A long-standing challenge in studying the global carbon cycle has been understanding the factors controlling inter–annual variation (IAV) of carbon fluxes related to vegetation photosynthesis and respiration, and improving their representations in existing biogeochemical models. Here, we compared an optimality-based mechanistic model and a semi-empirical light use efficiency model to understand how current models can be improved to simulate IAV of gross primary production (GPP). Both models simulated hourly GPP and were parameterized for (1) each site–year, (2) each site with an additional constraint on IAV (CostIAV), (3) each site, (4) each plant–functional type, and (5) globally. This was followed by forward runs using calibrated parameters, and model evaluations at different temporal scales across 198 eddy covariance sites. Both models performed better on hourly scale than annual scale for most sites. Specifically, the mechanistic model substantially improved when drought stress was explicitly included. Most of the variability in model performances was due to model types and parameterization strategies. The semi-empirical model produced statistically better hourly simulations than the mechanistic model, and site–year parameterization yielded better annual performance for both models. Annual model performance did not improve even when parameterized using CostIAV. Furthermore, both models underestimated the peaks of diurnal GPP in each site–year, suggesting that improving predictions of peaks could produce a comparatively better annual model performance. GPP of forests were better simulated than grassland or savanna sites by both models. Our findings reveal current model deficiencies in representing IAV of carbon fluxes and guide improvements in further model development.

In a model simulating dynamics of a system, parameters can represent system sensitivities and unresolved processes, therefore affecting model accuracy and uncertainty. Taking a light use efficiency (LUE) model as an example, which is a typical approach to estimate gross primary productivity (GPP), we propose a Simultaneous Parameter Inversion and Extrapolation approach (SPIE) to overcome issues stemming from plant-functional-type(PFT)-dependent parameterizations. SPIE refers to predicting model parameters using an artificial neural network based on collected variables, including PFT, climate types, bioclimatic variables, vegetation features, atmospheric nitrogen and phosphorus deposition and soil properties. The neural network was optimized to minimize GPP errors and constrain LUE model sensitivity functions. We compared SPIE with 11 typical parameter extrapolating methods, including PFT- and climate-specific parameterizations, global and PFT-based parameter optimization, site-similarity, and regression approaches. All methods were assessed using Nash-Sutcliffe model efficiency(NSE), determination coefficient and normalized root mean squared error, and contrasted with site-specific calibrations. Ten-fold cross-validated results showed that SPIE had the best performance across sites, various temporal scales and assessing metrics. None of the approaches performed similar to site-level calibrations(NSE=0.95), but SPIE was the only approach showing positive NSE(0.68). The Shapley value, layer-wise relevance and partial dependence showed that vegetation features, bioclimatic variables, soil properties and some PFTs are determining parameters. The proposed parameter extrapolation approach overcomes strong limitations observed in many standard parameterization methods. We argue that expanding SPIE to other models overcomes current limits and serves as an entry point to investigate the robustness and generalization of different models.