The hydroxyl radical (OH) is the main tropospheric oxidant and governs how quickly reduced species degrade in the atmosphere. Of recent interest is the impact of OH concentrations ([OH]) on methane oxidation, which determines the methane lifetime and thus its global warming potential . Chemical drivers of [OH] include VOC and nitrogen oxide (NOx=NO+NO2) emissions, with the former generally decreasing [OH] and the latter increasing or decreasing [OH] depending on the local chemical regime. Improved estimates of VOC and NOx emissions, as well as more detailed modeling of VOC-NOx chemical interactions, are crucial in constraining [OH], especially over remote tropical regions with few in-situ observations.

The most significant non-methane VOC by total emission flux is isoprene, with a total global flux of 440–660 Tg C yr⁻¹ . Isoprene is released by select species of trees and shrubs – particularly deciduous broadleaf trees – in response to light and heat (; ; ). Once in the atmosphere, isoprene is oxidized by OH on a timescale that depends on local [OH] (e.g., from 1–7 h across the OH levels of 0.4–1.6×106 molec. cm-3 detected during goAMAZON) . This oxidation can form other VOCs, organonitrates, and secondary organic aerosols via isoprene epoxydiol (IEPOX) formation, with the identity of these oxidation products depending on the local chemical regime .

Isoprene concentrations are controlled by biogenic emissions from plants (its source) and local [OH] (its primary sink), with higher [OH] increasing isoprene oxidation and thus decreasing isoprene concentrations. Isoprene emissions are often calculated via emission models, such as the Model of Emissions of Gases and Aerosols from Nature (MEGAN) which parametrizes isoprene emissions as a function of light, temperature, leaf area index, leaf age, soil moisture, and carbon dioxide . In turn, the amount of OH in a region depends on factors including specific humidity, actinic flux, and VOC and NOx concentrations . Thus, changes in isoprene emissions through temperature or light, and changes in OH through water, light, and VOC/NOx chemistry, will impact isoprene columns and their variability.

Significant amounts of isoprene are emitted in the remote tropics, frequently into a low-NOx atmosphere due to their sparse anthropogenic NOx sources. However, these low-NOx regimes can include substantial non-anthropogenic NOx sources, such as lightning, soil microbial activity, and biomass burning. For example, chemical interactions between lightning NOx and isoprene in convective plumes have recently been shown to be a source of new particle formation in the upper troposphere . Despite their importance in determining [OH] in remote regions, there remains large uncertainty and model-observation disagreement in tropical NOx emissions. For example, using the Yienger and Levy soil NOx parametrization on the Tapajos National Forest in the Amazon resulted in a 10× to 20× underestimation of soil NOx compared to observations . Validating modeled non-anthropogenic NOx fluxes and assessing their impacts on regional chemistry warrants continued investigation, especially as anthropogenic NOx emissions in the Northern Hemisphere decrease and non-anthropogenic (e.g. soil) NOx fluxes become more important .

Using isoprene retrievals from the Cross-track Infrared Sounder (CrIS) instrument on the Suomi-NPP satellite, processed analogously to the IASI ammonia retrievals described in , we can directly monitor global isoprene columns with daily to monthly temporal resolution, even over remote regions with few or no in-situ observations . Previous work with these CrIS retrievals has used them to identify the impacts of biomass burning on OH in New Guinea ; evaluate the impact of interannual variability of isoprene emissions on the atmosphere's oxidative capacity ; and perform or evaluate inversions on isoprene and NOx emissions . However, few studies have used these retrievals to analyze global isoprene variability and compare isoprene emissions and chemistry across different tropical regimes. Of special interest when investigating isoprene variability is the relationship between the El Niño-Southern Oscillation (ENSO) and isoprene emissions and columns. Previous studies have shown increased global isoprene emissions during El Niño and decreased emissions during La Nina largely due to changes in temperature and radiation, with the potential for strong El Niño to increase isoprene emissions for years after the event .

Here we aim to address the question: “what controls the variability of tropical isoprene?” In this study, we use CrIS isoprene retrievals to compare isoprene column variability across three tropical source regions: Amazonia, equatorial Africa, and the Maritime Continent due to their outsized influence in the global isoprene budget. We identify whether isoprene variability in these regions are largely controlled by changes in isoprene emissions (“emissions-controlled”) or changes in [OH] and NOx (“chemistry-controlled”).

