Optimization of China's terrestrial ecosystem productivity from a remote-sensing-driven ecosystem model: model calibration and validation

Process-based ecosystem models are essential for understanding the behavior of the terrestrial carbon cycle, but their capacity is largely limited by the uncertainty in prescribed parameter values. The Boreal Ecosystem Productivity Simulator (BEPS) is a well-established remote-sensing data-driven process-based ecosystem model, but its parameter uncertainties are not fully understood. Here we implemented a global parameter sensitivity analysis and calibration approach into BEPS for systematic optimization of critical model parameters by using data derived from 10 eddy covariance flux sites covering seven ecosystem types across China. We found that the photosynthesis and the soil water availability parameters were the most important parameters affecting the simulations of GPP. Their importance varied between plant function types, and local environmental conditions. In regions with persistent droughts and freezing-thawing, parameters related to soil water availability were controlling the GPP while for regions without climate constraints, the photosynthesis parameters were instead predominant. With the optimized parameters, the simulated GPP increased for most Chinese regions and for most plant functional types. The GPP response of grasses to drought in arid and semi-arid area was also improved. Furthermore, using three remote-sensing-derived leaf area index (LAI) products to drive the BEPS model, we found that the discrepancies between LAI products led to substantial uncertainties in simulated GPP, and it was shown that GLOBMAP was a relative comprehensive LAI product for BEPS. Our study suggests that calibrating process-based ecosystem models with observations from different ecosystems and sources could effectively improve regional carbon flux estimations.