Sensitivity of global greenhouse gas budgets to tropospheric ozone pollution mediated by the biosphere

Tropospheric ozone (O3), a harmful secondary air pollutant, can affect the climate via direct radiative forcing and by modifying the radiative forcing of aerosols through its role as an atmospheric oxidant. Moreover, O3 exerts a strong oxidative pressure on the biosphere and indirectly influences the climate by altering the materials and energy exchange between terrestrial ecosystems and the atmosphere. However, the magnitude by which O3 affects the global budgets of greenhouse gases (GHGs: CO2, CH4, and N2O) through altering the land–atmosphere exchange is largely unknown. Here we assess the sensitivity of these budgets to tropospheric O3 pollution based on a meta-analysis of experimental studies on the effects of elevated O3 on GHG exchange between terrestrial ecosystems and the atmosphere. We show that across ecosystems, elevated O3 suppresses N2O emissions and both CH4 emissions and uptake, and has little impact on stimulation of soil CO2 emissions except at relatively high concentrations. Therefore, the soil system would be transformed from a sink into a source of GHGs with O3 levels increasing. The global atmospheric budget of GHGs is sensitive to O3 pollution largely because of the carbon dioxide accumulation resulting from suppressed vegetation carbon uptake; the negative contributions from suppressed CH4 and N2O emissions can offset only ∼10% of CO2 emissions from the soil–vegetation system. Based on empirical data, this work, though with uncertainties, provides the first assessment of sensitivity of global budgets of GHGs to O3 pollution, representing a necessary step towards fully understanding and evaluating O3–climate feedbacks mediated by the biosphere.


Introduction
Tropospheric ozone (O 3 ) is formed from the photochemical oxidation of carbon monoxide (CO) and volatile organic compounds (VOCs), typically initiated by the hydroxyl radical (OH), in the presence of nitrogen oxides (NO x ≡ NOþNO 2 ) (Leighton 1961). Its concentration has significantly increased since the industrial revolution and is projected to either continue to rise or to decline over the course of this century with considerable spatial variability, driven largely by trends in fossil fuel combustion (source of NO x ) and vegetation growth ( All of these could largely contribute to changes in the atmospheric GHG budgets and affect climate (Sitch et al 2007). This paper provides the first assessment of the sensitivity of global budgets of CO 2 , CH 4 , and N 2 O to tropospheric O 3 abundance resulting from the alterations of land-atmosphere exchange based on a meta-analysis of published experimental studies of O 3 impacts on the exchange of these three gases from varying ecosystems.
There are a myriad of mechanisms at different organizational levels by which the oxidizing capacity of O 3 affects terrestrial ecosystems. Ozone can directly depress photosynthesis by disturbing cellular biochemistry or indirectly by damaging stomatal function; in either case it alters carbon and water exchange (Ainsworth et al 2012). Ozone can also change secondary metabolism and tissue chemistry, which can affect carbon mineralization (Findlay et al 1996, Loya et al 2003, Valkama et al 2007. At the community level, O 3 exerts indirect effects on species composition through mediation of competitive relations (Wang et al 2016). Since O 3 does not penetrate soil (Blum and Tingey 1977), impacts on the processes responsible for the production and/or consumption of CO 2 , CH 4 , and N 2 O in soils are mediated by these aboveground changes (figure 1). Changes in organic matter production, composition, and allocation, in turn, affect decomposition and carbon and nitrogen cycling, and alter rates of CO 2 and N 2 O production and CH 4 production/consumption (Agathokleous et al 2016).
Independent empirical studies from different ecosystems around the world have been conducted to measure changes in the sources or sinks of GHGs in response to elevated O 3 . These experiments feature different designs with regards to ecosystem type, fumigation method and duration, O 3 manipulation level, and other experimental factors (e.g. atmospheric CO 2 level). Such differences likely contribute to inconsistencies among different studies and complicate efforts to synthesize and interpret the results. Here, we conducted a meta-analysis (Hedges et al 1999, Gurevitch et al 2001 of the published literature to statistically synthesize and analyze all reported empirical findings and determine the sensitivity of CO 2 , CH 4 , and N 2 O exchange between terrestrial ecosystems and the atmosphere to changes in O 3 pollution. These results allow for the first systematic assessment of the sensitivity in the budgets of GHGs to tropospheric O 3 concentration elevation that are based on empirical data.  Environ. Res. Lett. 12 (2017) 084001 (Google, Mountain View, CA, USA) with a range of search terms. Only those studies that reported results on soil respiration, CH 4 uptake/emission, and/or N 2 O emission were selected. For example, the oftcited study by Findlay et al (1996) examined litter decomposition, but was not included here because CO 2 emission values were not reported. Similarly, studies that only measured the respiration of soil surrounding roots or root respiration were not included. We did not include lab incubation studies, except for CH 4 uptake with a lack of field measurements. These three GHGs show seasonal variations in their fluxes that appear to be independent of O 3 impacts. Therefore, to avoid the potential variability and bias caused by a limited number of sampling dates, only those studies that repeatedly measured the fluxes at least one year (growing season) are included. In these cases, we either extracted the annual sum, or calculated the mean values for a whole year.

