A carregar...
Projeto de investigação
EXPLAINABLE AI PIPELINES FOR BIG COPERNICUS DATA
Financiador
Autores
Publicações
Toward Robust Parameterizations in Ecosystem-Level Photosynthesis Models
Publication . Bao, Shanning; Alonso, Lazaro; Wang, Siyuan; Gensheimer, Johannes; De, Ranit; Carvalhais, Nuno; DCEA - Departamento de Ciências e Engenharia do Ambiente; John Wiley and Sons Inc.
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.
Narrow but robust advantages in two-big-leaf light use efficiency models over big-leaf light use efficiency models at ecosystem level
Publication . Bao, Shanning; Ibrom, Andreas; Wohlfahrt, Georg; Koirala, Sujan; Migliavacca, Mirco; Zhang, Qian; Carvalhais, Nuno; DCEA - Departamento de Ciências e Engenharia do Ambiente; Elsevier Science B.V., Amsterdam.
This study aims to (1) investigate whether two-big-leaf light use efficiency (LUE) models (TL) outperform big-leaf LUE models (BL) by incorporating different gross primary productivity (GPP) responses in sunlit and shaded leaves; (2) explore the robustness of using the leaf area index (LAI), clumping index (Ω) and spherical leaf angle distribution to partition canopies into sunlit and shaded leaves across canopy architectures; (3) identify optimal light response forms in LUE models. To exclude influences of drivers of GPP other than radiation, we collected various formulations of GPP response functions to temperature, vapor pressure deficit, CO2, soil water supply, light intensity and cloudiness index to construct 5600 BLs and 1120 TLs. These models were evaluated at 196 globally-distributed eddy covariance sites from the FLUXNET observational network using the Nash-Sutcliffe model efficiency (NSE), root mean squared error and Bayesian information criterion. Across all sites, the best big-leaf model (BL*; NSE=0.82) was statistically equal to the best TL (TL*; NSE=0.84). However, daily dynamics in GPP under hot and dry conditions were best described using TL* in 17% of sites, highlighting the local importance in separating sunlit and shaded leaves. Across approaches to represent effective LAI, the best approach relies on using normalized difference vegetation index with a spherical or flexible leaf angle distribution across sites rather than satellite LAI and Ω. We also observed similar performance between non-rectangular hyperbola and reciprocal light response functions across TLs. Models degrade when the maximum LUE is not differentiated between sunlit and shaded leaves, but not when light saturation levels are the same. Despite functional differences, the best five TLs agree in a larger contribution of shaded leaf area to total GPP, resulting from higher LAI and LUE. Overall, these results suggest marginal but robust selection of TL compared to BL.
Unidades organizacionais
Descrição
Palavras-chave
Contribuidores
Financiadores
Entidade financiadora
European Commission
Programa de financiamento
H2020
Número da atribuição
101004188