una estimación de la MSS
a lo largo de la traza desde las 7:29 UTC a las 16:57 UTC. Se realizó un posicionamiento
cinemático diferencial con distancias de base del orden de 35 km han sido aplicadas para el
cálculo de la posición de la boya.
Diferentes grupos con diferentes software de procesado de datos GPS obtuvieron el bias del
altímetro Alt-B del TOPEX:
5,1 +-6cm, campaña 1999, GIPSY-OASISII
6,8+-10cm, campaña 1999, KARS
4,7+-9cm/7,6+-7cm,cam.1999,GIPSY
3,7+-4,9cm, campaña 2000, GIPSY
P á g i n a | 210
Latin American Remote Sensing Week 2010
Figura 5. Area Geográfica del Cabo de Begur en donde se realizaron las campañas de
calibración altimétrica en 1999, 2000 y 2002.
En la Tabla 1 se indica el SSHBIAS estimado a partir del punto TOP-08 para el TOPEX-B y
TOP-11 for JASON-1
El objetivo de la experiencia española IBIZA 2003 (fig. 6), además de la implantación de la isla
de Ibiza como área permanente de calibración en el Mediterráneo Occidental, era adquirir un
conjunto amplio de datos GPS de nivel del mar según dos trazas nominales ascendente y
descendente del Jasón-1, suficientemente denso en espacio y tiempo, para obtener el geoide
local marino, teniendo en cuenta que el principal resultado no era determinar la altura del
geoide marino, sino la pendiente del geoide entre los datos altimétricos válidos en mar abierto
y el lugar donde se encuentran los mareógrafos costeros localizados en Ibiza y San Antonio.
Se utilizó un catamarán para realizar medidas continuas del nivel del mar. Su diseño fue
tomado del usado en Senetosa por P. Bonnefond. Llevaba dos antenas diferentes, Trimble
(CATL) y Leica (CATR) con resultados desiguales. Asimismo se utilizó una boya GPS para
P á g i n a | 211
Latin American Remote Sensing Week 2010
3.1 Agradecimientos
P á g i n a | 212
Latin American Remote Sensing Week 2010
Las campañas de calibración altimétrica del T/P y Jason-1 en el Cabo de Begur y la isla de
Ibizahan sido realizadas en el marco del Programa Nacional de Espacio I+D , CICYT,
Ministerio de Educación y Ciencia, ref:ESP97-1816-CO4 y ESP2001-4534-PE.
3.2 References
Watson C., R. Coleman, N. White, J. Church. And R. Govind. 2003. “Absolute Calibration of
TOPEX/Poseidon and Jason-1 using GPS Buoys in Bass Strait, Australia”, Marine Geodesy,
26(3-4), 285-304.
P á g i n a | 213
Latin American Remote Sensing Week 2010
b Department of Cartography, UNESP - Univ Estadual Paulista, 19060-900 Pres. Prudente, SP,
Brazil
Commission I, WG V/I
ABSTRACT: The use of digital cameras for photogrammetric applications, especially in aerial
mapping is a recognized trend due the advances in sensor resolution and other hardware and
software components. Nowadays, there are some models of cameras with 50 and 60
megapixels, with advantages because of their flexibility and cost effectiveness, which provide
ground coverage near to classic photogrammetric film cameras. This type of cameras can also
be integrated to GPS/INS systems, with specially designed mechanic mounts and electronics
devices that can control all components of the aerial mapping system. Several custom
designed systems have been used worldwide with excellent results for mapping applications.
In this context this paper presents the results obtained with SAAPI system in a huge project,
the digital mapping of Bahia State, Brazil. This system is an airborne acquisition system based
on professional Hasselblad digital frame cameras integrated to direct orientation sensors
(GPS/INS), electronic devices and hardware and software tools. In order to assess the
geometric accuracy of this mapping system a set of images acquired by SAAPI with a ground
sample distance (GSD of 80 cm was used. This block with 2997 images distributed in 48 strips
was processed with INPHO MATCH-AT. Several experiments were carried out to assess
accuracy improvements in the object space coordinates when considering in-situ calibration
bundle adjustment with direct georeferencing. Moreover, orthophotos and DTM products
were generated using the INPHO-MATCH-T software and analyzed. It was verified that
suitable accuracy was achieved rounding 1 to 1.5 GSD in the final products. These results
showed that this type of aerial system can be successfully used for mapping projects, provided
that rigorous photogrammetric processing workflow is used.
1. Introduction
With the recent developments in the technology of optical digital sensors, the use of
professional digital frame cameras emerged as an alternative for aerial photogrammetric
applications in the latest years. The main reasons are their flexibility and cost effectiveness,
when compared to film cameras and high-end digital systems. Nowadays, this practice has
been considered as an acceptable technique, and some countries (like EUA and Canada) are
preparing guidelines and specifications for the use of this category of camera in
photogrammetric works.
P á g i n a | 214
Latin American Remote Sensing Week 2010
Compared to classic film cameras (230 x 230 mm format) or high-end digital systems, the
professional digital frame cameras have smaller ground coverage area, although this scenario
is rapidly changing with the new versions of the sensor resolution (50-60-80 Mpixels). On the
other hand, the professional digital cameras allow the development of light weight and low
cost aerial photogrammetric systems.
Due to this favourable cost/benefit ratio, many companies have adapted with success
professional digital frame cameras to produce aerial mapping systems, as DigiCam (IGI), DSS
(Digital Sensor System – Applanix) and SAAPI (Lightweight Airborne Image Acquisition
System - Engemap). Others independents systems based on this category of camera for
different photogrammetric applications were previously implemented, as showed in Mostafa
and Schwarz (2000), Habib et al. (2002), Roig et al. (2006) and Petrie (2009).
Within this context, this paper presents the geometric results achieved for a large block of
images acquired by SAAPI in order to assess the geometric accuracy of this category of
mapping system.
2. SAAPI SYSTEM
The SAAPI system (Lightweight Airborne Image Acquisition System) was developed by
Engemap Company in Brazil, with UNESP partnership and FAPESP (The State of São Paulo
Research Foundation) grant in the project first phase. This system is composed by an
acquisition platform, control and power units (Figure 1).
- RGB and Infrared professional digital cameras at same resolution - 50 megapixels each
one (to be upgraded to 60 megapixels in the next months);
- Direct Georeferencing system;
- Specific rigid housing for the cameras;
- Autonomous system of triggering, high precision data synchronism and logging and
storage in SSD (Solid State Disk);
- Software for automatic flight plan generation (integrated to the GPS and Google Earth),
real time navigation and system control, and flight post processing data;
- Modular design: light weight (about 60 kg the complete system, including power unit)
and less power consumption needs (12V). These features allow high flexibility for
installing the system in different kinds of aircrafts, mainly in small aerial platforms.
P á g i n a | 215
Latin American Remote Sensing Week 2010
The technical specifications of the cameras that compose the SAAPI acquisition platform are
given in Table 1.
Focal length 50 or 35 mm
Pixel size 6 m
Radiometric
12 bit
resolution
8176 x 6132
Image size pixels (50
megapixels)
Approximated 46/59 - 50
field of view - mm lens
each camera
(along/across 63/78 - 35
track) mm lens
Several aerial survey projects were performed with success by Engemap Company in the
latest three years with the SAAPI system, including urban and environmental mapping,
highways, power lines and pipeline applications. These projects include cartographic mapping
products generation in different scales, from 1:10000 (image resolution of 60-80 cm) to
1:1000 (image resolution of 10-15 cm).
Nowadays, the main projects that are being carried out by Engemap Company in Brazil using
the SAAPI system are:
- Bahia state mapping, where digital imagery over an area of 600.000 km 2 are being
acquired with a GSD of 80 cm for DSM (Digital Surface Model) and Orthophoto
generation;
- Santa Catarina state mapping: RGB and infrared digital imagery of an area of 97.000 km2
with a GSD of 39 cm for DSM (Digital Surface Model), DTM (Digital Terrain Model),
Orthophoto generation and hydrography restitution.
3. BACKGROUND
In general, camera calibration is performed before aerial surveying and the IOP (Interior
Orientation Parameters) are usually estimated by laboratory methods or field method,
including close range camera calibration. These parameters are constrained in bundle block
triangulation, with the image coordinates being a priori corrected for the systematic errors
P á g i n a | 216
Latin American Remote Sensing Week 2010
(lens distortion, photogrammetric refraction and, less common, affine deformation). This
solution leads to a more simplified bundle adjustment model, with less parameters to be
estimated and without high correlations between the estimated IOP and EOP (Exterior
Orientation Parameter)
The camera IOP (focal length, principal point coordinates, lens distortions, affine
deformations) can be estimated during the bundle block adjustment based on the collinearity
equations with additional parameters that can be written by using different model. Even
knowing that the use of additional parameters can result in high correlations between
parameters, it is important to consider the differences in the environmental conditions
between the calibration field and the flight area. Moreover, the operation and the handling of
the cameras during the flight projects can modify the inner geometry of the cameras
(depending on the optical and sensor stability of the camera model) and this changes can
affects the bundle reconstruction.
