Вы находитесь на странице: 1из 111

DIGITAL TERRAIN

MODELING
Dr. S.K. GHOSH
Department of Civil Engineering
IIT Roorkee
scangfce@iitr.ernet.in

INTRODUCTION
Terrain information is essential to
applications in many disciplines:
Geography
Land surveying
Civil engineering
Landscape architecture
Earth environmental
Resource planning & management

Digital terrain modeling may be


approached in different ways.

The mathematical approach

aims to develop algorithms for terrain


representation, contour interpolation and
3D visualization.

The non-mathematical approach,

focus more on the applications of digital


terrain modeling techniques.

DEFINITIONS AND TERMINOLOGY

The term Digital Terrain Model (DTM)


was first introduced by Miller and La
Flamme (1958), who defined it as

a statistical representation of a
continuous surface of the ground by a
large number of selected points with
known X, Y and Z coordinates defined by
an arbitrary coordinate field.

Several terms have been coined to


describe
methods
and
processes
pertaining to digital terrain data.

These include

digital elevation model,


digital terrain elevation data,
digital terrain model, and several others.

Digital Elevation Model (DEM).

The world elevation in DEM is the


measurement of height above a datum.
It implies to the absolute altitudes or
elevations of the points contained in the
data.
The term digital elevation model and
its acronym DEM are particularly widely
used in North America to refer to data
sets
containing
regularly
spaced
elevations in the form of a lattice or
square grid.

Digital Terrain Elevation Data


(DTED).
This term is used primarily as a product
name for elevation data in the grid format
that are generated by the former United
States Defense Mapping Agency now a
part of the National Imagery and Mapping
Agency (NIMA).

DTM is a multi-step process that is made


up of the following sequence of tasks (Fig
1) Real world
Processing
Processing
Data
Dataacquisi
acquisi
tion
tion

Analysis
Analysis

DTM
DTM

Application
Application

Visualization
Visualization

Fig 1. The main tasks of digital terrain modeling in GIS.

DEM and TINS


Surface representation in GIS

Data types

Vector
[data models] A coordinate-based data model that
represents geographic features as points, lines, and
polygons. Each point feature is represented as a single
coordinate pair, while line and polygon features are
represented as ordered lists of vertices. Attributes are
associated with each vector feature, as opposed to a raster
data model, which associates attributes with grid cells.

Raster
[data models] A spatial data model that defines space as an
array of equally sized cells arranged in rows and columns,
and composed of single or multiple bands. Each cell
contains an attribute value and location coordinates. Unlike
a vector structure, which stores coordinates explicitly,
raster coordinates are contained in the ordering of the
matrix. Groups of cells that share the same value represent
the same type of geographic feature.

Vector data is represented by 3 geo-entities each of which have a unique


spatial dimension

Point
0D has a location, but no length

Line
1D an object having length composed of two or more 0D objects
Length
Direction
Intersection

Polygon
2D - an object having length and width bounded by at least three 1D line
segments
Area
Perimeter
Overlap
Adjacency

Vector data is discrete data

Used to represent features on a topographic


(planar) map

Roads
Hydrology
Vegetation
Fish locations
Etc.

NTS Map

How a surface is represented by


vector data

The contours on a topographic map shows


the relief of an area.

A hill is represented by lines of equal


elevation above mean sea level

Contours never cross

How a surface is represented by


vector data

Elevation values are printed


in several places along the
line

Steep slopes are represented


by closely spaced contours

Level ground is represented


by widely spaced contours

How a surface is represented by


vector data

The steepness or flatness of the


terrain will determine the contour
interval.
Flat ground will have less of a contour
interval.
E.g. 1m, 5m

Mountainous areas will have larger


contour intervals.
E.g. 100m, 1000m

Contour interval on a map is


determined by scale.

Digital Elevation
Model
DEM

What is a DEM

Digital representation of topography.

Cell based with a single elevation


representing the entire cell.