CrIS is an infrared Fourier-transform spectrometer with spectral resolution of 0.625 cm-1 in the longwave band (650–1095 cm-1) aboard the Suomi NPP satellite in sun-synchronous orbit with an equator overpass time of approximately 01:30 PM local time. This spectral range encompasses the bands where isoprene absorption is the strongest (ν28 and ν27) with minimal interference from other species . At nadir, CrIS has a footprint 14 km in diameter. The footprints are cloud-masked based on the difference between MERRA-2 surface temperatures and the 900 cm-1 brightness temperature. The choice of the cloud mask threshold provides less than 5 % uncertainty to the overall retrieval . In general, 20 %–60 % of the gridpoints within the tropical regions of interest contain non-cloudy scenes each day (Fig. S5 in the Supplement).

Similar to the IASI ammonia retrievals described in , the CrIS isoprene retrieval first calculates a hyperspectral range index (HRI) between 890–910 cm-1 for each CrIS footprint. The HRI, background spectrum, and covariance matrices are solved iteratively. The CrIS HRIs are then gridded and fed into a neural network trained on synthetic radiances simulated by the Earth Limb and Nadir Operational Retrieval (ELANOR) radiative transfer model, with the ELANOR inputs being temperature and water vapor profiles from GMAO, and GEOS-Chem isoprene profiles with Gaussian noise. To quantify isoprene columns from CrIS observations, the neural network uses thermal contrast, water vapor columns, surface pressures, and the viewing angle as inputs in addition to the HRI, and outputs daily isoprene columns from -60 to 60° latitude at 0.5°×0.625° spatial resolution. The final retrievals show general agreement with ground-based isoprene measurements, as demonstrated with Fourier-transform infrared measurements taken from Porto Velho in Brazil (r=0.47 using daily CrIS retrievals) . We remove October and November 2019 from our analysis due to potentially anomalous striping, as described in .

More information on the isoprene retrievals and their evaluation can be found in , , and . Although uses optimal estimation to retrieve isoprene columns, as opposed to the artificial neural network used in and , both methods show strong agreement and similar spatial distributions over Amazonia (r>0.9) .

Amazonia, the Maritime Continent, and equatorial Africa are regions of special interest, as these regions account for 50 % of the global isoprene column burden from the CrIS record (2012–2020). The tropics more generally account for 80 % of all isoprene emissions, making them the most important isoprene source regions in driving global variations in isoprene . Quantifying isoprene emissions has traditionally been done using satellite formaldehyde inversions; these inversions have shown that tropical isoprene emissions in Amazonia and Africa typically were overestimated by MEGAN . In this paper, we use direct isoprene retrievals from CrIS radiances to assess the drivers of isoprene column variability.

Based on the CrIS retrievals, the three outlined regions in Fig. a, which encompass approximately 48 % of the land in the tropics (-20 to 20° latitude) and approximately 15 % of the global land area, contain half of the total isoprene columns observed globally during this 8-year period (Fig. b). As a result, changes in isoprene columns within these three areas can have an outsized impact on global isoprene variability.

Figure c and d shows the isoprene column anomalies, calculated relative to 2012–2020 mean isoprene, globally and over these three tropical regions. In 2020, both Amazonia and Maritime Continent had broadly positive isoprene anomalies for most of the year until October 2020, when Amazonia dropped below its climatology while isoprene over the Maritime Continent stayed elevated. These responses resulted in some of the highest global isoprene anomalies over the 8-year period .

However, the largest regional isoprene anomalies do not occur in 2019–2020 but during the El Niño in 2015. This El Niño was characterized by higher isoprene over Amazonia – the largest regional isoprene anomaly in the 8 year record – and lower isoprene over the Maritime Continent. We observe the opposite response in the subsequent transition to La Niña conditions: the Maritime Continent had higher isoprene columns, while Amazonia had lower isoprene columns relative to their respective climatologies. Even outside of this 2015–2016 ENSO transition, Amazonian isoprene anomalies increase with El Niño and decrease with La Niña, while the inverse is true for the Maritime Continent (see Fig. a and g). We observe a weak positive relationship between ENSO and Amazonian isoprene anomalies (r=0.28; p<0.05) and a stronger negative relationship between ENSO and isoprene anomalies over the Maritime Continent (r=-0.52; p<0.05). In contrast to these two regions, isoprene anomalies over equatorial Africa do not correlate with ENSO (r=-0.04; p>0.05).

