Periodic Reporting for period 1 - CEROBI (Constraining forEst Responses to drOughts with carBon Isotopes)
Période du rapport: 2024-01-01 au 2025-12-31
The project, CEROBI: Constraining forEst Responses to droughts with carBon Isotopes, aims to address the need for reliable climate projections amidst the escalating crisis of drought-induced tree mortality in the mid-latitudes.
Overall Objectives
The core problem addressed by CEROBI is rooted in climate change, which is driving droughts to become more frequent and intense, thereby threatening the ability of forests to function as vital carbon sinks and producewood, fiber and biomass . Currently, Earth System Models (ESMs)—the most advanced tools for predicting future climate—rely on Land Surface Models (LSMs) that are demonstrably biased as they underestimate the severity of drought impacts on ecosystems.
Context and Core Problem
The scientific challenge lies in describing and parameterizing the complex physiological responses of trees to drought, which often leads to mortality through dual pathways: carbon starvation and hydraulic failure. The ORCHIDEE Land Surface Model (LSM) is particularly advanced in this regard: (1) . trees are categorized in different diameter classes to better capture effects related to tree demography and the long-term, delayed impacts of drought ("drought legacy effect») ;(2) Plant water stress calculations are based on the formalism of the hydraulic architecture. This allows the model to compute water availability by considering resistance along the water transport path (roots, sapwood, leaves), which is crucial for simulating hydraulic failure; and (3) Recent developments focused on refining the model's representation of the dynamics of non-structural carbohydrate (NSC) pools, to better simulate the coupled effects of hydraulic failure and carbon starvation.
Despite these advanced mechanistic process representations, the model’s large-scale output still comes with substantial uncertainty. This is because LSMs have over 200 internal parameters many of which are poorly constrained. This results in the insidious problem of equifinality: multiple parameter sets can match current observations but produce widely divergent and unreliable predictions for the future.
The CEROBI project is designed to reduce the issue of parameter uncertainty. By introducing carbon isotopes which is a tracer of plant water-use efficiency, the project will circumvent the equifinality problem. This allows the model to reduce the uncertainty of future projections not by redesigning the existing processes (like the hydraulic architecture), but by rigorously calibrating the internal parameters that govern them.
Pathway to Impact
The ultimate pathway to impact involves using the newly constrained model to perform long-term global simulations under various climate scenarios, finally enabling the identification of regions that are currently well watered but will in the future experience frequent droughts. Such regions should be prioritized for adaptation measures.
Overall Objectives
The core problem addressed by CEROBI is rooted in climate change, which is driving droughts to become more frequent and intense, thereby threatening the ability of forests to function as vital carbon sinks and producewood, fiber and biomass . Currently, Earth System Models (ESMs)—the most advanced tools for predicting future climate—rely on Land Surface Models (LSMs) that are demonstrably biased as they underestimate the severity of drought impacts on ecosystems.
Context and Core Problem
The scientific challenge lies in describing and parameterizing the complex physiological responses of trees to drought, which often leads to mortality through dual pathways: carbon starvation and hydraulic failure. The ORCHIDEE Land Surface Model (LSM) is particularly advanced in this regard: (1) . trees are categorized in different diameter classes to better capture effects related to tree demography and the long-term, delayed impacts of drought ("drought legacy effect») ;(2) Plant water stress calculations are based on the formalism of the hydraulic architecture. This allows the model to compute water availability by considering resistance along the water transport path (roots, sapwood, leaves), which is crucial for simulating hydraulic failure; and (3) Recent developments focused on refining the model's representation of the dynamics of non-structural carbohydrate (NSC) pools, to better simulate the coupled effects of hydraulic failure and carbon starvation.
Despite these advanced mechanistic process representations, the model’s large-scale output still comes with substantial uncertainty. This is because LSMs have over 200 internal parameters many of which are poorly constrained. This results in the insidious problem of equifinality: multiple parameter sets can match current observations but produce widely divergent and unreliable predictions for the future.
The CEROBI project is designed to reduce the issue of parameter uncertainty. By introducing carbon isotopes which is a tracer of plant water-use efficiency, the project will circumvent the equifinality problem. This allows the model to reduce the uncertainty of future projections not by redesigning the existing processes (like the hydraulic architecture), but by rigorously calibrating the internal parameters that govern them.
Pathway to Impact
The ultimate pathway to impact involves using the newly constrained model to perform long-term global simulations under various climate scenarios, finally enabling the identification of regions that are currently well watered but will in the future experience frequent droughts. Such regions should be prioritized for adaptation measures.
The CEROBI project developed a robust framework for simulating forest drought responses in the ORCHIDEE Land Surface Model through substantial advances in initialization and validation. These methodological improvements revealed that significant issues affecting forest carbon sink predictions must be addressed before isotopic constraints can be meaningfully applied to the model.
