HR: 0800h
AN: H31D-0645    [Abstracts]
TI: Impervious Surface Mapping of Jungnang-cheon Basin of Korea Using Remote Sensing Images
AU: * Kim, S
EM: sykim79@yonsei.ac.kr
AF: School of Civil and Environmental Engineering, Yonsei University, Shinchon-Dong, Seodaemun-Gu, Seoul, 120749, Korea, Republic of
AU: Heo, J
EM: jhheo@yonsei.ac.kr
AF: School of Civil and Environmental Engineering, Yonsei University, Shinchon-Dong, Seodaemun-Gu, Seoul, 120749, Korea, Republic of
AU: Heo, J
EM: jheo@yonsei.ac.kr
AF: School of Civil and Environmental Engineering, Yonsei University, Shinchon-Dong, Seodaemun-Gu, Seoul, 120749, Korea, Republic of
AB: Impervious surface is the important index for the estimation of urbanization and environmental change. In addition, impervious surface affects on various hydrological process such as the short-term rainfall runoff modeling, water balance analysis, and groundwater estimation in urban area. Therefore, the estimation of impervious surface is an important factor to analyze urban flood. The main objective of this study is the impervious surface mapping of case study area using remote sensing images. Case study area is Jungnang- cheon basin in South Korea. Remote sensing images for the impervious surface mapping are landsat-7 ETM+ and high resolution satellite image of Jungnang-cheon basin. Moreover, a tasseled cap transformation and NDVI transformation apply to landsat-7 ETM+ for considering various predicted parameters. Impervious surface is estimated by using regression tree algorithm which is a binary recursive partitioning process and a rule-based model for the prediction of continuous variables based on training data. Regression tree algorithm is applied to training data sets which are collected by overlaying between landsat-7 ETM+ and high resolution satellite image with different spatial resolution. Then, the predicted variables such as band 3(red), band 4(nearIR), band 5(midIR), and band 7(nearIR) of landsat-7 ETM+ and TC2(greenness) and TC3(wetness) of a tasseled cap transformed image and NDVI transformed image are selected for the efficient and fast prediction modeling. The independent variable of model is a continuous impervious index represented by percentage. The accuracy of variables combination is compared by the average error(AE), the relative error(RE), and correlation coefficient. As the results, the selected test composes with band 3, 4, 5 and 7 of landsat-7 ETM+, the greenness of a tasseled cap transformed image and NDVI. It shows the highest correlation coefficient(0.92) and the smallest the total average error(9.2). In addition, 10-folds cross-validation is used to evaluate the performance of regression tree for all tests. The suggested test has the highest correlation coefficient and the smallest error. Finally, the impervious surface mapping is performed by using the predicted variables from the selected prediction model.
DE: 1855 Remote sensing (1640)
SC: Hydrology [H]
MN: 2007 Fall Meeting