Академический Документы
Профессиональный Документы
Культура Документы
cn
International Journal of Remote Sensing
Vol. 29, No. 14, 20 July 2008, 4269–4276
Letter
H. XU*
College of Environment and Resources, Fuzhou University, Fuzhou, Fujian 350108,
China
A new index derived from existing indices – an index-based built-up index (IBI) –
is proposed for the rapid extraction of built-up land features in satellite imagery.
The IBI is distinguished from conventional indices by its first-time use of
thematic index-derived bands to construct an index rather than by using original
image bands. The three thematic indices used in constructing the IBI are the soil
adjusted vegetation index (SAVI), the modified normalized difference water
index (MNDWI) and the normalized difference built-up index (NDBI).
Respectively, these represent the three major urban components of vegetation,
water and built-up land. The new index has been verified using the Landsat
ETM + image of Fuzhou City in southeastern China. The result shows that the
IBI can significantly enhance the built-up land feature while effectively
suppressing background noise. A statistical analysis indicates that the IBI
possesses a positive correlation with land surface temperature, but negative
correlations with the NDVI and the MNDWI.
1. Introduction
Rapid development of urban areas has witnessed replacement of natural vegetation
cover with buildings and paved surfaces. This has brought about many negative
environmental repercussions to the world, for example, less precipitation, more
dryness and higher temperatures (Kaufmann et al. 2007), which contribute greatly
to global warming. The ability to monitor the built-up land dynamics and changes in
the urban extent in a timely and cost-effective manner is highly desirable for local
communities and decision makers alike. Fortunately, satellite remote sensing
technology offers considerable promise to meet this requirement and satellite
imagery has been used to discriminate built-up lands from non-built-up lands for the
last few decades. A popular method for the definition of built-up land areas began
with conventional multi-spectral classification. However, this may not produce
satisfactory accuracy, normally less than 80%, due to spectral confusion of the
heterogeneous built-up land class. Therefore, many studies have not only used a
single classification method to extract the built-up lands, but have also combined
different methods to improve the extraction. Masek et al. (2000) identified urban
built-up lands in the metropolitan Washington DC area, based on a normalized
difference vegetation index (NDVI) differencing approach with the assistance of an
unsupervised classification and achieved an overall accuracy of 85%. Xu (2002) and
Xian and Crane (2005) used logic tree algorithms to extract built-up land
information, but were only able to achieve an accuracy of around 85%. To date,
there are few simple processes of automatically mapping built-up lands, such as
using indices. The normalized difference built-up index (NDBI) of Zha et al. (2003)
is the only one proposed for this purpose. The development of the index was based
on the spectral response of built-up lands that have higher reflectance in the middle
infrared (MIR) wavelength range (such as TM 5), than in the near infrared (NIR)
wavelength range (see equation (1) below). However, studies have shown that the
reflectance for certain types of vegetation over the band pass of TM 5 increased as
leaf water content decreased (Cibula et al. 1992, Gao 1996), and the drier vegetation
can even have a higher reflectance in the MIR wavelength range than in the NIR
range (Gao 1996). Therefore, the extracted built-up land information using the
NDBI is often mixed with plant noise, and Zha et al. (2003) had to further use the
NDVI to filter out the noise. Obviously, simply using the original spectral bands to
construct an index is not applicable for the enhancement of built-up land due to its
complex spectral features. Therefore, the author proposes an index-based built-up
index (IBI) with the aim to provide a simple, but effective, method to enhance built-
up land presence in satellite imagery.
ðMIR{NIRÞ
NDBI~ , ð1Þ
ðMIRzNIRÞ
ðNIR{RedÞð1zl Þ
SAVI~ ð2Þ
ðNIRzRedzl Þ
and
ðGreen{MIRÞ
MNDWI~ , ð3Þ
ðGreenzMIRÞ
中国科技论文在线 http://www.paper.edu.cn
Remote Sensing Letters 4271
where MIR is a middle infrared band such as TM 5, NIR is a near infrared band
such as TM 4, Red is a red band such as TM 3 and Green is a green band such as
TM 2; l is a correction factor ranging from 0 for very high plant densities to 1 for
very low plant densities. The selection of the SAVI instead of the NDVI is because
the SAVI is more sensitive than the NDVI in detecting vegetation in the low-plant
covered areas such as urban areas. The SAVI can work in the area with plant cover
as low as 15%, while the NDVI can only work effectively in the area with plant cover
above 30% (Ray 2006). Therefore, the SAVI is more suitable for the urban area.
