Quantifying hydrothermal-driven dynamics of gross primary productivity in the Ta-Pieh mountains based on explainable artificial intelligence

Study region

The study area is the Ta-Pieh Mountains, located in central China at the northern margin of the East Asian monsoon region.

Study focus

This study integrates multi-source remote-sensing and meteorological data (2000 – 2022) to investigate the hydrothermal-driven dynamics of Gross Primary Productivity (GPP). We applied Theil–Sen trend analysis, Mann–Kendall tests, and least-squares cross-wavelet analysis to assess spatiotemporal variations in GPP and climate variables. A per-pixel sliding-window modeling framework was developed using four tree-ensemble algorithms—Random Forest (RF), eXtreme Gradient Boosting (XGBoost), Light Gradient Boosting Machine (LightGBM), and Category Boosting (CatBoost). TreeSHAP was employed to quantify the relative contributions of temperature, precipitation, Root-Zone Soil Moisture (RZSM), and Vapor-Pressure Deficit (VPD) to GPP.

New hydrological insights for the region

Results show that temperature consistently dominates GPP variability, with VPD contributions closely tracking temperature fluctuations. Precipitation exhibits a one-month lagged effect on GPP, while reduced precipitation and lower RZSM strongly limit carbon uptake during droughts, exemplified by the 2019 autumn drought. The sliding-window framework achieved high predictive accuracy and delineated the spatiotemporal influence of hydrothermal drivers across different climate scenarios. These findings highlight the critical role of hydrothermal variability in regulating ecosystem carbon uptake and provide a transferable toolkit for vegetation-climate interaction analysis and ecosystem management in mountainous and transitional regions.