1. Model Initialization and Demographic State Variables
Hydraulic architecture calculations are fundamentally dependent on tree height and stand density, as these structural variables determine hydraulic resistance within the soil-plant-atmosphere continuum and govern water transport dynamics. If tree heights and stand structure are unrealistic, the simulated hydraulic architecture becomes flawed, leading to inaccurate representations of plant water stress. Because isotopic fractionation is closely linked to stomatal conductance and transpiration—both controlled by water stress—realistic hydraulic architecture is essential for any subsequent isotopic analysis to be physiologically plausible. To address this, an initialization procedure called "blended spin-up" was developed to match simulated tree diameter distributions with European forest inventory data from 2002-2010. The method was refined to blend tree diameters separately for each Plant Functional Type (PFT) to ensure realistic species-specific size structures. However, this refinement initially violated carbon mass conservation because ORCHIDEE simulates wood product pools at the grid cell level rather than the PFT level. Code modifications were required to restore mass conservation while maintaining demographic realism, ensuring diameter adjustments remained synchronized with carbon and nitrogen pools in the vegetation, litter, soil, and wood products.
2. Impact on Carbon Flux Trajectories
To illustrate the impact of this model spin-up, two parallel 100-year simulations of European forests were conducted with identical climate forcing, comparing standard initialization against the refined PFT-specific diameter blended spin-up. The results revealed a substantial impact on long-term carbon cycle projections, where the non-initialized simulation showed European forests as a net carbon source over the century, while the simulation starting from the blended spin-up showed them acting as a net carbon sink. This reversal in the sign of the carbon flux has profound implications for climate mitigation assessments. This difference occurs because initial tree size distributions determine current biomass stocks, mortality vulnerability, growth potential, and demographic turnover rates. Consequently, incorrect initialization creates unrealistic demographic structures that propagate through decades of simulation, producing fundamentally wrong trajectories and highlighting that demographic initialization is a first-order control on model predictions rather than a technical detail.
3. Model Evaluation and Observational Uncertainty
Rigorous assessment of model performance was conducted to validate the initial model states and subsequent forest dynamics over Europe. Since a reasonable relationship is expected between forest age and structure in managed forests, the model's capability to reproduce observed age structure was evaluated against two forest age datasets. Spatial patterns of age were well captured, correctly representing the geographic distribution of young regenerating forests versus older stands across Northern Europe, though specific age values were sometimes over- or underestimated. Discrepancies between the two observational datasets were comparable in magnitude to the differences between the model and observations, emphasizing that model validation is constrained by substantial uncertainty in the benchmark data themselves. Furthermore, while the blended spin-up utilized ground-based diameter products, it was necessary to demonstrate that simulated diameters aligned with acceptable tree heights and stand densities. Aboveground biomass was used as an independent constraint since it is the product of diameter squared, height, stand density, and wood density. The initialization procedure resulted in a 50% reduction in errors for aboveground biomass across Europe compared to non-initialized simulations, though residual biases remained in regions like Scandinavia. These discrepancies stem from uncertainties in the forest inventory, inconsistencies between in-situ and satellite-derived estimates, and potential structural limitations in the model's allometry and biomass allocation schemes. To independently validate growth dynamics, forest state variables were initialized for 1990 and run forward to 2010, showing good agreement between simulated and observed diameter distributions.
4. Assessment of Forest Response to Drought
Forest resilience was evaluated by testing the ability of the ORCHIDEE model to simulate successional biomass recovery following drought, which was represented here as a clear-cut. The model demonstrated high accuracy in temperate forests, where stomatal closure and cavitation thresholds were identified as the primary physiological parameters governing the trade-off between carbon starvation and hydraulic failure. Ultimately, CEROBI established a foundation for future research by providing a validated model infrastructure with realistic initial conditions and identifying the key mechanisms governing how forests respond to drought.
1. Model Initialization and Demographic State Variables
Hydraulic architecture calculations are fundamentally dependent on tree height and stand density, as these structural variables determine hydraulic resistance within the soil-plant-atmosphere continuum and govern water transport dynamics. If tree heights and stand structure are unrealistic, the simulated hydraulic architecture becomes flawed, leading to inaccurate representations of plant water stress. Because isotopic fractionation is closely linked to stomatal conductance and transpiration—both controlled by water stress—realistic hydraulic architecture is essential for any subsequent isotopic analysis to be physiologically plausible. To address this, an initialization procedure called "blended spin-up" was developed to match simulated tree diameter distributions with European forest inventory data from 2002-2010. The method was refined to blend tree diameters separately for each Plant Functional Type (PFT) to ensure realistic species-specific size structures. However, this refinement initially violated carbon mass conservation because ORCHIDEE simulates wood product pools at the grid cell level rather than the PFT level. Code modifications were required to restore mass conservation while maintaining demographic realism, ensuring diameter adjustments remained synchronized with carbon and nitrogen pools in the vegetation, litter, soil, and wood products.
