Abstract: Knowing the changes of ecological landscape in lake areas of arid region is of particular importance to monitor land use change and adjust land use strategies. The current study focuses on Ebinur Lake region, Xinjiang, which suffers from severe salinization and desertification in recent years, to generate land use changing maps using Landsat-TM multispectral images of 1990, 2001 and 2011. Three commonly used classification approaches were compared, i.e., Maximum Likelihood Classification (MLC), Decision Tree Classification (DTC), and Support Vector Machine (SVM), to generate landscape metrics for stndying the landscape change during the past two decades in Ebinur Lake area. Three findings emerge from the study. First, among the three classification approaches, SVM produces the most accurate classification. For the year of 1990, 2001, and 2011, the classification accuracy rates are 97.312 3%, 97.778 2%, and 95.082 1%, respectively, Kappa coefficients are 0.963 2, 0.884 2, and 0.918 7, respectively. Second, for the past two decades, salinized and desertified lands are increasing drastically (salinized land increased 374.471 4 km²), along with increased cultivated land. While forest, grassland, water body and other land use types decrease in the same period, with grassland lost over 366.868 8 km². Third, there is a significant landscape difference among different periods. In general, calculated landscape metrics such as landscape density, maximum patch index, isolation index, Shannon diversity index and Shannon evenness index are increasing during the two decades. On the other hand, area weighted index, mean patch distance index, contagion index, interspersion juxtaposition index and patch cohesion index are decreasing. It seems that the landscape of study area becomes more fragmented during the study period, which leads to relatively higher diversity, and more even distribution of various land use types. Our study results also suggest that the current landscape distribution is relatively stable and even. However, the increasing salinization and desertification of land due to increased human activities (cultivation) requires close monitoring and constant evaluation.

Key words: remote sensing, landscape pattern change, SVM classification, Ebinur lake area