Generally polynomial models were used as additional parameters in bundle block adjustment.
In this approach the focal length and the principal point coordinates are constrained with its a
priori calibrated values and the polynomial coefficients aims at to absorb the residual
systematic errors related to the lens distortion, shrinkage and other non modeled
deformations. The process in which these models are used is generally known as self
calibration and was developed in the seventies; nevertheless some authors argue that this
term should be related to the camera calibration with a minimum set of constraints and even
without ground control and that a better term should be in-situ or on-the-job calibration
(Clarke and Fryer, 1998). Examples of groups of additional parameters are Ebner, Brown and
Grun models (Muray et al, 1984; Mikhail et al, 2001; Clarke and Fryer, 1998).
It is of crucial important to analyze the correct use of these additional parameters groups for
digital cameras, since that many errors existing in analog film cameras, like shrinkage or
errors due to comparator measurements are unsuitable for digital sensors. In order to
investigate this subject, some experiments using LPS (Leica Photogrammetry Suite) software,
in which triangulation module was set to use additional parameters, were carried out. Images
collected with a 39 MP Hasselblad digital camera were used. In the performed experiments it
was verified that using only the radial lens distortion provided better results on the object
space reconstruction than the polynomial models (Brown, Ebner, Bauer and Jacobsen). Details
of these tests are presented in Ruy et al (2008).
In this context the experiments presented in this paper were carried out with bundle block
adjustment performed by the INPHO package software (Match-AT, Match-T) with blocks of
images acquired by SAAPI system. In these experiments the camera calibration parameters
were computed in a small block with dense control points distribution and the entire block of
images was processed with different control points
configurations. Moreover it is presented the DSM quality achieved with this image block.
P á g i n a | 217
Latin American Remote Sensing Week 2010
4. EXPERIMENTAL ASSESSMENT
In this chapter a set of experiments with bundle block adjustment and DSM generation with
Bahia image block are presented. This block is composed by 2997 images distributed in 48
strips. The flight height was 4670 m, resulting in GSD (Ground Sample Distance) of 80 cm.
In this work it is also assessed an approach that relies on the calibration in a sub-block with a
dense control points distribution (See marked area in Figure 2). The IOP estimated in this sub-
block are then used in the whole block, with a reduced set of GCP. Figure 2 shows the
complete block of images related to the Bahia project as well as the ground control and check
points distribution. In the left-up region is show the sub-block with dense GCP distribution.
Figure 2. Complete block (2997 images) with control (127 points) and check points (26 points)
distribution and a sub-block with control and check points distribution (marked).
The Bahia project is a large aerial survey project (600,000 km2) that is being developed by
Engemap Company since 2009. The image block used in this study is located in south-west
part of Bahia state in Brazil, as showed in Figure 3.
P á g i n a | 218
Latin American Remote Sensing Week 2010
Brazil
Bahia state
Studied Block in
the Bahia project
(2997 images)
Figure 3. Geographic location of the studied block of images.
In order to assess the approach that uses a sub-block for computing the IOP an on-the-job
calibration using 344 images (marked area in Figure 2) and 154 control points was carried
out in the Match-AT INPHO software (In-Block module). The computed parameters are
presented in the Table 2.
P á g i n a | 219
Latin American Remote Sensing Week 2010
After that, a set of experiments with the complete block (2997 images) and different number
of control points was performed:
With all control points (127 points);
With 79 control points;
With 41 control points;
With 18 control points.
For all these tests the complete set of 26 check points were used for verifying the 3D
reconstruction quality
in each situation. Table 3 presents the RMS (Root Mean Square) error in the control and check
points for each experiment. In the last column of this table it is presented the expected
accuracy (e) of the object space coordinates for the block of images. This theoretical
accuracy is computed as a function of the measured point error, the block definition and the
geometry of the acquisition, as defined in Krauss and Waldhaus (1993). This expected
accuracy is only used as a reference value for the object space reconstruction.
Experiment
e
A B C D
0.22 0.20 0.20 0.14
X -
RMS 2 8 1 8
(m) 0.19 0.19 0.19 0.13
Y -
Control 9 5 0 7
Points 0.18 0.17 0.13 0.08
Z -
8 6 4 6
0.40 0.41 0.43 0.52 1.29
X
RMS 5 4 2 1 0
(m) 0.42 0.43 0.44 0.46 1.29
Y
Check 8 4 5 5 0
Points 0.75 0.77 0.71 0.81 2.42
Z
3 4 4 2 0
RMS X 0.5 0.5 0.5 0.6 1.6
(GSD) Y 0.5 0.5 0.6 0.6 1.6
Check
Z 0.9 1.0 1.0 1.0 3.0
Points
Table 3: RMS values of the discrepancies in the control and check points for the Bahia complete
block (2997 images).
The experiments were performed with the IOPs computed in the sub-block (344 images) in
order to assess the approach that computes the IOPs parameters using a sub-block of images
with a dense control point distribution and applying later these IOPs in the complete block
with reduced GCPs.
By analyzing Table 3 it can be verified that when the number of GCPs was reduced seven
times (experiment D), the quality of the solution in the object space was kept, when compared
to the experiment A that used all the GCPs. The accuracy in the 3D reconstruction was around
1/2 GSD in planimetry and one GSD in altimetry.
For the analysis of the experiments presented in this work it was verified that the IOPs
computed in the sub-block can be successfully applied to the complete block with a
considerable reduction of GCP, provided that the direct georeferencing system is used. In this
P á g i n a | 220
Latin American Remote Sensing Week 2010
case it was verified that one control point every 20 images on the same strip flight and one
control point every 8 strip flights are enough to guarantee the quality of the solution.
DSM generation
The main product of the Bahia project that is being generated by Engemap Company is the
DSM of the entire Bahia state with 5 m of resolution. Figure 4 shows an example of DSM
generated for the studied area presented in this paper.
The quality control of the products is being independently performed by the official
Cartographic Army agency in Brazil - DSG (Geographic Service Division). For this DSM
resolution (5 m) the Brazilian norms establishes that 90% of the altitude values of measured
points on the product must have error less than 2.5 m and the RMS errors have to be less than
1.8 m, when compared to the check points collected in the field.
For the quality control of the DSM related to the studied area presented in this paper, 111
check points were used by the DSG agency. The results are presented in Table 4.
P á g i n a | 221
Latin American Remote Sensing Week 2010
Considering the results presented in Table 4 it can be seen that the accuracy of the DSM is
clearly compatible with the specifications of the project, with the maximum individual error in
2.4 m and RMS error around 1 m.
By definition the DSM is related to the surface of the terrain and the absolute altitude or
elevation of the points above the terrain, as showed in Figure 5.
Point on the
DSM
terrain
Point on the
terrain
It was verified by the results of the quality control of DSM that there was a small tendency in
the values (see the average column in Table 4). This tendency was mainly positive, because
the values of heights measured on the DSM are higher than the corresponding ones obtained
by the geodetic instruments in field (check points). This bias can be explained, in part, by the
smoothing effect of DSM due to buildings, trees and others entities above the terrain artefacts
(see Figure 5). Then, if the check point for the quality control was collected near to these
features a difference in the DSM model can be expected in the quality analysis.
5. CONCLUSIONS
This paper presented an outline of the SAAPI system that is a digital acquisition platform for
photogrammetric applications composed by professional digital frame cameras in a modular
design, with flexibility and light-weight that can be installed in different kinds of aircrafts and
helicopters that can be used for fast mapping production.
The results obtained with the bundle block adjustment and DSM quality control showed that
this type of digital acquisition system can be successfully used for mapping projects, provided
that rigorous photogrammetric processing workflow is used.
One of the limitations of the digital professional frame cameras is the sensor dimension, which
results in an increase of the number of images and measured points in the photogrammetric
projects. Nowadays this limitation is becoming irrelevant due to the new sensors resolution of
digital cameras (50-60-80 mpixels), the GPS/INS integration and the high performance of
photogrammetric software available, like Inpho package, that allow the fully automation and
fast processing of the photogrammetric products.
P á g i n a | 222
Latin American Remote Sensing Week 2010
6. REFERENCES
Clarke, T. A.; Fryer, J. G. The development of camera calibration methods and models. The
Photogrammetric Record, 16 (91), pp. 51-66, 1998.
Habib, A., M. Morgan, and Y. Lee, 2002. Bundle adjustment with self-calibration using straight
lines, Photogrammetric Record, 17(100):635-650.
Kraus, K.; Waldhausl, P., 1993. Photogrammetry – Fundamentals and Standard Processes.
