Satellite NO2 trends reveal pervasive impacts of wildfire and soil emissions across California landscapes

Nitrogen dioxide (NO2) plays a pivotal role in the production of secondary pollutants, most importantly ozone (O3) and particulate matter. Regulatory controls have greatly reduced NO2 in cities, where most of the surface monitoring occurs, but the change in rural environments is less certain. Here, we present summertime (June–September) spatio-temporal patterns of NO2 concentrations using satellite and ground observations across California from 2009–2020, quantifying the differences in NO2 trends for five distinct land cover classes: urban, forests, croplands, scrublands (shrublands, savannas, and grasslands), and barren (minimally vegetated) lands. Over urban environments, NO2 columns exhibited continued but weakening downward trends (−3.7 ± 0.3%a−1), which agree fairly well with contemporaneous trends estimated from the surface air quality network (−4.5 ± 0.5%a−1). In rural (i.e., non-urban) parts of the state, however, secular trends are insignificant (0.0–0.4 ± 0.4%a−1) or in the case of remote forests are rapidly on the rise (+4.2 ± 1.2%a−1). Sorting the NO2 columns by air temperature and soil moisture reveals relationships that are commensurate with extant parameterizations but do indicate a stronger temperature dependence. We further find that rapidly rising temperatures and, to a lesser extent, decreasing precipitation in response to climate change are acting to increase soil NO x emissions, explaining about one-third of the observed NO2 rise in non-urban regions across California. Finally, we show that these trends, or their absence, can be attributed predominantly to the dramatic rise in wildfire frequency, especially since the turn of the 21st century.


Introduction
Nitrogen oxides (NO x = NO + NO 2 ) serve as important precursors to tropospheric ozone (O 3 ) and fine particulate matter (PM 2.5 ) with consequent adverse effects including premature death [1], cardiovascular mortality [2], respiratory diseases [3], and agricultural productivity losses [4]. The primary sources of NO x involve the thermogenic release during hightemperature combustion in air from vehicles [5] and power plants [6], lightning [7], biomass burning [8], and microbial emissions from soils [9]. A recent modeling study by Silvern et al [10] for the continental US (CONUS) estimates the proportion of total emissions from anthropogenic fossil fuel combustion to be only 42% in 2017 and falling. Nitrogen dioxide (NO 2 ) is a reactive gas with a daytime lifetime of a few hours with respect to its reaction with the hydroxyl radical (OH) [11], and can be observed from space due to its unique absorption spectra [12]. A number of satellites have been deployed to monitor the tropospheric NO 2 vertical column densities (NO 2 VCDs) with several semi-overlapping missions dating back to the 1990s [13][14][15][16][17]. In particular, the Ozone Monitoring Instrument (OMI) [16,18] aboard the National Aeronautics and Space Administration (NASA) Aura satellite provides a daily global record of NO 2 columns since 2004, which has been used extensively to infer the trends and sources of NO x emissions from regional to global scales [10,19].
Since the 1970s, the U.S. Environmental Protection Agency has prioritized policies and technologies to reduce NO x emissions from the combustion of fossil fuels, resulting in an array of benefits for the nation's air quality [20]. In California, a number of techniques, including OMI satellite retrievals, have been used to examine the efficacy of statewide NO x emission controls, although non-uniformly across the state [21]. The California Air Resources Board (CARB) emissions inventory reported a ∼43% reduction in anthropogenic NO x emissions over the period of 2009-2020 (−3.6%a −1 ) [22]. Based on OMI satellite measurements, Lamsal et al [23] estimated comparable reductions from 2005 to 2013 in Southern California (−5.2%a −1 ) and the Central Valley (−4.4%a −1 ). Similarly, Russell et al [24] estimated reductions of OMI NO 2 columns ranging from −4.6 to −5.7%a −1 in California urban regions in the period from 2005-2011. These and other satellite studies across the US [10,[25][26][27][28] have found a marked discontinuity in rates of decline around 2009, reporting summertime trends in remote regions that flatten (∼0%a −1 ) [10] or even rise (+2.0%a −1 ) [24]. Most have proposed reasons for these recent trends to be related to the growing relative importance of 'natural' or uncontrolled NO x sources like soil, lightning, and possibly wildfires. While many have assumed the abrupt change had to do with the economic downturn around 2009, Wang et al [27] use multivariate trend analysis to argue that just by meteorological chance the trends in the combined emissions from soils and lightning changed from −4%a −1 before 2009 to +0.6%a −1 afterwards.
Despite long-term progress, non-attainment of national air quality standards persists throughout California's Central Valley and other inland areas [29][30][31]. De Foy et al [32] [33] found that the Sierra Nevada Mountains occasionally experience O 3 exceedances during the summer months with 5 d in 2013 and 16 d in 2014. As substantial reductions in anthropogenic NO x emissions have been achieved, recent work has demonstrated that the O 3 formation regime has shifted to NO x -limited in these nonattainment inland areas since the 2010s [32], and that O 3 formation has become more sensitive to perturbations in free tropospheric background NO x attributed to biogenic emissions [10,34]. For example, Sha et al [35] estimated that soil emissions account for 40.1% of California's total NO x emissions for the month of July 2018, which significantly increase the surface NO 2 (+176.5%) and O 3 (+23.0%) concentrations in California. This finding is in agreement with other work that has identified soil NO x emissions as an important component of the overall budget [9,27,36]. Moreover, Pan and Faloona [37] reported that O 3 levels are enhanced, on average, by 10% in California's Central Valley during wildfire-influenced periods, which were identified 1 in 5 d during June-September from 2016-2020. Although lightning is much less of an influence in California than in most of the rest of the US [7], previous studies have identified the role of both persistent and fugitive NO x emissions over croplands and associated with wildfire smoke in the degradation of air quality in California's inland rural communities [29,38,39]. However, historical trend analyses of soil and wildfire NO x emissions in California have been largely neglected, especially over the last decade as surface temperatures and biomass burning have increased markedly (figure S1).
Here we use satellite NO 2 columns from the OMI and a surface monitoring network to examine the temporal trends of NO 2 across five land cover types in California during the summer months (June-September) from 2009 to 2020. For non-urban regions where soil and wildfire emissions could be dominant NO x sources, multiple gridded climate data sets are used along with a fire dataset to explore the influence of meteorological conditions and wildfires on the NO 2 levels and to make crude estimates of their growing importance by mid-century. Our study highlights the rise in wildfire and soil NO x emissions in inland California over the past decade and may provide guidance for understanding future air quality in the region.

