On the competing effects of contemporary land management vs. land cover changes on global air quality

Our work explores the impact of two important dimensions of land system changes, land use and land cover change (LULCC) and direct agricultural reactive nitrogen (Nr) emissions from soils, on ozone (O3) and fine particulate matter (PM2.5) air quality over contemporary (1992 to 2014) time scales. We account for LULCC and agricultural Nr emissions changes with consistent remote sensing products and new global emission inventories, respectively, estimating their impacts on global surface O3 and PM2.5 concentrations and Nr deposition using the GEOS-Chem global chemical transport model. Over this time 10 period, our model results show that agricultural Nr emission changes cause reduction of annual mean PM2.5 level over Europe and northern Asia (up to -2.1 μg m), while increasing PM2.5 level India, China and eastern US (up to +3.5 μg m). Land cover changes induce small reductions in PM2.5 (up to -0.7 μg m) over Amazonia, China and India due to reduced biogenic volatile organic compounds (BVOC) emissions and enhanced deposition of aerosol precursor gases (e.g. NO2, SO2). Agricultural Nr emission changes only lead to minor changes (up to ± 0.6 ppbv) in annual mean surface O3 level, mainly over China, India 15 and Myanmar. Meanwhile, our model result suggests a stronger impact of LULCC on surface O3 over the time period Across South America, the combination of changes in dry deposition and isoprene emissions results in -0.8 to +1.2 ppbv surface ozone changes. The enhancement of dry deposition reduces surface ozone level (up to -1 ppbv) over southern China, eastern US and central Africa. The enhancement of soil NOx emission due to crop expansion also contribute to surface ozone changes (up to +0.6 ppbv) over sub-Saharan Africa. In certain regions, the combined effects of LULCC and agricultural Nr emission changes 20 on O3 and PM2.5 air quality can be comparable (> 20%) to that of anthropogenic emission changes over the same time period. Finally, we calculate that the increase in global agricultural Nr emissions leads to a net increase in global land area (+3.67 ×10 km) that potentially faces exceedance in critical Nr load (> 5 kgN ha yr). Our result demonstrates the possible impacts of contemporary LULCC and agricultural Nr emission changes on PM2.5 and O3 air quality, which also implies the potential importance of land system changes on air quality over multi-decadal timescales. 25

Given the large spatial scale of LULCC (e.g. Hansen et al., 2013;Li et al., 2018) and agricultural emission changes (e.g. Crippa et al., 2018;Hoesly et al., 2018;Xu et al., 2019) over recent decades, these two land system changes could be contributing 65 substantially to global trends in O3 and PM pollution. While changes in land cover and agricultural emissions actually occur contemporaneously across the globe, they are rarely considered together in simulations of air quality from chemical transport models. As a result, it is not clear to what extent LULCC may either amplify or offset the impacts of some of the associated agricultural emission changes, how this may vary regionally, and to what extent these land system impacts may compare to concomitant changes resulting from other direct anthropogenic emissions (e.g. emissions from industrial and transport sectors). 70 The recent availability both of consistent long-term land records of land cover derived from satellite remote sensing observations, and global anthropogenic emission inventories, opens an opportunity for a more holistic and observationallyconstrained assessment of the impacts on global O3 and PM air quality from contemporary changes in LULCC and agricultural emissions simultaneously, and a comparison of these with the effects from direct anthropogenic emissions. In this study, we 75 model the effects of contemporary LULCC and agriculture emissions changes on global surface O3 and PM2.5 levels, and gauge their importance relative to changes in other direct anthropogenic emissions over the same period of time. We also highlight the effect of agricultural emissions changes on nitrogen deposition on land ecosystems. Through our chemical transport model predictions, we aim to identify potential global hotspots of contemporary land changes that may be substantially altering trends in air quality and nitrogen deposition. 80

