Predicting and Explaining Spatial and Inter–Annual Variability in Inversion-Derived Maximum Light Use Efficiency

Long-term eddy-covariance measurements reveal considerable inter–annual variability (IAV) in carbon fluxes. Although models reproduce gross primary productivity (GPP) well from subdaily to seasonal scales, they often fail to capture its IAV. Previous studies showed that annual calibration of a light use efficiency model improves annual GPP predictions, motivating efforts to predict inter–annual variability in maximum light use efficiency ( \varepsilon_{\mathrm{max}} ). Here, we predicted spatial and inter–annual variability in \varepsilon_{\mathrm{max}} across 671 site-years from 96 sites using meteorological, climatic, vegetation, soil, and remote-sensing-based predictors. We evaluated three tree-based models, two linear regression models, and a prior-data-fitted network (TabPFN) within a nested cross-validation (CV) framework. All models captured spatial variability more accurately than IAV in site-level CV, where the validation set contained sites not present in the training set. TabPFN performed best, explaining 25% of the spatiotemporal variability in \varepsilon_{\mathrm{max}} on an out-of-fold dataset. Post hoc analyses of the tree-based models and TabPFN showed that the strongest predictors were mostly climatic variables, such as annual statistics of temperature, cloudiness index ( \overline{\mathit{CI}_{s,t}} ), and water availability index ( \overline{\mathit{WAI}_{s,t}} ), along with soil bulk density ( \mathit{BDOD}_s ). In a separate five-fold CV repeated ten times (50 folds total), symbolic regression most frequently selected \mathit{BDOD}_s and \overline{\mathit{WAI}_{s,t}} in the best equations balancing complexity and accuracy. Fitting multiple linear regressions across 50 folds always revealed a positive association between \overline{\mathit{CI}_{s,t}} and \varepsilon_{\mathrm{max}} , and assigned the highest median absolute coefficient to \overline{\mathit{WAI}_{s,t}} . Although IAV remained difficult to predict, we identified key predictors of spatial variability in \varepsilon_{\mathrm{max}} and assessed their transferability to sites excluded from model training. These predictors, along with potential predictors currently lacking observations, such as nutrient pools and deposition, might help explain the spatiotemporal acclimation and adaptation of plant traits, which are often represented by model parameters.

Authors

Institutions

Publication Details

Published
2026-11-18
DOI
https://doi.org/10.22541/essoar.15012454/v1
Primary Topic
Plant Water Relations and Carbon Dynamics
Type
preprint

Funders

Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

Predicting and Explaining Spatial and Inter–Annual Variability in Inversion-Derived Maximum Light Use Efficiency

Ranit De, Alexander Brenning, Nuno Carvalhais, Siyuan Wang et al.
1 links
Plant Water Relations and Carbon Dynamics
preprint

Predicting and Explaining Spatial and Inter–Annual Variability in Inversion-Derived Maximum Light Use Efficiency

Ranit De, Alexander Brenning, Nuno Carvalhais, Siyuan Wang, Markus Reichstein
preprint en

Abstract

Long-term eddy-covariance measurements reveal considerable inter–annual variability (IAV) in carbon fluxes. Although models reproduce gross primary productivity (GPP) well from subdaily to seasonal scales, they often fail to capture its IAV. Previous studies showed that annual calibration of a light use efficiency model improves annual GPP predictions, motivating efforts to predict inter–annual variability in maximum light use efficiency ( \varepsilon_{\mathrm{max}} ). Here, we predicted spatial and inter–annual variability in \varepsilon_{\mathrm{max}} across 671 site-years from 96 sites using meteorological, climatic, vegetation, soil, and remote-sensing-based predictors. We evaluated three tree-based models, two linear regression models, and a prior-data-fitted network (TabPFN) within a nested cross-validation (CV) framework. All models captured spatial variability more accurately than IAV in site-level CV, where the validation set contained sites not present in the training set. TabPFN performed best, explaining 25% of the spatiotemporal variability in \varepsilon_{\mathrm{max}} on an out-of-fold dataset. Post hoc analyses of the tree-based models and TabPFN showed that the strongest predictors were mostly climatic variables, such as annual statistics of temperature, cloudiness index ( \overline{\mathit{CI}_{s,t}} ), and water availability index ( \overline{\mathit{WAI}_{s,t}} ), along with soil bulk density ( \mathit{BDOD}_s ). In a separate five-fold CV repeated ten times (50 folds total), symbolic regression most frequently selected \mathit{BDOD}_s and \overline{\mathit{WAI}_{s,t}} in the best equations balancing complexity and accuracy. Fitting multiple linear regressions across 50 folds always revealed a positive association between \overline{\mathit{CI}_{s,t}} and \varepsilon_{\mathrm{max}} , and assigned the highest median absolute coefficient to \overline{\mathit{WAI}_{s,t}} . Although IAV remained difficult to predict, we identified key predictors of spatial variability in \varepsilon_{\mathrm{max}} and assessed their transferability to sites excluded from model training. These predictors, along with potential predictors currently lacking observations, such as nutrient pools and deposition, might help explain the spatiotemporal acclimation and adaptation of plant traits, which are often represented by model parameters.

American Society for Photogrammetry and Remote Sensing (US), Faculdade de Tecnologia e Ciências (BR), Max Planck Institute for Biogeochemistry (DE), Friedrich Schiller University Jena (DE), Technische Universität Dresden (DE), Universidade Nova de Lisboa (PT)
HORIZON EUROPE Framework Programme
Climate action, Life on land
Plant Water Relations and Carbon Dynamics
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon