The Simplified Simple Biosphere Model version 4 coupled with the Top-down Representation of Interactive Foliage and Flora Including Dynamics Model (SSiB4/TRIFFID) is used in this study to investigate the response of terrestrial ecosystems to stomatal uptake of O3. SSiB4/TRIFFID is a process-based land surface model integrated with dynamic global vegetation model. The model is designed to simulate the energy, water, and carbon cycles at the terrestrial surface, incorporating well-established mechanisms for calculating multi-timescale vegetation dynamics and land surface characteristics including vegetation cover and structure (Zhang et al., 2015; Liu et al., 2019). The photosynthesis and stomatal processes are based on the framework developed by Collatz et al. (1991) and incorporated by Zhan et al. (2003), with a leaf-to-canopy scaling strategy introduced by Sellers et al. (1996). To date, SSiB4/TRIFFID employs 7 plant functional types (PFTs) including evergreen broadleaf forest (EBF), evergreen needleleaf forest (ENF), deciduous broadleaf forest (DBF), shrubland (Shrub), C3 grassland (C3), C4 grassland (C4), and tundra (Tundra) (Liu et al., 2019). In this study, all of the SSiB4/TRIFFID simulations are conducted at a spatial resolution of 1°×1° and a temporal interval of 3 h.

Six O3–vegetation damage schemes were implemented in SSiB4/TRIFFID, including: (1) the Lombardozzi et al. (2015) scheme used in CLM4.5 (hereafter L2015), (2) Li et al. (2024) scheme applied in CTSM 5.2 (hereafter Li2024), (3) the Sitch et al. (2007) scheme used in MOSES (hereafter S2007), (4) re-calibrated S2007 method by Ma et al. (2023) (hereafter CS2007), (5) LMA-based approach by Ma et al. (2023) using gridded global LMA map (hereafter LMAgrid), and (6) prescribed PFT-specific LMA values (hereafter LMApft). Although these schemes differ in their treatment of O3 damage, they are all based on stomatal O3 uptake. Therefore, we first define the instantaneous stomatal O3 flux fO3, which represents the O3 uptake through stomata at each modelling time step: 1fO3=[O3]r+kO3×rs where [O3] is the O3 concentration at the top of the canopy, r denotes the aerodynamic and boundary layer resistance between leaf surface and the reference level, kO3 is the ratio of leaf resistance for O3 relative to that for water vapor (1.67), and rs refers to the stomatal conductance for H2O (ms-1). Based on fO3, the excessive instantaneous O3 flux above a threshold y (nmolm-2s-1) is defined as: 2U=max⁡(fO3-y,0) where y is the area-based flux threshold. U represents the portion of instantaneous stomatal O3 uptake that exceeds the critical threshold and is therefore assumed to trigger phytotoxic damage, and is used differently among parameterizations. In S2007 and CS2007, U is used directly at each model timestep to calculate the O3 damage factor. Whereas in L2015 and Li2024, U is accumulated into a prognostic cumulative O3 uptake metric to represent chronic O3 exposure and its impacts. In the LMA-based approaches, the area-based stomatal O3 uptake is reformulated on a leaf-mass basis. The key characteristics and differences among the six O3 damage schemes are summarized in Table 1, and the detailed formulation of each scheme is described below.

Sitch et al. (2007) proposed a semi-mechanistic parameterization to represent the transient O3-induced damage to plants through the coupling between stomatal conductance and photosynthetic rate (S2007). The O3 modification factor on photosynthesis (F) is formulated as a function of the excessive instantaneous stomatal O3 flux: 3FO3=1-αPFT×U where U is the excessive instantaneous stomatal O3 flux defined in Eq. (2), and αPFT is the PFT-specific O3 sensitivity coefficient derived from observations (Sitch et al., 2007). The coupled formulation of Eqs. (1)–(3) produces a quadratic in FO3 that can be solved analytically at each timestep. The resulting FO3 is then applied as a reduction factor on net photosynthetic rate, while stomatal conductance is subsequently calculated using the semi-empirical Ball–Berry formulation driven by the O3-modified photosynthetic rate (Collatz et al., 1991; Zhan et al., 2003).

Within the same framework, Ma et al. (2023) further recalibrated the sensitivity parameters of S2007 using observed dose–response relationships compiled from a wide range of field and experimental studies. This recalibrated version is denoted as CS2007 in this study. Although CS2007 is not structurally independent from S2007, it differs in the updated PFT-specific coefficients and thresholds. This distinction allows us to evaluate how updated observational constraints on plant O3 sensitivity influence simulated O3 vegetation damage within the original S2007 framework. It should be noted that both S2007 and CS2007 provide low-, moderate-, and high-sensitivity parameter sets to represent uncertainty in plant O3 responses. In this study, we used the moderate-sensitivity parameter sets as default configuration for both schemes to ensure a consistent comparison with the other parameterizations. The SSiB4/TRIFFID implementation of S2007 and CS2007 is summarized in Table S1 in the Supplement, and the corresponding sensitivity parameters are shown in Table S2. More details of the two schemes are provided in Sitch et al. (2007) and Ma et al. (2023).