California land cover classification
We use the Terra and Aqua combined Moderate Resolution Imaging Spectroradiometer (MODIS) Land Cover Type (MCD12Q1) Version 6 data product with a spatial resolution of 500 m to classify the land cover types across California [40]. In our trend analysis, we merge the MODIS land covers into five distinct types: urban, forests, croplands, scrublands, and barren based on the land cover distributions ( figure  S2(a)). The subordinate types within the forests or scrublands categories have similar NO 2 concentrations (difference <0.16 × 10 15 molec cm −2 ), annual NO 2 trends (difference <1.8%a −1 ), and responses to temperature and soil moisture (figure S2, table S1). The distribution of the five simplified land covers is shown in figure 1(a). Overall, California's land cover is primarily made up of scrublands (48.4%), forests (22.8%), and barren lands (15.3%). Agriculture is also an important land use in California, with the Central Valley being one of the most productive agricultural regions in the world, where nitrogenrich fertilizers are applied to the vast cropland areas (9.5%) [36]. Urban areas account for a relatively small part (4.0%), mainly located in the state's coastal regions.

NO 2 column retrievals
The OMI measures backscattered solar radiation from the earth in the ultraviolet and visible wavelength range from 270 to 500 nm [16]. It provides daily global measurements of various trace gases, including NO 2 with a 2600 km wide swath and a spatial resolution of 13 × 24 km 2 for nadir pixels. We use three versions of OMI NO 2 column retrievals: the NASA NO 2 column retrieval [41], the Berkeley High-Resolution (BEHR) retrieval [42], and the Quality Assurance for Essential Climate Variables (QA4ECV) retrieval [43,44]. The area-weighted average algorithm described by Jin et al [45] is used to grid the daily Level-2 swaths to the monthly mean NO 2 columns over California with a spatial resolution of 0.1 • × 0.1 • . To ensure the quality and stability of the satellite data, we select observations with cloud fraction <0.3, solar zenith angle <80 • , surface albedo <0.3, and no 'row anomaly (RA)' data [46]. The pixels at the swath edge (first and last five rows) are also removed for QA4ECV and NASA retrievals.