Data collection
For effect size calculations, data including mean, standard deviation or standard error, and number of replicate for both control and treatment were extracted from text, tables, or graphs. The basic principles for data extraction from each individual study were as follows: To keep the independence of each observation, continuous observations within a single study were subjectively excluded to extract only one observation (i.e. the last year or growing season). However, different treatments (e.g. multiple manipulated levels of O 3 , different plant community, and elevated CO 2 level) in one study were treated as independent observations (Gurevitch et al 2001), and thus were included. The papers with data presented only as graphs were digitized using Engauge Digitizer (http://digitizer.sourceforge.net/). In addition, the control and treatment O 3 level (ppb) were extracted either directly if reported or indirectly by converting the cumulative O 3 exposure measures (e.g. accumulative ozone exposure above a threshold of 0 and 40 ppb: AOT0 and AOT40). Categorical variables included ecosystem type (forest, cropland, rice paddy, grassland, and peat land), fumigation method (free air concentration enrichment-FACE and open top chambers-OTCs), fumigation duration, and CO 2 elevation. The categorical variables and the corresponding levels used by different response variables (soil respiration, CH 4 flux, and N 2 O flux) depended on the data availability. The detailed information of these compiled data and the corresponding literature are listed in supporting information S1 available at stacks.iop.org/ERL/12/084001/mmedia.

Meta-analysis and meta-regression
First, the response ratio (RR) of O 3 elevation for each individual observation was calculated by response ratio: of which the sampling variance was computed by: where X c is the control mean, X t is the treatment mean, SD c is the control standard deviation, SD t is the treatment standard deviation, N c is the control replication number, and N t is the treatment replication number.
The normalization of effect sizes under different O 3 manipulation levels to a same DO 3 followed a linear transformation procedure: where X nt is the normalized value, X c is the control mean, X t is the treatment mean, DO 3 is the O 3 level difference between the treatment and control, and DO 3 Ã represents the O 3 level difference to which the effect sizes are normalized (i.e. a certain O 3 concentration interval between the current troposphere O 3 level and an elevated O 3 level: 10, 20, 30 and 40 ppb). Meta-analysis of RR was performed by fitting to the random-effects model via the restricted maximum-likelihood (REML) method. The inverse-variance weighted mean of RRs and the standard error of the weighted mean were calculated. The detailed calculation procedures are fully described in references Hedges et al (1999) and Koricheva et al (2013). The unlogged mean RR and corresponding 95% confidence interval (CI) were further obtained. Percentage change due to elevated O 3 was calculated from the unlogged mean RR. The O 3 effect on a response variable was considered significant if the 95% CI did not overlap 1.
The meta-regression of RR with moderators was performed by fitting to the mixed-effects model via the REML method as well. An omnibus test with a chisquare distribution (Q M ) (Koricheva et al 2013) was conducted to show whether the moderator has a significant contribution to the total heterogeneity. For categorical moderator with a significant Q M , the categorical group was subdivided and the inversevariance weighted mean RR for each level was calculated. An omnibus test was further conducted, with a significance of Q M indicating that the mean RR were different among the levels of this categorical moderator.
In addition, during the meta-analysis publication bias was tested by funnel plots (Jennions et al 2013), and Egger's regression test was further performed to test the plot asymmetry (Egger et al 1997). Q-Q normal plots were created to test whether the effect sizes fulfill the normality requirement (Wang and Bushman 1998). All calculation procedures described above were executed by R language using the 'metafor' package (Viechtbauer 2010).