UMMLER/BONN, v.1.
Mikhail, E. M.; Bethel, J. S. Mcglone, J. C., 2001. Introduction to Modern Photogrammetry. Inc.
New York : John Wiley & Sons, 479p.
Mostafa, M.M.R., and K.P. Schwarz, 2000. A Multi-Sensor System for Airborne Image Capture
and Georeferencing. Photogrammetric Engineering & Remote Sensing, 66(12):1417-1423.
Muray, S.; Matsuoka, R.; Okuda T., 1984. A study on Analytical Calibration for non Metric
Camera and Accuracy of Three Dimensional Measurement. In: The International Archives of
Photogrammetry and Remote Sensing - XVth ISPRS Congress – Commission 5, Rio de Janeiro,
Vol.25, Part V pp. 570-579.
Petrie, G., 2009. Systematic Oblique Aerial Photography Using Multiple Digital Frame Camera,
Photogrammetric Engineering & Remote Sensing, 75(2):102-107.
Roig, J., M. Wis, and I. Colomina, 2006. On the geometric potential of the NMC digital camera,
Proceedings of the International Calibration and Orientation Workshop – EuroCOW 2006, 25-27
January, Castelldefels, Spain, unpaginated CDROM.
Ruy, R. S.; Tommaselli, A. M. G.; Galo, M.; Hasegawa, J. K.; Menossi, R. C., 2008. Fototriangulação
com parâmetros adicionais para câmaras digitais: uma avaliação experimental. Boletim de
Ciências Geodésicas, Curitiba, vol.14, n. 4, pp. 571-587.
P á g i n a | 223
Latin American Remote Sensing Week 2010
Resumen: El conocimiento del factor topográfico es un insumo básico para comprender gran
parte de los procesos ocurridos en la superficie de la tierra. Es por esto, que para facilitar el
procesamiento computacional de la información se ha generado lo que se conoce como
Modelo de Elevación Digital (DEM). Las crecientes necesidades de contar con datos de
elevación con una baja demanda de tiempo, procesamiento y a un costo accesible, ha
incentivado el desarrollo de diversas plataformas satelitales destinadas a generar
información de la superficie terrestre en forma de DEM, sin embargo, a pesar de que los
rangos de precisión están en general bien descritos para cada plataforma, diversos autores
recomiendan que las validaciones deben realizarse a nivel local para tomar mejores
decisiones. El objetivo de este trabajo es realizar un análisis cuantitativo de la calidad de los
DEM SRTM y ASTER comparados con puntos de control terrestre bajo diversos tipos de
relieve y cobertura vegetal. El análisis comparativo se baso en la definición de cuatro
tipologías de superficie terrestre en las cuales se capturaron datos de altitud GPS para su
posterior contraste con los DEM. Los resultados indican que existen diferencias significativas
entre la distribución de los errores en cada tipología de superficie, siendo comparativamente
menores en sectores de llanura, aunque indistintamente del tipo de cobertura vegetal. En
relación a la calidad de ambos modelos, SRTM se presenta con mejores resultados en cada
uno de los parámetros evaluados tanto a nivel global como por tipologías.
Abstract: Knowledge of the topographic factor is a basic input to encompass most of the
processes occurring at the surface of the earth. Therefore, in order to facilitate computer
processing of information has created what is known as Digital Elevation Model (DEM). The
growing need for elevation data with a low demand time, processing and affordable, has
encouraged the development of various satellite platforms designed to generate information
of the earth's surface in the form of DEM, however, despite precision ranges are generally well
described for each platform, several authors recommend that validation must be done locally
to make better decisions. The aim of this paper is presents a quantitative analysis of the
quality of the SRTM and ASTER DEM compared with ground control points under various
types of terrain and vegetation. The comparative analysis was based on the definition of four
P á g i n a | 224
Latin American Remote Sensing Week 2010
types of land area in which was captured GPS altitude data to contrast with the DEM. The
results indicate that significant differences exist between the distribution of errors in each
type of surface and comparatively lower in areas of plain, but regardless of the type of
vegetation. In relation to the quality of both models, SRTM is presented with the best results
in each of the parameters assessed both globally and by type.
Palabras clave: Modelo de elevación digital, evaluación de la calidad DEM, SRTM, ASTER.
1. INTRODUCCIÓN
El conocimiento del factor topográfico es un insumo básico para comprender gran parte de los
procesos ocurridos en la superficie de la tierra (Hook, 1998). Se utiliza en análisis de ecología,
hidrología, riesgos naturales, entre muchos otros, como una variable de base para explicar y
predecir ciertos comportamientos mediante la modelación (Jarvis et al., 2004). En este
contexto, para facilitar el tratamiento y análisis computacional de esta información, surge el
concepto de Modelo de Elevación Digital (DEM). Un DEM es una estructura numérica de datos
que representa de forma cuantitativa y continua las características topográficas del territorio
(Felicísimo, 1998; Chaplot et al., 2006).
En la actualidad, existe una creciente demanda de datos de elevación tanto por parte de
organismos públicos, privados, centros de investigación y universidades, los que pueden ser
generados a partir de mapas topográficos, fotografía aérea, sistemas GPS, datos de sensores
láser como el LIDAR, sensores radares o mediante imágenes satelitales estereoscópicas. Sin
embargo, en estudios que requieren información sobre extensas superficies, la ventaja que
presentan los sistemas satelitales es ampliamente superior. Esto dado principalmente por la
relación de costo/tiempo/beneficios que presentan, además de la consistencia de los datos ya
que son capturados en un mismo momento y bajo iguales condiciones atmosféricas y de
calibración del sensor (Roa & Kamp, 2008). Es por esto que en las últimas décadas se han
desarrollado una serie de plataformas satelitales que capturan datos de la superficie terrestre
generando DEMs.
Entre los sistemas satelitales que ofrecen datos topográficos, se encuentran plataformas cuya
información está disponible de manera gratuita, de fácil acceso y con una disponibilidad a
nivel global. Entre estos, está el sistema radar Shuttle Radar Topographic Mission (SRTM),
uno de los más ampliamente utilizados, y el sistema ASTER (Advance Space Borne Thermal
Emisión and Reflection Radiometer), que destacan por su alta resolución espacial y
disponibilidad global. Si bien, ambos sistemas presentan rangos de precisiones conocidas, son
muchos los autores que señalan que esta dependerá finalmente de las condiciones locales
presentes en el área de estudio (Kiamehr & Sjöberg, 2005; Schumann et al., 2008; Li & Wong,
2009). Por otra parte, nuestra capacidad de poder entender y modelar de mejor forma los
distintos procesos dependerá en gran medida de la calidad de la información topográfica que
tengamos disponible, en este caso de los DEM.
P á g i n a | 225
Latin American Remote Sensing Week 2010
El objetivo de este trabajo es realizar un análisis cuantitativo de la calidad de los DEM SRTM y
ASTER, comparados con puntos de control terrestre, evaluando la distribución y magnitud de
los errores en función del tipo de relieve y la cobertura vegetal.
2. METODOLOGÍA
El área de estudio corresponde a la cuenca del río Queule, localizada entre los 39° 3’ y 39° 25’
de Latitud Sur y los 72° 51’ y 73° 13’ de Longitud Oeste. Posee una superficie de 69.143
hectáreas distribuidas en tres principales cuencas costeras: la del río Queule inferior, río
Boroa y río Boldo, cuyo relieve se caracteriza por cordones montañosos, plataformas de
erosión marina y extensas planicies fluviomarinas (Peña-Cortes et al., 2006). El clima según Di
Castri y Hajek (1976) es oceánico con influencia mediterránea con un promedio anual de
precipitaciones de 1.200 mm a 1.600 mm.
Los datos de elevación fueron obtenidos de las fuentes libres disponibles en internet. Para el
caso de SRTM, la imagen fue descargada del sitio http://srtm.csi.cgiar.org/ obteniendo el
producto estándar SRTM-v4 de 90 mt. El DEM ASTER fue descargado desde el sitio
https://wist.echo.nasa.gov. En ambos casos las imágenes fueron trabajadas con un tamaño de
pixel de 30 metros, proyección UTM y Datum WGS84, con una altitud considerada al nivel
medio del mar con el modelo geoidal EGM96, mismo sistema utilizado para la captura de
puntos de control en terreno.
Para evaluar las diferencias en la calidad de ambos DEM, se trabajo con cuatro principales
tipologías de superficies basadas en las condiciones de relieve y la cubierta vegetacional
presente. Estas tipologías consideraron los rangos más extremos entre ambas variables
consideradas, es decir, para el caso del relieve se consideraron unidades de cordones
montañosos y llanuras, mientras que para la vegetación se consideraron sectores de praderas
y zonas boscosas, con las tipologías y replicas indicadas en la tabla 1. En cada una de las
combinaciones generadas se realizó la toma de puntos de control mediante equipos GPS
L1/L2 con método absoluto (Ocup. 2’). Los puntos fueron procesados y convertidos en una
capa Shape vectorial para extraer el correspondiente valor dado por los DEM SRTM y ASTER.