Surface NO 2 measurements
The surface NO 2 measurements during summer (June-September) from 2009 to 2020 are collected from the CARB. The hourly data are averaged over the afternoon hours (12:00-16:00) to temporally match the OMI measurements. There are 68 CARB NO 2 sites with continuous records between 2009 and 2020 which are selected to compare with the satellite data over five metropolitan air basins designated by CARB (South Coast, San Diego County, San Francisco Bay, SJV, and Sacramento County) as well as other, more remote regions. The distribution of measurement sites is shown in figure 2(a).

Meteorological datasets
To explore the meteorological effects on the NO 2 trends, we use the reanalysis data from the fifthgeneration European Centre for Medium-Range Weather Forecasts (ERA5, ERA5-Land) [47,48], Parameter-elevation Relationships on Independent Slopes Model (PRISM) climate group [49], and the NCEP North American Regional Reanalysis (NARR) [50]. The variables considered in this study include 2 m air temperature, soil moisture, precipitation, planetary boundary layer height, horizontal wind speed, downward shortwave radiation, and cloud cover. The meteorological datasets used in this study are summarized in table S2.

Wildfire records
The California statewide database of fire history is obtained from the Fire and Resource Assessment Program (FRAP), which compiles fire perimeters from CAL FIRE, the United States Forest Service Region 5, the Bureau of Land Management, and the National Park Service. This data includes the California fire events that occurred since 1950 along with fire alarm dates, containment dates, area burned, and the causes of ignition. This work focused on the wildfire events that happened during the summer months (June-September) from 2009 to 2020, covering a total burned area of ∼7.8 million acres across 3117 separate fires.

Temperature/soil moisture dependence
For non-urban regions where the soils have been identified as a major source of NO x emissions [9], the parameterization of Berkeley-Dalhousie soil NO x emissions (BDSNP) [51], which is commonly used in chemistry transport models, is adopted to capture the relationship between meteorological variables and NO 2 columns. In the BDSNP scheme, soil NO x emissions increase exponentially with temperature in the temperature-sensitive regime (0 • C-30 • C) and remain constant when the temperature is above 30 • C for wet soils [51,52]. Similarly, an exponential function is fitted to the relationship between NO 2 columns and temperature in this study: where [NO 2 ] represents the NO 2 columns, T is the temperature (in degrees Celsius, only above freezing), T 0 is the upper limit of the temperature-sensitive regime. Unlike the BDSNP scheme which fixes T 0 to 30 • C for all wet soils, we customize different T 0 values across land types based on their observed temperature dependencies (discussed in section 3.3). First, the exponential function in equation (1) is fitted to the relationship between NO 2 VCDs and air temperature in all non-urban regions (the sum of forests, croplands, scrublands, and barren) (figure S9). Optimal values for the three parameters (a, b, and c) are calculated based on the fit. Parameter b is then assumed constant and applied to the fittings over all land types, with the optimal parameters a and c acquired for each land type. This procedure corresponds to what was done in the original study by Yienger and Levy [53], but allows us to determine these parameters independently.
The non-linear relationship between soil NO x emissions and soil moisture is typically described by a Poisson function, with the lowest emissions at both extremely dry and wet conditions [51,54]. A Poisson distribution is fit for the relationship between OMI NO 2 VCD and volumetric soil moisture (VSM) as shown in equation (2), with the optimized parameters (a, b, and c) selected based on the same criteria as the temperature fittings. Where [NO 2 ] represents the NO 2 VCD and θ is the VSM.
Although T and θ are inversely correlated across the landscape ( figure S6(b)), largely due to altitude, we make the fits to equations (1) and (2) independently to allow all six parameters to be solved for. Figure 2 and tables S1, S3, and S4 show the trends of summertime NO 2 concentrations during 2009-2020 in California from the Quality Assurance for Essential Climate Variables (QA4ECV) OMI retrievals [43] and ground measurements. We present here the summertime (June through September) trends because they are most relevant to persistent O 3 issues [29][30][31][32] and were found to have more pronounced flat or increasing trends than other seasons, and we begin our analysis in 2009 to avoid the discontinuity in trends found before that time [26]. Comparisons with trends detected using the NASA standard product, BEHR [42], and TROPOMI [55] retrievals (table S3 and figure  S3) show excellent agreement. Continued, significant reductions in the NO 2 columns are observed in California's urban areas from 2009-2020 at a rate of −3.7 ± 0.3%a −1 , which is slightly smaller than the range (−3.9 to −5.7%a −1 ) of diminution reported in previous studies (table S4), but this difference is likely attributable to slower rates of change in the years subsequent to 2009 [26]. We further compared the OMI NO 2 column trends to surface NO 2 concentrations derived from the CARB air quality network. The afternoon (12:00-16:00) surface NO 2 concentrations are closely correlated (R 2 = 0.70, N = 816) with the NO 2 columns across the state ( figure 2(a)). At the air basin scale over the South Coast, both the satellite (−4.5 ± 0.5%a −1 ) and ground measurements (−4.6 ± 0.4%a −1 ) demonstrate a coherent decline in NO 2 (p < 0.05) ( figure S4). However, significantly slower rates of decline were found by satellite in the other urbanized air basins across California relative to the surface measurements ( figure S4). This is consistent with the findings of Lamsal et al [23] in relatively lower NO x environments, and the interested reader is referred to that work for a thorough discussion of possible reasons for discrepancies between surface and satellite trends.

