Abstract: Landslide is one of the main geological disasters which cause huge economic losses and casualties in Nepal. Remote sensing provides great potential to identify, extract, and monitor landslide information. Based on the average impurity reduction algorithm of random forest model, this study uses indices including NDBI, MNDWI, NDVI, and red band from Landsat images to extract landslide information. Further extraction of landslide information is obtained by visual interpretation. In this study, data grid method is used to divide the study area into 0.05°×0.05° grids. Based on this method, the distribution of landslides and correlation and partial correlation of landslides with rainfall and temperature are analyzed for different types of land cover. The results are shown as follows. of research show that:1) The landslides in central Nepal are mainly distributed at elevation between 1 000 and 2 500 m and at the slope gradient between 20° and 40°. 2) Compared with other types of land cover, the correlation coefficient between the number of landslides and the rainfall is 0.671 5 in forest land, while the number of landslides has the strongest partial correlation with temperature in grassland and the coefficient is 0.436 1. 3) The extensive distributions of gneiss and slate, together with earthquake, the fault zone, and the human activities, are destroying the stability of the slope and exposing the study area to be highly prone to landslide disasters.

Key words: landslides, random forests, partial correlation analysis