P á g i n a | 226
Latin American Remote Sensing Week 2010
N°
Tipología de Superficie
Puntos
Zonas boscosas en cordón montañoso (ZBCM) 14
Zonas boscosas en llanura (ZBL) 5
Zonas de pradera en cordón montañoso (ZPCM) 12
Zonas de pradera en llanura (ZPL) 13
Para la comparación entre las distintas fuentes de información se realizó una primera
aproximación general de los datos, la que consistió en un análisis de varianza entre los errores
observados en cada tipología para comprobar si efectivamente existen diferencias
significativas por efectos de la configuración de la superficie.
Los criterios de análisis empleados para establecer la calidad de cada DEM, se basaron en las
diferencias de altitud entre cada punto de control GPS (Valor observado) y los obtenidos de
los modelos de elevación (valor esperado). Para esto se consideraron los siguientes
parámetros: Error cuadrático medio (RMSE), Sesgo (BIAS), Error medio absoluto (MAE) y el
Coeficiente de correlación (R2). Los resultados se agruparon por tipología evaluando la
distribución y magnitud en cada una de ellas.
Finalmente se elaboró un DEM de diferencias entre los modelos SRTM y ASTER, en el cual se
indica la magnitud de estas en cada píxel.
3. RESULTADOS Y DISCUSION
Los DEM analizados presentaron diferencias significativas (p-valor 0,026; α 0,05) en relación
a la distribución de errores entre las cuatro tipologías de superficie estudiadas. Lo anterior se
puede apreciar con detalle en la tabla 2, en la cual tanto para el caso de SRTM como de ASTER
se observan valores comparativamente más bajos de RMSE y MAE en tipologías de bajo
relieve. Respecto al tipo de cobertura predominante, esta presentó una leve influencia en
sectores de cordón en donde las zonas boscosas obtuvieron valores más altos de error que en
coberturas con praderas.
Tabla 2. Distribución y magnitud de los errores entre puntos de control terrestre y los
DEM SRTM y ASTER. RMSE = Error cuadrático medio; BIAS = Sesgo; MAE = Error medio
absoluto; R2 = Coeficiente de correlación.
P á g i n a | 227
Latin American Remote Sensing Week 2010
Zonas boscosas en llanura 3,2 1,2 3 0,151 4,3 2,6 2,9 0,179
Zonas de pradera en cordón montañoso 12,3 -7,2 9,9 0,998 19,8 -13,9 14,7 0,997
Zonas de pradera en llanura 5,2 3,4 3,6 0,989 7,3 0,8 5 0,592
Por otra parte, si bien el ajuste global de los datos derivados de los distintos DEM se
correlacionó fuertemente con los puntos de control, al efectuar el análisis por tipologías estas
correlaciones tendieron a bajar levemente en algunos sectores, siendo las zonas boscosas en
llanura las que presentaron los valores más bajos. Lo anterior puede deberse a la escaza
cantidad de puntos de muestreo que se lograron capturar en esas áreas producto de que se
trataba de bosques pantanosos de difícil acceso vial e inundados la mayor parte del año.
La comparación pixel a pixel entre ambos DEM presentó valores que oscilan entre los -70 y los
156 mt, siendo las diferencias positivas las que concentran la mayor parte de los pixel. La
figura 1 presenta la magnitud y distribución de estas disimilitudes.
P á g i n a | 228
Latin American Remote Sensing Week 2010
Como se observa, las mayores diferencias se encuentran en sectores de cordón montañoso, los
que en el área de estudio se caracterizan por poseer predominantemente coberturas boscosas
ya sea naturales o plantaciones. En este sentido para el caso de SRTM están descritos los
efectos de sobre estimación que generan este tipo de cubiertas en la captura de los datos, lo
cual explica las altas diferencias positivas que se presentan en estos sectores en relación al
sensor ASTER. A pesar de lo anterior, para la cuenca costera estudiada se recomienda el uso
de datos de elevación SRTM ya que presentó un mejor ajuste tanto en términos globales como
por cada tipología.
4. BIBLIOGRAFÍA
Chaplot, V., Darboux, F., Bourennane, H., Leguédois, S., Silvera, N., and Phachomphon, K., 2006.
Accuracy of interpolation techniques for the derivation of digital elevation models in relation
to landform types and data density. Geomorphology, Vol. 77, p. 126-141.
Di Castri, F., and Hajek, E., 1976. Bioclimatología de Chile. Ediciones Universidad Católica de
Chile, Santiago, Chile, 128 p.
Felicísimo, A., 1998. Modelos Digitales del Terreno. Introducción y aplicaciones en las ciencias
ambientales. 122 p.
Jarvis, A., Rubiano, J., Nelson, A., Farrow, A., and Mulligan, Mark., 2004. Practical use of SRTM
data in the tropics – Comparisons with digital elevation models generated from cartographic
data. Centro Internacional de Agricultura Tropical (CIAT), Cali, Colombia, N° 198, 35 p.
Kiamehr, R., Sjöberg, L.E., 2005. Effect of the SRTM global DEM on the determination of a high-
resolution geoid model: a case study in Iran. J Geod, Vol. 79, p. 540-551.
Li, J., & Wong, D., 2009 (Article in Press). Effects of DEM sources on hydrologic applications.
Computers, Environment and Urban Systems, p. 1-11.
Roa, J., & Kamp, U., 2008. Modelos de elevación digital (MED) a partir de sistemas satelitales.
Una introducción y análisis comparativo en la cordillera de Mérida-Venezuela. Revista
Geográfica Venezolana, Vol. 49 (1), p. 11-42.
Peña-Cortés, F., Rebolledo, G., Hermosilla, K., Hauenstein, E., Bertrán, C., Schlatter, R., & Tapia,
J., 2006. Dinámica del paisaje para el periodo 1980-2004 en la cuenca costera del Lago Budi,
Chile. Consideraciones para la conservación de sus humedales. Ecología Austral, Vol. 16, p.
183-196.
P á g i n a | 229
Latin American Remote Sensing Week 2010
Schumann, G., Matgen, P., Cutler, M.E.J., Black, A., Hoffmann, L., and Pfister, L., 2008.
Comparison of remotely sensed wated stager from LiDAR, topographic contours and SRTM.
Journal of Photogrammetry & Remote Sensing, Vol. 63, p. 283-296.
5. AGRADECIMIENTOS
También se agradece el valioso apoyo de Elías Andrade, alumno del Magister en Planificación
y Gestión Territorial de la Universidad Católica de Temuco.
P á g i n a | 230
Latin American Remote Sensing Week 2010
dcastro@ambiente.gob.ec
www.ambiente.gob.ec
CONDESAN
condesan@condesan.org
www.condesan.org
Abstract: The ministry of Environment is executing the project “Map of Ecosystems”, which
has the prime objective to represent cartographically the most important ecosystems that are
within the continental Ecuador, for this reason the efforts of the project have been directed to
the construction of a legend of multi-scale ecosystem, as to the reinforcement of the processes
of interpretation of satellite images to determine the ecosystems, contrasting y validating this
information with floral data retrieved in the field.
1. ANTECEDENTES
El Ministerio del Ambiente del Ecuador (MAE), como autoridad ambiental nacional, está
encargado de diseñar las políticas ambientales y coordinar las estrategias y proyectos para el
cuidado de los ecosistemas y el aprovechamiento sostenible de los recursos naturales del país.
En este contexto, el MAE ha priorizado el desarrollo del “Mapa de Ecosistemas del Ecuador
Continental” con el objetivo de contar con información espacial actualizada de los
ecosistemas, su remanencia, su nivel de representatividad en el Sistema Nacional de Áreas
Protegidas y una identificación de áreas prioritarias para conservación y restauración.
Adicionalmente, este Proyecto contribuirá a la estructuración de un sistema de monitoreo
P á g i n a | 231
Latin American Remote Sensing Week 2010
2. INTRODUCCION
3. MATERIALES Y METODOS
BIOGEOGRAFIA
COBERTURA DE LA BIOCLIMA
TIERRA
PISO ALTITUDINAL DE
ECOSISTEMA VEGETACION
REGIMEN DE GEOFORMA
INUNDACION
FENOLOGIA
P á g i n a | 232
Latin American Remote Sensing Week 2010
3.2. Biogeografía
La biogeografía estudia las áreas de distribución de las especies, los patrones resultantes de
estos estudios permiten identificar los centros de origen y dispersión de las mismas, así como
comprender los patrones espaciales de la diversidad biológica.