Surface air temperature can explain 20 % of the isoprene column variability in Amazonia, followed by direct photosynthetically active radiation (PAR) (Fig. b and c). Given that isoprene emissions increase with both temperature and photosynthetic photon flux density, this relationship suggests that isoprene emissions are the strongest driver of Amazonian isoprene column variability when spatially averaged . This temperature dependence drives a positive correlation with ENSO, which is consistent with previous modeling studies that show an increase in isoprene emissions during El Niño over Amazonia, as well as globally .

The other two tropical regions do not exhibit the same correlation with temperature, although the highest temperature anomalies over equatorial Africa, namely 2015–2017, correspond with high isoprene column anomalies. Outside of equatorial Africa in 2015–2017 and 2019, isoprene column anomalies over equatorial Africa and the Maritime Continent do not correlate with temperature, which is likely due to the smaller dynamic range in temperature over these regions. For instance, the spatially-averaged temperature anomalies (10th–90th percentiles) over the Maritime Continent never exceed 1 K over the eight-year period, compared to Amazonia's >2 K temperature anomalies in 2015.

We do note that certain isoprene emission drivers, namely leaf area index, may temporally lag environmental variables, and a Granger causality test shows a statistical significant relationship between isoprene and direct PAR in Africa (lag =3 months; p=0.01) and isoprene and temperature in the Maritime Continent (lag =3 months; p=0.03). However, even with time lags, the relative magnitude of direct PAR and temperature variability in 2014–2015 is not sufficient to explain the magnitude of the isoprene variability during that period. For example, the largest negative isoprene anomaly over the Maritime Continent occurred in late 2015. By lagging temperature by three months, this nadir in isoprene coincided with a weak negative temperature anomaly. However, a lagged negative temperature anomaly of comparable magnitude in late 2014 coincided with a significantly smaller isoprene anomaly, indicating inconsistent magnitudes between the potential drivers and the observed isoprene anomalies. Due to the lower dynamic range in temperature, factors other than isoprene emissions may control most of the isoprene column variability in the Maritime Continent and equatorial Africa.

The Maritime Continent shows a statistically significant inverse relationship between isoprene column anomalies and ENSO (r=-0.52, p<0.05), while Amazonia has a positive correlation (r=0.28, p<0.05). If driven solely by emissions, the observed relationship between isoprene columns and ENSO over the Maritime Continent contrasts with previous model results, which show higher isoprene emissions during El Niño . Furthermore, three-day resampled isoprene column anomalies positively correlate with the second principal component of the Outgoing Longwave Radiation Madden-Julian Oscillation Index (Fig. a). For positive values of the second principal component (i.e. Phases 3–6), the MJO’s convection is highest above the Maritime Continent, as opposed to Africa or the Western Hemisphere . Thus, CrIS isoprene columns increase while the convective phase of the MJO moves over the Maritime Continent.

There is a weak positive correlation between the regional isoprene anomalies and temperature, suggesting that isoprene emissions do have some impact on the region's isoprene column anomalies. However, the seasonal to annual variation governed by ENSO and the subseasonal variation governed by the MJO are likely due to the strong positive correlation between isoprene column anomalies and precipitation, with higher precipitation coinciding with higher isoprene anomalies. Variability in precipitation then translates to changes in MERRA-2 soil moisture (Fig. b, c). We now ask the question, “why do we observe a positive relationship between isoprene anomalies and precipitation over the Maritime Continent?”

Here we investigate the positive correlation between isoprene columns and precipitation in the Maritime Continent. Although precipitation may increase isoprene emissions via indirect changes in leaf area index (LAI), isoprene columns and MODIS LAI do not correlate over the Maritime Continent (Fig. S8). Thus, if driven by isoprene emissions, this relationship has not been observed in previous literature and runs in opposition to our expectation and that of some earth system models . In general, in-situ isoprene emission flux measurements do not show a direct relationship between precipitation/moisture and isoprene emissions outside of drought conditions . MEGAN does include soil moisture in its parametrization, but its impact on isoprene emissions is only relevant when the soil moisture drops below the wilting point (0.01–0.138 m3m-3 depending on soil type), which is significantly lower than the soil moistures observed in the Maritime Continent (Fig. a). Terpene emissions have been observed to increase with rainfall, but the same has not been shown with isoprene , and the indirect impact of terpenes on isoprene via [OH] depletion is likely small due to their significantly lower emission fluxes. We highlight the need for additional observations of isoprene fluxes in the Maritime Continent to determine whether isoprene emissions may increase with rainfall, but given the lack of a correlation between LAI and isoprene columns that would support an increase in isoprene emissions with rainfall, we focus our attention to other potential hypotheses.