Method
To simulate global changes in surface O3 and PM2.5 concentrations due to LULCC, agricultural emissions, and direct anthropogenic emissions over 1992 to 2014, we use the GEOS-Chem chemical transport model (version 12.7.0, https://doi.org/10.5281/zenodo.3634864). We choose our timeframe due to the availability of consistent high-resolution remote sensing products (PFT and LAI maps) and concurrent global emission inventories. We define "direct anthropogenic" 85 and "agricultural" emissions separately in more detail below.
We perform five sets of simulations summarized in Table 1 The role of direct anthropogenic emission changes can be evaluated by comparing simulation (1) and (2); the additional role 95 played by land cover changes over this time period is evaluated by comparing simulation (2) and (3); and finally the additional impact of agricultural emission changes is evaluated by comparing simulation (3) and (4). The latter two effects will be the focus of this paper, but we compare these to the role of direct anthropogenic emission changes for context. Since changes in surface ozone and PM2.5 should be sensitive to NOx-VOC ratio and availability of NO3and SO4 2ions, the sensitivity of the effects from land cover change and agricultural emission changes to anthropogenic emission changes can be quantified by 100 evaluating simulation (5).
We use assimilated meteorological fields from Modern-Era Retrospective analysis for Research and Applications Version 2 (MERRA-2) (Gelaro et al., 2017) to drive GEOS-Chem. All simulations are carried out at 2° latitudes by 2.5° longitudes resolution over the globe, using identical meteorological fields from 2011 to 2014 in order to exclude meteorological variability from the analysis. The output from 2011 is discarded as spin-up. The GEOS-Chem model simulates O3 chemistry with a 105 comprehensive HOx-NOx-VOC-O3-BrOx chemical mechanism Mao et al., 2013). Gaseous dry deposition follows Wang et al. (1998) and Wesely (1989), while particle deposition follows Zhang et al. (2001). Wet deposition is described by Liu et al. (2001) with updates from Amos et al. (2012) and Wang et al. (2011Wang et al. ( , 2014. The recent update from Luo et al., (2019) on wet deposition parameterization is also included to improve model-observation agreement for sulfate-nitrateammonium (SNA) aerosol. The thermodynamics and gas-aerosol partitioning of the NH3-H2SO4-HNO3 system is simulated 110 by ISORROPIA II module (Fountoukis and Nenes, 2007). A simple yield-based secondary organic aerosol (SOA) estimate is also included (Kim et al., 2015). Other types of aerosol represented in the model include sea salt, dust, primary black carbon (BC) and organic carbon (OC). The total PM2.5 mass is then calculated at 35% relative humidity for consistency with the measurement standard in US.

115
We use anthropogenic and agricultural emissions based on the Community Emission Data System (CEDS) inventory (Hoesly et al., 2018), which contains the estimates of anthropogenic NOx, non-methane volatile organic compounds (NMVOCs), CO, BC, OC, SO2 and NH3 harmonized from a wide range of global and regional inventories. In this inventory, emissions are from six major sectors: energy production, industry, transportation, RCO (residential, commercial, other), agriculture, and waste.
For this study, "agricultural emissions" specifically refer to NOx and NH3 emitted from fertilizer application and manure 120 management, which correspond directly to agricultural nitrogen input. We do not consider the changes in agricultural of other trace species (e.g. CH4, SO2, CO).
Biogenic volatile organic compound emissions are calculated by Model of Emissions of Gases and Aerosols from Nature (MEGAN) v2.1 (Guenther et al., 2012). Soil NOx emission follows Hudman et al. (2012), with fertilizer emissions zeroed out 125 to avoid double counting with the agricultural NOx emission in CEDS inventory. Fire (Global Fire Emissions Database v4.1, Van Der Werf et al., 2017) and lightning (Murray et al., 2012)  We use the European Space Agency Climate Change Initiative (ESA CCI) land cover map  to characterize LULCC and drive the biosphere-atmosphere emission fluxes in our simulations. The ESA CCI land cover map is a consistent 130 global annual land cover time series derived from the satellite observations from the AVHRR, MERIS, SOPTVGT and PROBA-V instruments. It has a native spatial resolution of 300 m following the United Nations Land Cover Classification System. Time-consistent land surface characterization also requires leaf area index (LAI) data. We use the Global Land Surface Satellite (GLASS) product  (retrieved from http://globalchange.bnu.edu.cn/), which is a global LAI time series combining AVHRR and MODIS observation. 3-year average (1991-1993average LAI for 1992land cover, 2013 2015 average LAI for 2014 land cover) is used as input for LAI to GEOS-Chem to reduce the possible effect of interannual variability.
This satellite-derived land surface characterization on its own is not directly compatible with the input to the vegetation-related modules in GEOS-Chem, thus requires further harmonization (dry deposition, BVOC emissions, soil NOx emissions), which is a common problem for simulations involving land change (e.g. Geddes et al., 2016). We first aggregate and process the 140 ESACCI land cover map with the tool and crosswalk table provided with the land cover product to derive the percentage coverage of plant functional type (PFT) at 0.05° resolution, which is the native resolution of GLASS LAI. The dominant surface type can be readily mapped to the 11 deposition surface type in the Wesely dry deposition model. We adopt the approach of Geddes et al. (2016) to replace roughness length (z0) from assimilated meteorology with that prescribed for each deposition surface type. To derive the MODIS-Koeppen type land map (Steinkamp and Lawrence, 2011) required for soil NOx 145 module, we first use translate the PFT map according to International Geosphere-Biosphere Programme (IGBP) land cover classification system (http://www.eomf.ou.edu/static/IGBP.pdf). We use global monthly temperature climatology (Matsuura and Willmott, 2012) to further differentiate the land types by climate with criteria outlined by Kottek et al. (2006). Finally, the ESA CCI PFT map is converted to Community Land Model (CLM) PFT map, which is required for MEGAN BVOC emissions module, by the temperature criteria specified by Bonan et al. (2002). As the method of deriving C3:C4 grass ratio was 150 subsequently updated (Lawrence and Chase, 2007), this ratio is directly taken from CLM land surface data set.
In the Supplementary Material, we provide an evaluation of the annual mean simulated SNA aerosol mass concentration and surface O3 mixing ratios from the Simulation 4 (representative of 2014 conditions) with globally available observations from the same time period. In general, the model captures the regional annual means of individual SNA species reasonably (Fig. S1  155 and Table S1), especially over US and Europe where the bias is within ±30%. The model underestimates all SNA species over China in a relatively uniform fashion (36 -55%). Over the region covered by Acid Deposition Monitoring Network in East Asia (EANET) (Japan, Korea and southeast Asia) the model underestimates the negatively charge ions (36% for sulphate and 16% for nitrate) while overestimating ammonium by 14%. Figure S2 shows the reasonable agreement on annual mean surface O3 between our model output and the gridded observation dataset from Sofen et al. (2016)  absolute error = 3.97 ppbv). Our model therefore captures the present-day annual means of surface SNA and O3 concentrations, providing basis for our subsequent analyses. Table 2 shows the changes in global coverage of the major land cover types from 1992 to 2014 derived by the ESA CCI land cover product. The coverage of managed grass (including cropland and pasture) and built-up area, both of which are 165 unmistakably related to human activities, have increased mainly at the expense of forest coverage, indicative of a global trend in deforestation over this period. Figure 1 shows the spatial distribution of changes in fractional coverage of the major land cover types. Expansion of agricultural land at the expense of broadleaf forest coverage is most notable in southern Amazonia and Southeast Asia, which is well-documented in other studies based on remote sensing (Hansen et al., 2013) and national surveys (Keenan et al., 2015). The expansion of agricultural land over this time period is also observed in central Asia, African 170 savannah, southeastern Australia, South China and North Africa, but usually at the expense of land types other than broadleaf forests (mainly primary grassland and needleleaf forests). Meanwhile, transitions from agricultural land to forests and builtup areas is observed in northern China and eastern Europe, consistent with the findings of Potapov et al. (2015) and Lai et al. (2016).