Digital Surface Models (includes buildings and vegetation)


Digital Terrain Models (a somewhat more accurate term)

From planer to 3D

DSM of a city

DEM data sources


a. DEMs by digitising contours

digitising contours on maps into contour layers (eg NTS maps to NTDB layer). This
is 'second hand', as the lines are abstract and may result in artifacts in the model.

stereo photos -> contour lines -> digitised lines -> interpolate to raster
GRID
b. Digital Stereo photogrammetry:
From the real surface, as captured from aerial photographs (mass points)

stereo photos -> mass points -> interpolate to raster GRID

DEM data versus DEMs


Mass points or Contour layers can be used to:
store elevation data and to create DEMs
. but they are not DEMs themselves

Lines

Points

Raster GRID (DEM)

How DEM data is stored

DEM is a matrix of elevations with a uniform


cell size

How DEM data is stored

Xmax and Ymax

Cell index number x


cell size defines
position relative to
Xmin, Ymin and Xmax,
Ymax and infers an
exact location

Xmin and ymin are in projected


units

1
2

m
3

4
5
6a

f
i

f
h

z
b

Interpolation in DEM and TIN.

In the fig, for example, the estimated


elevation values for m (Zm) and (Zn) in
the DEM and TIN may be calculated as
Zm = (Z3b+Z4b+Z3c+Z4c)/4

Where Z3b, Z4b , Z4c and Z3c are elevation


values for grid intersection 3b, 4b, 3c,
and 4c respectively and
Zn=(Zc+Zb+Zc)/3
Where Zc, Zb and Zc are elevation values
for vertices c, b, and e respectively.

In practice, however, a surface-fitting


algorithm is commonly used to improve
the result of terrain modeling.
A surface-fitting algorithm may be
linear or nonlinear, depending on the
order of the polynomial equations used
for estimating the elevation of the
required point.
The general polynomial equation for
surface fitting is given in Table where zi
is the elevation value of an individual
point i; xi and yi are its coordinates, and
a0, a1, a2,. are coefficients of the
polynomial.

Individual Terms
Z=aa
+a1x+a2y
+a3x2+a4y2+a5xy
+a6x3+a7x3+a8x2y+a9xy2
+a10x4+a11y4+a12x3y+a13x2y2+a14xy2
+a15x5+

Other of Descriptive No. of


Terms
Terms
Term
s
Zero
First
Second
Third
Fourth
Fifth

Planar
Linear
Quadratic
Cubic
Quartic
Quintic

1
2
3
4
5
6

To find the elevation values for point m


in Fig using linear interpolation, for
example, the first step is to set up a set
of equations to solve for the coefficients
a0, a1, and a2.
Z3b=a0+a1x3b+a2y3b
Z4b=a0+a1x4b+a2y4b
Z3c=a0+a1x3c+a2y3c
Z4c=a0+a1x4c+a2y4c

Since there are more equations than the


number of unknowns, this is a leastsquares solution.

Once the values of a0, a1, and a2 have


been determined, the elevation value of
m can be found by
Zm=a0+a1xm+a2ym
where xm and ym are known coordinates
of point m.

This method of interpolation is known as


the local approach.

The global approach of interpolation


removes such a shortcoming of the local
approach by utilizing all or most of the
elevation data to characterize the
surface at a point, thus allowing
prediction to be made on the trend of
the surface.
Fig.
illustrates
conceptually
the
difference
between
these
two
approaches of interpolation.
For the terrain point p, the estimated
elevation is h1 when interpolated locally
using only the values of the two
neighboring points 3 and 4 (Fig a ).
If interpolated globally using the values of
points 1 through 6 (Fig b), the estimated
elevation value of p become h2, which is
greater than h1.

p 4

h1

(a) Local interpolation, elevation of point


p = h1
p 4
3
5
6
2
1
h2
(b) Global interpolation with surface
fitting, elevation of point p = h2,
1 where h2,>h1 in (a) 6
5
2
3

4
p
h3

(c) Global interpolation with surface


fitting, elevation of point p = h3, where
h3<h1 in (a)
Globally interpolated surfac
Locally interpolated surface

The estimated elevation values of p will


be the same in (Fig b) and (c) if
interpolated locally, assuming that the
neighboring data points used in both case
have the same elevation values.
If the effects of non-neighboring points
are taken into account in global
interpolation, the resulting elevation
values of p in the two cases will be
different.
The examples of Fig (b) and (c) clearly
demonstrate that the estimated elevation
value of a terrain point is affected not
only by its immediate neighbors, but also
by those points that are located away
from it.

Therefore,
global
interpolation
approaches
obviously
have
the
advantage
of
preserving
terrain
continuity and smoothness.

