Forecasting Forest Product Yields in China Based on a Random Forest Model: Interaction Between Climate Change and Socio-Economic Factors

ABSTRACT: This study presents a comprehensive projection of China’s forest product yield dynamics (encompassing commodity timber and logs) through 2100, employing an innovative integration of machine learning and economic modeling. We developed a hybrid analytical framework combining random forest algorithms with Cobb-Douglas production functions to assess multi-dimensional drivers, including climatic variables, socio-economic indicators, and demographic trends. Our multi-model validation demonstrated strong predictive performance (R² are 0.86 and 0.92), particularly in quantifying climate-production interactions, with sensitivity analysis identifying surface downward shortwave radiation (RSDS), population density (POP), and mean annual temperature (MAT) as dominant predictors explaining 68% of yield variance. Future yields exhibited significant spatial and temporal variations under different SSP scenarios, especially under SSP126, where yields were more stable, and under SSP245 and SSP370, where yields showed a moderate increasing trend. The SSP585 shows higher fluctuations and a decreasing trend in yields due to climate change. Geospatial modeling uncovered critical regional disparities, suggesting potential production migration from traditional southern bases to north-eastern/northwestern frontiers under climate stress. The southern subtropical belt emerged as particularly vulnerable to thermal extremes and precipitation variability, while northern regions demonstrated greater climate resilience but require substantial silvicultural adaptation. These results provide a scientific basis for developing more precise forest management policies and sustainable development strategies to help meet the challenges posed by future demand for forest products and climate change.