P á g i n a | 233
Latin American Remote Sensing Week 2010
3.4. Geoformas
PENDIENTE ABSOLUTA
HIPSOGRAFIA
EDICION E
INTERPRETACION
TEXTURA TOPOGRAFICA
El régimen de inundación puede definir en gran medida la cantidad de nutrientes que reciben
las plantas, lo cual determina niveles de productividad primaria, y a su vez, la energía que se
distribuye en las cadenas tróficas. Esta combinación de la química del agua y la frecuencia con
la que inunda ciertos sitios puede tener un efecto profundo en los ecosistemas.
MAPA HIDROGRAFICO DATOS DE CAMPO AGUAS CLARAS, AGUAS BLANCAS, AGUAS NEGRAS
P á g i n a | 234
Latin American Remote Sensing Week 2010
UNIDADES UNIDADES
BIOCLIMATICAS GEOMORFOLOGICAS
LEYENDA DE UNIDADES
ECOSISTEMAS AMBIENTALES
P á g i n a | 235
Latin American Remote Sensing Week 2010
4. RESULTADOS PRELIMINARES
P á g i n a | 236
Latin American Remote Sensing Week 2010
5. CONCLUSIONES
Con el fin de afinar el modelo expuesto, el Ministerio del Ambiente del Ecuador se encuentra
realizando un proceso de actualización y mejoramiento de cada una de las variables,
interactuando con las instituciones generadoras de información oficial en cada una de las
temáticas relacionadas con la representación cartográfica de Ecosistemas.
Con el objetivo de dar sostenibilidad en el tiempo el Ministerio del Ambiente del Ecuador
implementará un sistema de monitoreo de Ecosistemas que permita obtener un instrumento
para el planteamiento de directrices que conlleven a una mejor utilización del territorio.
P á g i n a | 237
Latin American Remote Sensing Week 2010
6. BIBLIOGRAFIA
Josse, C., et al., Ecological Systems of Latin America and the Caribbean: A Working
Classification of Terrestrial Systems. 2003, NatureServe: Arlington, VA. p. 47. 6.
P á g i n a | 238
Latin American Remote Sensing Week 2010
a a
Valerie Ussyshkin , Livia Theriault
a
Optech Incorporated, 300 Interchange Way, Vaughan, Ontario, Canada L4K 5Z8
valerieu@optech.ca, liviat@optech.ca
ABSTRACT:
Over the past decade airborne lidar technology has seen the development of new systems
capable of providing highly accurate geospatial data with unprecedented efficiency for a
variety of applications. Unlike two-dimensional aerial imagery, the elevation component of
airborne lidar data provides the ability to represent complex vertical structures of ground
targets with very high precision, which is a prerequisite to many lidar applications focusing
on the analysis of elevated features, particularly, in forestry. However, many conventional
airborne lidar systems used in the commercial lidar sector for mapping complex 3D targets
such as vegetation and forested areas have been limited in their ability to collect data of
sufficient quality for sophisticated data analysis in the vertical domain.
This paper presents an overview of the evolution of specific features of airborne lidar
technology concerning 3D mapping applications, particularly in forestry and vegetation
mapping, which are considered to be of special importance in Latin America. Several key
performance characteristics of commercial airborne lidar systems are discussed in the
context of precise mapping of complex vertical targets, and illustrated by the most advanced
capabilities of the ALTM Orion, a new airborne lidar mapper manufactured by Optech
Incorporated. It is demonstrated that the Orion’s unique combination of exceptional elevation
precision with enhanced vertical target discrimination and vegetation penetration
capabilities produces lidar data of enhanced quality, which can represent complex 3D targets
such as vegetation structure equivalent in some aspects to the content of full waveform (FW)
data. It is shown that the recent advances in conventional airborne lidar technology bear the
potential to create a new application niche where top-quality dense point clouds, enhanced
by fully recorded intensity for multiple returns, may provide sufficient information for
modeling some vegetation parameters which have traditionally been derived mainly from full
waveform data.
1. INTRODUCTION
Airborne lidar technology has been successfully used for forestry and vegetation surveying
and mapping application for more than two decades beginning with several demonstration
studies back in the early eighties. One of the first attempts to use airborne lidar for these
purposes was reported by Arp and co-authors on a custom-designed lidar, which was
P á g i n a | 239
Latin American Remote Sensing Week 2010
successfully used to map tropical forest in Central America (Arp at all, 1982). In the mid
1980s, the Canadian Forestry Service demonstrated the applicability of profiling lidar for the
estimation of stand heights, crown cover density and ground elevation below the forest
canopy (Aldred and Bonner, 1985). These and following studies showed that lidar data
profiling vegetation canopy could be used to quantify many other forest attributes including
above-ground biomass, basal area, mean stem diameter, vertical foliar profiles and canopy
volume, and others characteristics (Means et al., 1999; Nelson et all, 1984, Lefsky et al., 1999;
Dubayah and Drake, 2000).
In the beginning most of the attempts to use airborne lidar for mapping such complex 3D
targets as tropical forest or vegetation canopy were done mostly on a non-commercial basis
using specially designed airborne lidar systems. Later when airborne lidar sensors became
widely commercially available, numerous studies on forestry applications based on the use of
commercial lidar systems have been reported. (Kopela, 2010, Wagner, 2008)
However, the use of airborne lidar technology for forestry applications has been only
partially commercialized for a number of reasons; some of them are listed below (Nelson et
all, 2003):
● Laser data post-processing requires the attention of specialists and often customized
proprietary software;
● It is only in the last few years that researchers have demonstrated beyond a reasonable
doubt that lidar data can be used to create reliable maps of tree height, canopy density, bio-
mass, and volume;
Although these points seem to reflect the general trend in the use of airborne lidar for
forestry and other vegetation mapping applications, they also imply, to some extent, a certain
misapprehension about different types of airborne lidar sensors, and the way data are
collected and interpreted. Discrete return lidar systems—those most commonly used in
commercial lidar surveying—have long been offered with commercialized software
processing tools, and the workflow established in most mapping applications has become
widely accepted, and does not require specialists or customized software tools. Full
waveform lidar sensors, on the other hand, widely used for forestry research, are often
custom-built on a non-commercial basis and the data workflow and interpretation is very
much customized as well. As forestry research has mainly been focused on full waveform
lidar data analysis, there is a tendency to consider full waveform lidar as the de facto type of
airborne lidar used in forestry applications. This paper will show that commercially available
discrete return airborne lidar sensors are not only useful for vegetation mapping and forestry
applications, but also have evolved to a new level, which in some aspects, might be
comparable with full waveform technology traditionally used for forestry research.
P á g i n a | 240
Latin American Remote Sensing Week 2010
This paper focuses mostly on specific aspects of commercialized airborne lidar technology
relevant to forestry and other 3D mapping applications.
An airborne lidar system’s ability to map complex 3D vertical structures and generate high-
quality data comprising complex targets is determined by its hardware design. A vast
majority of airborne lidar sensors currently used in the industry, and particularly those used
in forestry research applications, can be categorized as either discrete return systems (DR) or
full waveform (FW) systems (Ussyshkin and Theriault, 2010b). They differ from one another
mainly with respect to the range sampling measurement method, which produces a more or
less detailed representation of vertical 3D structures in complex targets. The vertical
sampling of lidar systems relates to the number of range measurements recorded for each
emitted laser pulse. Discrete return lidar systems commercially available these days
generally allow for a few, typically four, multiple returns to be recorded for each emitted
laser pulse. A full waveform lidar measures the full profile of a return signal by sampling it in
fixed time intervals and providing a quasi-continuous distribution of the reflected energy for
each emitted laser pulse.
The information content of the returned laser signal also depends on characteristics of the
horizontal sampling area, which is typically a circular or elliptical laser footprint. The size of
the laser footprint on the ground, or a target close to the ground, is determined by the laser
beam divergence, and is directly proportional to the flying altitude. The distribution of the
laser pulse energy in the spatial and temporal domain within the illuminated area is
determined by the lidar system design and is critically important in determining the quality
of 3D mapping of complex targets required in forestry applications.
The most common type of commercial lidar sensors are small-footprint, discrete return
systems that record two to four returns for each emitted laser pulse. Some lidar system
manufacturers offer airborne lidar systems capable of both operational modes, where
conventional discrete return operation is provided by the main sensor, while full waveform
data collection is supported by an optional unit, which may or may not be used during data
collection missions (Optech, 2010).
Each data collection mode, whether full waveform or discrete return, has distinct advantages
and disadvantages that vary, depending upon the applications. Most conventional discrete
return systems can provide extremely high ground point density. This enables the high-
resolution representation of complex targets in the horizontal plane, but with a somewhat
P á g i n a | 241
Latin American Remote Sensing Week 2010
coarsely resolved elevation structure. These characteristics make the discrete return system
a perfect choice for mapping.
