Indirect contributions of global fires to surface ozone through ozone–vegetation feedback

Fire is an important source of ozone (O₃) precursors. The formation of surface O₃ can cause damage to vegetation and reduce stomatal conductance. Such processes can feed back to inhibit dry deposition and indirectly enhance surface O₃. Here, we apply a fully coupled chemistry–vegetation model to estimate the indirect contributions of global fires to surface O₃ through O₃–vegetation feedback during 2005–2012. Fire emissions directly increase the global annual mean O₃ by 1.2 ppbv (5.0 %) with a maximum of 5.9 ppbv (24.4 %) averaged over central Africa by emitting a substantial number of precursors. Considering O₃–vegetation feedback, fires additionally increase surface O₃ by 0.5 ppbv averaged over the Amazon in October, 0.3 ppbv averaged over southern Asia in April, and 0.2 ppbv averaged over central Africa in April. During extreme O₃–vegetation interactions, such a feedback can rise to >0.6 ppbv in these fire-prone areas. Moreover, large ratios of indirect-to-direct fire O₃ are found in eastern China (3.7 %) and the eastern US (2.0 %), where the high ambient O₃ causes strong O₃–vegetation interactions. With the likelihood of increasing fire risks in a warming climate, fires may promote surface O₃ through both direct emissions and indirect chemistry–vegetation feedbacks. Such indirect enhancement will cause additional threats to public health and ecosystem productivity.

1 Introduction

Tropospheric ozone (O₃) is a toxic air pollutant with detrimental effects on vegetation (Yue and Unger, 2014; Juráň et al., 2021). Plant stomatal uptake of O₃ decreases both chlorophyll and Rubisco contents and increases the deformity rate of chloroplasts (Booker et al., 2007; Akhtar et al., 2010; Inada et al., 2012), which further reduces the leaf area index (LAI) and gross primary productivity (GPP) of ecosystems (Karnosky et al., 2007; Ainsworth et al., 2012). Modeling studies estimated that O₃ damage reduces global GPP by 1.5 %–3.6 % with regional maximum reductions of 8 %–20 % over eastern US, western Europe, and eastern China (Yue and Unger, 2014; Lei et al., 2020; Zhu et al., 2021). In turn, vegetation damage also influences both the sources and sinks of O₃ through biogeochemical and biogeophysical feedbacks (Curci et al., 2009; Heald and Geddes, 2016; Fitzky et al., 2019). The damaged vegetation decreases isoprene emissions and stomatal conductance (Wittig et al., 2009; Feng et al., 2019), which influence O₃ production and dry deposition. Moreover, weakened leaf-level transpiration following O₃ damage modulates meteorological parameters, such as surface air temperature and atmospheric relative humidity, leading to substantial biogeophysical feedbacks on surface O₃ (Lombardozzi et al., 2012; Sadiq et al., 2017).

Interactions between air pollution and terrestrial ecosystems remain challenging due to limited process-based knowledge and the separate development of chemistry and vegetation models (He et al., 2020). At present, the feedbacks from O₃-damaging vegetation on O₃ have only been examined by four papers (Sadiq et al., 2017; Zhou et al., 2018; Gong et al., 2020; Zhu et al., 2021). Sadiq et al. (2017) implemented a parameterization of O₃ vegetation damage into a climate model and quantified online O₃–vegetation coupling. Simulations showed that surface O₃ could be enhanced by up to 4–6 ppbv over Europe, North America, and China through comparable effects from biogeochemical (decreased dry deposition and increased isoprene emissions) and biogeophysical (changes in meteorological variables following reduced transpiration rate) feedbacks from O₃–vegetation interactions. Similar conclusions were drawn by Zhu et al. (2021), who investigated the effects of O₃–vegetation interaction in China using a two-way coupled land–atmosphere model. By including O₃ damage to isoprene emissions in a fully coupled global chemistry–carbon–climate model, Gong et al. (2020) highlighted that such O₃–vegetation positive feedbacks were mainly driven by reduced dry deposition following O₃ damage to photosynthesis. Different from the three studies above, Zhou et al. (2018) implemented steady-state O₃-induced LAI changes into the GEOS-Chem model and quantified only the influences of O₃–vegetation biogeochemical feedbacks because the model is driven with prescribed meteorological fields. Results showed that O₃-induced damage to LAI can enhance O₃ by up to 3 ppbv in the tropics, eastern North America, and southern China through changes in dry deposition and isoprene emissions. All studies revealed a strong positive O₃–vegetation feedback to surface O₃, though the magnitudes are different due to discrepancies in O₃-damaging schemes, as well as differences in the models.

