Spatiotemporal monitoring of grasshopper habitats in China: integrating ensemble machine learning, degree-day modeling, and multi-source remote sensing data

Grasshopper outbreaks pose serious ecological and economic threats to temperate grasslands, reducing vegetation productivity and destabilizing steppe ecosystems. To enhance monitoring and prediction, this study integrated remote-sensing indicators with ensemble machine learning and a degree-day (DD) model to assess grasshopper habitat suitability in the Xilingol steppe of Inner Mongolia from 2018 to 2022. Field-based occurrence data of Oedaleus decorus asiaticus and Dasyhippus barbipes were combined with 28 environmental variables encompassing meteorological, vegetative, soil, topographic, and landscape factors. Following multicollinearity screening using tolerance, variance inflation factor (VIF), and correlation thresholds, independent predictors were used to train four algorithms—Random Forest, Multilayer Perceptron, XGBoost, and MaxEnt—whose weighted integration formed the ensemble model. Spatial autocorrelation and hotspot analyses were employed to examine distributional clustering, while suitability maps were classified into low, moderate, and high categories. The ensemble achieved the highest predictive accuracy (AUC = 0.923–0.945), outperforming individual models. Results indicated persistent spatial clustering of grasshopper occurrences, with stable hotspots around Xilinhot and East and West Ujumqin and notable expansions during favorable years (2020, 2022). Minimum temperature during the egg stage (22.6%) and fractional vegetation cover during the nymph stage (19.8%) emerged as dominant environmental drivers, followed by precipitation, elevation, and above-ground biomass. These findings underscore the central role of temperature and vegetation dynamics in shaping grasshopper habitat suitability. The integration of DD-based phenological modeling with ensemble learning provides a robust, ecologically coherent framework for regional pest monitoring, early warning, and sustainable grassland management in semi-arid ecosystems.