However, as terrain modeling always


involves a very large number of points,
global methods are computationally very
demanding.

A widely used global surface-fitting


method is trend surface analysis.

In this method the terrain surface is


approximated
by
a
polynomial
expansion of the coordinates of the
sampled points.

The coefficients of the polynomial


function are determined by solving a set
of
simultaneous
equations,
which
include the sums of powers and cross
products of the x, y, and z values, by
the method of least squares.
This ensures that the sum of the squared
deviations from the trend surface is a
minimum.
Once the coefficients have been
computed, the polynomial function can
be evaluated at any point within the
geographic space under investigation.

Commonly used polynomials are


as follows:
Linear
Z = a+bx+cy
Quadratic Z = a + bx + cy + dx 2 + exy +
fy2
Cubic
Z=
a+bx+cy+dx2+exy+fy2+gx3+hx2y
+ixy2+jy3
where Z is the estimated height, x and y
are geographic coordinates, and a, b, c,
., j are polynomial coefficients (next Fig
).

(a
)

(b
)

(c
)

Graphical representation of the three forms of trend


surface produced by (a) linear (b) quadratic and (c)
cubic polynomial equations.

Trend surface analysis is also a useful


data exploratory tool.

It is relatively simply to create a grid


matrix of values by substituting the
coordinates of the grid nodes into the
polynomial and calculating an estimate
of the surface for each node.

Inverse distance weighted (IDW)


interpolation
is
the
simple
local
interpolation method most commonly
used.
it is usually implemented in the form of
moving window to define the zone of
influence.
In vector-based processing, a circular
moving window is often used instead.
This method is based on the assumption
that the influence of each input on the
interpolated value at the center of the
window is inversely proportional to a
power (p) of its distance from the center
as expressed in the following formula

z(u 0 )

n
p
z ( u i )d ij
n

i 1

p
d ij
1

Where z(uo) is the estimated value of the point


at an unsampled location, z(ui) are the data
points, dij is the distance between each data
point at an unsampled location and p is a
parameter.
Typically, p = 2; in other words, the weights are
usually inversely proportional to the squared
distance between the data point and the point
at the unstampled location.
The quality of the resulting surface, as indicated
by the fitting between the surface and the input
points, is dependent on the density, distribution
and accuracy of the input points.

Moving window
Sample
point
Point where
value is
estimated

Distance
between
sample point
and
center of
moving window

Moving average interpolation using inverse distance


weighting

A shortcoming of this interpolation


method is that it does not reproduce the
local shape implied by the data and
tends to produce local extreme at the
data points.

Just like trend surfaces, IDW surfaces


seldom go the through the input points.

In other words, IDW is not an exact


interpolation method.

Finally,
an
optimum
local
interpolation method know as kriging,
originally developed for ore reserve
estimation in geostatistics is now widely
used in DTM software packages because
it is flexible and can handle any type of
data.

Kriging, named after the pioneering work of


Danie Krige (1951), is a generic name for a
family of least-squares linear regression
algorithms that are used to estimate the
value of a continuous attribute (such as
terrain height) at any unsampled location
using only attribute data available over the
study area.

Kriging treats the continuous attribute to be


interpolated as a regionalized variable.

In one approach known as simple


kriging, it is assumed that the expected
value of a random variable at a specified
location (i.e., the estimate) equals the
mean value in the study area.
In this case, a variogram which
represents the relationship between the
mean-square difference between sample
values and their distances, can be used
to determine the value at the unsampled
location.
Mathematically, the variogram (2r) or
semi-variogram (r) is defined by.
N N
r [z(u i d) z( u i )]2
2 i 1

where d is the distance between two


sample points z(ui), i = 1,2,., N represent
the z data available over the study area
say the N data and N is the total number of
data points.
Graphically this can be represented in Fig
below.
Squared
difference

C0

Sill
C1
Range

An example of a simple variogram

Scale in DEMs

Scale determines resolution (cell size)


This depends on the source data

Resolution determines use of DEM and what


spatial features are visible

COPG DEM shows all buildings and features


were extracted and turned into polygons.

