Toward Robust Parameterizations in Ecosystem‐Level Photosynthesis Models.

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 for estimating 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. Taking site‐level calibrations as a benchmark (NSE = 0.95), SPIE performed with an NSE of 0.68, while all the other investigated approaches showed negative NSE. The Shapley value, layer‐wise relevance and partial dependence showed that vegetation features, bioclimatic variables, soil properties and some PFTs determine parameters. SPIE 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. Plain Language Summary: Parameters can represent ecosystem properties and sensitivities of ecosystem processes to environmental changes, affecting model accuracy and output uncertainties. Therefore determining parameters is of great importance for applying models. Current ecosystem‐level models mostly determine parameters according to biomes. For example, light use efficiency (LUE) models, a typical tool to estimate gross primary productivity (GPP), use plant‐functional‐type (PFT)‐specific parameters. However, PFT‐specific parameters cannot represent the spatial variance of GPP sensitivities to environmental conditions within PFT and introduce significant estimation errors. To overcome these issues, we propose a Simultaneous Parameter Inversion and Extrapolation method (SPIE). Taking an LUE model as an example, we estimated model parameters using SPIE based on the input features representing vegetation, climate and soil properties at 196 observational sites. We compared SPIE with 11 other parameter extrapolating methods and all of these methods with site‐specific calibrations based on full‐time‐series observed GPP. The results were validated and showed that SPIE was the best method to extrapolate parameters across various temporal and spatial scales. According to the importance of input features, vegetation features, bioclimatic variables, soil properties and some PFTs are dominating spatial changes of parameters. Overall, SPIE is a robust method that overcomes strong limitations in many standard parameter extrapolating methods. Key Points: We propose a machine‐learning‐based parameterization method to model gross primary productivityThe method overcomes significant biases in extrapolations using assumptions of biome‐dependent and globally fixed parameterizationsVegetation, climate and soil properties explain most variability in model parameters, challenging current prescriptions of their patterns [ABSTRACT FROM AUTHOR]