2. Impact on Carbon Flux Trajectories
To illustrate the impact of this model spin-up, two parallel 100-year simulations of European forests were conducted with identical climate forcing, comparing standard initialization against the refined PFT-specific diameter blended spin-up. The results revealed a substantial impact on long-term carbon cycle projections, where the non-initialized simulation showed European forests as a net carbon source over the century, while the simulation starting from the blended spin-up showed them acting as a net carbon sink. This reversal in the sign of the carbon flux has profound implications for climate mitigation assessments. This difference occurs because initial tree size distributions determine current biomass stocks, mortality vulnerability, growth potential, and demographic turnover rates. Consequently, incorrect initialization creates unrealistic demographic structures that propagate through decades of simulation, producing fundamentally wrong trajectories and highlighting that demographic initialization is a first-order control on model predictions rather than a technical detail.
3. Model Evaluation and Observational Uncertainty
Rigorous assessment of model performance was conducted to validate the initial model states and subsequent forest dynamics over Europe. Since a reasonable relationship is expected between forest age and structure in managed forests, the model's capability to reproduce observed age structure was evaluated against two forest age datasets. Spatial patterns of age were well captured, correctly representing the geographic distribution of young regenerating forests versus older stands across Northern Europe, though specific age values were sometimes over- or underestimated. Discrepancies between the two observational datasets were comparable in magnitude to the differences between the model and observations, emphasizing that model validation is constrained by substantial uncertainty in the benchmark data themselves. Furthermore, while the blended spin-up utilized ground-based diameter products, it was necessary to demonstrate that simulated diameters aligned with acceptable tree heights and stand densities. Aboveground biomass was used as an independent constraint since it is the product of diameter squared, height, stand density, and wood density. The initialization procedure resulted in a 50% reduction in errors for aboveground biomass across Europe compared to non-initialized simulations, though residual biases remained in regions like Scandinavia. These discrepancies stem from uncertainties in the forest inventory, inconsistencies between in-situ and satellite-derived estimates, and potential structural limitations in the model's allometry and biomass allocation schemes. To independently validate growth dynamics, forest state variables were initialized for 1990 and run forward to 2010, showing good agreement between simulated and observed diameter distributions.
4. Assessment of Forest Response to Drought
Forest resilience was evaluated by testing the ability of the ORCHIDEE model to simulate successional biomass recovery following drought, which was represented here as a clear-cut. The model demonstrated high accuracy in temperate forests, where stomatal closure and cavitation thresholds were identified as the primary physiological parameters governing the trade-off between carbon starvation and hydraulic failure. Ultimately, CEROBI established a foundation for future research by providing a validated model infrastructure with realistic initial conditions and identifying the key mechanisms governing how forests respond to drought.
The CEROBI project provides a methodological template for LSMs with demography that can be adopted by other modeling groups, along with a quantification of initialization impacts. The evaluation establishes a Land Surface Model baseline performance to reproduce the European forest structure and carbon sink, documenting its strengths and limitations for European applications. In addition, the project established a systematic benchmarking workflow for Land Surface Models regarding forest response to disturbances, such as drought, at multiple sites, while open-source tools enable the community to replicate and build upon this work. It allows for the identification of critical parameters responsible for hydraulic failure as a necessary step to constrain them.
The proof of concept of the blended-spinup focused on European forests only. Application to other continents would require reliable inventory data structures for each region or continent. More accurate inventories of age are also needed to better constrain the age-carbon sink relationship of the forest.
Beyond CEROBI, the findings highlight a systemic issue in Earth System Model development: increasing process complexity—including demography, hydraulics, and dynamics—requires commensurate attention to initialization. Many so-called "next-generation" LSMs with demographic features may be producing unreliable projections due to inadequate initialization. The community needs protocols for demographic initialization analogous to established soil carbon spin-up procedures, making this work essential for improving the credibility of forest carbon sink projections.
In line with the research tradition within the Earth System Modeling community, all model developments are open source under the CeCiLL license. The open-source tools and data products include:
An isotope-enabled land surface model with carbon pools in equilibrium
Python code of the initialization procedure with corrected carbon conservation and PFT-specific initialization capability
Initialized forest state files for European simulations
Evaluation datasets and analysis scripts for reproducible research
The proof of concept of the blended-spinup focused on European forests only. Application to other continents would require reliable inventory data structures for each region or continent. More accurate inventories of age are also needed to better constrain the age-carbon sink relationship of the forest.
Beyond CEROBI, the findings highlight a systemic issue in Earth System Model development: increasing process complexity—including demography, hydraulics, and dynamics—requires commensurate attention to initialization. Many so-called "next-generation" LSMs with demographic features may be producing unreliable projections due to inadequate initialization. The community needs protocols for demographic initialization analogous to established soil carbon spin-up procedures, making this work essential for improving the credibility of forest carbon sink projections.
In line with the research tradition within the Earth System Modeling community, all model developments are open source under the CeCiLL license. The open-source tools and data products include:
An isotope-enabled land surface model with carbon pools in equilibrium
Python code of the initialization procedure with corrected carbon conservation and PFT-specific initialization capability
Initialized forest state files for European simulations
Evaluation datasets and analysis scripts for reproducible research