Spatially heterogeneous NO 2 trends
By contrast, NO 2 columns outside of California's urban centers show either non-significant (unchanging) or positive trends from 2009 to 2020 (figure 2). In particular, forested areas, which have been much more frequently affected by wildfires over the past two decades [56], exhibit a strongly increasing trend (+4.2 ± 1.2%a −1 ) of NO 2 columns. Much more muted trends are seen in scrublands (+0.4 ± 0.4%a −1 ), and statistically non-significant changes are found in barren (+0.2 ± 0.4%a −1 ) and cropland regions (0.0 ± 0.4%a −1 ).

Anthropogenic effects
Given the short NO x lifetime during summer [57], we infer that regulations targeting fossil fuel combustion within and near urban areas would have a more limited impact on NO 2 levels in more remote regions during the summer months. To examine this hypothesis, we compare the weekly cycle of NO 2 columns between different land types ( figure S5). This analysis reveals that NO 2 columns on weekends (Sundays) is on average 42% (1.8 × 10 15 molec cm −2 ) lower than on weekdays (Tuesdays-Fridays) in urban areas because of the well-documented pattern of human activities [32,58], consistent with the hypothesis. Non-urban areas do not present significant declines on weekends, with only small, detectable decreases found in croplands (−0.3 × 10 15 molec cm −2 ) and scrublands (−0.2 × 10 15 molec cm −2 ). It is important to note that the observed NO 2 cycles are not influenced by the wildfires and soil emissions because the weekend-weekday variations in the burned area and temperature are less than 0.2% (table S5). Since the weekend-weekday effects on the trends of NO 2 columns are minimal in the forests and barren lands, which tend to be the farthest from urban centers (figure 1), we assume that fossil fuel sources affect these two land types negligibly.