Uses of a DEM

Determine the characteristics of terrain

Slope
Aspect
Viewshed
Hillshade
Watershed
Drainage networks
Stream Channels

Estimating slopes in a DEM


Slope are calculated locally using a
neighbourhood function based on a moving
3x3 window
Distances are different in horizontal and
vertical directions vs diagonal

1.41

1.41

1.41

1.41

Slopes
elevations

Difference/distan
ce

340

335

330

8/42.47

3/30

2/42.47

337

332

330

5/30

-3/30

330

328

320

-2/42.47

-2/30

12/42.27

Aspect
Once slope is determine aspect can be
determined
Aspect can be thought of as the slope
direction. The values of the output raster
will be the compass direction of the aspect.
Aspect is expressed in positive degrees
from 0 to 359.9, measured clockwise from
north.

Viewshed

What can be seen from a specific viewpoint

Hillshade

Degree of illumination =
315

Triangulated Irregular
Network
TIN

Triangulated Irregular Networks (TIN):

vector polygons

A series of triangles capturing the topography .. x, y, z at nodes


Each triangle has a defined slope and direction (aspect)
The development of the TIN model is often attributed to Dr. Tom Poiker (SFU)
Advantages: variable data density depending on landscape
significant points or lines can be encoded e.g. peaks, ridges, valleys
Disadvantages: more complex, needs more processing to generate
triangle facets are often evident in processing
Summary: most DEM data today are stored and processed as raster GRIDS
Surface models can be either (raster) DEMs or TINs

CHARACTERISTCS OF THE TIN


In the TIN model, the terrain is recorded
as a continuous surface made up of a
mosaic of non-overlapping triangular
facets formed by connecting selectively
sampled elevation points using a
consistent
method
of
triangle
construction.
Mosaic of nonoverlapping planer triangular
facets
A vertex
An edge

Fig 4. Elements of a triangulated Irregular Network (TIN).

For simpler mathematical interpolation,


most TIN models assume planar
triangular facets defined by three edges
(sometimes referred to as legs).
Each edge is bounded by two vertices
(sometimes referred to as nodes).
In a TIN, these edges depict linear
terrain features (e.g., breaks, ridges,
and river channels) while the vertices
describe nodal topographic features
(e.g., peaks, pits, and passes).

a)

b)

In addition to the difference in the


method of sampling elevation points,
the TIN data model is distinct from the
DEM data model in two important
ways.
Each and every sample point in a TIN
has an (x, y) coordinate and an
elevation or Z where as in DEM, value
(the locations of elevation points are
implicit in the data model).
The TIN data model may include
explicit
topological
relationships
between points and their proximal
triangles.

Topological
relationships
play
a
significant role in the TIN data model.
By
building
these
topological
relationships, the totally unstructured
elevation points as collected are
converted into a properly organized
geographic database suitable for terrain
modeling applications.
The process of triangulating thousands
of discrete points is no trivial task, both
conceptually and computationally.

As illustrated below, a given set of


elevation points may be triangulated in
many ways.
+120

+90

+90

+100

+120

Elevation points
.. Contour Lines
+

+120

(a) Elevation Points in TIN

(b) Different ways of triangulating the elevation points

(c) Contour lines interpolated forms TIN in (b)

The impact of different ways of triangulation on the


application of a TIN for contour interpolation.

If the resulting TINs are used for


contouring, very drastically different
maps will be generated.

The primary requirement in the TIN data


model, therefore, is to develop the
necessary procedures that will ensure
the producing of a unique TIN for a
given set of elevation points.

Delaney triangulation commonly used


method of creating a network of triangles
from sampled points over a particular
geographic space for a TIN as it allows
well-shaped triangles to be generated.
It is closely associated with proximal
regions
also called as Nearest
Neighborhood
regions,
Thiessen
polygons, or Voronoi polygons.
Proximal regions are created as the result
of subdividing a particular geographic
domain or space encompassing a set of
points into a set of convex polygons.

Given three points P1, P2, and P3 in a


plane s (Fig a), it is possible to partition
the plane using the perpendicular
bisectors to the segments P1P2, P2P3, and
P1P3, into three proximal regions V1, V2,
and V3.
P1

P1

V1

P3

P3

P2

Plane S

(a) Points in a
plane

P2

V2

V3

(b)
Subdivision
into
proximal
regions

Each of these regions contains one of the


original points called anchor points (Fig
b).