Unlike S2007 and CS2007, which diagnose O3 damage from instantaneous excessive stomatal O3 uptake, L2015 and Li2024 calculate O3 damage based on cumulative uptake of O3 (CUO). According to Lombardozzi et al. (2015), CUO is calculated as a prognostic cumulative O3 exposure metric that evolves with O3 uptake, decay, and healing process. 4CUOt=CUOt-1(1-D)+U(1-H)5D=Δtlleaf×3600×24×365,evergreen0,else6H=max⁡1-LAIt-1LAIt,0,for all PFTs where CUOt and CUOt-1 denote the cumulative O3 stress at the current and previous model time steps, respectively. U is the excessive instantaneous stomatal O3 flux defined in Eq. (2), representing the newly added O3 uptake at the current time step. D denotes the decay factor applied to the previously accumulated CUO, and is parameterized as a function of the leaf longevity lleaf (year). H denotes the healing factor, which is derived from the change in leaf area index (LAI) between the current and previous time steps. It represents the attenuation of previously accumulated O3stress by new leaf growth, assuming that newly formed leaves are initially free of O3 damage. Following Lombardozzi et al. (2015), CUO is accumulated only when LAI≥0.5. When LAI falls below this threshold, CUO is reset to zero to represent the end of the growing season and O3 stress accumulation.

Based on the CUO formulation described above, Lombardozzi et al. (2015) developed separate empirical response functions for photosynthesis and stomatal conductance using experimental data compiled from the peer-reviewed literature. Accordingly, the O3 modification factors for photosynthesis (FO3_A) and stomatal conductance (FO3_g) are expressed as linear functions of CUO: 7FO3_A=ap+bp×CUO8FO3_g=ag+bg×CUO where ap and bp are the intercept and slope coefficients for the photosynthetic response, and ag and bg are the corresponding coefficients for the stomatal conductance, respectively. The detailed response functions and PFT-specific parameter settings are shown in Table S2.

Building on the cumulative O3 stress framework of L2015, Li2024 updates the response functions using a substantially expanded observational database. Using an expanded O3 fumigation database compiled from the peer-reviewed literature, with a sample size approximately six times larger than that used in L2015, Li et al. (2024) developed PFT-specific response functions for photosynthesis and stomatal conductance. Unlike the linear or constant response functions that are used in L2015, Li2024 considers both linear and nonlinear functions to better capture the observed physiological responses across a wider range of vegetation types. Another difference between L2015 and Li2024 lies in the treatment of decay and healing process in the CUO calculation. In L2015, the healing factor is applied to the newly added uptake term. In contrast, following the logic of Li et al. (2024), both decay and healing are applied to the previously accumulated O3 stress for different PFTs. In our implementation, CUO is calculated as: 9CUOt=CUOt-1(1-D)(1-H)+U10D=Δtlleaf×3600×24×365evergreen0else11H=0evergreenmax⁡1-LAIt-1LAIt,0else. The main difference between this formulation and L2015 is how the healing process is applied. In L2015, H modifies the newly added uptake term. In Li2024, however, it is applied to the previously accumulated CUO, representing the attenuation of existing O3 stress by newly formed leaves. The two schemes also differ in their CUO accumulation periods. In Li2024, U contributes to CUO only during daytime, which is defined as timesteps when absorbed solar radiation is greater than zero. In contrast, L2015 retains full-day accumulation. The key formulations and parameter settings for the L2015 and Li2024 schemes implemented in SSiB4/TRIFFID are provided in Table S1.

Following the stomatal flux-based O3 damage framework of S2007, Ma et al. (2023) proposed a trait-based O3 damage parameterization in which the plant O3 sensitivity is represented using leaf mass per area. By using LMA to convert the area-based stomatal O3 flux into a mass-based flux, this conversion allows a unified representation of O3 sensitivity with a single PFT-independent parameter a across different plant species. The relationship between the original area-based sensitivity coefficient αPFT and the mass-based sensitivity parameter a is expressed as: 12a=αPFT×LMA.