On the other hand, commercially available full waveform, small-footprint airborne lidar
systems capture full profiles of the laser backscattered energy for each emitted laser pulse as
a function of time (distance) with a typical sampling rate of 1 ns, which is equivalent to a one-
way distance of 30 cm, and results in a 15-cm range measurement rate. They provide much
more detailed information about the vertical structure of 3D targets, information that could
be used as the most valuable input for sophisticated scientific analysis, including deriving
target physical properties (Chauve et al., 2009a). It has been demonstrated that full waveform
data provide a more complete and accurate assessment of the surface, canopy and potential
obstruction detection than discrete return data (Magruder et al., 2010). Moreover, full
waveform lidar data capture gives the user much more flexibility and control in the data
processing and interpretation steps (Chauve et al., 2009b). However, dealing with full
waveform data sets takes lidar data management to a drastically higher level of complexity
compared with conventional 3D point cloud data. First, the volume of full waveform data is
overwhelming (Chauve et al., 2009b). Moreover, there are neither commercial nor open-
source toolkits to handle full-waveform lidar data, but only custom-made solutions typically
designed for specific sensors (Bretar et al., 2008). Therefore, handling, interpretation and
analysis of full-waveform lidar data is a very challenging and expensive task. This limits the
commercial use of full waveform lidar data, confining it mainly to research institutions.
On the other hand, the rich content of full waveform data acquired at a high sampling rate,
which is obviously important for analysis of complex 3D targets and forestry, has limited or
no value for broad area topographic mapping where only one or two discrete returns are
needed to create a digital elevation map or bare earth map of the surveyed terrain. That is
why the main application focus of full waveform technology remains in forestry and
vegetation mapping, and it remains unclear if it might be useful outside these applications
(Petrie, 2010)
P á g i n a | 242
Latin American Remote Sensing Week 2010
As mentioned above, the distribution of laser pulse energy in the spatial and temporal
domain within the illuminated area over the target is critically important for the quality of 3D
mapping of complex targets required in forestry applications, irrespective of whether the
sensor operates in FW or DR mode. Certain laser pulse characteristics including wavelength,
pulse width, shape and pulse energy, determine the penetration capabilities of the emitted
pulse, and some characteristics of the return signal, which are obviously important for
vegetation mapping. In addition, certain parameters in the sensor receiver electronics
influence the signature of the signal return so that the recoded lidar profile can, in fact,
represent a convolution of waveform characteristics of the lidar sensor itself with scattering
properties of the complex 3D target.
Therefore, the number of multiple returns in a discrete return sensor, or the sampling rate of
a full waveform sensor are not the only determinants to accurately map targets with a
complex vertical structure; lidar system parameters that determine signal penetration,
detection sensitivity, range precision, and minimal vertical target discrimination distance are
all equally critical factors. The latter parameter, which characterizes the ability of the lidar
sensor to resolve two separate targets in the vertical domain, is solely determined by the
lidar hardware design— mostly by characteristics of the emitted laser pulse and receiver
electronics—and it would be quite the same regardless of whether the measurements are
taken in discrete return mode or full waveform mode. Along with range precision and other
characteristics of the lidar sensor listed above, vertical target discrimination distance
determines the number of details and the quality of their representation in the data
comprising complex 3D targets, particularly, trees and vegetation.
In most commercial lidar systems the minimal vertical target discrimination distance is
about 2-3.5 m. This means that vertical structure details separated by any distance less than
this minimum, in general, cannot be resolved by consecutive range measurements in the
vertical domain, even if the data were collected by a full waveform unit integrated with the
main sensor.
Table 1 shows the distribution of multiple returns for high- and low-canopy vegetation in
several data samples collected by an older ALTM model operating in DR mode with a minimal
vertical target discrimination distance of about 2.5 m.
Table 1: Distribution of multiple returns for high and low-canopy vegetation for a DR lidar
with minimal vertical target discrimination distance of about 2.5 m
For a tall vegetation canopy of 22-27 m, all four discrete returns are recorded, including the
last one from the ground beneath the vegetation canopy; the percentage of third and fourth
P á g i n a | 243
Latin American Remote Sensing Week 2010
returns is substantial. This shows excellent pulse penetration capabilities in this system.
However, for low-canopy vegetation of 6-7 m, the system cannot detect all four returns
because the minimal vertical target discrimination limit is 2.5 m, and most third and fourth
returns are not recorded as the system cannot resolve the vertical vegetation structure
within a distance less than 2.5 m. This example also shows that for some traditional DR lidar
systems the number of returns from the vegetation canopy are often limited not only by the
density of vegetation, but also by the vertical target discrimination distance. That is why
some of the observations of low percentages in third and fourth returns from vegetation (Lim
et all, 2003) may have been the result of the system’s limited ability to resolve vertical
vegetation structure rather than the characteristic of an “optimal number” of echoes per
pulse in a DR lidar being limited to three. This example shows that the same discrete return
sensor can detect a substantial percentage of returns as third and fourth echoes for tall
vegetation but cannot do the same for medium-height vegetation because of limited
resolution in consecutive discrete return measurements in the vertical domain.
Until recently, the numbers characterizing minimal vertical target separation distances were
not typically specified in the data sheets of most commercial lidar systems, but could be
found in more detailed specification documents, or provided to users upon request. This
situation created some misunderstanding in the lidar community when the users of discrete
return systems expected to get multiple returns from low- or medium-height vegetation
without considering this parameter. Not accounting for the minimal target discrimination in
the vertical domain may also lead to misinterpreting multiple return data, and even gross
systematic errors caused by faulty interpolation.
With full waveform sensors, the high number of measured intervals per vertical target (i.e.,
high sampling rate) is often misinterpreted as the parameter that determines the amount of
detail present in the recoded profile of a vegetation canopy. Since the sensitivity of a full
waveform sensor is very low and the measurement sampling rate is very high, the user may
expect even small volumes of vegetation to influence the shape of the return waveform, and
small structural details to be detected (Lim et all, 2003). However, this applies only if small
volumes of vegetation changes are separated by a distance that the lidar sensor can resolve,
which is not always the case. The sampling rate of 1-ns (15 cm) typically offered by
commercial FW lidar sensors would generally not affect the capability of the sensor itself to
resolve two separate targets in the vertical domain, and the level of details in vegetation
structure recorded in the canopy profile. Very complicated decomposition algorithms applied
to full waveform data might help to go beyond the vertical resolution limit of the lidar system,
but it would still remain the most fundamental parameter in determining the resolution of
vertical structure details, not the sampling rate.
Since the minimum vertical target discrimination distance is one of the most important
performance parameters to characterize the ability of an airborne lidar system to represent
complex vertical 3D targets, especially in forestry applications, it is very important for users
to understand how it determines the number of details represented in the recorded data
profile, either in FW or DR mode.
P á g i n a | 244
Latin American Remote Sensing Week 2010
The following sections illustrate the importance of this parameter and its evolution over the
years, as well as the potential use of discrete return data for waveform analysis in forestry
applications.
Early commercial airborne lidar systems, such as Optech’s ALTM 1020, 1210 and 1225
models manufactured between 1993 and 1998, had the ability to capture only two returns
(first and last) for each emitted laser pulse. This feature, though seemingly modest compared
with today’s advanced lidar systems, already provided enriched information for sophisticated
analysis of both returns for applications such as feature extraction in forest (Hopkinson et al.,
2004; Roberts et al.,
2005) or urban areas
(Alharthy and Bethel, 2002). With the further evolution of lidar technology, more advanced
ALTM models capable of capturing four range and four intensity returns became
commercially available, and for the last decade, the maximum number of multiple returns per
emitted laser pulse has stabilized at this limit.
Figures 1-2 present typical examples of multiple return data for the ALTM 3100 and Gemini
systems. In both cases the laser beam penetrated through 16-20 m of vegetation, and the last
return with strong intensity clearly indicates the signal reflected from the ground.
An example of a four-return record for one emitted laser pulse with a minimum pulse
separation distance of 2.14 m.
P á g i n a | 245
Latin American Remote Sensing Week 2010
Figure 2. ALTM Gemini: An example of a four-return record for one emitted laser pulse
with a minimum pulse separation distance of 1.45 m.
Although these two examples represent leading-edge discrete return airborne lidar
technology, it is apparent that full waveform technology with a 1-ns sampling rate would
provide much more detailed information about the complex vertical structure of vegetation.
It is for this reason that, in the last decade, full waveform technology has been the only choice
for lidar applications requiring analysis of complex vertical targets with fine structure,
particularly, in forestry. However, the ALTM Orion, a new breakthrough in discrete return
lidar technology, has changed this situation.
