Environmental Engineering Reference
In-Depth Information
TABLE 17.5 Error matrix of object-based land cover classification.
Reference class
Object-oriented classification
Impervious
Forest
Water
Non-forest
Shadow
Producer's accuracy (%)
Impervious
7904
3
0
490
36
93.7
Forested area
0
1569
0
226
0
87.4
Water
0
0
935
0
14
98.5
Non-forested area
399
0
0
2752
0
87.3
Shadow
0
43
0
0
1799
97.7
User's accuracy (%)
95.2
97.2
100
79.4
97.3
Overall accuracy (%) = 92.5 Kappa (%) = 88 . 7
five classes. The final derived impervious surface classification
map shows homogeneous patterns (Fig. 17.8b). Nevertheless, for
small impervious patterns such as the single-family residential
buildings, the object-based method tends to amalgamate imper-
vious buildings and some surrounding lawn areas to the same
object. A separate accuracy assessment using 200 randomly sam-
pled impervious points indicated a 94% overall accuracy for the
impervious/non-impervious map.
adjacent regions in an image as long as they have significant
contrast. Successful segmentation should create image objects
that have optimal information for further extraction of land
cover information. Our study confirms the outcome of image
segmentation is directly related to the user defined parameters of
scale and shape. However, defining the optimal scale and shape
factor is an intricate task since different land cover classes demon-
strate different characteristics in relation to the modifications of
these parameters. Moreover, this study indicates shadow prob-
lem can be addressed with class-related features in object-based
classification. To obtain accurate impervious surface map, we
have to differentiate tree shadows, mostly occurring in pervious
areas, from building shadows that may occur in both pervious
and impervious areas.
Conclusions
This chapter reviews the two groups of high resolution imper-
vious surface estimation models: pixel-based and object-based
methods. For pixel-based models, this chapter explored whether
two popular impervious surface estimation methods developed
for coarse-resolution remote sensing data can be transferred
to IKONOS imagery. Further, to partially address the prob-
lems associated with the SMA and RT approaches, this paper
developed an integrated model to improve estimation accuracy.
Results suggest both SMA and RT models can be successfully
applied to high-resolution remote sensing imagery with similar
estimation accuracy, while the performance of the SMA model is
slightly better. Moreover, the integrated model produces the best
estimation accuracy, with the lowest RMSE and MAE ,andthe
highest correlation coefficient between the modeled and 'true'
impervious surface fractions. In addition, when hard classifica-
tion is considered, the accuracy of overall classification is around
90%. These results suggest that these automatic methods have
the potential to replace the labor-intensive and time-consuming
digitizing process in creating high-resolution impervious surface
information.
Although the results from pixel-based models are promising
(with the RMSE about 10%and correlation coefficient about 0.9),
it is still necessary to explore whether object-based methods can
further improve the estimation accuracy. Through this case study,
we found the object-based classification can produce accurate
impervious surface map. The QuickBird imagery enabled map-
ping a complex urban areawith high spatial variation. In addition,
it is very convenient to refine the object-oriented classification
result in eCognition
References
Adams,J.B.,Sabol,D.E.,Kapos,V., et al . (1995) Classification
of multispectral images based on fractions of endmembers:
Application to land cover change in the Brazilian Amazon.
Remote Sensing of Environment , 52 , 137-154.
Baatz, M., Benz, U., Dehghani, S. et al . (2004) eCognition User
Guide 4. Definiens Imaging.
Bauer, M.E., Loeffelholz B.C. and Wilson, B. (2007) Estimating
and mapping impervious surface area by regression analysis
of Landsat imagery, Remote Sensing of Impervious Surfaces (Q.
Weng, editor), CRC Press, Boca Raton, pp. 3-20.
Benz, U.C., Hofmann, P., Willhauck, G., Lingenfelder, I. and
Heynen, M. (2004) Multi-resolution object-oriented fuzzy
analysis of remote sensing data for GIS-ready information.
ISPRS Journal of Photogrammetry and Remote Sensing , 58 ,
239-258.
Breiman, L., Friedman, J.H., Olshen, R.A. and Stone, C.J. (1984)
Classification and Regression Trees , Wadsworth, Inc., Mon-
terey.
Dougherty,M.,Dymond,R.L.,Goetz,S.J.,Jantz,C.A.andGoulet,
N. (2004) Evaluation of impervious surface estimates in a
rapidly urbanizing watershed. Photogrammetric Engineering
and Remote Sensing , 70 (11), 1275-1284.
since that the classification process is itera-
tive. Nevertheless, the object-based method tends to amalgamate
small buildings and surrounding lawn areas together. In object-
oriented image analysis, multiresolution segmentation separates
Search WWH ::




Custom Search