Accordingly, the O3 modification factor FO3 in Eq. (3) is reformulated as: 13FO3=1-a×max⁡fO3LMA-x,0 where a is the PFT-independent mass-based O3 sensitivity parameter (nmol-1sg) within each LMA-based implementation. Specifically, we set to a=3.5 for LMAgrid and 2.82 for LMApft as suggested by Ma et al. (2023). x is the mass-based flux threshold. Following Feng et al. (2018) and Ma et al. (2023), we set x=0.019nmolg-1s-1 based on the observations. In this study, we implement two versions of the LMA-based scheme with the same mathematical formulation but different sources of LMA information. The LMAgrid scheme uses a spatially explicit global LMA map from Moreno-Martínez et al. (2018), allowing O3 sensitivity to vary among grid cells. In contrast, the LMApft scheme uses prescribed PFT-specific LMA values, thereby keeping O3 sensitivity spatially uniform within each PFT (Table S2). Details of the LMA dataset and processing are described in Sect. 2.3.2.

For regional-scale model evaluation over China, we used independent satellite- and data-driven products of leaf area index (LAI) and gross primary productivity (GPP) to assess the simulated vegetation dynamics. For LAI, we obtain the Global Land Surface Satellite (GLASS) LAI product, which is generated using general regression neural networks and remote-sensing observations. Compared to other long-term LAI products, the GLASS LAI is characterized by high accuracy and temporal stability (Liang et al., 2021). In this study, the original GLASS LAI data at 0.05° for 2011–2020 were aggregated to 1° to validate and evaluate the SSiB4/TRIFFID simulations.

We used the GOSIF and FLUXCOM GPP product as independent benchmarks to evaluate the simulated GPP over China. GOSIF GPP is a satellite-based product derived from the Orbiting Carbon Observatory-2 (OCO-2) solar-induced chlorophyll fluorescence (SIF) data using a light-use-efficiency framework. GOSIF GPP provides seamless global estimates at 0.05° resolution since 2000 (Li and Xiao, 2019). The FLUXCOM RS+METEO GPP product is derived using machine-learning approaches that integrate eddy-covariance carbon flux measurements with satellite remote sensing and meteorological data, which is available at 0.5° resolution for 1979–2018 (Jung et al., 2020). We obtain GOSIF and FLUXCOM GPP data from 2011–2020 (2018 for FLUXCOM) and interpolate these products to a spatial resolution of 1° to match the model configuration. Both of the datasets have been widely applied in regional and global assessments of ecosystem productivity and model evaluations.

To ensure a consistent intercomparison of vegetation responses across different O3 damage parameterizations, we estimate PFT-specific O3 sensitivities from the dose–response relationships between relative GPP change (RGPP) and PODy. Here, PODy refers to the phytotoxic O3 dose over a flux threshold of y (nmolm-2s-1). As a flux-based diagnostic metric, PODy directly links to biologically relevant stomatal O3 uptake by accounting for O3 concentration, stomatal conductance, and exposure duration (Mills et al., 2011; Pleijel et al., 2022). In this study, PODy is calculated from the simulated stomatal O3 flux as: 16PODy=∫max⁡(fO3-y,0)dt where fO3 is the instantaneous stomatal O3 flux defined in Eq. (1), and y is the prescribed flux threshold. We obtain the annual PODy by accumulating the above-threshold stomatal O3 flux within each simulation year. Different from CUO, which represents O3 stress after accounting for decay and healing, PODy is used as a diagnostic dose metric to compare vegetation responses among different parameterizations on a consistent basis. The dose–response relationship between RGPP and PODy can be expressed as: RGPP=b×PODy+c where RGPP denotes the relative change in GPP under O3 exposure compared with the O3OFF experiment, b is the slope of the linear relationship, and c is the intercept. Following Ma et al. (2023), the slope b is used to quantify the PFT-specific O3 sensitivity for each scheme, with higher negative value indicating stronger GPP reduction per unit PODy. These sensitivities are compared across different PFTs and parameterization schemes.

Observations of O3 sensitivity are assembled from recent literature to evaluate the performance of the simulated dose–response relationships. We obtain the observational results from Ma et al. (2023) and Li et al. (2024), both of which compiled field measurements and O3 fumigation experiments across a range of literature sources. For Ma et al. (2023), PFT-specific O3 sensitivities were compiled from published studies based on regressions between biotic indicators and PODy. We use the median value for each PFT as the observed sensitivity and the reported minimum-maximum range as an uncertainty interval. In contrast, Li et al. (2024) provided a comprehensive database of >4000 paired observations of photosynthesis and PODy. We therefore derive the slopes of PFT-specific dose–response relationships directly from the raw data as the observed O3 sensitivity (Fig. S3). Uncertainty was quantified using bootstrap resampling (1000 iterations), with the 2.5th and 97.5th percentiles of the slope distribution taken as the confidence intervals (CIs).