The ALTM Orion represents a radical departure from previous generations of airborne lidar
instruments. First, the physical form factor—size, weight and displacement—has been
reduced by a whole order, making the Orion the first ultra-compact complete lidar solution
(Hussein et al., 2009). Second, the lidar data produced by the Orion has established a new
benchmark in the industry for data quality, accuracy and precision (Ussyshkin and Theriault,
2010a). It has been shown that the outstanding performance characteristics of the ALTM
Orion include a highly efficient system design that provides the best combination of
maximum area coverage rate, exceptional ground data accuracy and precision, and the sub-
centimeter resolution of data comprising small complex targets such as the thinnest wires in
power line corridors.
The third advantage provided by both ALTM Orion models is the revolutionary small
minimal pulse discrimination distance, which is of particular importance in complex target
mapping applications such as forestry and vegetation mapping. Figure 3 shows an example of
ALTM Orion-M data collected over 6-m high vegetation, with four returns for one emitted
laser shot and a minimum pulse separation of 73 cm.
P á g i n a | 246
Latin American Remote Sensing Week 2010
Figure 3. ALTM Orion-M: An example of a four-return record for one emitted laser pulse
with a minimum pulse separation distance of 0.73 m.
Such a small sub-meter pulse separation has never been achieved by any discrete return
lidar, and in combination with the system’s excellent ability to detect weak partial signal
returns from low-canopy layers of vegetation, it provides unprecedented data quality with
exceptionally rich content.
Figure 4 shows another advantage of the ALTM Orion’s exceptionally small pulse separation
distance. The data in this example was collected over a dense cornfield 2.2 m in height; the
ALTM Orion was still able to detect three consecutive pulse returns, with the last showing
strong intensity representing the ground return. Table 2 presents a statistical summary of
average and minimum pulse separation distances in multiple returns for three randomly
selected samples within the cornfield dataset.
The unique capability of the ALTM Orion to generate data so rich in content with fine sub-
meter elevation resolution from a dense cornfield enables the user to take data analysis to a
different level with highly accurate calculations of various attributes of vegetation properties
(biomass, height, density, structure, etc). This capability was not previously possible with
discrete return lidar data without using full waveform measurements. As one can see, these
sub-meter vertical target discrimination characteristics provide the quality of mapping
complex vegetation structures similar to that expected only from full waveform data.
P á g i n a | 247
Latin American Remote Sensing Week 2010
Figure 4. ALTM Orion-M: Three consecutive pulse returns with a minimal sub-meter pulse
separation are detected over a dense cornfield; the last return represents the signal
reflected from the ground.
P á g i n a | 248
Latin American Remote Sensing Week 2010
Figure 6 (a, b) gives another illustration of this evolution by showing images of vegetation
data collected by two ALTM models over the same area around the same time of the year; the
average canopy height is 15-17 m in both datasets; the type of vegetation is the same (mixed
forest).
a b
Figure 6 a shows data collected by an older ALTM model with a minimum vertical target
discrimination distance of about 2.5 m. Figure 6 b shows data collected over the same area by
an ALTM-Orion with a minimum vertical target discrimination distance of about 60-70 cm.
Both systems operated at the same laser pulse repletion frequency (100 kHz), and at an
2
altitude close to 1 km, which produced similar ground point density of 1.5–2.0 ppm . This
example clearly shows the advantage of the ALTM Orion’s much smaller vertical target
discrimination distance, which produces a much better representation of the vegetation
canopy structure as substantially more multiple returns are detected from the lower layers of
P á g i n a | 249
Latin American Remote Sensing Week 2010
vegetation. It is also important to note that there is virtually no gap between the ground and
the low canopy returns in ALTM Orion data that are typically seen in similar data images
produced by other discrete return lidar systems, either older or current models.
The next section shows that, by combining range and intensity information from multiple
return data collected by ALTM Orion, a simplified waveform analysis can be applied to
discrete return data in a way similar to that of full waveform data.
The simplified analysis of discrete multiple returns presented in this section is based on the
approach used for full waveform data analysis described by Chauve and co-authors (Chauve
et al., 2007). They used a parametric approach either with simple (three parameters) or
generalized Gaussian (four parameters) or a Lognormal function to model extracted relevant
peaks as echoes. Since the discrete return data used in our analysis represent the peaks of
partial signal returns, we assumed that the entire reflected laser pulse energy could be
decomposed into a sum of components while each one would be represented by a single
discrete return:
Figure 7 shows a graphic representation of our approach, where the peak of each discrete
return is modeled by a simple Gaussian (2) while a, µ, and δ were used as fitting parameters
so that the amplitude of each peak would be proportional to the recorded intensity value.
Furthermore, we assumed that the superposition (1) of all four simple Gaussian functions
representing the waveforms of the discrete partial returns would represent the total optical
receiver power Pr, which can be modeled through the lidar equation (Measures, 1984).
Considering partial signal returns Pi, the intensity of each one was modeled using the lidar
equation in the form derived by Jelalian (1992):
Here:
P á g i n a | 250
Latin American Remote Sensing Week 2010
Here the reflective properties of each target for each partial return Pi are described by the
backscattering cross-section ςi, which is proportional to the target reflectance ρi and the i-
fraction of the total received power Pr in each return:
Based on the approach described by equations (1-4), and using the known characteristics of
the emitted laser pulse and lidar system hardware, it was possible to model a waveform from
each discrete return (Figure 8-9) and estimate the effective reflectivity of consistent complex
vegetation targets such as cornfields and coniferous trees. This work is still in progress and
requires more detailed analysis, but the preliminary results partly presented here
demonstrate the huge potential of discrete return technology, the evolution of which has
achieved a level approaching in some aspects that of full waveform technology. The discrete
return data analysis described above resembles the procedures applied to full waveform data
analysis, and might be used in applications similar to those which to date, have been
considered as belonging solely to full waveform technology.
Figure 8. Illustration of the modeling for cornfield data collected by ALTM Orion-M re
Figure 4.
P á g i n a | 251
Latin American Remote Sensing Week 2010
The modeling of discrete signal profiles in vegetation data presented above could be
compared with the analysis of full waveform data collected over similar vegetation targets
(Wagner et al., 2004). An example of a coniferous tree profile with a total length of 35 ns
recorded with a 1-ns sampling rate showed three Gaussian-shaped peaks with a target
separation distance of about 1.5 m. Comparing these numbers with the ALTM Orion data
presented in Figure 3, one can conclude that discrete return lidar data of enhanced quality
can provide equivalent representation of vegetation structures at a level similar to that of full
waveform data. Another example (Wagner et al., 2004) of full waveform data collected over a
wheat field of 2.5 m height can be compared with the cornfield data collected by the ALTM
Orion (Figures 4 and 8), where three discrete return data with excellent target separation
characteristics provide equivalent or even better input for Gaussian modeling of the crop and
ground signals.
Figure 9. Illustration of the modeling for high-canopy vegetation data collected by ALTM
3100 re Figure 1.
This can be considered a potential fusion of two types of airborne lidar data on the
application side when similar approaches and tools are used for the analysis of both data
types. However, full waveform technology will continue to be essential and irreplaceable for
applications considering the analysis of pulse-broadening effects associated with laser beam-
target interaction and interception geometry (Schaer et al., 2007). In these cases the high
sampling rate of the FW approach would give a much more reliable result for modeling
waveforms and data analysis. Full waveform technology would also be a preferable choice for
forestry research applications that require the analysis of tall vegetation canopies (e.g.,
mature forest) where four returns offered by advanced DR lidar systems would not provide a
continuous profile of the vertical canopy structure. Figure 9 illustrates the limitations of the
discrete return approach to map the vertical structure of tall vegetation canopy. In this case,
only four multiple discrete returns would leave wide gaps in the mapped vertical canopy
profile, while FW technology would much better represent the vertical structure with a
continuous profile of the entire canopy. However, with further development of DR and data
handling technologies, when more than four discrete returns are supported by the lidar
system hardware, and the format of output data, it may be possible that the DR and FW
P á g i n a | 252
Latin American Remote Sensing Week 2010
approach will give similar results in terms of representing vegetation or forest canopy for a
wide variety of applications.
5. CONCLUSION
For the last two decades the use of airborne lidar technology in forestry applications has
focused mostly on analyzing canopy profiles recorded by full waveform sensors—either
custom-designed or commercial. The use of discrete return lidar has been limited mostly
because of its coarse resolution in the vertical domain. The evolution of discrete return lidar
technology has achieved a new level, with capabilities that approach those of full waveform
technology in forestry applications. The trade-off between the high complexity and costs
associated with handling full waveform data on the one hand, and conventional discrete
return data of enhanced quality on the other hand, has the potential to create a new
application niche in the lidar industry, and commercialize some aspects of forestry research.
In this niche, top-quality dense point clouds, with fully recorded intensity information for
each of the multiple returns, may provide sufficient information for modeling the received
waveforms and quantitative characterization of forest and vegetation.