Fire plays an important role in disturbing the terrestrial carbon budget (Bond-Lamberty et al., 2007; Amiro et al., 2009; Turetsky et al., 2011; Yue and Unger, 2018). Global fires directly emit 2–3 Pg (1 Pg=10¹⁵ g) carbon into the atmosphere every year (van der Werf et al., 2010). Moreover, fires contribute to the production of tropospheric O₃ by emitting a substantial number of precursors (Cheng et al., 1998; Kita et al., 2000; Oltmans et al., 2010; Jaffe et al., 2013; Lu et al., 2016). Globally, fires account for 3 %–5 % of the total tropospheric O₃ (Bey et al., 2001; Ziemke et al., 2009; Jaffe and Wigder, 2012). Regionally, especially in the Amazon and central Africa, fires can enhance surface O₃ by 10–30 ppbv through emissions of NO_x and volatile organic compounds (VOCs) during fire seasons (Yue and Unger, 2018; Pope et al., 2020). Over these regions, strong O₃–vegetation interactions are expected because of high fire O₃ concentrations and dense vegetation cover. Previous studies showed that fire O₃ causes a large GPP reduction of 200–400 Tg C yr⁻¹ over the Amazon and central Africa (Pacifico et al., 2015; Yue and Unger, 2018). With likely increased wildfire activity due to global warming, surface O₃ will be further enhanced by wildfires (Amiro et al., 2009; Balshi et al., 2009; Wang et al., 2016; Yue et al., 2017), leading to more severe O₃ damage on vegetation. Although the feedback of vegetation damage on surface O₃ has been comprehensively explored on global (Sadiq et al., 2017; Zhou et al., 2018; Gong et al., 2020) and regional (Zhu et al., 2021) scales, these studies all focused on O₃–vegetation interactions from combined anthropogenic and natural sources. Therefore, quantification of the O₃–vegetation interactions associated with fire emissions is very important for a comprehensive understanding of the effects of fires on surface O₃.

Here, we apply a fully coupled chemistry–vegetation model (GEOS-Chem-Yale Interactive terrestrial Biosphere – hereafter referred to as GC-YIBs) to examine the indirect contributions of fires to surface O₃. Fire-induced O₃ affects plant photosynthesis and stomatal conductance. In turn, predicted changes in LAI and canopy stomatal conductance influence both the sources and sinks of tropospheric O₃. Such O₃–vegetation interactions result in the additional enhancement of surface O₃ caused by fire emissions (Fig. 1). Section 2 describes the GC-YIBs model and sensitivity experiments conducted in this study. Section 3 quantifies the feedbacks of fire-induced O₃ vegetation damage on surface O₃ concentrations. The last section summarizes the findings and discusses the uncertainties.

Figure 1 Diagram of the impacts of fires on surface O₃ through direct emissions and O₃–vegetation feedback.

2 Materials and methods

2.1 The GC-YIBs model

GC-YIBs is a coupled chemistry–vegetation model developed by implementing the Yale Interactive terrestrial Biosphere (YIBs) model into GEOS-Chem version 12.0.0 (Lei et al., 2020). GEOS-Chem is a widely used global 3-D chemical transport model (CTM) for simulating atmospheric composition and air quality (Yue et al., 2015; Yan et al., 2018; David et al., 2019; Lu et al., 2019). This model uses a detailed ${\mathrm{HO}}_x-{\mathrm{NO}}_x-\mathrm{VOC}-{\mathrm{O}}_{\mathrm{3}}$–halogen–aerosol tropospheric chemistry to simulate tropospheric O₃ fluxes (Barret et al., 2016; Gong and Liao, 2019), while a simplified linearized Linoz chemistry mechanism is applied to simulate stratospheric O₃ (McLinden et al., 2000). Aerosols simulated in GEOS-Chem include secondary inorganic aerosols, secondary organic aerosols, primary organic aerosols, black carbon, dust, and sea salt (Dang and Liao, 2019; Li et al., 2019). The gas–aerosol partitioning of the sulfate–nitrate–ammonium system is computed by the ISORROPIA v2.0 thermodynamic equilibrium model (Fountoukis and Nenes, 2007). The atmospheric emissions from different sources, regions, and species on a user-defined grid are calculated through the online Harvard–NASA Emissions Component (HEMCO) module (Keller et al., 2014). HEMCO is highly customizable in that it can automatically combinate, overlay, and update emission inventories and scale factors specified by the users. In general, the GEOS-Chem model overestimates summer surface O₃ concentrations in the eastern US and China (Zhang et al., 2011; Travis et al., 2016; Schiferl and Heald, 2018).