Six O3–vegetation damage parameterizations were implemented in SSiB4/TRIFFID against field observations at two flux-tower sites with O3 flux measurements. Both sites have been widely used in previous studies to investigate O3–vegetation interactions and dry deposition (Ducker et al., 2018; Li et al., 2020; Lin et al., 2020). Figure 1 presents the simulated O3 flux and GPP for US-Ha1 in 1998 and FI-Hyy in 2009. The same evaluation for US-Ha1 in 1999 and FI-Hyy in 2008 is provided in Fig. S4 as well. Overall, all schemes agree well with observations in both amplitude and seasonality of stomatal O3 flux and GPP. However, stomatal O3 flux tends to be underestimated in November-February, likely due to the simulated weak photosynthetic activity in winter (Fig. 1).

Among the five schemes, L2015 and Li2024 generally exhibit smaller biases in simulated O3 flux, followed by LMApft and CS2007. The S2007 scheme consistently simulates the lowest O3 uptake (Fig. 1b and e). All schemes indicate O3-induced reductions in GPP at both sites, while the simulated magnitude varies substantially across parameterizations. At US-Ha1, L2015 produces the weakest O3-induced GPP reduction, followed by CS2007 and LMApft, whereas Li2024 and S2007 yield the largest reductions (Fig. 1c). At FI-Hyy, the simulated O3 impacts on GPP is weaker with a smaller inter-scheme spread. The CS2007 and LMApft simulate relatively smaller reductions, whereas L2015, Li2024, and S2007 produce larger GPP losses (Fig. 1e). We further quantified inter-scheme differences by computing the mean absolute error (MAE) of simulated stomatal O3 flux and GPP relative to site observations (Fig. S5). For stomatal O3 flux, Li2024 consistently yields the lowest MAE across all four site-years. For GPP, however, the MAE varies across site-years: L2015 shows the smallest MAE at FI-Hyy, while Li2024 and CS2007 produce the lowest errors in US-Ha1. Although the magnitude of simulated GPP response differs among the schemes, accounting for O3 effects consistently reduces GPP bias relative to the O3OFF experiment. Overall, all schemes reproduce the observed magnitude and seasonal cycle of both stomatal O3 flux and GPP values. Among the schemes, Li2024 shows the best agreement with observations for stomatal O3 flux, as indicated by the lowest MAE across sites.

In regional simulation experiments, we first quantified the baseline performance of SSiB4/TRIFFID over China under the control experiment without O3-induced vegetation damage (i.e. O3OFF). Simulated GPP and LAI are compared with satellite measurements, including GPP from GOSIF and LAI from GLASS. The results indicate that SSiB4/TRIFFID is capable of realistically reproducing vegetation growth and photosynthetic activity over China, with high spatial correlations of 0.94 for GPP and 0.89 for LAI in the growing season (April–September), respectively (Figs. 2 and S6). Consistent with satellite observations, high GPP values are found in southern China and eastern China. At the national scale, the O3OFF experiment generally overestimates GPP and LAI compared to observations, with an annual positive bias of 14.88 % in GPP and 22.78 % in LAI, respectively.

Using hourly surface O3 concentrations from HrSOD, we quantified O3-induced vegetation damage over China with the six O3–vegetation damage parameterizations implemented in SSiB4/TRIFFID model (Fig. 3). Across all schemes, ambient O3 pollution substantially suppresses vegetation photosynthesis nationwide. The largest reductions in GPP occur in southern China, where densely-distributed vegetation coincides with elevated O3 exposure. In eastern China, despite the comparatively lower vegetation distribution, severe O3 pollution still leads to pronounced photosynthetic damage. Based on the multi-scheme assessment, we estimate an average O3-induced GPP loss over China by 20.76 %, with a spread of 16.81 %–26.63 % among the six schemes during growing season (April–September). As a result, the national annual GPP decreases from 8.90 Pg C in O3OFF to 7.07 Pg C, with an inter-scheme range of 6.27–7.51 Pg C. The seasonal evolution of O3-induced GPP loss exhibits a clear unimodal pattern, with the largest reductions occurring during summer, coincident with peak O3 concentrations and maximum vegetation growth.

Overall, ambient O3 pollution leads to substantial reductions in GPP across China. Comparisons against observations further indicate that implementation of O3 damage schemes into land surface model can improve the simulation of carbon fluxes over China. Meanwhile, the six parameterizations yield marked discrepancies in O3-induced GPP damage, as evidenced by the large inter-scheme spread as shown by the shading in Fig. 3c. These divergences are particularly pronounced during the late growing season. At the annual scale, the inter-scheme spread is comparable in magnitude to the GPP loss itself, which indicates a large uncertainty in assessing O3-induced GPP reductions in China. Given the severe O3 burden in China and the pronounced differences among current parameterizations, a comprehensive investigation is needed to diagnose inter-scheme divergences and to evaluate their performance in China, which is crucial for narrowing the uncertainties in simulated O3 impacts on vegetation and improving regional GPP simulations over China.