Changes in Land Cover, and Biospheric Fluxes, and Agricultural Emissions
175 Figure 2 shows the global changes in 3-year (2012-2014 minus 1991-1993) annual mean LAI calculated from the GLASS LAI data set. Over South China, Paraguay and northern Argentina, the area with regionally consistent deforestation experience general increase in LAI, while the opposite effect is observed in African savannah and central Asia. In Europe, LAI decreases in Ukraine, Poland, Germany, but increases in most other parts despite a fairly consistent retraction of agricultural land is observed over the whole Europe. The agricultural expansion and deforestation over Southeast Asia is mostly concurrent with 180 the LAI decreases. LAI increases notably in northern China where agricultural land decreases. The fact that LAI change can be driven by factors other than changes in land cover type (e.g. temperature, precipitation, atmospheric CO2 level) (e.g. Zhu et al., 2016) may explain the regionally divergent trends response of LAI to agricultural land use change. We note that since the relationship between satellite-derived surface reflectance and retrieved LAI depends on land cover, the use of static land cover map in long-term LAI retrievals (Claverie et al., 2016;Xiao et al., 2016;Zhu et al., 2013) may not fully capture the effect of 185 LULCC on LAI (Fang et al., 2013).
These changes in land cover produce changes in the biogenic fluxes of reactive trace gases between the Earth's surface and atmosphere derived by GEOS-Chem. Figure 3a shows the calculated changes in annual mean isoprene emission due to land cover change over 1992 to 2014, and suggests that global isoprene emission could have decreased by 5.12 Tg/yr (-1.5 %). The 190 largest local reductions in isoprene emissions are observed in southern Amazonia, Paraguay and northern Argentina, where deforestation from highly isoprene-emitting broadleaf forests is most strongly observed, causing decreases in isoprene https://doi.org/10.5194/acp-2021-132 Preprint. Discussion started: 23 March 2021 c Author(s) 2021. CC BY 4.0 License. emissions by up to 30%. We note that the decrease of isoprene emission simulated in Southeast Asia does not agree with the result from Silva et al. (2016), since our remote sensing data does not have separate land cover class for oil palm plantations which have expanded dramatically in the region. Our model may not therefore capture the full effects of LULCC on isoprene 195 emission, and its effect on PM2.5 and O3 over the region. Elsewhere in the world, the signals of land cover change on isoprene emissions are mostly small and follow the local patterns of changes in LAI. Changes in monoterpenes (< 5 ng m -2 s -1 ) and sesquiterpene (< 1 ng m -2 s -1 ) emissions are relatively small. Figure 3b shows the changes in annual mean soil NO emission due to LULCC, which represent the change in soil emission 200 driven purely by LAI and land cover changes (i.e. without considering the changes in nitrogen input). LULCC leads to a small signal of +0.04 TgN/yr (+0.6%) in global soil NO emission. The magnitude of changes in soil NO emission induced by LULCC is comparable to that in agricultural NO emissions inventory (see below) over certain regions (e.g. South America, Australia, Sub-Saharan Africa). Relatively large increases in soil NO is simulated over western Africa.
205 Figure 3c shows the changes in annual mean O3 dry deposition velocity (vd), which also closely follow the pattern of LAI changes. Slight increases of vd are observed in China, India, Southeast US, Mexico, northern Amazonia, Europe and southern Africa (SAf). In Southeast Asia vd decreases concurrently with deforestation and reduction in LAI. In central Brazil, the increase in LAI is offset by the deforestation of tropical evergreen broadleaf forests that have higher vd than other land types (Song-Miao Fan et al., 1990;Wang et al., 1998), leading to small overall change in vd. Likewise, despite deforestation observed 210 in the savannah and grassland over Rio de la Plata basin, these losses are offset by strong increases in LAI so that vd increases by up to 0.1 cm s -1 . Significant changes in the vd of O3 due to LAI also imply that vd of other relevant trace gases (e.g. NO2, SO2) would also be perturbed by land cover change in our model, which will be discussed briefly in the subsequent section. Figure 4 shows the changes in agricultural NH3 and NOx emissions between 1992 and 2014, which consists mostly of emissions 215 from fertilizer application and manure management (Hoesly et al., 2018). According to the CEDS inventory, global direct agricultural NH3 emissions increased by 7.6 Tg N/yr since 1992, equivalent to a 19 % increase in total anthropogenic NH3 emissions. Direct agricultural soil NOx emissions increased by 0.37 Tg N/yr since 1992, and while this is a substantial increase in agricultural soil NOx emissions (26 %), it represents only a 1 % increase in total anthropogenic NOx emissions.