YIBs is a vegetation model designed to dynamically simulate the changes in LAI and tree height based on carbon assimilation, respiration, and allocation processes (Yue and Unger, 2015). The model computes carbon uptake for nine vegetation types, including evergreen needleleaf forest, deciduous broadleaf forest, evergreen broadleaf forest, shrubland, tundra, ${\mathrm{C}}_{\mathrm{3}}/{\mathrm{C}}_{\mathrm{4}}$ grasses, and ${\mathrm{C}}_{\mathrm{3}}/{\mathrm{C}}_{\mathrm{4}}$ crops. The canopy is divided into an adaptive number of layers (typically 2–16) for light stratification. The YIBs model applies a well-established Michaelis–Menten enzyme kinetics scheme to compute the leaf photosynthesis (Farquhar et al., 1980; Von Caemmerer and Farquhar, 1981), which is further upscaled to the canopy level by the separation of sunlit and shaded leaves (Spitters, 1986). The LAI and carbon allocation schemes are from the TRIFFID model (Clark et al., 2011). Previous studies have shown that the YIBs model has good performance in simulating the spatial pattern of and temporal variability in GPP and LAI based on site observations and satellite products (Yue and Unger, 2015, 2018).

The GC-YIBs model links atmospheric chemistry and vegetation in a two-way coupling. As a result, changes in chemical components or vegetation will simultaneously feed back to influence the other systems. In this study, the GC-YIBs model is driven with the meteorological fields from the Modern-Era Retrospective analysis for Research and Applications, version 2 (MERRA2) with a horizontal resolution of 4^∘ latitude by 5^∘ longitude, as well as 47 vertical layers from the surface to 0.01 hPa. Within GC-YIBs, the online-simulated surface O₃ in GEOS-Chem affects photosynthesis and canopy stomatal conductance; in turn, the online-simulated vegetation parameters, such as LAI and stomatal conductance, in YIBs affect both the sources and sinks of O₃ by altering precursor emissions and dry deposition at the 1 h integration time step. An earlier study evaluated the GC-YIBs model and showed good performance in simulating surface O₃, GPP, LAI, and O₃ dry deposition (Lei et al., 2020).

2.2 Scheme of O₃ vegetation damage

The GC-YIBs model calculates the impacts of O₃ exposure on photosynthesis based on a semi-mechanistic scheme (Sitch et al., 2007):

$$\begin{array}{}\text{(1)}& A^{\prime }=\mathit{\alpha }\cdot A,\end{array}$$

where A^′ and A represent the O₃-damaging and original leaf photosynthesis, respectively. The O₃ damage factor is represented by α; O₃ can cause damage to photosynthesis only if α<1. The factor α is calculated as a function of excessive O₃ flux and damaging sensitivity coefficient (β):

$$\begin{array}{}\text{(2)}& \mathit{\alpha }=-\mathit{\beta }\cdot max\left(F_{{\mathrm{O}}_{\mathrm{3}}}-T_{{\mathrm{O}}_{\mathrm{3}}},\mathrm{0}\right).\end{array}$$

The coefficient β can have two values for each vegetation type (Table S1 in the Supplement), indicating low to high O₃-damaging sensitivities (Sitch et al., 2007). $T_{{\mathrm{O}}_{\mathrm{3}}}$ represents the O₃ flux threshold, reflecting the O₃ tolerance of different vegetation types. $F_{{\mathrm{O}}_{\mathrm{3}}}$ represents the stomatal O₃ flux and is calculated based on ambient [O₃], aerodynamic resistance (r_a), boundary layer resistance (r_b), and stomatal resistance (r_s):

$$\begin{array}{}\text{(3)}& F_{{\mathrm{O}}_{\mathrm{3}}}={\displaystyle \frac{\left[{\mathrm{O}}_{\mathrm{3}}\right]}{r_{\mathrm{a}}+r_{\mathrm{b}}+k\cdot r_{\mathrm{s}}^{\prime }}}.\end{array}$$