However, in the area where the plant cover is more than 30%, the NDVI can be
used:
ðNIR{RedÞ
NDVI~ : ð4Þ
ðNIRzRedÞ
After producing SAVI, MNDWI, and NDBI images, a new image was created,
which used these three new images as three bands. The change from the original
multi-band image into the three-thematic-band image largely reduces redundancy
between original multi-spectral bands, and the three new bands are negatively
correlated with each other (see table 1). Consequently, the spectral clusters of the
three major urban components are well separated (see figure 1 and table 2).
Table 2 shows the mean and standard deviation (std. dev.) values of the three
urban land-use classes in the new three-index-derived images. A unique feature is
that the mean value of built-up land is greater than those of vegetation and water in
the NDBI band. Furthermore, the mean value of built-up land in the NDBI band
exceeds its values in the SAVI band and the MNDWI band. According to these
distinct features, the IBI can be created as follows:
½NDBI{ðSAVIzMNDWIÞ=2
IBI~ : ð5Þ
½NDBIzðSAVIzMNDWIÞ=2
The index can enhance the built-up land feature easily because the subtraction of
the SAVI band and the MNDWI band from the NDBI band will result in positive
values for built-up land pixels only. The index takes advantage of the condition
where the features with higher NDBI values but lower SAVI and MNDWI values
will be enhanced. Obviously, the IBI is a normalized difference index and thus has
such features as: (1) a ratio-based index, (2) values ranging from 21 to + 1 and (3)
enhanced information has positive values, while the suppressed background noise
generally has zero to negative values. Dividing by two in the equation is to avoid
getting too small values of IBI. Before calculating the IBI using equation (5), the
Table 1. Comparison of correlation values between the three new bands and the three
original ETM + bands representing the visual light band group, the near infrared band and
the middle infrared band group, respectively.
values of the NDVI, the NDBI and the MNDWI should be added to 1 or rescaled
within 0–255 to convert negative values of the indices into positive values.
The IBI is distinguished from conventional indices by its first-time use of thematic
index-derived bands, instead of original image bands, to construct an index. When
the NDVI is used instead of the SAVI in equation (5), the IBI can be rewritten,
based on equations (1), (3), (4) and (5), as:
2MIR=ðMIRzNIRÞ{½NIR=ðNIRzRedÞzGreen=ðGreenzMIRÞ
IBI~ ð6Þ
2MIR=ðMIRzNIRÞz½NIR=ðNIRzRedÞzGreen=ðGreenzMIRÞ
Equation (6) allows the use of the IBI in just ‘one touch’ without having to make
three indices before calculating the IBI.
3. Validation of IBI
The IBI has been verified using the Landsat ETM + image of Fuzhou City in
southeastern China (see figure 2(a)), acquired on 29 May 2003. Even if a raw image
can be directly used for computing the IBI, the image was radiometrically corrected
before the calculation to avoid a dataset-specific result. The correction employed the
Figure 2. Landsat ETM + image of Fuzhou City: (a) false colour image (RGB:432), (b) IBI
image, enhanced built-up land features are in a light-grey to white tone and suppressed
background noise is in a dark-grey to black shade and (c) built-up land extraction image.
中国科技论文在线 http://www.paper.edu.cn
Remote Sensing Letters 4273
Table 2. Statistics of the three main urban land-use classes of the new three-band image.
Table 3. Statistics of the three main urban land-use classes of the IBI image.
for a raise in land surface temperature and a decrease in water and vegetation
covers.
The scatter plot of the IBI versus the MNDWI shows a characteristic triangular-
shaped envelope of pixels (figure 3(c)). The correlation between the IBI and the
MNDWI is not as strong as the other two because the random sampling procedure
sampled not only built-up land and water pixels, but also many vegetation pixels.
The vegetation pixels have low values in both IBI and MNDWI. As a result, their
Figure 3. Scatter plots showing the relationship of the IBI with the LST, the NDVI and the
MNDWI.
中国科技论文在线 http://www.paper.edu.cn
Remote Sensing Letters 4275
points accumulate in the lower left part of the scatter plot and thus cause the lowering of
the correlation between the IBI and the MNDWI. Nevertheless, a declined and sharply
defined upper edge of the triangular envelope can be seen in the figure. This may
represent the real relationship between the IBI and the MNDWI, implying a strong
negative correlation between the two indices. This upper edge may correspond to the
maximum water proportion at a given level of built-up land cover. In figure 3(b), the
points scattered in the lower left part of the scatter plot represent pure water or water-
dominated pixels, which have low values in both the IBI and the NDVI, and therefore
have also lowered the correlation between the IBI and the NDVI to some extent.