By definition, a proximal region is a


polygon within which every point is
closer to its own anchor point than the
anchor points of all other regions.

This means that any point located in V1,


for example, is closer to P1 than to P2
and P3.

The process of systematically subdividing a


geographic domain in this way is called as
Dirichlet
tessellation
in
computational
geometry.
The resulting surface is known as a Voronoi
diagram or Thiessen polygons (Fig c ).

(c) A Voronoi diagram

If all the pairs of anchor points sharing a


common edge of a polygon in the Voronoi
diagram are connected, a network of triangles
may be obtained (Fig d).
The process of creating a network of triangles in
this way is known as Delaunay triangulation.

(d) Delaunay
triangulation

The triangles obtained by this method have a


unique property, i.e., the circumcircle that passes
through the vertices of a particular triangle contains
no other point in its interior.

(e) Delaunay triangles and the empty-circle criterion

This property, which is called the empty-circle


criterion,
is used as
a mechanism to
automatically construct a TIN from a set of points
in Delaunay triangulation.
Delaunay triangulation may be applied to 2D data
as well as 3D data.

Creation of DEM

Creation of DEMs
Conversion of contour lines
Photogrammetry
Satellite Stereo
Radar Stereo
Radar Inferometry
Laser Altimetry

GLY560: GIS and RS

7/19/16

Conversion of printed contour


lines
Existing map plates are scanned
Resulting raster is vectorized and edited
contours are "tagged" with elevations
Additional elevation data are created from
the hydrography layer (e.g. shorelines
provide additional contours)
Algorithm is used to interpolate elevations
at every grid point from the contour data

GLY560: GIS and RS

7/19/16

Aerial photography and


Lidar data collection
collect DEM information
using various digital
techniques.

Photogrammetry
Manually: an operator looks at a pair of
stereophotos through a stereoplotter and
must move two dots together until they
appear to be one lying just at the surface of
the ground
Automatically: an instrument calculates the
parallax displacement of a large number of
points (e.g. for USGS 7.5 minute quadrangles,
the Gestalt Photo Mapper II correlates
500,000 points)
Correction of elevation from photographs:
water bodies are assumed to be flat.

GLY560: GIS and RS

7/19/16

Stereo Satellite

7/19/16

Two satellite
passes are
combined to
get effective
stereo view

GLY560: GIS and RS

Stereo Radar

7/19/16

Works like
photogrammetry
but use radar wave
instead.
Can be done from
space or airborne
with side-lookingairborne-radar
(SLAR)
Can penetrate
vegetation canopy
GLY560: GIS and RS

DEM sources: global

GTOPO30 ~1993 1km

http://edc.usgs.gov/products/elevation/gtopo30/gtopo30.html

SRTM (Shuttle Radar Topographic Mission): 90 metre pixels, 60S-60N

Shuttle Radar Topography


Mission (SRTM)
1 arc-second elevation
data for the United
States, 3 arc-second
data for the globe
Produced by radar
measurements from a
Shuttle mission, Feb
11-22, 2000
http://srtm.usgs.gov/d
ata/obtainingdata.html

DEM

83

DEM

84

Santa Barbara, California

http://srtm.usgs.gov/srtmimagegallery/index.html
DEM

85

San Andreas Fault, California

http://srtm.usgs.gov/srtmimagegallery/index.html
DEM

86

Salt Lake City, Utah

http://srtm.usgs.gov/srtmimagegallery/index.html

DEM

87

Mt Kilimanjaro, Tanzania

http://srtm.usgs.gov/srtmimagegallery/index.html

DEM

88

Interferometry used by
SRTM

In interferometry,
two images are
taken from different
vantage points of
the same area. The
slight difference in
the two images
allows scientists to
determine the
height of the
surface.

DEM

89

Radar Interferometry
Use phase
difference in two
radar signals to
measure elevation
differences
Signals are from two
receivers so called
synthetic aperture
radar or SAR.

7/19/16

GLY560: GIS and RS

Laser Altimetry

7/19/16

Fly laser over area, and time


reflection of laser.