Here k represents the ratio of leaf resistance of O₃ to leaf resistance of water vapor. Parameters r_a and r_b are calculated by the GEOS-Chem model. O₃-damaging leaf photosynthesis (A^′) is then integrated over all canopy layers to generate O₃-damaging GPP:

$$\begin{array}{}\text{(4)}& {\text{GPP}}^{\prime }={\int }_{\mathrm{0}}^{\text{LAI}}{A}^{\prime }\mathrm{d}L.\end{array}$$

The O₃-damaging stomatal resistance ($r_{\mathrm{s}}^{\prime }$) is calculated based on the model of Ball and Berry (Baldocchi et al., 1987):

$$\begin{array}{}\text{(5)}& {\displaystyle \frac{\mathrm{1}}{r_{\mathrm{s}}^{\prime }}}=g_{\mathrm{s}}^{\prime }=m{\displaystyle \frac{A_{\text{net}}^{\prime }\cdot \text{RH}}{c_{\mathrm{s}}}}+b,\end{array}$$

where m and b represent the slope and intercept of empirical fitting to the Ball–Berry stomatal conductance equation, respectively. $A_{\text{net}}^{\prime }$ represents O₃-damaging net leaf photosynthesis, RH represents the relative humidity, and c_s is the ambient CO₂ concentration. Previous studies have shown that this scheme within the framework of YIBs can reasonably capture the response of GPP and stomatal conductance to surface [O₃] based on hundreds of global observations (Yue et al., 2016; Yue and Unger, 2018).

2.3 Fire emissions

Fire INventory from the National Center for Atmospheric Research (FINN) version 1.5 is used by GC-YIBs to simulate fire-induced perturbations in O₃. FINN provides daily global emissions of many chemical species from open biomass burning at a resolution of 1 km² (Wiedinmyer et al., 2011). The inventory estimates fire locations and biomass burned using satellite observations of active fires and land cover, together with emission factors and fuel loadings. For each land type, emission factors for different gaseous and particulate species are taken from measurements (Andreae and Merlet, 2001; Andreae and Rosenfeld, 2008; Akagi et al., 2011). Daily fire emissions for 2002–2012 are available at https://www.acom.ucar.edu/Data/fire/, last access: 1 August 2021). In GC-YIBs, all biomass burning emissions occur in the atmospheric boundary layer. Such a configuration might slightly overestimate regional O₃ formation as observations suggested ∼20 % of fire plumes reached the height above the boundary layer (Val Martin et al., 2010) and consequently enhanced the surface O₃ level in the downwind regions (Jaffe and Wigder, 2012). The FINN inventory has been widely used in regional and global chemical transport models (e.g., WRF-Chem and GEOS-Chem) to quantify the impacts of fires on air quality and weather (Jiang et al., 2012; Nuryanto, 2015; Vongruang et al., 2017; Brey et al., 2018; Watson et al., 2019).

2.4 Site-level measurements

Measurements of surface [O₃] in the US are provided by Air Quality System (AQS; https://www.epa.gov/aqs, last access: 1 August 2021), and those over Europe are provided by the European Monitoring and Evaluation Programme (EMEP; https://emep.int, last access: 1 August 2021). The observed [O₃] at the Manaus (Brazil), Tanjung Malim (Malaysia), and Welgegund (South Africa) sites are from earlier studies (Ahamad et al., 2014; Laban et al., 2018; Pope et al., 2020).

2.5 Model simulations

In this study, eight simulations (Table 1) are performed to examine both the direct and indirect contributions of fires to surface O₃. These simulations can be divided into two main groups:

1. CTRL_FIRE and CTRL_NOFIRE are the control runs using the same emissions except that the latter omits fire emissions. These runs calculate and output offline O₃ damage, which decreases instantaneous leaf photosynthesis but does not feed back to affect plant growth and O₃ dry deposition.

2. O3CPL_FIRE and O3CPL_NOFIRE are the sensitive experiments that consider online coupling between O₃ and vegetation. These runs include online O₃ damage to plant photosynthesis, which feeds back to affect both vegetation and air pollution. The two simulations apply the same emissions, except that the latter omits fire emissions.

Table 1 Summary of simulations using the GC-YIBs model.