Soil emissions effects
Soil emissions, modulated by soil microbes, weather conditions, and reactive nitrogen (N) amounts, have been identified as a significant source of NO x in California, particularly in warm agricultural regions (croplands) with high fertilizer applications [9,36,59]. Figure S6 shows temporal correlations between NO 2 columns and a set of potential drivers over nonurban areas in California. The temperature (r = 0.81) and soil moisture (r = −0.70), which are two major factors driving soil emissions, are highly correlated with the NO 2 columns, though the correlations vary regionally (figure S7). For example, Northern California, dominated by forests and scrublands, has the strongest positive correlations with temperature (r > 0.8) and is inversely correlated with soil moisture (r < −0.6), while opposite relationships are found in much of the southeastern barren areas. This deviation highlights the variability of NO 2 dependencies upon soil conditions, and a fine-scale analysis is critical for accurately quantifying the impacts of soil emission on long-term NO 2 trends in California. Following the BDSNP [51], we apply a truncated exponential function (equation (1)) and a Poisson function (equation (2)) to depict the non-linear responses of NO 2 columns to temperature and VSM from the ERA5-Land reanalysis dataset (figures 3 and S8). High-fire months in the top 75th percentile of burned area (figure S9) and the highest 5% of NO 2 columns in each non-urban land type are excluded to eliminate the wildfire effects and remove any elevated NO 2 concentrations transported from urban centers [25] on rare occasions. The exponential function performed well in capturing the temperature dependence of NO 2 columns (figure 3, R 2 > 0.95), consistent with responses of soil NO x emissions reported in previous studies [51][52][53]. We find that in croplands and scrublands our results agree with the BDSNP parameterization assuming a soil NO x emission that increases exponentially followed by a plateau at 30 • C; however, in forests and barren lands we observe NO 2 columns reach their maxima at air temperatures closer to ∼33 • C. This is possibly due to the difference in air vs. soil temperatures, but that effect would most likely work in differing directions during midday in summer over forests (cooler soils) than over nonvegetated lands (warmer soils). Another possibility for the difference could be microbial adaptations to high temperatures or the importance of deep, cooler soil layers to the total NO x emissions [9].
All land types tend to yield a very similar background concentration (parameter c in equation (1)) of ∼6.6 ± 0.6 × 10 14 molec cm −2 , which is comparable to the values consistently observed well offshore by TROPOMI ( figure 1(b)) [58] and OMI retrievals (∼5.5 ± 0.6 × 10 14 molec cm −2 ), which moreover exhibit no significant trend during 2009-2020 (figure S10) [60]. Further, the pre-exponential factors (parameter a in equation (1)) in figure 3(a) are related in magnitude in a manner consistent with the 'wet' emission factors outlined in the original Yienger and Levy [53] parameterization with croplands being the largest (due to the higher available reactive N), scrublands (including open and closed shrublands, savannahs, and grasslands) are similar but smaller, and forests (including deciduous, evergreen, and mixed) an approximate order of magnitude smaller still. On the other hand, the larger temperature coefficients (parameter b in equation (1)) in our exponential fits indicate a much stronger temperature dependence than those of Yienger and Levy [53] and the other common parameterizations [35,51] across all landscapes. After removing the background concentrations, the ratio of soil emissions for our data between 30 • C and 20 • C is a factor of 6.9 larger, whereas the traditional parameterizations predict an increase by a factor of 2.8. A recent study focusing on high-temperature agricultural soils over the US [27] proposed a stronger soil temperature dependence between 20 • C -30 • C (a factor of 3.9 greater) and a higher threshold temperature at which emissions plateau (40 • C) relative to prevailing parameterizations. Both of these characteristics are qualitatively supported by our results across all landscapes in California illustrated in figure 3(a).
By controlling the ratio of oxygen to water in the soil pore space, soil moisture is another important factor regulating soil NO x emissions [61]. Distinctly different VSM dependences of NO 2 columns are observed for the four non-urban land types ( figure 3(b)), although they have similar patterns to the typical dependence of soil emission parameterizations [51,61]. That is, the NO 2 columns are low in both extremely dry and wet conditions with the highest NO 2 columns observed at VSM values between 0.03-0.14, which can be accurately represented by the fit Poisson function (R 2 > 0.73). Many soil NO x studies [35,51,62] use water-filled pore space (WFPS) to characterize soil moisture, which is equal to VSM divided by the soil porosity. Most soils in California have a porosity in the range of 0.4-0.6 [63] so our VSM peaks correspond to WFPS values in the approximate range of 0.05-0.35. A recent study by Huber et al [64] updated the BDSNP parameterization to fix the soil NO x emission peaks at the observed median VSM for spring/summer. Figure 3(b) shows the June-September median values of VSM by the dotted lines indicating that they are fairly similar to our observed peaks (difference <0.03) for each land type in California.
Using the empirically fit relationships shown in figure 3, we calculate the trends of NO 2 columns that are driven by temperature and VSM changes to estimate how much of the observed trends can be ascribed to soil NO x emissions ( figure 4(a)). We use the sum of temperature and VSM-driven NO 2 column changes to represent the soil emission effects. The results reveal that soil emissions can explain about one-third of the OMI NO 2 trends over all non-urban areas but this ranges from nearly all in scrublands to less than 10% of the observed change in forested regions. This important environmental soil NO x enhancement is dominated by the temperaturedriven response, rather than the VSM-driven response, mostly because temperature has exhibited a more significant trend (∼+1.2 • C decade −1 ) than soil moisture (∼−1% decade −1 ) in the reanalysis data over the study period ( figure S7). In particular, scrublands show the largest increase (0.48 × 10 13 molec cm −2 a −1 ), followed by forests (0.28 × 10 13 molec cm −2 a −1 ), and then croplands (0.17 × 10 13 molec cm −2 a −1 ), with barren areas indicating no net change from changing soil conditions because of the opposing temperature and VSM effects. As a reminder, in our fits we have only considered periods that are not substantially impacted by wildfire or urban emissions. Thus, the difference between the summed results and the observed trends likely indicates the changing influence of wildfires (discussed later) and urban emissions. It is worth noting that although the temperature dependence of NO 2 columns is strongest in croplands in the temperature range between 20 • C -30 • C, the predicted soil emission change is small because of the assumed plateau above ∼30 • C. This is because only 6.6% of the summertime temperatures over croplands fall below the threshold in the temperature-sensitive regime during OMI overpass time (∼13:30 LT) (figure S11(a)), thereby muting the climate warming effects. Nearly half of all scrublands, on the other hand, appear to be at temperatures below its plateau temperature making their overall emissions much more susceptible to rising temperatures ( figure S11(b)). Because of their higher average temperature plateau threshold and elevations, forest soils are mostly all susceptible to increasing temperatures even though their absolute emissions are smaller than those in croplands and scrublands.

