Dynamic monitoring of grassland land cover types of Inner Mongolia 1990–2023 and testing the causal relationship between meteorological data and grassland area

The area of global grassland ecosystems has been continuously declining over the past few decades. For this purpose, this study compared the classification performance of random forest (RF) and support vector machine (SVM) algorithms in large-scale land cover mapping using Landsat series of remotely sensed imagery since 1990 and selecting Inner Mongolia in China's continental cold and arid climate zone as a typical study area. We then applied the Shapley additive explanation (SHAP) method to analyze the sensitivity of remotely sensed features to grassland land types and accurately detected grassland disturbance dynamics using the LandTrendr algorithm. We explored the intrinsic relationship between major climate factors (temperature and precipitation) and grassland area changes using the convergent cross-mapping method. The results show that (1) the classification accuracy of RF model is better than SVM, and the average accuracy of grassland identification reaches 86.6%; (2) the chlorophyll index green, BRE soil index, ratio vegetation index, SR_B7 and altitude have a significant effect on the classification accuracy of grassland remote sensing; and (3) the average annual temperature of the grasslands around the meteorological station increases significantly by 2.13 °C. With the increase of annual mean temperature, the grassland area decreased significantly.