Assimilation of Satellite‐Based Forest Biomass, Productivity and Leaf Area Index in a Global Land Model Incorporating Parameter Spatial Variations and Age Effects

Abstract

Disturbances profoundly impact forest capacity for sequestering and storing carbon, and the recovery rates after disturbances vary geographically. Many terrestrial carbon models inadequately simulate the effects of forest disturbances on biomass due to their reliance on equilibrium assumptions, the absence of historical disturbance data, as well as homogeneous vegetation parameters. In this study, we developed a new hybrid machine learning framework to optimize the spatial distribution of the parameters of a process‐based model, the integrated biosphere simulator (IBIS), across global forest regions. High‐resolution satellite‐derived products of gross primary productivity (GPP), leaf area index (LAI), forest age, and biomass were used as references to optimize parameters of both fast and slow carbon processes in the IBIS model. By integrating age maps and spatially explicit growth curves for biomass accumulation, our model was able to better represent the biomass dynamics in regenerating forests after disturbances. Our findings underscored the role of spatially optimized parameters in accurately simulating GPP, LAI, and biomass across global forests, with particular gains from accounting for age structures. The optimized IBIS model shows superior performance in reproducing biomass gradients across climate zones, compared with global satellite‐derived products and simulations from multiple dynamic global vegetation models (DGVMs). The optimization revealed a greater carbon sequestration potential in regrowing forests after disturbances. Our framework provides a new strategy for using forest age data to improve the accuracy of large‐scale forest carbon simulations in DGVMs.