Abstract
Environmental factors lead mainly to the uncertainty of gross primary productivity estimation in most light use efficiency (LUE, ε) models since the simple physical formulas are inadequate to fully express the overall constraint of diverse environmental factors on the maximum ε (ε(max)). In contrast, machine learning has the natural potential to detect intricate patterns and relationships among various environmental variables. Here, we presented a hybrid model (TL-CRF) that utilizes the random forest (RF) technique to incorporate various ecological stress factors into the two-leaf LUE (TL-LUE) model, meanwhile, seasonal differences in the clumping index (CI) on a global scale are considered to adjust seasonal patterns of canopy structure. The comprehensive integration of complex environmental variables based on this RF submodule is conducive to scaling theoretical ε(max) to actual ε as much as possible. The proposed TL-CRF model considerably improves global GPP estimation by complementing innate advantages between the process-based and data-driven models.•The seasonal CI averages in different stages of the leaf life cycle are estimated for different vegetation types on a global scale.•Various environmental stress factors are integrated via the RF technique.•The RF submodule is embedded into the TL-LUE model to establish a hybrid model.