Estimation of global GHG budget sensitivity
The meta-analysis results provide the average changes of each gas across the terrestrial ecosystems over the planet under a certain O 3 concentration interval between the current troposphere O 3 level and an elevated O 3 level (i.e. DO 3 Ã in ppb). We can give an approximate sensitivity estimate of the global GHG changes in response to this tropospheric O 3 level increase relative to today in a way as follows: As for the determination of net flux of CO 2 into the atmosphere, our work only conducted a metaanalysis of soil respiration and thus the change in carbon uptake by terrestrial vegetation is necessary to complete the analysis. Previous meta-analysis of plant photosynthesis response to O 3 elevation for trees ( Moreover, the synthesis work by Lombardozzi et al (2013) did not find a significant correlation between photosynthesis response and O 3 level and agreed with the average 20% effect. Therefore, a constant 20% value is applied to constrain the global net primary productivity (NPP) to quantity the reduction of vegetation carbon uptake caused by different O 3 elevation levels. It is noteworthy that our metaanalysis of soil respiration also includes the autotrophic respiration by plant roots, which, however, is not included in the NPP term. This represents a small uncertainty. Note that for the CO 2 flux change a 50% reduction is further applied assuming 50% percent of the net accumulation of CO 2 from the terrestrial biosphere is sequestered by the ocean (Sabine et al 2004).
To describe the global GHG budget change, GWP (global warming potential) was used to calculate the GHG balance in units of CO 2 equivalents as follows: where F CO 2 ÀC, F CH 4 ÀC, and F N 2 OÀN are annual changes of total soil CO 2 efflux, N 2 O, and CH 4 as calculated by the above equation. The fractions 44/12, 16/12 and 44/28 were used to convert the mass of CO 2 ÀC, CH 4 ÀC and N 2 OÀN into CO 2 , CH 4 , and N 2 O. GWP CH 4 (Pg CO 2 equivalent (equiv.) per Pg CH 4 ) and GWP N2O (Pg CO 2 equiv. per Pg N 2 O) are constants indicating integrated radiative forcing of CH 4 (28) and N 2 O (265) in terms of a CO 2 equivalent unit over a period of 100 years (Ciais et al 2013).

Results
Collectively, 96 effect sizes (RRs) for soil respiration (CO 2 emission), N 2 O emission, and CH 4 emission and uptake were compiled from peer-reviewed studies (supporting information S1). These studies were conducted in the Northern Hemisphere for ecosystems including forest, grassland, agricultural land, and peat land (figure S1 in supporting information S2). The data do not show publication bias based on funnel plots and Egger's tests (figure S2 supporting information S2) and fulfill the requirement of normality according to Q-Q plots (figure S3 in supporting information S2). Moreover, methodological differences, i.e. fumigation method and duration, among studies do not make a significant contribution to RR variability, with the exception of fumigation method for CH 4 emission (table S2 in supporting information S2). For these reasons, we conducted the analyses across the entire dataset.
Differing O 3 manipulation levels were adopted by different studies. We choose the absolute difference in experimental O 3 concentration between the treatment and control (hereafter referred to as DO 3 ) as a variable to describe the different studies. Meta-regressions between DO 3 and RR of different gases indicate that the RR of CO 2 and CH 4 (both emission and uptake) hold a significantly positive and negative linear relationship with DO 3 , respectively, while the RR of N 2 O is uncorrelated with DO 3 (figure 2). These results suggest that the magnitude of CO 2 and CH 4 responses depend on O 3 levels in the lower atmosphere. As for N 2 O, based on limited data we can not exclude its dependence on O 3 levels. In this study, we temporarily treat N 2 O response as being independent O 3 level. Therefore, to assess the sensitivity of terrestrial ecosystems in terms of GHG exchange to O 3 pressure, we test four DO 3 levels-10, 20, 30, and 40 ppb-to which RR were linearly normalized, except for N 2 O. Meta-analyses were further conducted on these normalized RR.
