[Prediction of Global Warming Potential of Qilian Mountain Grassland Ecosystem Based on Ensemble Machine Learning]

The Qilian Mountain grassland ecosystem serves as an important carbon reservoir， and accurate prediction of its global warming potential （GWP） is essential for climate-adaptive management. Based on multi-source data from 1980 to 2024， an ensemble machine learning model integrating XGBoost， LightGBM， CatBoost， and RF was developed to predict GWP dynamics under different climate scenarios during 2025-2060. The results indicate that： ① The ensemble machine learning model achieved the best performance， with a coefficient of determination （R²） of 0.938 and a root mean square error （RMSE） significantly lower than those of individual models， effectively capturing the complex nonlinear relationships of the grassland ecosystem. ② Soil clay and silt contents， together with grazing intensity， were identified as the key driving factors of GWP， among which grazing intensity exhibited a threshold effect. ③ Future scenario simulations suggest that under the SSP5-8.5 high-emission pathway， GWP shows a continuous upward trend with intensified interannual fluctuations， and the risk in high-grazing areas increases substantially， exceeding 42 g·（m²·a）^-1 by 2060， while under the SSP1-2.6 low-emission pathway， GWP remains relatively stable. In terms of spatial patterns， GWP demonstrates a "higher in the southeast and lower in the northwest" distribution， with the southeastern high-value areas expanding further under high-emission scenarios. These findings provide theoretical support and technical guidance for global warming risk assessment， ecosystem management zoning， and low-carbon pathway development in typical alpine grasslands.