Abstract: The alpine steppe is a major type of the grassland ecosystem in the Qinghai-Tibet Plateau and plays an important role in soil erosion control and wild animal conservation. In recent years, the desertification of alpine steppe is expanding because of the global climate change and human disturbance. Therefore, it is very important to monitor the area and extent of grassland desertification at a spatial-temporal scale for control. The study used two models of unmanned aerial vehicle (DJ Phantom 3 and Matrice100) and ground survey technology to investigate the desertification status of alpine steppe of Maduo County in Sanjiangyuan National Park, which located in Qinghai Province. The purpose of this study is to select the proper vegetation indices of UAV that suit to build desertification model and classification criteria for alpine steppe desertification. The results showed as following:(1) Based on the respective correlation between Visible-Band Difference Vegetation Index (VDVI), Enhanced Normalized Difference Vegetation Index (ENDVI), Normalized Green-Red Difference Index (NGRDI) and Grassland Desertification Index (GDI), the optimal vegetation index of UAV is VDVI (R=0.9055). (2) Built the grassland desertification model, VDVI=0.3024GDI²-0.0335GDI+0.0119(R²=0.9326), the relative error is 1.779% (RMSE=0.165, R²=0.7447), which means the higher fitting precision. (3) The desertification of the alpine steppe in the study area is divided into five grades, involving no obvious desertification(VDVI>0.2247), mild desertification(0.1493 < VDVI < 0.2246), moderately desertification(0.0924 < VDVI < 0.1492), severely desertification(0.0692 < VDVI < 0.0923), and extremely desertification(VDVI<0.0692).

Key words: UAV, desertification, vegetation index, alpine steppe