Figure 4 shows the spatial distribution of O3-induced GPP loss simulated by the six O3–vegetation damage parameterizations. All schemes reproduce a similar spatial pattern, with the strongest losses in southern China, followed by eastern China, and weak O3 impacts are found in northwestern China where vegetation is sparsely distributed and baseline GPP is low. In contrast to the consistent spatial pattern, the simulated GPP loss differs substantially among the six schemes: Li2024 and S2007 yield the strongest O3-induced GPP reductions, while L2015 and CS2007 simulate the weakest responses, and the LMA-based schemes (LMAgrid and LMApft) show intermediate reductions. The O3-induced suppression in photosynthesis reduces carbon assimilation of plants, which in turn constrains vegetation growth and results in a modest decrease in LAI (Fig. S7). The schemes also diverge in simulated stomatal O3 uptake and associated impacts on stomatal conductance, with L2015 and Li2024 yielding the highest O3 uptake and S2007 the lowest, which consists with the site-level evaluation.

However, although higher O3 flux generally correlates with increased photosynthetic and stomatal damage within a single scheme, this relationship is not consistent across different parameterizations (Figs. 4, S8 and S9). This mismatch occurs because stomatal O3 uptake represents the amount of O3 entering the leaf, whereas GPP damage depends on how stomatal O3uptake is converted into photosynthetic reduction within each parameterization. The conversion differs among schemes due to discrepancies in response functions, phytotoxic thresholds, sensitivity coefficient settings, and the treatment of photosynthesis–stomatal coupling. For example, L2015 applies separate response functions for photosynthesis and stomatal conductance, and the response functions are either constant or linear with relatively small coefficients for several PFTs. Therefore, high stomatal O3 uptake does not necessarily lead to a strong GPP response. In contrast, the strong GPP reductions simulated by S2007 under relatively low O3uptake mainly reflect its parameter settings, including the prescribed sensitivity coefficients and flux thresholds. After recalibration, CS2007 yields weaker GPP reductions within the same stomatal flux-based framework. These variations indicate substantial differences in vegetation O3 sensitivity among schemes and highlight the need for a systematic assessment of scheme-specific O3 sensitivity. In the subsequent analysis, we quantify the vegetation O3 sensitivity of the six schemes and evaluate their performance against observational constraints.

Based on the simulated O3-induced vegetation damage over China, we derive the response relationships between GPP and O3 uptake for each of the six parameterizations. Specifically, we evaluate the RGPP as a function of PODy at the annual scale. As shown in Fig. 5, significant discrepancies arise among the schemes in the representation of RGPP–PODy relationship. Under the L2015 scheme, RGPP exhibits little sensitivity to increasing O3 uptake. This is because the O3 damage factor in L2015 remains constant for many PFTs, lacking a response to additional O3 absorption. As a result, while O3-induced GPP damage is simulated, the RGPP response to PODy is not captured. In contrast, S2007 displays the highest O3 sensitivity, which explains the substantial reductions in GPP under lower stomatal O3 uptake. The remaining schemes demonstrate moderate sensitivities with a small inter-scheme discrepancy: Li2024 shows a higher sensitivity than the LMA-based schemes, and the O3 sensitivity of CS2007 is slightly lower.

In addition to inter-scheme differences, significant variation in O3 sensitivity is observed across PFTs. According to our results, C3 vegetation, shrubs, and deciduous broadleaf forests (DBF) exhibit relatively high O3 sensitivity, whereas evergreen broadleaf forests (EBF), evergreen needleleaf forests (ENF), and C4 vegetation show greater tolerance to O3 exposure (Fig. 5). To quantify vegetation O3 sensitivity at the PFT level, we used the slope of the RGPP–PODy relationships in Fig. 5 as an indicator of O3 sensitivity for each PFT. These modelled sensitivities were then compared against observations from Ma et al. (2023) and Li et al. (2024), who synthesized field measurements and experimental results from peer-reviewed literature. The detailed methodology used to derive O3 sensitivity from simulation and observation is provided in Sect. 2.

Figure 6 compares the modelled PFT-specific O3 sensitivities with observation datasets derived from Ma et al. (2023) and Li et al. (2024). Given the differences in observed O3 sensitivity between these two sources, we present the results from both datasets. Parameterizations exhibiting O3 sensitivities within the confidence intervals of both datasets or between them are considered the most reliable. Schemes that fall within the confidence intervals of only one dataset are considered moderately reliable, while those deviate from both ranges are regarded less reliable. For EBF, the Li2024, CS2007, and the LMA-based schemes fall within the confidence intervals of Ma et al. (2023) and exhibit sensitivities between the two observational datasets. L2015 substantially underestimates O3 sensitivity. As for S2007, a positive slope has been fitted to the RGPP–PODy relationship, which is inconsistent with expected behaviour and likely caused by sparse sampling and its distribution. For ENF, Li2024, S2007, and LMAgrid agree well with the two observational estimates, while L2015, CS2007, and LMApft fall outside the uncertainty range. For C3 and C4 grasses, sensitivity estimates remain highly uncertain due to the limited availability of observational data. For C3 grasses, Li2024 and the LMA-based schemes agree well with Ma et al. (2023), whereas for C4 vegetation, S2007 exhibits the closest agreement with both observational datasets. For shrubs, Li2024, CS2007, and the LMA-based schemes remain consistent with observations. For DBF, the LMA-based schemes show the strongest consistency, while Li2024, S2007, and CS2007 also fall within the observational confidence intervals.