For each of these four configurations, two runs are conducted with either high (HS) or low (LS) O₃-damaging sensitivities. All simulations are performed from 2002 to 2012 using the GC-YIBs model driven by MERRA2 meteorological fields. The first 3 years are used as spinup, and the results of the last 8 years are analyzed. For the same configurations, the results from low and high O₃-damaging sensitivities are averaged. The differences between CTRL_NOFIRE and O3CPL_NOFIRE represent the surface O₃ enhancements through O₃–vegetation feedback without fire emissions. The differences between CTRL_FIRE and CTRL_NOFIRE, named O3OFF, represent the direct contributions of fires to surface O₃. The differences between O3CPL_FIRE and O3CPL_NOFIRE, named O3CPL, represent both direct and indirect contributions of fires to surface O₃. The differences between O3CPL and O3OFF represent the indirect contributions of fires to surface O₃ through O₃–vegetation interactions. It should be noted that only biogeochemical feedbacks from O₃ vegetation damage on surface O₃ are considered in this study because GC-YIBs uses prescribed meteorology (MERRA2).

3 Results

3.1 Model validation

Simulated surface daily maximum 8 h average O₃ concentrations (MDA8 [O₃], short for [O₃] hereafter) are evaluated using measurements from the AQS and EMEP datasets over the period of 2005–2012 (Fig. 2). The model captures well the observed spatial distribution of annual [O₃] in the US and Europe, with a high correlation coefficient of 0.51 (p<0.01). Although GC-YIBs overestimates the [O₃] in the eastern US while underestimating it in western Europe, the normalized mean bias (NMB) is only 4.0 %, with a root mean square error (RMSE) of 5.4 ppbv. Therefore, the simulated O₃ vegetation damage in our study is slightly overestimated in the eastern US but underestimated in western Europe.

Figure 2 Spatial pattern of (a) simulated and (b) observed surface [O₃]. (c) Scatter plot of surface [O₃] over measurements in two regions. The black line shows the linear regression between the observed and simulated [O₃]. The regression fit, correlation coefficient (R), root mean square error (RMSE), and normalized mean bias (NMB) are shown in the bottom panel with an indication of site numbers (N) used for statistics.

3.2 Direct contributions of fires to O₃

Without fire emissions, the simulated global mean [O₃] is 23.9 ppbv, with a grid maximum of 63.7 ppbv over the Beijing–Tianjin–Hebei region averaged for 2005–2012 (Fig. 3a). Most high [O₃] is distributed in the Northern Hemisphere, where anthropogenic emissions make the dominant contributions. The inclusion of fire emissions increases global annual [O₃] by an average of 1.2 ppbv (5.0 %). Regionally, the largest enhancement of [O₃] by 5.9 ppbv (24.4 %) is averaged over central Africa, with smaller enhancements of 5.7 ppbv (38.2 %) averaged over the Amazon and 3.8 ppbv (10.2 %) averaged over southern Asia. Smaller enhancements of 1.1 ppbv (2.2 %), 0.9 ppbv (2.1 %), and 0.8 ppbv (2.2 %) are averaged respectively over eastern China, western Europe, and the eastern US (Fig. 3b). The predicted fire-induced enhancements in [O₃] agree well with the simulations using the same model but with fire emissions from the Global Fire Emissions Database (GFED) version 3 (Yue and Unger, 2018).

Figure 3 Annual surface [O₃] from (a) non-fire sources and (b) fire sources alone. The six subregions are marked with black boxes: eastern US (EUS; 30–50^∘ N, 95–70^∘ W), western Europe (WEU; 40–60^∘ N, 0–40^∘ E), eastern China (ECH; 20–35^∘ N, 108–120^∘ E), Amazon (AMZ; 25^∘ S–0^∘, 80–50^∘ W), central Africa (CAF; 10^∘ S–10^∘ N, 10–40^∘ E), and southern Asia (SAS; 10–30^∘ N, 95–110^∘ E).

We further evaluated the model performance in simulating fire-induced Δ[O₃] at three sites across biomass burning regions (Fig. S1 in the Supplement). Without fire emissions, the [O₃] is obviously underestimated with NMBs of −25.5 % at Tanjung Malim, −53.6 % at Manaus, and −21.3 % at Welgegund. As a comparison, simulations with fire emissions show NMBs in fire seasons of −8.7 % at Tanjung Malim, −1.4 % at Manaus, and −15.1 % at Welgegund, suggesting that O₃ simulations were improved by including fire emissions.

