Estimating impervious surface base on comparison of spectral mixture analysis and support vector machine methods

Impervious surface percentage(ISP) is the key parameter for urban regional environment research.This paper compares the ISP estimate-performance of spectral mixture analysis(SMA) and support vector machine(SVM) on TM image.The SVM model establish the non-linear relationship between spectral feature of TM pixels and corresponding ground sample ISP values and then be implied on without-sample TM pixels for ISP estimation.On the TM image of Tianjin urban area,we first select high resolution image from Quickbird classification results,including college,industrial and residential districts as training sample(7020 items) and then test sample(1500 items).The toot mean square error(RMSE) of SVM model is 15.4%,which is better than SMA with 19.4%.Additionally,after adding "greenness" of tasseled cap transform and "high-albedo" of SMA,the RMSE decreases to 12%.The results of the study indicate that SVM model is suitable for large area ISP mapping without insufficient samples because of the non-linear characteristic and good performance of small-sample generalization.By adding spectral feature vector having significant relation with ISP,it can adjust the value of ISP estimation where the land cover types is lack of training samples and improve the overall accuracy of regional ISP estimation.