First, GHGs from different ecosystems present differing sensitivities to changes in O 3 (table S3 in Environ. Res. Lett. 12 (2017) 084001 supporting information S2). The CO 2 effluxes from soils in both cropland and peat land, though with a similar magnitude, show a low sensitivity to increased O 3 (figure S4 in supporting information S2). By contrast, CO 2 effluxes from grassland soils display strong sensitivity to elevated O 3 , with CO 2 flux diminished by from more than 10% (10 ppb) to more than 60% (20 ppb), and the forest shows a smaller but still significant response of 5% (10 ppb) to 30% (40 ppb) (figure S4 in supporting information S2). For CH 4 emissions, rice paddies present a significant reduction from almost 7% (10 ppb) to 25% ( Integrating the responses across these different ecosystems, we derived the average responses for each gas. For soil CO 2 efflux, the RR of CO 2 is positively correlated with O 3 level ( figure 2(a)), but does not show a significant sensitivity to O 3 enrichment until DO 3 of 30 and 40 ppb, when soil respiration is stimulated by an average of about 12% and 15%, respectively ( figure 3(a)). By contrast, CH 4 emissions are sensitive to O 3 across the full range of treatments, with an average of 6% (10 ppb), 11% (20 ppb), 14%  Figure 2. Meta-regression between RR and DO 3 . (a)-(d) denote soil respiration, CH 4 emission, CH 4 uptake, and N 2 O emission, respectively. The RR of N 2 O is not statistically correlated with DO 3 (d). The size of the bubble is the relative weight of the RR in the random-effects meta-regression. Larger bubbles indicate study outcomes that contributed a greater overall weight in meta-regressions.
Environ. Res. Lett. 12 (2017) 084001 (30 ppb), and 16% (40 ppb) ( figure 3(b)). Similarly, CH 4 uptake is sensitive to the dose of elevated O 3 with a large reduction seen initially but with low sensitivity to further increases in O 3 (figure 3(c)). Finally, the RR of N 2 O emissions does not depend on O 3 manipulation level (figure 2(d)); elevated O 3 significantly decreases N 2 O emission by an average of 22% regardless of the DO 3 level ( figure 3(d)). Our metaanalysis suggests that the soil respiration of terrestrial ecosystems is rather insensitive to O 3 pressure, stimulating CO 2 release to the atmosphere only after large O 3 changes. These responses of CO 2 , CH 4 , and N 2 O exchange occur primarily because of reduced carbon availability from inhibited photosynthesis and slower decomposition and nitrogen return from altered detritus quality ( Scaling up these average ecosystem-scale responses, we estimate the sensitivity of global budget of the three gases to enhanced O 3 (figure 4) As shown in figure 4(a), the contribution by suppressed N 2 O flux to the global budget is consistently ∼1.23 Pg CO 2 equiv. yr À1 . Increased O 3 of 10 and 20 ppb induce a decline of the current terrestrial natural CH 4 emission flux by an average of 6% and 11%, which contributes to a reduced balance of 0.6 and 1.0 Pg CO 2 equiv. yr À1 , while suppressed CH 4 uptake positively contributes 0.8 and 0.05 Pg CO 2 equiv. yr À1 . Global soil CO 2 efflux is insensitive to these small changes in O 3 (DO 3 = 10 and 20 ppb). By contrast, in response to larger increases in O 3 (ΔO 3 = 30 and 40 ppb) stimulated soil CO 2 emissions contribute an average of 4.4 and 5.5 Pg CO 2 yr À1 , respectively ( figure 4(a)). At the same DO 3 , suppressed CH 4 emissions contribute 1.3 and 1.5 Pg CO 2 equiv. yr À1 , while suppressed CH 4 uptake contributes only 0.07 and 0.09 Pg CO 2 equiv. yr À1 ( figure 4(a)). This sensitivity analysis clearly shows that the soil system can be transformed from a sink into a source with O 3 level continuously increasing because of enhanced soil CO 2 emissions outweighing suppressed CH 4 and N 2 O emissions. Taking into consideration O 3 -suppressed CO 2 uptake by vegetation (which is about 44.0 Pg CO 2 yr À1 ), plus the oceans' CO 2 sequestration, the O 3 -induced net global GHGs budget change on average is approximately 21.0 (5.7), 19.8 (5.4), 24.0 (6.6) and 24.9 Pg CO 2 yr À1 (6.8 Pg C yr À1 ) for the 10, 20, 30, and 40 ppb change in O 3 , respectively (figure 4(b);  4(b)). Therefore, the sensitivity of terrestrial vegetation productivity to elevated O 3 predominate the global GHG budget change sensitivity.

Discussion
The overall exchange of GHGs between terrestrial ecosystems and the atmosphere is sensitive to tropospheric O 3 pollution, and significantly contrib-utes to the atmospheric GHG accumulation. This matches the conclusion of a large buildup of CO 2 in the atmosphere derived from the modelling study by Sitch et al (2007), which, however, only considered the CO 2 . For understanding the full range of interactions between the atmosphere and terrestrial processes, we need an integrative perspective (e.g. Tian et al 2016).
