Grassland vegetation changes can be assessed based on vegetation growth dynamics and spatial characteristics of vegetation growth or land cover patches. However, previous studies have focused mainly on assessing changes in grasslands based only on vegetation growth condition, with limited consideration of spatial characteristics, which may lead to inaccurate identification of grassland dynamics. Here, a decision tree approach was developed to classify changes in grassland vegetation based on remote sensing data considering not only temporal variation in vegetation cover at the grid cell scale but also temporal variation in both spatial heterogeneity of vegetation cover and spatial configuration of bare land patches at the local landscape scale. The decision tree approach was verified using field and historical data and used to classify vegetation changes in the Xilingol League, Inner Mongolia, China, during the 2000-2019 period. According to the results, the study area was occupied mainly by grasslands with an insignificant change in vegetation growth, followed by those with significant improvement, and it was least occupied by those with significant decrease (but still covering 344 km2). Compared to vegetation cover, a combination of spatial characteristics was more effective in identifying changes in vegetation, including improvement, fragmentation, and degradation, and subtle changes within each process. A slight change in spatial heterogeneity of vegetation cover (such as the spatial coefficient of variation of the normalized difference vegetation index) may have overall impacts on the grassland. Each state during grassland retrogression, from moderate degradation through severe degradation to desertification, can be inferred by changes in spatial configuration (such as connectivity) of bare land patches, as well as the current bare land area. Overall, spatial characteristics can provide important theoretical information to facilitate assessment of grassland ecosystem dynamics and stability and serve as a critical early warning indicator for potential desertification.