Support vector machine-based decision tree for snow cover extraction in mountain areas using high spatial resolution remote sensing image

Abstract

Snow cover extraction in mountain areas is a complex task, especially from high spatial resolution remote sensing (HSRRS) data. The influence of mountain shadows in HSRRS is severe and normalized difference snow index-based snow cover extraction methods are inaccessible. A decision tree building method for snow cover extraction (DTSE) integrated with an efficiency feature selection algorithm is proposed. The severe influence of terrain shadows is eliminated by extracting snow in sunlight and snow in shadow separately in different nodes. In the feature selection algorithm, deviation of fuzzy grade matrix is proposed as a class-specific criterion which improves the efficiency and robustness of the selected feature set, thus making the snow cover extraction accurate. Two experiments are carried out based on ZY-3 image of two regions (regions A and B) located in Tianshan Mountains, China. The experiment on region A achieves an adequate accuracy demonstrating the robustness of the DTSE building method. The experiment on region B shows that a general DTSE model achieves an unsatisfied accuracy for snow in shadow and DTSE rebuilding evidently improves the performance, thus providing an accurate and fast way to extract snow cover in mountain areas.