Our results also suggest that suppressed CH 4 and N 2 O emissions can offset a portion of the increased CO 2 emissions. This study clearly pinpoints the necessity for assessing surface-atmosphere exchange processes comprehensively in air quality-climate feedback analyses. Therefore, other possible feedback pathways mediated by biogenic volatile organic compounds (BVOCs) and soil nitrous oxides (NO x ) emissions should be investigated and fully incorporated into assessment of O 3 -climate feedbacks (e.g. Hickman et al 2010).
It is important to note that these estimates of global GHG flux sensitivities come with uncertainties. First, GHGs from different ecosystems present differing sensitivities to changes in O 3 due to fundamental differences in vegetation composition. However, this study, as a pioneering work, integrates the different systems in an aggregate way, using the average responses disregarding the variations (though they are small relative to the sensitivities themselves). Moreover, uncertainties in the baseline estimates of the GHG fluxes will necessarily increase the magnitude of uncertainty of GHG fluxes sensitivity to O 3 exposure. Ecosystem-specific assessments are expected for future studies to tackle these uncertainties. A related question that should be addressed involves the impact of using different measures of O 3 dosage (e.g. AOT40 and O 3 flux) rather than O 3 concentration on estimates of ecosystem sensitivity. Second, the studied ecosystem types and independent observations that can be included in this meta-analysis are limited. For example, upland forest ecosystems generally act as a sink of CH 4 (e.g. Yavitt et al 1990), but no study is Environ. Res. Lett. 12 (2017) 084001 currently available of the O 3 -sensitivities of these ecosystems. There is also a striking lack of experimental data from the Southern Hemisphere (figure S1 in supporting information S2). This problem is particularly important because many regions in the Southern Hemisphere are predicted to industrialize in the future, which could lead to higher O 3 levels in the vicinity of cities and hemisphere-wide. Clearly, more field measurements on CO 2 , N 2 O, and CH 4 fluxes around the world should be undertaken. Third, exchange of GHGs between the atmosphere and terrestrial ecosystems can be simultaneously influenced by other global change agents. To more fully and realistically evaluate O 3 impacts on GHGs, the combined impacts with other global change agents, e.g. nitrogen deposition and climate warming, need additional investigation.

Conclusion
By fully accounting for the three gases simultaneously, we find that with the O 3 level continuously increasing the whole soil system would be transformed from a sink into a source of GHGs. With an increase of O 3 concentration by 10 ppb, the global annual net atmospheric budgets would on average increase by ∼12% (i.e. ∼5 PgC yr À1 ). However, in the global atmospheric budget suppressed CH 4 and N 2 O emissions can offset a small portion (at most 10%) of the net CO 2 increase derived from changes in soil respiration and vegetation carbon fixation in terms of CO 2 -equivalent. The sensitivity of global net GHGs budgets is still predominantly determined by the high sensitivity and thus a large magnitude of O 3 -induced reduction in terrestrial vegetation carbon sequestration capability. Natural systems with a high biodiversity, however, are increasingly suggested to be resilient to O 3 pressure. For example, the productivity and carbon stock of the temperate deciduous forest in the Eastern United States are shown to be unsuppressed by O 3 pressure (Wang et al 2016). The FACE study by Zak et al (2011), which included both O 3 -tolerant and O 3 -sensitive species or genotypes, also indicated unsuppressed net primary productivity after a longterm fumigation (11 years). Grassland ecosystems including temperate (Volk et al 2011), calcareous (Thwaites et al 2006, and alpine types (Bassin et al 2007) have also shown that the productivity is insensitive to elevated O 3 . Hence, we postulate that the indirect role of O 3 (that is via altering the landatmosphere exchange) on the Earth's radiative balance might be most significant on those managed systems that are of low diversity, e.g. agricultural and forest plantation systems. This means agricultural production practices would play a key role in mitigating O 3 's indirect effects on global climate, reinforcing the conclusion that adoption of best practices in human- Methane consumption in two temperate forest soils Biogeochemistry 9 39-52 Young P J et al 2013 Pre-industrial to end 21st century projections of tropospheric ozone from the Atmospheric Chemistry and Climate Model Intercomparison Project (ACCMIP) Atmos. Chem. Phys. 13 2063-90 Zak D R, Pregitzer K S, Kubiske M E and Burton A J 2011 Forest productivity under elevated CO 2 and O 3 : positive feedbacks to soil N cycling sustain decade-long net primary productivity enhancement by CO 2 Ecol. Lett. 14 1220-6 Environ. Res. Lett. 12 (2017) 084001