Finally, we averaged the observational O3 sensitivities from the two datasets and benchmarked the simulated sensitivities across PFTs. Scheme performance was assessed using the MAE between modelled and observed sensitivities. Figure 7 shows that Li2024 matches observations most closely (mean bias=0.234; MAE=0.292), followed by LMAgrid and LMApft with similar average biases. CS2007 consistently underestimates O3 sensitivity across all PFTs. S2007 exhibits significant PFT-dependent errors, which results in a large MAE despite a modest mean bias. L2015 shows very weak O3 sensitivity, implying that vegetation responses to additional O3 uptake are not realistically represented. Overall, our assessment based on vegetation O3 sensitivity offers a clear, observation-based benchmark for evaluating reliability of O3–vegetation damage parameterization schemes. Among the six schemes, Li2024 shows the most consistent agreement with observed sensitivities across PFTs, and the LMA-based approaches show similarly good performance with modest bias. This benchmark helps identify the sources of inter-scheme differences in simulated O3 damage and provides a consistent basis for scheme comparison that is less affected by model-dependent biases in simulated GPP.

Although the Li2024 scheme performs well in the inter-scheme comparison, our SSiB4/TRIFFID implementation shows strong CUO accumulation, which affects the simulated seasonal cycle of GPP. Unlike S2007 and the LMA-based schemes, which diagnose O3 damage from instantaneous stomatal O3 flux, Li2024 links the O3 damage factor to CUO, which is updated each timestep through uptake, decay, and healing process (see Sect. 2.2). In the original implementation, the LAI-based healing term reduces CUO mainly during leaf area expansion, while deciduous PFTs lack a continuous decay pathway after LAI reaches its seasonal peak. As a result, the accumulated CUO can lead to a persistent increasing suppression of photosynthesis over the growing season (Fig. 8a). Consequently, this carry-over effect from early-season CUO accumulation leads to a pronounced late-season GPP reduction and biases the simulated seasonal cycle (Fig. 8c).

To address these issues, we implement two adjustments to Li2024. First, we extend the decay term of CUO to all PFTs, instead of restricting it to evergreen species. Second, we updated the key parameter lleaf by prescribing PFT-specific values using observed leaf lifespan estimates. These updates are designed to reduce persistent CUO carry-over during the mid-to-late growing season. As shown in Fig. 8a, the revised formulation (Li2024_modify) shows a pronounced reduction in CUO accumulation with a similar instantaneous stomatal O3 uptake, indicating a altered CUO dynamics rather than changes in stomatal absorption. The reduced CUO weakens O3-induced GPP damage, lowering the annual GPP MAE by 15.9 % and increasing the spatial correlation with observations (Fig. 8b). Notably, the improvement is stronger under high-GPP conditions, where the original Li2024 scheme exhibits pronounced underestimations, implying that the overestimation of GPP damage is largely reduced in densely vegetated regions and seasons. As for seasonality, the reduced CUO limits the carry-over of accumulated O3 stress. Consequently, GPP increases later in the growing season, particularly in summer and autumn with high O3 exposures and high ecosystem productivity, which improves the simulated seasonal cycle of GPP compared with observations (Fig. 8c). Spatially, the strongest GPP recovery occurs over southern China (Fig. 8d), a hotspot of stomatal O3 uptake and O3-induced GPP reductions (Fig. 8d). Overall, the trait-constrained CUO attenuation treatment reduces the late-season CUO carry-over and moderates the associated GPP suppression, with the largest effects concentrated in densely vegetated areas of southern China.

In this study, we implemented six up-to-date O3–vegetation damage parameterizations into SSiB4/TRIFFID to assess the impact of surface O3 pollution on vegetation in China. Our analysis revealed significant variation in the simulated O3 sensitivity across these schemes, largely attributed to discrepancies in how vegetation O3 sensitivity is represented within each scheme.