3.3 Fire-induced O₃ damage to GPP

Surface O₃ causes strong damage to ecosystem productivity (Fig. 4). Without fire emissions, surface O₃ reduces global annual GPP by 1.7 % (3899.8 Tg C yr⁻¹; Fig. 4a and c). Regional maximum reductions of 10.9 % (372.0 Tg C yr⁻¹), 6.1 % (366.1 Tg C yr⁻¹), and 4.9 % (323.8 Tg C yr⁻¹) are averaged respectively over eastern China, the eastern US, and western Europe; these reductions are attributed to the high ambient [O₃] level and the large stomatal conductance over these regions. The patterns of O₃-induced GPP reductions agree with previous estimates using the same O₃ damage schemes (Sitch et al., 2007; Yue and Unger, 2015). However, compared to simulations using another scheme (Lombardozzi et al., 2012; Zhou et al., 2018; Zhu et al., 2021), this study estimates smaller GPP reductions. Such a discrepancy indicates that there are large uncertainties in O₃ vegetation damage schemes, and more observations should be developed to evaluate different schemes in future studies.

Figure 4 Annual percentage of reductions in GPP caused by O₃ from (a, c) non-fire sources and (b, d) fire sources alone with (a, b) high and (c, d) low O₃ sensitivities. Please note the differences in color scales.

The inclusion of fire emissions causes additional GPP reductions. Globally, fire-induced ΔO₃ decreases annual GPP by 0.4 % (1312.0 Tg C yr⁻¹; Fig. 4b and d). Regionally, the largest GPP reduction of 1.4 % (370.3 Tg C yr⁻¹) is averaged over the Amazon due to the largest enhancement of [O₃] caused by fires. Furthermore, fire Δ[O₃] causes additional annual GPP reductions of 1.3 % (358.0 Tg C yr⁻¹), averaged over central Africa, and 1.0 % (77.1 Tg C yr⁻¹), averaged over southern Asia. In contrast, limited damage is found in eastern China, western Europe, and the eastern US due to low fire Δ[O₃]. Following the changes in GPP, fire-induced O₃ damage to LAI shows a regional maximum of 0.3 %–0.7 % in central Africa and a global reduction of 0.02 %–0.5 % (Fig. S2).

Figure 5 Annual feedback to surface O₃ caused by O₃ vegetation damage with (a, b) high and (c, d) low O₃ sensitivities. Panels (a) and (c) represent feedback by O₃ from non-fire sources, and panels (b) and (d) represent feedback by O₃ from fire emissions alone. Please note the differences in color scales.

3.4 Indirect contributions of fires to O₃

Vegetation parameters such as LAI and stomatal conductance play important roles in modulating surface [O₃]. The O₃-induced changes in these variables interactively feed back to alter local [O₃] (Fig. 5). Without fire emissions, the annual Δ[O₃] from O₃–vegetation interactions is limited in eastern China by 0.5 ppbv, the eastern US by 0.3 ppbv, and western Europe by 0.2 ppbv. The largest grid positive feedback of up to 0.8 ppbv is found in the eastern US (Fig. 5a and c). Sensitivity experiments further show that such an enhancement of surface [O₃] mainly results from the inhibition of stomatal conductance following reduced photosynthesis because of O₃ damage (Fig. S3a). Consequently, large Δ[O₃] values (Fig. 5a and c) are collocated with areas enduring high levels of O₃ vegetation damage (Fig. 4a and c). As a comparison, the feedback of LAI changes is generally small (Fig. S3b), which is mainly attributed to limited O₃ damage to LAI (Fig. S2). The enhancement of [O₃] from fires causes additional feedback to the surface [O₃]. The largest annual Δ[O₃] of 0.13 ppbv due to O₃–vegetation feedback is averaged over the Amazon (Fig. 5b and d), where the highest GPP reductions by fire-induced O₃ are predicted (Fig. 4b and d). Such feedbacks additionally enhances local [O₃] by 0.12 ppbv, averaged over central Africa, and 0.09 ppbv, averaged over southern Asia. However, limited O₃–vegetation feedback is found in the eastern US, eastern China, and western Europe because of either low fire-induced Δ[O₃] (Fig. 3b) or low ΔGPP (Fig. 4b and d). The changes in O₃ dry deposition velocity broadly match the pattern of O₃–vegetation feedback (Fig. S4), suggesting that reduced dry deposition velocity due to O₃-induced inhibition of stomatal conductance is the dominant driver for the enhanced surface [O₃].