Wildfire effects
A recent study of the background NO 2 in the US revealed that soil emissions, lightning, and meteorological changes simulated by the GEOS-Chem model could not fully explain the observed decadal rise in remote NO 2 during summer [28] and further suggested that deficiencies in the model's treatment of wildfire NO x could be the culprit. Another study by Wang et al [27] suggested that the abrupt deceleration in NO x reductions observed in 2009 was due to a coincident change in the soil and lightning emissions at that time, conceding that wildfire emissions while rising rapidly (∼6.5%a −1 ) were still too small of a source to influence total NO x trends accounting for only 1.9% of the US inventory in the period from 2005-2019. Although this work and a number of other studies have demonstrated an underestimation of soil NO x emissions [9,27,35,36] in current models, we further infer that the dramatically increasing biomass burning activity over the past two decades [56], in California in particular, is a critical factor contributing to a dramatic shift in background NO 2 trends. Figure 4(b) shows the differences in trends of NO 2 columns after removing the upper 75th percentile wildfire-impacted months as measured by statewide area burned ( figure S9). This upper quartile threshold was selected because there were months in this range distributed across most of the study period (figure S9). Statistically significant decreased rates are observed in all urban (−2.1 × 10 13 molec cm −2 a −1 reducing the trend from −3.7 to −4.2%a −1 ) and non-urban land cover types ( figure 4(b)), with the most pronounced declines found in forests (−1.8 × 10 13 molec cm −2 a −1 ), followed by croplands (−1.6 × 10 13 molec cm −2 a −1 ) and scrublands (−1.4 × 10 13 molec cm −2 a −1 ), and a relatively minor impact in barren regions (−0.7 × 10 13 molec cm −2 a −1 ). Our results further reveal that the offshore, background NO 2 column trends are not affected by wildfires (figures S10(b) and (c)). Considering that the warmer and drier conditions during high-fire months may also enhance the soil NO x emissions, we further compare the predicted temperature and VSM-driven NO 2 changes with the OMI NO 2 columns to isolate the influences from wildfires and soil emissions ( figure 4(c)).
Our results reveal that these two factors have comparable influences on the NO 2 columns across the entirety of non-urban regions, although their contributions varied significantly in different land covers. For example, the NO 2 columns in croplands and scrublands show similar enhancements, with wildfires being the major contributors (84%) in croplands, while soil emissions have a greater impact on scrublands (63%). Recall from figure S5 that the fossil fuel effects are minimal in the remote forest and barren regions, and so we estimate the effects of wildfires on the NO 2 trends in these two surface types by subtracting the influence of temperature and soil moisture from the OMI NO 2 columns. The difference is not an exact quantification however because we are only able to eliminate some arbitrary top percentile (in this case the 75th) of wildfire impacts. The results nevertheless show that the substantial rise in summertime NO 2 concentrations in forest habitats is mostly driven by increasing wildfire emissions, accounting for ∼90% of the OMI NO 2 trend ( figure 4(a)). One can also discern this effect in the observations of greater mean VCDs across the Northern California mountains and the Central Sierra Nevada in the summertime relative to the wintertime despite the much shorter photochemical lifetime (figure S12). This hypothesis is further supported by the fact that the NO 2 columns in forest regions show a much weaker trend (0.4 ± 0.8 × 10 13 molec cm −2 a −1 ) during the winter months (December-March) when the NO 2 columns are not significantly affected by the wildfires (figure S13). For barren lands, wildfire is also the major driver of their moderately increasing trends. Figure S14 graphically demonstrates the increase in California burned areas which can be represented by an exponential function since the 1970s (R 2 = 0.32) [65] with a doubling time scale of about a decade. Given the linear responses of NO 2 columns to burned areas (R 2 = 0.14, figure S15) [66], and the NO 2 trends driven by temperature and VSM as shown in figure 4(a), we crudely predict the approximate NO 2 changes in 2050 for the remote forests and barren regions by simply extrapolating these wildfire and soil emission effects (table 1). Our results indicate that the NO 2 columns in the forests could be more than four times larger than current levels in 2050 if the fire activities continue to increase in the coming decades as expected in California [65]. However, there are many feedbacks that remain highly uncertain such as the accumulation of N in soils as wildfire emissions continue to rise leading to soil microbial changes and lag effects that may further influence future soil emissions. In any event, significant rises in background NO 2 will likely resist the continuation of gains from fossil fuel NO x controls and represent a growing challenge for future air quality management across the state and potentially across the entire western US.

