Improved SVM for Extracting Snow Cover in Northern Xinjiang
基于改进SVM的新疆北部地区积雪面积反演研究——以天山山区中段为例

Snow is the most active natural land surface factor and has implications for the global climate and hydrological environment. In Xinjiang, seasonal snow accumulation and melt are important to water resource management and the sustainable development of oases. Snow area identification and mapping comprise fundamental environment research in cold, arid and semi-arid regions. Here, we extract snow area from MOD02 HKM images, one of the three MODIS L1B products (MOD02 QKM, MOD02 HKM, MOD02 1KM) at 500m spatial resolution for the Tianshan Mountains. We test the SVM method combined with a snow component of spectral mixture analysis and texture feature extraction by GLCM using MODIS data to extract snow area information. Results indicate that using the snow component of spectral mixture analysis achieves superior results compared to traditional SVM classification results; classification accuracy was increased by 0.2702%. Combining the texture features extracted with GLCM for classification improved overall accuracy by 1.081% and mapping accuracy of 99.01%. This suggests that the SVM method combined with the snow component of spectral mixture analysis and texture feature extraction by GLCM is effective for snow area extraction using MODIS data at low spatial resolution. The classification method proposed in this paper can adapt to the nonlinear relationship between features. By adding spectral feature vectors with a significant relationship to snow area, and texture features like the SVM input vector, our method adjusts snow area extraction when land cover types lack training samples and improves overall accuracy.