Finally, the relationship between the IBI, the NDVI and the MNDWI can be further
examined in a three-dimensional spectral feature space (figure 3(d)). This reveals a
saddle-shaped scatter plot, the top of which is composed of built-up land-dominated
pixels, while the two feet represent water and vegetation pixels, respectively.
4. Conclusions
Many studies have demonstrated that the built-up land class cannot be efficiently
enhanced using an index constructed simply of original multi-spectral bands because
the class has a heterogeneous characteristic. The proposed IBI is distinguished from
conventional indices by the first-time use of thematic index-derived bands to
construct an index, rather than by using original image bands. The new image,
composed of three thematic-index bands, the SAVI, the NDBI and the MNDWI,
can greatly reduce the data dimensionality and redundancy of the original multi-
spectral image and thus substantially avoids the spectral confusion between land-use
classes. Consequently, the IBI can effectively suppress background noise while
retaining built-up land features in satellite imagery.
Acknowledgements
This work is supported by the National Natural Science Foundation of China
(no. 40371107) and the Natural Science Foundation of Fujian Province, China
(no. 2007J0132).
References
CHANDER, G. and MARKHAM, B., 2003, Revised Landsat-5 TM radiometric calibration
procedures and postcalibration dynamic ranges. IEEE Transactions on Geoscience and
Remote Sensing, 41, pp. 2674–2677.
CHAVEZ, P.S. Jr., 1996, Image-based atmospheric corrections – revisited and revised.
Photogrammetric Engineering and Remote Sensing, 62, pp. 1025–1036.
CIBULA, W.G., ZETKA, E.F. and RICKMAN, D.L., 1992, Response of thematic bands to plant
water stress. International Journal of Remote Sensing, 13, pp. 1869–1880.
GAO, B.C., 1996, NDWI – a normalized difference water index for remote sensing of
vegetation liquid water from space. Remote Sensing of Environment, 58, pp. 257–266.
KAUFMANN, R.K., SETO, K.C., SCHNEIDER, A., LIU, Z., ZHOU, L. and WANG, W., 2007,
Climate response to rapid urban growth: evidence of a human-induced precipitation
deficit. Journal of Climate, 20, pp. 2299–2306.
LOWRY, J., KIRBY, J. and LANGS, L., 2004, SWReGAP Land Cover Mapping Methods
Documentation, Available online at: ftp.nr.usu.edu/swgap/data/landcover/map_
methods/UT/UT5_Methods.pdf (accessed 3 October 2007).
MASEK, J.G., LINDSAY, F.E. and GOWARD, S.N., 2000, Dynamics of urban growth in the
Washington DC metropolitan area, 1973–1996, from Landsat observations.
International Journal of Remote Sensing, 21, pp. 3473–3486.
中国科技论文在线 http://www.paper.edu.cn
4276 Remote Sensing Letters
MCFEETERS, S.K., 1996, The use of normalized difference water index (NDWI) in the
delineation of open water features. International Journal of Remote Sensing, 17,
pp. 1425–1432.
RAY, T.W., 2006, Vegetation in remote sensing FAQs. In ER Mapper Applications, pp. 85–97
(Perth, Australia: ER Mapper Ltd.).
RIDD, M.K., 1995, Exploring a V-I-S (vegetation-impervious surface-soil) model for urban
ecosystem analysis through remote sensing: comparative anatomy for cities.
International Journal of Remote Sensing, 16, pp. 2165–2185.
WU, C. and MURRAY, A.T., 2003, Estimating impervious surface distribution by spectral
mixture analysis. Remote Sensing of Environment, 84, pp. 493–505.
XIAN, G. and CRANE, M., 2005, Assessment of urban growth in the Tampa Bay watershed
using remote sensing data. Remote Sensing of Environment, 97, pp. 203–205.
XU, H., 2002, Spatial expansion of urban/town in Fuqing and its driving force analysis.
Remote Sensing Technology and Application, 17, pp. 86–92.
XU, H., 2006, Modification of normalized difference water index (NDWI) to enhance open
water features in remotely sensed imagery. International Journal of Remote Sensing,
27, pp. 3025–3033.
ZHA, Y., GAO, J. and NI, S., 2003, Use of normalized difference built-up index in
automatically mapping urban areas from TM imagery. International Journal of
Remote Sensing, 24, pp. 583–594.