Despite the diverse physical processes emphasized by the individual schemes, all flux-based O3 damage parameterizations consistently recognize that vegetation O3 sensitivity varies substantially across PFTs. This variability is influenced by multiple factors, including foliar structural traits, stomatal regulation, antioxidant capacity, detoxification pathways, and underlying differences in photosynthetic metabolism, such as C3 versus C4 pathways (Hayes et al., 2007; Li et al., 2017; Agathokleous et al., 2020). Empirical studies have shown that evergreen species generally exhibit greater resilience to O3-induced oxidative stress compared to deciduous species, likely due to higher antioxidant enzyme activities and larger leaf mass per area (LMA) in evergreen leaves (Li et al., 2017). Leaves with high LMA are typically characterized by thicker or denser palisade mesophyll layers, which reduce O3 exposure per unit leaf mass and enhance antioxidant capacity through a larger apoplastic compartment (Feng et al., 2018). Therefore, LMA has been identified as a key factor in governing O3 sensitivity at the leaf level in recent studies (Agathokleous et al., 2020; Ma et al., 2023; Feng et al., 2021). In addition, species with shorter leaf lifespans tend to exhibit higher O3 sensitivity. Fast-growing species, such as grasses and crops, typically have lower leaf toughness and higher specific leaf area, which results in weaker structural and biochemical defences against environmental stressors, including O3 exposure (Ma et al., 2023). Moreover, recent studies have observed greater O3 sensitivity in C3 compared to C4 plants (Li et al., 2023). This difference can be attributed to the CO2-concentrating mechanism in C4 plants, which improves photosynthetic efficiency while reducing stomatal conductance. The lower stomatal conductance in C4 plants limits O3 uptake, while the higher photosynthetic efficiency supports enhanced detoxification processes and antioxidant biosynthesis, thereby reducing O3-induced damage and increasing O3 tolerance.

Consistent with these findings, our results indicate that shrubs, C3 grasses, and deciduous broadleaf trees exhibit higher O3 sensitivity, whereas C4 grasses, evergreen broadleaf forests, and evergreen needleleaf forests tend to be less sensitive (Fig. 6). Notably, the insufficient observation evidence of C3 and C4 species limited the ability to accurately assess O3 sensitivity between schemes, resulting in a broad confidence interval when comparing simulations with observed data. By comparing the results of all the schemes, we found that the recently-developed Li2024 scheme shows the closest agreement with observations in terms of simulated O3 sensitivity. This advantage can be attributed to its nonlinear formulation of plant responses to CUO, which is built upon an extensive database of more than 4000 experimental measurements (Li et al., 2024). Similarly, the LMA-based approaches also produced relatively accurate O3 sensitivity estimates due to its uniformed expression based on PFT- or grid-specific LMA, a key trait that governs O3 sensitivity across diverse plant species and functional types (Agathokleous et al., 2020; Feng et al., 2018). In contrast, S2007 and L2015 schemes exhibited larger deviations from the observational O3 sensitivity for several PFTs. For S2007, this likely reflects the selected sensitivity parameter set rather than a deficiency in the stomatal flux-based framework. For L2015, the deviations are more closely related to its linear or constant response functions and the relatively limited observational basis available when the scheme was proposed (Li et al., 2024). Consequently, these earlier schemes are less reliable in simulating O3 sensitivity compared to the more recent Li2024 and LMA-based parameterizations.

In summary, our study applies a sensitivity-based comparison approach that offers a reliable and robust framework for cross-scheme evaluation. This approach avoids the influence of model-specific biases and mitigates potential disturbance caused by model configurations. Our results underscore the importance of incorporating both observational constraints and mechanistic understanding from field experiments into O3 sensitivity parameterizations. This is essential for improving the realism of O3-damage simulations and for developing more robust parameterizations that can represent the diverse responses of terrestrial ecosystems to O3 exposure.

Our evaluation shows that Li2024 provides the closest agreement with observed O3 sensitivity among the tested schemes. However, when implemented in SSiB4/TRIFFID, its CUO-based formulation also highlights the importance of how accumulated O3 stress is attenuated through the growing season. This is particularly important for deciduous vegetation, because the LAI-related healing term mainly operates during periods of canopy expansion. Once LAI reaches its seasonal peak and becomes stable or begins to decline, this healing pathway becomes ineffective to attenuate the accumulated O3 stress. To reduce this carry-over of CUO, we extended the decay term to all PFTs to provide a gradual attenuation for accumulated O3 stress. As shown in Sect. 3.4, the refined treatment reduces this carry-over effect and improves the simulated magnitude and seasonality of GPP. These results indicate that treating decay and healing as complementary processes is important for representing the seasonal evolution of CUO.

