Spatiotemporal pattern and prediction model of normalized difference vegetation index in the Yellow River Source Zone

The Yellow River Source Zone is a critical ecological barrier for the Yellow River Basin, playing a vital role in regional water conservation, climate regulation and biodiversity protection. We integrated MODIS remote sensing images, meteorological, hydrological and terrain data from 2000-2020, used the methods including Sen slope method, partial correlation analysis, variance inflation factor analysis, interaction detection, as well as random forests and geographically weighted random forests models to comprehensively analyze the spatiotemporal variations and driving mechanisms of the normalized difference vegetation index (NDVI) in the Yellow River Source Zone and constructed a prediction model. The results showed that the annual NDVI of the zone increased significantly at a rate of 0.0028·a^-1 from 2000 to 2020, rising from 0.3301 to 0.3924, with an overall growth rate of 18.9%. The spatial distribution exhibited an increasing trend from the northwest to the southeast. Through partial correlation analysis, variance inflation factor screening, and interaction detection using a geographical detector, we found that wind speed, precipitation, and minimum temperature were the main driving factors of NDVI changes. There were complex interaction relationships among these factors, jointly affecting vegetation growth. The geographically weighted random forest model (coefficient of determination, R²=0.976, root mean square error, RMSE=0.017, mean absolute error, MAE=0.013) outperformed the random forest model (R²=0.465, RMSE=0.082, MAE=0.063) in revealing the spatial heterogeneity and local driving mechanisms of NDVI changes. It could assign reasonable feature weights to different regions, effectively improving the prediction accuracy.