220
The increases in agricultural emissions are most substantial over the Indian Subcontinent, followed by China, Egypt, southeast We note that the hotspots of change in managed land cover and of change in agricultural emissions are not always overlapping. 230 For example, agricultural emissions increase significantly over northern China and northern India, while the cropland coverage over those regions does not increase correspondingly over this same period. Similarly, agricultural emissions have declined over Kazakhstan, while the area of managed land has not decreased significantly. This highlights a degree of independence between land management and LULCC, with both being components of land change but having potentially distinct spatial patterns and impacts on air quality. This also highlights the importance of treating both in our chemical transport model 235 simulations as they occur contemporaneously around the globe, and may have different impacts on air quality. 4 Impact of LULCC and agricultural emission changes on surface PM2.5 Figure 5 shows the modeled impacts of LULCC, changes in agricultural emissions, and the combined effects of both, on annual mean surface PM2.5 (under 2014 anthropogenic emissions). We have calculated the impacts of LULCC on PM2.5 ("∆PM2.5, LULCC" as the difference in PM2.5 predicted by Simulation 2 and Simulation 1; the impacts of agricultural emission changes on 240 PM2.5 ("∆PM2.5, agr_emis") as the difference in PM2.5 predicted by Simulation 3 and Simulation 2; and the impacts of these combined ("∆PM2.5, LULCC+agr_emis") as the difference in PM2.5 predicted Simulation 3 and Simulation 1 (see Table 1).
The effect of LULCC on PM2.5 (Fig. 5a) is mainly through perturbing BVOC emissions as they are a precursor to SOA. Over southern Amazonia and maritime southeast Asia, where isoprene emissions drop significantly due to deforestation, PM2.5 is 245 reduced by up to 0.7 μg m -3 . Land cover changes also lead to changes in the dry deposition velocity of some SNA precursor gases where stomatal uptakes is an important deposition pathway (e.g. NO2 and SO2, Fig. S4). Indeed, over northeastern India and eastern China, where our model suggests high levels of SNA aerosol precursors, contemporary LULCC enhances dry deposition of these constituents which reduces PM2.5 overall by up to 0.3 μg m -3 , similar to the finding of Fu et al. (2016).

