Ecosystem‐Specific Climatic and Biophysical Controls on Gross Primary Productivity Across China: Machine Learning and Spatial Causal Inference Approach

Abstract

Gross Primary Productivity (GPP) is a fundamental component of the terrestrial carbon cycle, yet its long‐term dynamics and controlling mechanisms remain uncertain at continental scales, particularly where ecosystems respond differently to climate change. We investigated spatiotemporal trends and process‐level controls on annual GPP (AGPP) across China's major terrestrial ecosystems from 2002 to 2020 using an integrated Earth‐system diagnostic framework. Spatial trend analysis, machine learning prediction, functional response analysis, and spatial causal inference ‐ Geographical Pattern Causality (GPC) are combined to distinguish predictive associations from dominant mechanistic constraints. Nationally, AGPP increased marginally (+11.3 gC m−2 yr−1), masking strong ecosystem divergence: wetlands exhibited rapid gains (+72.6 gC m−2 yr−1), while forest AGPP declined significantly (−12.9 gC m−2 yr−1). Spatial causal analysis identifies a consistent national control hierarchy, with leaf area index (LAI), photosynthetically active radiation (PAR), and atmospheric CO2 (ρc,yr) exerting the strongest influence, while evaporative demand, precipitation, and temperature play secondary roles. Causal mechanisms are strongly ecosystem‐specific. Cropland productivity is dominated by PAR, LAI, and CO2, with no detectable causal role of temperature. Grasslands are co‐limited by radiation, canopy structure, CO2, and soil nitrogen. Forest productivity is thermally constrained but shows strong negative responses to atmospheric drying, indicating increasing hydraulic stress. Wetland productivity depends primarily on radiation and CO2, with negligible influence from moisture or canopy structure. By separating prediction from spatially inferred causation, this study clarifies ecosystem‐specific controls and vulnerabilities within China's terrestrial carbon uptake.