Inter–Annual Variability of Model Parameters Improves Simulation of Annual Gross Primary Production

作者：Ranit De, Alexander Brenning, Markus Reichstein, Ladislav Šigut, Borja Ruiz Reverter, Mika Korkiakoski, Eugénie Paul‐Limoges, Peter D. Blanken, T. Andrew Black, Bert Gielen, Torbern Tagesson, Georg Wohlfahrt, Leonardo Montagnani, Sebastian Wolf, Jiquan Chen, Michael J. Liddell, Ankur R. Desai, Sujan Koirala, Nuno Carvalhais · 发表于：Journal of Advances in Modeling Earth Systems · 年份：2026 · DOI：10.1029/2025ms005116 · 研究领域：Plant Water Relations and Carbon Dynamics、Remote Sensing in Agriculture、Hydrology and Watershed Management Studies

Abstract Parametric uncertainty can hinder land surface models (LSM) from accurately simulating carbon fluxes, such as gross primary production (GPP). These models generally cannot capture inter–annual variability (IAV) of fluxes well due to missing processes, and temporally varying parameters can partially alleviate this limitation. We evaluated this assumption using two models: a light‐use efficiency (LUE) model with several environmental response functions, and an optimality‐based model that includes parameter acclimation and drought stress. De et al. (2025, https://doi.org/10.1029/2024MS004697 ) concluded that calibrating all parameters per site–year improves annual performance. As a follow‐up, we now inverted parameters of each environmental response function annually at a time, while simultaneously estimating year‐invariant optima for all other parameters, applying this across 198 eddy‐covariance sites. The IAV of GPP in arid sites was substantially improved when hydrological parameters varied annually, both for herbaceous and forest ecosystems. However, for tropical, temperate and boreal sites, IAV improved from annual variation of parameters controlling the GPP responses to temperature, light or atmospheric dryness. Given the paucity of arid and semi‐arid sites, allowing year‐specific parameters for vapor pressure deficit and atmospheric carbon dioxide effects yielded an overall median annual normalized Nash‐Sutcliffe efficiency of 0.733. Re‐evaluating some experiment...