250
We find that the agricultural emissions generally have larger impact on annual mean surface PM2.5 level (Fig. 5b) than LULCC.
The largest increases in annual mean surface PM2.5 due to changes in agricultural emissions over 1992 to 2014 occur across China (+0.7 μg m -3 ) and India (+1.6 μg m -3 ). Over some hot spots in the two countries (e.g. northwestern India and North China Plain), the local changes in PM2.5 exceed >3.5 μg m -3 , supporting the previously emphasized importance of controlling NH3 emissions on PM air quality of China (Fu et al., 2017), but potentially India as well. Some moderate increases (< 2 μg m -255 The largest decreases in annual mean PM2.5 due to changes in agricultural emissions are simulated in eastern Europe, particularly over Ukraine (-2.1 μg m -3 ), Bulgaria (-1.5 μg m -3 ) and Romania (-1.2 μg m -3 ). In Russia, similarly large decreases 260 are observed in the western and south Siberian part of the country. Despite comparable reductions in agricultural NH3 emissions, decreases in PM2.5 over the Benelux region are smaller because of weaker sensitivity of SNA aerosol to NH3 emissions, which is consistent with the finding of Lee et al. (2015) and Pozzer et al. (2017). In general, reductions in annual mean PM2.5 due to agricultural emission changes simulated over western Europe are weaker than over eastern Europe.
265 Figure 5c shows the combined effect of agricultural emissions and LULCC on annual mean surface PM2.5, which we have already shown is mostly dominated by the effect of agricultural emissions. Nevertheless, we find that the effects of LULCC are able to partially offset the increase in PM2.5 due to agricultural emissions changes over eastern China and northeastern India. These offsets are occurring in densely populated areas, so that the effects on population-weighted average PM2.5 concentrations (see below), and therefore potentially exposure, may be noteworthy. This is discussed in further detail below. 270 We note that the difference between Figure S5a Table 3 summarizes the simulated effects of these land change phenomena on PM2.5 air quality, and compares their magnitudes with the concomitant effects from direct anthropogenic emission changes ("∆PM2.5, anth") over the same time period. We additionally compare area-averaged and population-weighted global and regional metrics. While the resolution of our 280 simulations does not capture urban-scale gradients and non-linearities in urban chemistry, the use of population weighting allows us to explore whether signals of change in land cover or land management are concentrated over areas of high population, or whether they are primarily observed over less populated areas. The country definitions of each region in Table   2 are provided in Table S2 ( This suggests that the increase in agricultural emissions over NAm has partially canceled out the effects of other emission controls on PM2.5, though this effect is small so far (~5%). In other regions, population-weighted ∆PM2.5, LULCC+agr_emis is 295 generally on the order of 5% to 12% of changes due to direct anthropogenic emissions (e.g. in CEU and WEU). Notably, over FSU, the Middle East (ME), and central America (CAm), ∆PM2.5, LULCC+agr_emis are much more comparable to the effect of anthropogenic emission changes (24%, 42%, and 208% respectively).
Our result shows that the impact of LULCC and land management changes on PM2.5 is mainly from the agricultural emission 300 changes, while LULCC can result in additional impacts in regions with high SNA precursor emissions (e.g. India, China) through modulating dry deposition. The magnitude of population-weighted ∆PM2.5, LULCC+agr_emis suggests that land change may contribute significantly to regional and global changes in human PM2.5 exposures and that the effects of these changes are not isolated to low population regions. Particularly, over the regions experiencing rapid change in land use intensity (e.g. FSU) or slow change in anthropogenic emissions (e.g. CAm, ME), the effects of land changes on particulate air pollution could be 305 comparable (24% to 208%) to the effects of direct anthropogenic emission changes. Figure 6 shows the modeled impacts of LULCC, changes in agricultural emissions, and the combined effects of both, on annual mean surface O3 (under 2014 anthropogenic emissions). These changes are calculated identically as for PM2.5 above: the impacts of LULCC on O3 ("∆O3, LULCC") is the difference in O3 predicted by Simulation 2 and Simulation 1; the impacts of 310 agricultural emission changes on O3 ("∆O3, agr_emis") is the difference in PM2.5 predicted by Simulation 3 and Simulation 2; and the impacts of these combined ("∆O3, LULCC+agr_emis") is the difference in PM2.5 predicted Simulation 3 and Simulation 1 (see Table 1). We also use predictions of surface HNO3/H2O2 ratios ( Figure S6) as a proxy of VOC-vs NOx-sensitive chemical O3 production (Peng et al., 2006;Sillman, 1995) in our discussion of the results.