Acknowledgements
The authors are grateful to Brent Smith, Eric Yang, Mike Sitar and Helen Guy-Bray for fruitful
discussions.
References
Aldred, A. and Bonner, M. 1985: Application of airborne lasers to forest surveys. Canadian
Forestry Service, Petawawa National Forestry Centre, Information Report PI-X-51, 62 pp.
Alharthy, A., Bethel, J., 2002. Heuristic filtering and 3D feature extraction from lidar data.
Proceedings of the International Archives of Photogrammetry, Remote Sensing and Spatial
Information Sciences, 34 (Part 3), Graz, Austria.
Arp, H., Griesbach, J., Burns, J., 1982. Mapping in tropical forests: a new approach using the
laser APR. Photogrammetric
Bates, P.D., Pappenberger, F., Romanowicz, R., 1999. Uncertainty and risk in flood inundation
modeling. In: K. Beven and J. Hall (Editors), Flood Forecasting. Wiley & Co, New York.
Bretar, F., Chauve, A., Mallet, C., Jutzi, B., 2008. Managing full waveform lidar data: a
challenging task for the forthcoming years. The International Archives of the
Photogrammetry, Remote Sensing and Spatial Information Sciences, 37 (Part B1), pp. 415-
420.
Chauve, A., Mallet, C., Bretar, F., Durrieu, S., Pierrot-Deseilligny, M., Puech, W., 2007.
Processing full-waveform lidar data: modelling raw signals. In: The International Archives of
P á g i n a | 253
Latin American Remote Sensing Week 2010
Photogrammetry, Remote Sensing and Spatial Information Sciences, Espoo, Finland, Vol. 36,
Part 3/W52, pp. 102–107.
Chauve, A., Vega, C., Bretar, F., Durrieu, S., Allouis, T., Pierrot-Deseilligny, M., Puech, W., 2009.
Processing full-waveform lidar data in an alpine coniferous forest: assessing terrain and tree
height quality. International Journal of Remote Sensing, 30 (19), pp. 5211-5228.
Chauve, A., Bretar, F., Pierrot-Deseilligny, M., Puech, W., 2009. Full Analyze: A research tool
for handling, processing and analyzing full-waveform lidar data, Proceedings of the IEEE
International Geoscience and Remote Sensing Symposium (IGARSS), Cape Town, South Africa.
Dubayah, R.O. and Drake, J.B. 2000: Lidar remote sensing for forestry. Journal of Forestry 98,
44–46.
Evans, J.S., Hudak, A.T., Faux, R., Smith, A.M.S., 2009. Discrete return lidar in natural
resources: Recommendations for project planning, data processing, and deliverables, Remote
Sens., 1, pp. 776-794.
Hopkinson, C., Sitar, M., Chasmer, L., Treitz, P., 2004. Mapping snowpack depth beneath forest
canopies using airborne lidar. Photogrammetric Engineering & Remote Sensing, 70(3), pp. 323-
330.
Hudak, A.T., Evans, J.S., Smith, A.M.S., 2009. Review: LiDAR Utility for Natural Resource
Managers. Remote Sens., 1, pp. 934-951.
Hussein, M., Tripp, J., Hill, B., 2009. An ultra compact laser terrain mapper for deployment
onboard unmanned aerial vehicles, Proc. SPIE, Vol. 7307, 73070B.
Jelalian, A.V., 1992. Laser Radar Systems. Artech House, Boston, Massachusetts.
Jutzi, B., Stilla, U., 2006. Range determination with waveform recording laser systems using a
Wiener Filter. ISPRS Journal of Photogrammetry and Remote Sensing, 61, pp. 95-107.
Korpela, I., Ørka, H.O., Maltamo, M., Tokola, T. & Hyyppä, J., 2010. Tree species classification
using airborne LiDAR – effects of stand and tree parameters, downsizing of training set,
intensity normalization, and sensor type. Silva Fennica, 44(2), pp. 319–339 and previous
publications.
Lefsky, M.A., Harding, D., Cohen, W.B., Parker, G. and Shugart, H.H. 1999a: Surface LiDAR
remote sensing of basal area and biomass in deciduous forests of Eastern Maryland, USA.
Remote Sensing of Environment 67, 83–98 and Lefsky, M.A., Cohen, W.B., Acker, S.A., Parker, G.G.,
Spies, T.A. and Harding, D. 1999b: Lidar remote sensing of the canopy structure and biophysical
properties of Douglas-fir western hemlock forests. Remote Sensing of Environment
70, 339–61
P á g i n a | 254
Latin American Remote Sensing Week 2010
Lim, K., Treitz, P., Wulder, M., St-Onge, B., Flood, M., 2003. LiDAR remote sensing of forest
structure, Progress in Physical Geography, 27(1), pp. 88–106
Magruder, L.A., Neuenschwander, A.L., Marmillion, S.P., Tweddale, S.A., 2010. Obstruction
detection comparison of small-footprint full-waveform and discrete return lidar, Proc. SPIE, Vol.
7684, 768410
Means, J.E., Acker, S.A., Harding, D.J., Blair, D.B., Lefsky, M.A., Cohen, W.B., Harmon, M.E. and
McKee, W.A. 1999: Use of largefootprint scanning airborne LiDAR to estimate forest stand
characteristics in the Western Cascade of Oregon. Remote Sensing of Environment 67, 298–
308
Measures, R.M., 1984. Laser Remote Sensing, Fundamentals and Applications. Wiley
Interscience, New York.
Nelson, R., Krabill, W. and Maclean, G. 1984: Determining forest canopy characteristics using
airborne laser data. Remote Sensing of Environment 15, 201–12
Nelson, R., Parker, G., Hom, M., 2003. A Portable Airborne Laser System for Forest Inventory,
Photogrammetric Engineering & Remote Sensing, 69(3), pp. 267–273.
Roberts, S.D., Dean, T.J., Evans, D.L., McCombs, J.W., Harrington, R.L., Glass, P.A., 2005.
Estimating individual tree leaf area in loblolly pine plantations using LiDAR-derived
measurements of height and crown dimensions, Forest Ecology and Management, 213(1-3),
pp. 54-70.
Schaer, P., Skaloud, J., Landtwing, S., Legat, K., 2007. Accuracy estimation for laser point cloud
including scanning geometry. ISPRS - The 5th International Symposium on Mobile Mapping
Technology, Padua, Italy, May 29-31.
Ussyshkin, V., Sitar, M., 2009. Applications and Benefits of Airborne Lidar Technology for
Transmission Line Asset Management, CIGRE Canada Conference on Power Systems, Toronto,
Ontario, October 4-6 (on CDROM).
Ussyshkin, V., Theriault, L., 2010a. Precise mapping: ALTM Orion establishes a new standard
in airborne lidar performance, ASPRS Annual Conference, San-Diego, California, April 26-30
(on CDROM).
P á g i n a | 255
Latin American Remote Sensing Week 2010
Ussyshkin, V., Theriault, L., 2010b. Ussyshkin, V., Theriault, L., 2010. ALTM-Orion: Bridging
conventional lidar and full waveform digitizer technology. ISPRS TC VII Symposium – 100
Years ISPRS, Vienna, Austria, July 5-7. IAPRS, 38(7B), pp. 606-611
Wagner, W., Ullrich, A., Melzer, T., Briese, C., Kraus, K., 2004. From single-pulse to full-
waveform airborne laser scanners: Potential and practical challenges. Proceedings of the
International Society for Photogrammetry and Remote Sensing 20th Congress, 35(Part B/3)
6-12, Commission 3, Istanbul, Turkey, July 12-23.
Wagner, W., Ullrich, A., Ducic, V., Melzer, T., Studnicka, N., 2006. Gaussian decomposition and
calibration of a novel small-footprint full-waveform digitising airborne laser scanner. ISPRS
Journal of Photogrammetry and Remote Sensing, 60(2), pp. 100-112
Wagner, W., Hollaus, M., Briese, C., Ducic, V., 2008. 3D vegetation mapping using small-
footprint full-waveform airborne laser scanners, International Journal of Remote Sensing
archive, 29(5), 3D Remote Sensing in Forestry, pp. 1433-1452.
Wu, J., van Aardt, J., Asner, G.P., Mathieu, R., Kennedy-Bowdoin, T., Knapp, D., Wessels, K.,
Erasmus, B.F., Smit, I., 2009. Connecting the dots between laser waveforms and herbaceous
biomass for assessment of land degradation using small-footprint waveform lidar data.
Proceedings of the International Geoscience and Remote Sensing Symposium (IGARSS-2009),
Cape Town, South Africa
P á g i n a | 256
Latin American Remote Sensing Week 2010
Contacto:
viviana.barrientos@saf.cl
Servicio Aerofotogramétrico
Para consultas:
lars@saf.cl
www.lars.cl
www.saf.cl
P á g i n a | 257
www.lars.cl