Abstract: This study aimed to reveal the spatiotemporal variations and the influencing factors of vegetation in the Qiangtang grassland during 2001-2020, and to predict the change trends of vegetation under climate change scenarios. Based on the data of MODIS NDVI, temperature, precipitation, and wind speed, we explored the relationship between vegetation changes and meteorological factors. Furthermore, NDVI prediction models were establish with three machine learning algorithms of random forest, support vector machine, and random gradient descent regression. The optimal model with the best simulation accuracy was selected to simulate vegetation changes under multiple scenarios. We found that NDVI of the Qiangtang grassland showed a slight increasing trend with a growth rate of 0.0003 a^-1 from 2001 to 2020. The response of NDVI to temperature lagged by 3 months, precipitation lagged by 0-1 months. NDVI was negatively correlated with wind speed without lag. The random forest algorithm had the highest simulation accuracy (Adjusted R²=0.958). The scenario for improvement of vegetation coverage in the future included 1.0 ℃ increase in temperature, 25% increase in precipitation, and 25% decrease in wind speed. This study contributed to early warning of vegetation degradation, which would help vegetation conservation under climate change.

Key words: Qiangtang grassland, normalized difference vegetation index, random forest, multi-scenario prediction