315
The modelled response of surface O3 to LULCC (∆O3, LULCC) (Fig. 6a) involves several distinct processes (dry deposition, soil NOx, and BVOC emissions). Over eastern US and central Mexico, the increase in dry deposition velocity (vd) reduces annual mean surface ozone by up to 0.5 ppbv overall. In central Brazil, deforestation of tropical rainforests leads to significant reduction is isoprene emissions, reducing surface ozone by up to 0.8 ppbv in this NOx-limited environment (Fig. S6). In contrast, modelled surface ozone increases by up to 1.2 ppbv over Bolivia and northern Argentina, where strong increases in 320 LAI lead to largely increases vd. The modelled reduction of surface ozone (up to 1 ppbv) over central African rainforests is also likely attributable to increased vd as neither soil NOx nor isoprene emissions change much in the region. However, in other parts of sub-Saharan Africa, up to 0.6 ppbv of surface ozone increases are simulated, mainly because of the relatively large increase in soil NO emission. In southern China, up to 0.5 ppbv reduction in surface ozone is simulated, which is likely attributable to the increase in vd, and slightly offset by the small increase in isoprene emission under this NOx-saturated 325 environment (Fig. S6).
Overall, the role of agricultural emission changes in fertilizer-associated NOx plays a minor role in surface O3 changes (Fig.   6b). An exception to this is observed in the large increase in agricultural NOx emissions in parts of Asia which reduce surface O3 by up to 0.6 ppbv over NOx-saturated northern India and eastern China, but increase surface O3 in NOx-limited Myanmar 330 by similar magnitude. Slight increases in surface O3 level due to increased agricultural NOx emissions are also simulated over East Africa and southern Brazil. Whether the effect of agricultural emissions strengthens (e.g. eastern China and Sahel) or offset (e.g. over southern Brazil and northern India) is largely region-dependent. As shown in Fig. 6c, LULCC tend to dominate the impacts on surface O3 over most regions in the world (unlike PM2.5 where the effects of agricultural emission changes dominate). 335 Similar to PM2.5, we find that the changes in anthropogenic emission background over 1992 to 2014 is strong enough to alter the sensitivity of O3 to land change. As indicated by Fig. S6, Asia was less NOx-saturated, while western Europe and coastal United States were more NOx-saturated in 1992 than in 2014. For example, the increase in soil NO emission over India is more like to increase rather than decrease surface ozone concentration ( fig. S7a), leading to different modelled effect on surface 340 ozone. Table 4 shows the change in area and population-weighted annual mean afternoon surface O3 due to the effects of anthropogenic emissions ("∆O3, anth", Fig. S7b), ∆O3, LULCC, ∆O3, agr_emis and ∆O3, LULCC+agr_emis. In most regions, ∆O3, LULCC+agr_emis is positive. However, this is offset by the negative population-weighted average ∆O3, LULCC+agr_emis over the most populous 345 regions (South Asia and China), resulting in very small globally averaged population-weighted ∆O3, LULCC+agr_emis.
The magnitudes of population-weighted ∆O3 (within ±0.5 ppbv) display less regional variability than that of ∆PM2.5. Over Eastern Africa (EAf), Western Africa (WAf) and Southern Africa (SAf), area-averaged ∆O3, agr_emis+land_cover generally has similar magnitudes to population-weighted ∆O3, agr_emis+land_cover. In other regions, the differences between area and population-350 weighted ∆O3, LULCC+agr_emis are more substantial. The largest discrepancies between area and population-weighted ∆O3, LULCC+agr_emis is found over China, where increases in surface O3 are predicted over less populated western China, while reductions in surface O3 are simulated over more densely-populated eastern China. In South America (SAm), there are large sub-regional signals of ∆O3, LULCC+agr_emis, but these the positive and negative largely offset each other, resulting both in small area-weighted and population-weighted ∆O3, LULCC+agr_emis. 355 Over China, FSU, ME, WAf, EAf and SAf, the magnitudes of population-weighted ∆O3, LULCC+agr_emis are more than 20% of that of ∆O3, anth, implying that contemporary land system changes could be a regionally important component in contemporary trends of surface O3. The effects of agricultural emission changes and LULCC can either noticeably enhance (e.g. over ME, JK, China) or offset (e.g. over, SAs) each other because of the dependence of ∆O3, agr_emis+land_cover on regional NOx-VOC 360 chemistry and details of LULCC, indicating the complexity of diagnosing the effect of land change on surface O3 at regional and global scale.
Our result suggests that contemporary agricultural emission changes and LULCC each have distinct effects on surface O3, with LULCC generally stronger in magnitude. Both of the effects are dependent on local NOx-VOC chemistry, as agricultural 365 emission changes perturb NOx emissions, while LULCC tends to affect BVOC emissions. In addition, LULCC is also able to affect surface O3 (and other precursors) directly through dry deposition through LAI changes over our period of concern. These effects are found to affect O3 pollution over densely populated regions (e.g. China) and could be comparable to the magnitudes of O3 changes due to anthropogenic emissions over specific regions (e.g. FSU, EAf, WAf), indicating the importance of land change in studying long-term changes in surface O3. 370

Impact on Nitrogen Deposition
Finally, we estimate the effect of these land changes on nitrogen deposition estimates. Figure 7 shows the global impact of LULCC and agricultural emission changes on total nitrogen deposition (∆Ndep), and Table 5 summarizes the regional and global results. The largest increase and decrease in nitrogen deposition (Ndep) are simulated over SAs (+1.91 TgN yr -1 ) and FSU (-1.28 TgN yr -1 ), respectively. Notable increases in Ndep are also simulated over China (+1.55 TgN yr -1 ), SAm (+1.24 375 TgN yr -1 ), NAm (+0.66 TgN yr -1 ), WAf (+0.39 TgN yr -1 ) and EAf (+0.41 TgN yr -1 ). Figure 7 also illustrates the simulated changes over 1992 to 2014 in area with nitrogen deposition (Ndep) exceeding 5 kgN ha -1 yr -1 , which is a proxy of possible exceedance of critical Ndep loads for terrestrial and fresh water (Moriarty, 1988).
Globally, there is a net increase in land area with Ndep > 5 kgN ha -1 yr -1 of 3.67×10 6 km 2 . The increase is mostly simulated 380 over the Americas, Africa, ME and China, which is partially offset the large decrease over FSU. Meanwhile, despite agricultural changes that lead to notable ∆Ndep, over most of Europe, eastern US, China, South Asia and Southeast Asia, nitrogen input from other sources are large enough that this signal alone does not lead substantial changes in Ndep exceedances of 5 kgN ha -1 yr -1 . However, over the periphery of North American Great Plain, southeastern part of South America, Nile River Delta, western China and African Savannah, agricultural changes are simulated to increase Ndep from below to above 5 kgN 385 ha -1 yr -1 . This implies these natural ecosystems at the edge of these areas are at risk of nitrogen exceedances due to agricultural changes. In contrast, the substantial reduction of Ndep in southern Russia may have significantly reduce the risk of nitrogen exceedance of natural ecosystem from agricultural sources.