Figure 6 Seasonal variations in O₃–vegetation feedback driven by (a) non-fire sources and (b) fire sources alone. The blue, red, and green bars in (a) represent the O₃–vegetation feedback in eastern US (EUS), eastern China (ECH), and western Europe (WEU), respectively. The blue, red, and green bars in (b) represent the O₃–vegetation feedback in the Amazon (AMZ), central Africa (CAF), and southern Asia (SAS), respectively. The error bars represent low to high O₃-damaging sensitivities.

Figure 6 shows seasonal variations in O₃–vegetation feedback. Without fire emissions, O₃–vegetation feedback in eastern China, the eastern US, and western Europe shows similar seasonal variations, increasing from January to July and then decreasing (Fig. 6a). For these regions, surface [O₃] and stomatal conductance reach maximums during the growth season (May–October), resulting in instantaneous O₃ uptake. Therefore, O₃–vegetation interactions are expected to be stronger during the growth season in the Northern Hemisphere. However, O₃–vegetation feedback driven by fires in the Amazon and southern Asia reaches a maximum during August–December and February–June, respectively. Moreover, double peaks are shown in central Africa, with maximums during February–April and July–September (Fig. 6b). The distinct seasonal variations in biomass burning regions are attributed to fire emissions. At low latitudes, stomatal conductance shows limited seasonal variations. Therefore, O₃–vegetation feedback driven by fires is mainly dependent on fire-induced Δ[O₃].

Fire-induced O₃ shows stronger interactions with vegetation under favorable meteorological conditions. We sort daily Δ[O₃] from O₃–vegetation feedback and calculate the average of Δ[O₃] above the 95th percentile (Fig. S5). The spatial pattern of Δ[O₃] during extreme O₃–vegetation feedback is broadly consistent with that of the annual average, albeit with much stronger O₃–vegetation feedback. Without fire emissions, O₃–vegetation feedback enhances [O₃] by 2.0 ppbv averaged over eastern China, 1.8 ppbv averaged over the eastern US, and 1.1 ppbv averaged over western Europe (Fig. S5a and c). Fire emissions alone enhance [O₃] through O₃–vegetation interactions by 1.1 ppbv averaged over the Amazon, 0.8 ppbv averaged over southern Asia, and 0.6 ppbv averaged over central Africa during extreme O₃–vegetation feedback (Fig. S5b and d).

3.5 Indirect vs. direct contributions of fires to O₃

We further compare the indirect and direct contributions of fire emissions to surface [O₃]. Here, the direct contributions indicate Δ[O₃] caused by fire emissions of chemical precursors, while the indirect contributions represent additional Δ[O₃] from O₃–vegetation interactions caused by fire-induced O₃. Without fire emissions, O₃–vegetation interactions cause an enhancement of [O₃] by 1.0 % averaged over eastern China, 0.8 % averaged over the eastern US, and 0.5 % averaged over western Europe (Fig. 7a and c). Compared to non-fire sources, fire emissions cause larger relative perturbations in surface [O₃] through O₃–vegetation interactions (Fig. 7b and d). The ratios of indirect to direct annual Δ[O₃] are 3.7 % averaged over eastern China, 2.0 % averaged over the eastern US, and 1.6 % averaged over western Europe. For these regions, the absolute Δ[O₃] from direct fire emissions is usually lower than 1 ppbv (Fig. 3b). However, the high level of background [O₃] (all sources except fire emissions; Fig. 3a) provides such a sensitive environment that the moderate increases in [O₃] from fires can cause significant feedback to regional surface [O₃] through vegetation damage. For fire-prone regions, the ratios of indirect to direct annual Δ[O₃] are 2.6 % averaged over southern Asia, 1.9 % averaged over the eastern US, and 1.4 % averaged over central Africa.

Figure 7 Annal ratios of indirect Δ[O₃] to ambient [O₃] from (a, c) non-fire emissions and the ratios of indirect to direct Δ[O₃] from (b, d) fire emissions alone with (a, b) high and (c, d) low O₃-damaging sensitivities. Please note the differences in color scales.