Conclusions
Using long-term (2009-2020) observations from the OMI satellite measurements, this study depicts spatial patterns in summertime NO 2 trends across California. Our results reveal a significant NO 2 improvement within cities, while the non-urban regions show either no detectable changes or steadily increasing NO 2 columns (in forested regions). We then quantitatively evaluate the impacts of the major NO x sources, focusing on three main sources: fossil fuel emissions, soil emissions, and wildfire emissions. In general, the rising soil and wildfire emissions combined are offsetting the anthropogenic NO x decline over non-urban portions of the state, and the wildfires play a more important role in determining the trends of NO 2 columns than do the soil emissions, although the relative contributions of these two sources vary a lot in distinct land types. With continued progress in regulating fossil fuel NO x emissions, these less-understood NO x sources will become increasingly important to air quality control strategies in California. Our results point to opportunities for different sets of policies and technologies to assist in reducing NO 2 concentrations in rural and economically disadvantaged areas of California, but will require a concerted effort to better understand the exact environmental dependence of soil and wildfire emissions.

Acknowledgments
I C Faloona's effort was supported by the USDA National Institute of Food and Agriculture, [Hatch Project CA-D-LAW-2481-H, 'Understanding Background Atmospheric Composition, Regional Emissions, and Transport Patterns Across California']. The authors are also grateful to F Boersma for helping us with quality assurance of the different satellite data sets, and we would like to thank D D Parrish, C D Cappa, and M Kleeman for a series of fruitful discussions regarding this work. We thank A T O'Geen for help interpreting the soil properties found in his California soil resource lab's SoilWeb database. The inputs of two anonymous reviewers also helped to improve the manuscript.