Discussion and Conclusions
In this work, we have explored how changes in the global land system, through LULCC and agricultural emission changes, 390 may have impacted contemporary global air quality over 1992 to 2014. We model the effects of contemporary LULCC and agricultural emission changes, individually then in combination, on surface O3 and PM2.5 using the GEOS-Chem CTM with a uniquely consistent framework that is able to integrate direct information from global emission inventories (CEDS) with updated land surface remote sensing products (ESACCI land cover and GLASS LAI) on surface O3 and PM2.5, allowing us to avoid invoking extra assumptions on land management practices (e.g. constant Nr input, emissions or emission factors over 395 time) and biophysical properties of PFTs (e.g. constant PFT-specific LAI over time).
We find that changes in agricultural emissions are simulated to increase the annual mean surface PM2.5 concentrations in China and India by up to 3.5 μg m -3 and to decrease in Europe by up to 3.5 μg m -3 . Our simulation suggests that though ΔPM2.5 is mainly attributable to changes in agricultural emissions in global scale, LULCC over northeastern India and eastern China can 400 lead to enhanced dry deposition of certain PM2.5 precursor gases (SO2 and NO2), thus partially offsetting (~10 %) the increase in PM2.5 from agricultural regions. This implies a potentially important role of LULCC in determining SNA aerosol level over certain heavily polluted regions. Also, LULCC reduces BVOC emissions over Amazonia which lead to reductions in PM2.5 by up to 0.7 μg m -3 . In a future with decreasing anthropogenic NOx and SO2 emissions, which could diminish the importance of agricultural emissions on PM2.5 formation (Bauer et al., 2016), LULCC may become increasingly important in the overall 405 effect of land change on PM2.5. Noticeable changes (> 1 μg m -3 ) population-weighted ∆PM2.5, LULCC+agr_emis are simulated over China (+1.45 μg m -3 ), SAs (+1.71 μg m -3 ), CEU (-1.00 μg m -3 ) and FSU (-1.01 μg m -3 ), indicating the potential impact of land change on long-term public health through modulating PM2.5 level at regional scale. Our results suggest that contemporary (1996)(1997)(1998)(1999)(2000)(2001)(2002)(2003)(2004)(2005)(2006)(2007)(2008)(2009)(2010)(2011)(2012)(2013)(2014) changes contribute to changes in PM2.5 at regional and global scales that range from on the order of 5 to 10% of changes in PM2.5 resulting from direct anthropogenic emissions over the same time period, and up to ~25% or more in FSU, 410 ME and CAm specifically.
In contrast, the effect of LULCC is generally stronger than that of agricultural emission change in simulations of surface O3.
We find that the role of LULCC over 1992 to 2014 is regionally significant enough to induce changes in BVOC emissions and dry deposition which affect surface O3, but that the overall effects largely offset each other on the global scale, leading to very 415 small population-weighted ∆O3, LULCC+agr_emis. Both the effects of agricultural emission changes and LULCC, through NOx and BVOC emissions, are sensitive to regional ozone production regime. The increase in agricultural emissions reduces O3 over NOx-saturated China and SAs by up to 0.6 ppbv, while the reduction in BVOC emissions increases surface O3 over VOClimited Amazonia by up to 1.2 ppbv; enhancements of dry deposition reduce O3 over Rio de la Plata Basin, eastern China, and eastern US by up to 1.2 ppbv. Overall, the largest population-weighted ∆O3, LULCC+agr_emis is simulated over WAf (+0.42 ppbv) 420 and EAf (+0.47 ppbv). We find that the ratio between ∆O3, LULCC+agr_emis and ∆O3, anth varies widely depending on region, with https://doi.org/10.5194/acp-2021-132 Preprint. Discussion started: 23 March 2021 c Author(s) 2021. CC BY 4.0 License. some having ∆O3, LULCC+agr_emis that are comparable (>20%) to ∆O3, anth. These results show the complexity and importance of land change on mediating long-term changes in surface O3.
We also find that both the modelled ΔO3, LULCC+agr_emis and ΔPM2.5, LULCC+agr_emis are sensitive to the changes in anthropogenic 425 emissions suggested by CEDS inventory over 1992 and 2014, as the changes in NOx, SO2 and VOC emissions are large enough to perturb atmospheric HNO3 and H2SO4 production, and ozone production regime considerably in many regions (e.g. Asia and western Europe). This highlights the necessity of accurate and relevant emission inventories when evaluating the impacts of land change on air quality (e.g. Bauer et al., 2016).