3.6 Aggravated O₃ damage to GPP through O₃–vegetation feedback

The additional O₃ enhancement can exacerbate the damaging effects on vegetation. Without fire emissions, online O₃ causes a global annual GPP reduction of 0.2 % (299.6 Tg C yr⁻¹; Fig. S6a and c) from the offline O₃. Regionally, additional reductions are mainly found in eastern China, the eastern US, and western Europe, where GPP is further decreased by 27.1, 40.8, and 28.4 Tg C yr⁻¹, respectively. For fire emissions, the online fire-induced ΔO₃ results in a higher GPP reduction of 25.0 Tg C yr⁻¹ averaged over the Amazon, 24.3 Tg C yr⁻¹ averaged over central Africa, and 7.1 Tg C yr⁻¹ averaged over southern Asia compared to the offline fire-induced ΔO₃ (Fig. S6b and d). Such spatial patterns are broadly consistent with Δ[O₃] induced by O₃–vegetation feedback (Fig. 5).

4 Conclusions and discussion

Many studies have explored the direct contributions to surface O₃ by fire emissions. However, the feedback of fire-induced O₃ vegetation damage to surface [O₃] remains unquantified. In this study, we find that fire-induced O₃ causes a positive feedback to surface [O₃] mainly because of the inhibition effects on stomatal conductance. Regionally, O₃–vegetation feedback driven by fires enhances surface annual [O₃] by 0.13 ppbv averaged over the Amazon, 0.12 ppbv averaged over central Africa, and 0.09 ppbv averaged over southern Asia. Such feedbacks exhibit large seasonal variations, with the maximums of 0.5 ppbv averaged over the Amazon in October, 0.3 ppbv averaged over southern Asia in April, and 0.2 ppbv averaged over central Africa in April. During extreme O₃–vegetation interactions, the feedback can rise to >0.6 ppbv in these fire-prone areas. Although direct formations of O₃ from fires are limited in eastern China and the eastern US, the feedback of O₃–vegetation coupling results in the additional enhancement of surface [O₃] by 3.7 % and 2.0 % on the fire-induced Δ[O₃]. Such large ratios in these regions are attributed to the high level of ambient [O₃] that provides a sensitive environment in which moderate increases in [O₃] from fires can cause large indirect contributions to regional [O₃] through vegetation damage.

Some uncertainties may affect the conclusions of this study. (i) The GC-YIBs simulations do not consider the direct fire damage to vegetation and the consequent long-term recovery of forests. In our study, we focus only on the feedbacks of fire-induced O₃–vegetation interactions to surface O₃. (ii) Fires can decrease VOC emissions from biogenic sources by damaging vegetation directly. However, compared to the VOCs emitted by fires, the VOC loss from burned vegetation is generally smaller (Fig. S7). Therefore, the influence of reduced VOCs from vegetation loss on surface [O₃] can be ignored. (iii) There is evidence that O₃ exposure may cause “sluggishness” that delays the stomatal responses to O₃ damage (Huntingford et al., 2018). However, we do not include sluggishness in our scheme because its net impacts on stomatal conductance remain uncertain. For example, observations found that the increased short-term water loss (delayed stomatal responses) may be offset by the decreased long-term water loss (lower steady-state stomatal conductance) with the stomatal sluggishness (Paoletti et al., 2019). (iv) We employed a model resolution of $\mathrm{4}{}^{\circ }\times \mathrm{5}{}^{\circ }$ due to the limitations in computational resources. We performed a 1-year sensitivity simulation at a $\mathrm{2}{}^{\circ }\times \mathrm{2.5}{}^{\circ }$ resolution. The comparisons show that fire-induced direct O₃ enhancement is very similar between the simulations at low and high resolutions, although the former runs predict slightly higher changes in [O₃] than the latter (Fig. S8). (v) Different biomass burning datasets may affect the estimated O₃–vegetation feedback in our study. At present, the FINNv1.5 and GFEDv4.1 inventories are available in the public release of GEOS-Chem v12.0.0. Compared with the FINNv1.5 inventory, simulations using the GFEDv4.1 inventory predict a lower O₃–vegetation feedback in the Amazon (Fig. S9a) and southern Asia (Fig. S9c) but a higher O₃–vegetation feedback in central Africa (Fig. S9b).

Despite these uncertainties, we present the first estimate of O₃ enhancement by fire emissions through O₃–vegetation interactions. Such an enhancement is not limited to fire-prone regions but is also significant over downwind areas with high ambient [O₃] levels. Although the absolute perturbations may be moderate for the whole fire season, O₃–vegetation interactions can largely increase surface O₃ during extreme O₃–vegetation interactions, leading to additional threats to public health and ecosystem productivity.