GLY560: GIS and RS

Available Resolutions of
DEM

GLY560: GIS and RS

7/19/16

7.5-minute USGS DEM


Resolution:
10 m or 30 m
Coverage:
7.5 x 7.5 min Quad (1:24,000)
Vertical Accuracy:
7m

GLY560: GIS and RS

7/19/16

10m and 30m DEM


Availability

GLY560: GIS and RS

7/19/16

7.5-minute USGS DEM Data


Classification Levels:
Level-1 Scanned from National High Altitude
Photography (NHAP)/NAPP photography. A vertical
RMSE of 7 meters is the desired accuracy standard.
(Most 7.5 DEMs are Level-1)
Level-2 Processed or smoothed for consistency and
edited to remove systematic errors. A RMSE of 1/2
contour interval.
Level-3 DEMs are derived from DLG data by
incorporating hypsography (contours, spot
elevations) and hydrography (lakes, shorelines,
drainage). A RMSE of 1/3 contour interval.

GLY560: GIS and RS

7/19/16

30-minute USGS DEM


Resolution:
2 arc seconds (~60m)
Coverage:
30x30 min block ( 1:100,000)
Vertical Accuracy:
25 m

GLY560: GIS and RS

7/19/16

1- degree USGS DEM


Resolution:
3 arc seconds (~100m)
Coverage:
1x1 deg block ( 1:250,000)
Vertical Accuracy:
25 m

GLY560: GIS and RS

7/19/16

Seamless DEM Data

Buy DEM from USGS already in mosaic form


by National Mapping Program

GLY560: GIS and RS

7/19/16

Global DEMs
GTOPO30
Aster
Shuttle Mapping Program
IKONOS

GLY560: GIS and RS

7/19/16

GTOPO30
Resolution:
30-second (~1 km)
Coverage:
50 deg lat x 40 deg long
Vertical Accuracy:
30 m

GLY560: GIS and RS

7/19/16

ASTER DEM
Advanced Spaceborne Thermal Emission
and Reflection Radiometer
Off-Nadir pointing allows DEM
First international DEM of decent quality
available.
Must request generation of DEM (slow
turnaround) but FREE!

GLY560: GIS and RS

7/19/16

Aster DEM

ASTER
(30 m2/pixel)

GTOPO30
(1 km2/pixel)

7/19/16

GLY560: GIS and RS

DEM info

C. Direct image grid DEM (~ new millenium)

Satellite sensors creating DEM grids directly have multiplied since 2000

D. LiDAR

e.g. (Stonehenge)

http://www.youtube.com/watch?v=pzYUx4l80m8

DEM grid

(.tif format or ESRI grid)


<- displayed as grayscale
elevation in metres

Hillshading

(NW light source 315. 45)

Stereo Air Photos (-> contours)

->

DEM

->

Shaded relief

Contour lines can be interpolated from DEM heights or pre-exist from digitised maps, and are a standard layer in digital databases
and online web mapping for display.

DEM data availability


DEMs have been created at a variety of scales by different agencies.
Some can be downloaded free

AGENCY

SCALE

TYPICAL
RESOLUTION
(metres)

Municipal

1: 5,000

1(usually contours)

Provincial

1: 20,000

25

BC TRIM (see lrdw.ca)

Federal

1: 50,000

50

-> www.geobase.ca

Federal

1: 250,000

Global

1: 1,000,000

500-1000

SRTM

1: 100,000

90

200

DEM derived layers

DEM elevation (usually) in metres

Calculated in degrees or %
(between adjacent pixels or from TIN)
Slope (degrees) = angle opposite rise/run
0 (flat) -> 90 (vertical);.. 1 to 1 slope = 45
Slope (percent) = rise/run * 100
0 (flat) to (cliff); .. 1 to 1 slope = 100%

The compass direction in which a slope is


facing, measured by azimuth (0-360)
Cardinal directions are 90 (E), 180 (S),
270 (W), 0 and 360 (N).
Flat slopes are given a unique integer for
aspect (e.g. -1 or 9999);

3D perspectives and fly-throughs


e.g. Google Earth and ArcGIS ArcScene
User specifies:
> viewer position
> azimuth / vertical angle of view
> vertical exaggeration
> 'draped' layers e.g. imagery,
map scan, vector layers
terrainmap and virtual terrain project

ASTER image: Svalbard, Norway

(15 metre resolution)

Ikonos 4metre resolution

Whistler Brandywine Meadows

Whistler Brandywine Meadows (terrain + shaded relief)

THANK
YOU

Вам также может понравиться