430
The increased atmospheric reactive nitrogen (+7.20 Tg/yr) due to agricultural emissions is mostly found to deposit near to source region as the atmospheric lifetime of NH3 is generally short, which implies the potential risk of excessive nitrogen input over the natural ecosystems near to the regions with increases in agricultural emissions.
Our work suggests that, at contemporary timescales (on the order of ~20 years), the effect of land change on air quality is 435 contributes changes in air quality that can sometimes be important relative to changes induced by trends in direct anthropogenic emissions. We also find that agricultural emission changes have stronger effects on PM2.5, while LULCC have stronger effects on O3. This finding is comparable to that from Heald and Geddes (2016), which suggest a much more comparable changes in biogenic SOA (mostly induced by LULCC) and particulate nitrate (mostly induced by agricultural emission changes), and stronger surface ozone changes induced by land change over 1850 -2000. This shows that both the magnitudes and relative 440 contributions from different components of land change effects on air quality vary significantly as timescale of study, and its potential importance at longer timescales (e.g. multidecadal, centennial), despite the relative small signal that we obtain here.
We note several important limitations and opportunities for development. We were only able to evaluate our simulation extensively over Europe, North American and East Asia. In most other regions where such evaluation of SNA speciation is 445 not feasible, the sensitivity of SNA formation to NH3 emissions can be a major source of uncertainty. Given the changes in agricultural emissions have occurred in global scale, effort of monitoring SNA speciation outside North America and Europe (e.g. Weagle et al., 2018) is necessary for understanding the sensitivity of PM2.5 to agricultural emissions in global extent.
Better understanding of both the sources and sinks of HNO3 (e.g. Heald et al., 2012;Holmes et al., 2019;Luo et al., 2019;Petetin et al., 2016) and nitrate partitioning (e.g. Vasilakos et al., 2018) are important for modelling SNA aerosol and its 450 sensitivity to NH3 emissions. NH3 emissions estimates also carry large uncertainty (Crippa et al., 2018). The inherent inconsistency of long-term LAI time series derived from reflectance measured by different instruments (Jiang et al., 2017) and the use of static land cover maps also introduce uncertainty in the LAI retrieval (Fang et al., 2013)  Our study helps demonstrate the possible magnitudes and regional patterns of the impacts of contemporary LULCC and agricultural emission changes on PM2.5 and O3, and suggests that the combination of these factors should not be neglected in the study of regional and global air quality changes over multi-decade timescales. Our results confirm the potential importance of controlling agricultural emissions on improving PM2.5 air quality, which could be practical as numerous feasible options 460 exist for reducing agricultural emissions through optimizing livestock and crop production system (e.g. Ti et al., 2019).
Furthermore, as increasing reactive nitrogen input and land use change are the two of the main strategies to meet the global demand for biomass-based products in the future (Foley et al., 2011), the distinct yet significant impacts of agricultural emissions and land use change on O3, PM2.5 and nitrogen deposition should be investigated as part of the overall environmental impacts of land system changes, especially when tradeoff between increasing land input and cropland expansion exists (e.g. 465 Lotze-Campen et al., 2010;Mauser et al., 2015).

Code Availability
The source code of GEOS-Chem model is publically available (https://doi.org/10.5281/zenodo.3634864). The GEOS-Chem model output and other source code used in the project can be obtained by contacting the corresponding author (jgeddes@bu.edu). 470

Competing interests
The authors declare that they have no conflict of interest.

Author Contributions
AYHW and JAG developed the ideas for this study, formulated the methods, and designed the model experiments together.
AYHW performed the chemical transport model simulations and data analysis, with input and feedback from JAG. Manuscript 475 preparation was performed by AYHW, reviewed, edited, and approved by JAG.

Acknowledgement
This work was funded by an NSF CAREER grant (ATM-1750328) to project PI J.A. Geddes. We also thank the Global    Table S2.  Table S2.   Table 5. Changes in total nitrogen deposition (∆Ndep) and land area that has nitrogen deposition > 5 kgN ha -1 yr -1 (∆Areacrit), which is a proxy of potential risk of critical nitrogen deposition load exceedance. Only regions with significant ∆Ndep (> 0.25 TgN yr -1 ) or ∆Areacrit (> 10 5 km 2 ) are shown. 775 † The definitions and abbreviations of all regions can be found in Table S2.