Inter-annual parameter variability improves the predictability of inter-annual GPP but is weakly explained by climate
Abstract
Forests cover about one-third of the global land surface and constitute the largest terrestrial carbon sink. Accurate representation of forests in Earth system models is critical for constraining land–atmosphere carbon exchange. Yet increasing climatic variability perturbs forest functioning and induces pronounced interannual variation (IAV) in forest gross primary productivity (GPP), contributing substantially to uncertainty in the global carbon budget. Here we quantify the contribution of vegetation climatic response to IAV of GPP in temperate deciduous broadleaf forests using long-term eddy-covariance data from nine sites with a model–data fusion approach. Simulations with site-level optimal parameters explained 2-27% of the observed GPP IAV across the nine flux tower sites, while allowing parameters to vary annually improved model performance substantially, with the mean R2 increasing from 0.14 to 0.94. Notably, comparable performance was achieved by optimizing only two to four key traits. Traits related to leaf allocation (specific leaf area, SLA), photosynthetic capacity (leaf carbon-to-nitrogen ratio, LCNR, and electron transport to carboxylation capacity ratio, aJ2V ), and atmospheric drought sensitivity (b1) were the most influential across the nine sites, whereas phenological traits, such as the critical temperature for leaf growing degree days (TGDD) and leaf longevity (LL), contributed little to GPP IAV. However, despite the strong influence of temporal parameter variations on GPP, the inferred parameter–environment relationships remained generally weak and site-dependent. This mismatch underscores the need to incorporate empirical trait–environment relationships into model processes to better capture vegetation climatic response and reduce uncertainty in future model simulations. Plain Language Summary Forests absorb large amounts of carbon dioxide from the atmosphere and play a key role in regulating climate, but their carbon uptake varies strongly from year to year due to changing climatic drivers. This interannual variability remains poorly captured by Earth system models, leading to uncertainty in global carbon cycle projections. In this study, we examined whether incorporating forest functional trait adaptation to climate variability can explain interannual variation of forest carbon uptake. Using a model–data fusion approach across nine long-term observed temperate deciduous forest sites, we found that models with fixed plant traits failed to reproduce observed year-to-year variation in forest productivity. Allowing key traits to vary annually substantially improved model performance. Only a small subset of traits was needed to explain most variability, particularly those related to leaf structure, photosynthetic capacity, and sensitivity to atmospheric drought. In contrast, traits associated with leaf phenology and longevity played a minor role. Although temporal trait variation strongly improves forest productivity simulations, the model-derived parameter–environment relationships were generally weak and inconsistent across sites. These results suggest that improving predictions of future forest carbon uptake under climate change requires incorporating empirically observed trait–environment relationships into model processes. Key points • We investigated forest trait–climate responses and evaluated their impacts on interannual variation in forest carbon uptake • Temporal trait variation can improve forest productivity simulations, but parameter–environment relationships are weak and site-dependent • Improving future forest carbon uptake predictions require incorporating field observed trait–environment relationships into model processes
Related
- Masting Breakdown in European Beech Reduces Fitness Benefits of Masting, Partly Explained by Climate Change
- Accounting for Variability Across Years, Depths, and Non‐Target Reads Improves Quantification of Root Community Composition in Plant Mixtures
- Regional climate variability and rice yields prediction in Bangladesh using machine learning and deep learning models
- MI-PEFT: Mixture-of-Experts Integrated Parameter-Efficient Fine-Tuning Protein Language Models Improves Acidophilic Proteins Classification
- Active sensing explains cortical variability and improves visual information
BibTeX
@misc{pan2026interannual,
title = {{Inter-annual parameter variability improves the predictability of inter-annual GPP but is weakly explained by climate}},
author = {Quan Pan and Marc Peaucelle and Marijn Bauters and Steven de Hertog and Félicien Meunier and Vladislav Bastrikov and Philippe Peylin and Nicolas Viovy and Hans Verbeeck},
howpublished = {HAL},
year = {2026},
month = jul,
doi = {10.22541/essoar.15005595/v1},
url = {https://hal.science/hal-05781534v1},
note = {CC BY-NC 4.0},
}
RIS
TY - UNPB
TI - Inter-annual parameter variability improves the predictability of inter-annual GPP but is weakly explained by climate
AU - Quan Pan
AU - Marc Peaucelle
AU - Marijn Bauters
AU - Steven de Hertog
AU - Félicien Meunier
AU - Vladislav Bastrikov
AU - Philippe Peylin
AU - Nicolas Viovy
AU - Hans Verbeeck
PY - 2026
DA - 2026/07/02
PB - HAL
DO - 10.22541/essoar.15005595/v1
UR - https://hal.science/hal-05781534v1
AB - Forests cover about one-third of the global land surface and constitute the largest terrestrial carbon sink. Accurate representation of forests in Earth system models is critical for constraining land–atmosphere carbon exchange. Yet increasing climatic variability perturbs forest functioning and induces pronounced interannual variation (IAV) in forest gross primary productivity (GPP), contributing substantially to uncertainty in the global carbon budget. Here we quantify the contribution of vegetation climatic response to IAV of GPP in temperate deciduous broadleaf forests using long-term eddy-covariance data from nine sites with a model–data fusion approach. Simulations with site-level optimal parameters explained 2-27% of the observed GPP IAV across the nine flux tower sites, while allowing parameters to vary annually improved model performance substantially, with the mean R2 increasing from 0.14 to 0.94. Notably, comparable performance was achieved by optimizing only two to four key traits. Traits related to leaf allocation (specific leaf area, SLA), photosynthetic capacity (leaf carbon-to-nitrogen ratio, LCNR, and electron transport to carboxylation capacity ratio, aJ2V ), and atmospheric drought sensitivity (b1) were the most influential across the nine sites, whereas phenological traits, such as the critical temperature for leaf growing degree days (TGDD) and leaf longevity (LL), contributed little to GPP IAV. However, despite the strong influence of temporal parameter variations on GPP, the inferred parameter–environment relationships remained generally weak and site-dependent. This mismatch underscores the need to incorporate empirical trait–environment relationships into model processes to better capture vegetation climatic response and reduce uncertainty in future model simulations. Plain Language Summary Forests absorb large amounts of carbon dioxide from the atmosphere and play a key role in regulating climate, but their carbon uptake varies strongly from year to year due to changing climatic drivers. This interannual variability remains poorly captured by Earth system models, leading to uncertainty in global carbon cycle projections. In this study, we examined whether incorporating forest functional trait adaptation to climate variability can explain interannual variation of forest carbon uptake. Using a model–data fusion approach across nine long-term observed temperate deciduous forest sites, we found that models with fixed plant traits failed to reproduce observed year-to-year variation in forest productivity. Allowing key traits to vary annually substantially improved model performance. Only a small subset of traits was needed to explain most variability, particularly those related to leaf structure, photosynthetic capacity, and sensitivity to atmospheric drought. In contrast, traits associated with leaf phenology and longevity played a minor role. Although temporal trait variation strongly improves forest productivity simulations, the model-derived parameter–environment relationships were generally weak and inconsistent across sites. These results suggest that improving predictions of future forest carbon uptake under climate change requires incorporating empirically observed trait–environment relationships into model processes. Key points • We investigated forest trait–climate responses and evaluated their impacts on interannual variation in forest carbon uptake • Temporal trait variation can improve forest productivity simulations, but parameter–environment relationships are weak and site-dependent • Improving future forest carbon uptake predictions require incorporating field observed trait–environment relationships into model processes
ER -