Experimental and modelling studies consistently indicate that vegetation responses to O3 exposure can be moderated over time through multiple physiological and canopy-level processes (Heath et al., 2009; Felzer et al., 2009). On the one hand, exposure to O3 can severely impair photosynthesis and stomatal functions, with the resulting stomatal closure acting as negative feedback that constrains further O3 uptake (Li et al., 2017). On the other hand, elevated O3 exposure alters hormone regulation and increases apoplastic antioxidant capacity (e.g., ascorbate, phenolic, glutathione), which can directly react with O3 and scavenge ROS, thereby enhancing the O3 tolerance (Castagna and Ranieri, 2009; Heath et al., 2009; Li et al., 2023). Recent evidence from tree species suggests that higher O3 doses can stimulate antioxidant production, which helps mitigate oxidative stress and enhance O3 tolerance (He et al., 2025). O3 is also known to damage leaf cells and accelerate senescence, while plants may partly offset these effects through cellular repair and the production of new leaves (Heath and Taylor, 1997; Grulke and Heath, 2020). Taken together, these mechanisms indicate that O3 damage is dynamically moderated over time and cannot be fully represented by a fixed PODy threshold alone (Heath et al., 2009). Motivated by this evidence, we apply the decay term for all PFTs as a simplified representation of the gradual attenuation of O3-induced oxidative stress, accompanied by the healing term to account for the dilution from emergence of new, undamaged leaves (Felzer et al., 2009; Li et al., 2024).

In our refined treatment, the decay term should not be interpreted as an explicit expression of a specific physiological mechanism. Instead, it provides a simplified representation of the gradual attenuation of accumulated O3 stress within the existing CUO framework. Observational leaf longevity is used here to provide a practical trait-based constraint for this process, as it represents the typical timescale over which O3 exposure can affect leaves before leaf fall or replacement, after which the accumulated O3 effect should be removed from the canopy. With these modifications, CUO accumulation is reduced through the growing season, which improves GPP simulations in both spatial patterns and seasonal cycles. This trait-based decay refinement acknowledges that O3 impacts on vegetation are dynamically regulated by physiological processes and feedbacks, rather than solely controlled by the rise in O3 uptake. Despite these improvements, an important uncertainty remains regarding the physiological interpretation of the gradual decay term. Direct observations and experimental evidence are currently insufficient to determine how canopy CUO evolves during leaf senescence, or to constrain the timescale over which accumulated O3 stress declines. Therefore, Li2024_modify is not intended to explicitly resolve individual physiological processes, such as antioxidant regulation, detoxification, or cellular repair. Instead, it adopts a simplified, trait-informed treatment in which PFT-specific leaf longevity is used to inform the characteristic timescale of CUO attenuation. During leaf fall, a decrease in canopy-level CUO may reflect the turnover and removal of previously exposed foliage. The improvement in GPP simulations shows that the refinement is useful at the model level, but it does not directly validate its physiological interpretation. Further experimental and observational studies are therefore needed to quantify how antioxidant regulation, detoxification, cellular repair, and foliage turnover jointly regulate the persistence and attenuation of O3 effects over the leaf life cycle. Such knowledge would help clarify how vegetation moderates chronic O3 stress over time and provide a stronger mechanistic basis for representing O3 stress dynamics and long-term vegetation responses in numerical models.

In this study, we integrated six flux-based O3–vegetation damage parameterizations into SSiB4/TRIFFID, and quantified the impacts of tropospheric O3 on terrestrial vegetation in China in the 2010s. Based on comprehensive evaluation of inter-scheme differences and refinement of O3 decay representations, our work provides a more comprehensive assessment of O3-induced GPP losses in China. Nevertheless, there are some limitations that should be acknowledged.

First, nitrogen deposition was not considered in this study. Given that nitrogen fertilization can alleviate nutrient limitation and thereby partially offset O3-induced productivity losses (Ren et al., 2011), this exclusion therefore introduces uncertainty into the estimated O3 damage, especially for nitrogen-limited regions. Moreover, although the effects of CO2 and vapor pressure deficit (VPD) on vegetation were included in our simulations, their interactions with O3 damage were not explicitly separated. Both elevated CO2 and increased VPD can regulate stomatal conductance and thereby modify stomatal O3 uptake. Therefore, the simulated O3 responses may incorporate indirect effects of these concurrent drivers, introducing additional uncertainty in the attribution of O3 impacts. In addition, our offline framework is not able to capture feedbacks mediated by land–atmosphere coupling. Previous studies indicate a positive O3–vegetation feedback, whereby O3-induced physiological and biophysical responses can both reduce O3 removal and enhance photochemical production. In particular, O3-driven stomatal closure weakens canopy O3 uptake and deposition, while the associated surface warming accelerates biogenic VOC emissions and O3 formation (Jin et al., 2023; Cao et al., 2024). Accordingly, the absence of land–atmosphere-chemistry feedbacks in our offline framework may contribute to a slightly weaker response and a modest underestimation of O3-induced vegetation damage relative to coupled simulations (Zhu et al., 2022; Jin et al., 2023). Our results should therefore be interpreted as conservative with respect to fully coupled modelling, and our intercomparison and refinement efforts offer a solid baseline for quantifying O3–vegetation interactions under coupled frameworks.