Grazing intensity estimation in temperate typical grasslands of Inner Mongolia using machine learning models

Grasslands are essential components of terrestrial ecosystems and provide a broad range of biodiversity and ecosystem services for humans. Grasslands have been degraded to varying degrees owing to overgrazing. A scientific understanding of grazing intensity (GI) is essential for the continuous development of meadow ecosystems. This study aimed to estimate the GI of temperate typical grasslands in Inner Mongolia using machine learning algorithms. We used the extreme gradient boosting (XGBoost), random forest (RF), mogrifier long short term memory (MogLSTM), and batch attention module MogLSTM Kolmogorov–Arnold network (BMogLSTM-KAN) models improved by MogLSTM, modeled using field-measured GI and GI influencing factors. The GI impact factors included the aboveground biomass (AGB), normalized difference vegetation index (NDVI), enhanced vegetation index (EVI), digital elevation model (DEM), slope, aspect, monthly mean temperature and precipitation, distance to river (DTR), and distance to settlement (DTS). The results indicated that all four machine learning algorithms performed effectively in estimating the steppe GI, with determination coefficients (R²) exceeding 0.84. The BMogLSTM-KAN model demonstrated the highest performance, with an R² of 0.93, and a root mean square error (RMSE) of 0.68, indicating its high accuracy and stability in estimating the steppe GI. In addition, the GI spatial distribution map revealed that the GI of temperate typical grasslands in Inner Mongolia was low in the northeast and high in the southwest. These results offer a significant reference for the scientific protection and sustainable management of temperate typical grasslands in Inner Mongolia.