In addition to isoprene emissions, variability in local [OH] (e.g. from NOx sources) can affect isoprene columns. found that biomass burning NOx modulated OH and isoprene columns over New Guinea. However, the impact of biomass burning is episodic: in New Guinea, it was largely relegated to January–May 2016, thus not covering the entire ENSO period . Therefore, due to its episodic nature, biomass burning is unlikely to explain all of the isoprene variability observed over the Maritime Continent, although it is still an important driver of isoprene columns over the region.

To explain the continuous positive correlation between isoprene and precipitation over the entire 8-year record, we focus on the following three hypotheses:

Soil NOx sources in oil palm plantations vary with precipitation

We detail each potential hypothesis for this relationship and ultimately suggest that soil NOx, biomass burning NOx, and/or a combination of lightning and convection are the most likely causes for this unexpected relationship.

Unlike the other two regions, isoprene columns in equatorial Africa do not strongly correlate with ENSO. However, isoprene anomalies correlate weakly with surface air temperature. This relationship with temperature is strongest between 2015–2017 and in 2019, where peaks in temperature coincide with peaks in isoprene anomalies (Fig. b). Temperature and its impact on isoprene emissions thus only explain part of the column variability, and only during anomalously hot periods (>0.5 K above climatology). For the rest of the 8-year period, which exhibits cooler temperatures and smaller temperature variability than 2015–2017 and 2019, isoprene over equatorial Africa shows a stronger negative correlation with biomass burning, which is quantified here with GFED4 total burned dry matter (Fig. a and c). This anticorrelation exists both for isoprene anomalies, as well as for the seasonal cycle in isoprene columns (Fig. S10).

Total isoprene columns negatively correlate with GFED4 burned dry matter, a correlation not observed in the other two tropical regions (Fig. S12). In Amazonia and the Maritime Continent, both soil and biomass burning NOx have a seasonal peak in the second half of the year that coincides with a peak in isoprene emissions, columns, and temperature. On the other hand, isoprene columns and emissions are consistently out-of-phase with both NOx sources in equatorial Africa, but particularly with biomass burning (Figs. S7 and S8).

Although there is some spatial heterogeneity between NOx and isoprene sources, seasonally-averaged 850 hPa winds are in the correct orientation to carry air masses from regions with heavy biomass burning toward isoprene source regions throughout the year, but especially during the dry season south of the equator (June–September) (Fig. 7). Based on the GFED4 fire emission inventory , Equatorial Africa has the highest biomass burning emission fluxes out of all three tropical regions, and the region of interest is smaller in total land area compared to the Amazonia bounding box. In fact, the seasonal peak in biomass burning dry matter fluxes in equatorial Africa are 3× as high as the fluxes observed in Amazonia and the Maritime Continent. Thus, the prevalence of both biomass burning regions and forested isoprene source regions within a smaller area may lead to increased isoprene-NOx co-location compared to Amazonia.

We hypothesize that as a result of this colocation, isoprene column variability in equatorial Africa is driven by biomass burning-derived NOx emissions reacting to produce tropospheric ozone and OH downwind of the fire, which then modulates isoprene oxidation and loss. Biomass burning in sub-Saharan Africa emits 1.9±0.6 Tg NO during the dry season (June–October), which can contribute 40 %–60 % of the total NOx budget in the region . These emissions of NOx can undergo chemistry to produce OH . It is important to note that NOx emissions are highest in flaming fires versus smoldering fires, and thus the intensity of the fire and its combustion material can impact plume emissions and chemistry . In general, southern African woody savannah fires are more flaming in the early season (May–July) and become more smoldering with time, which may be due to changes in fuel type or precipitation . May–July is when GFED4-derived biomass burning emissions are highest in equatorial Africa and the surrounding savannahs, which generally coincides seasonally with flaming fires with higher NOx emission factors compared to fires later in the year.

The impact of biomass burning on [OH] concentrations may depend on other factors beyond direct NOx emissions. Biomass burning produces smoke aerosols, which can change both diffuse and direct PAR and thus isoprene emission fluxes. These aerosols can also decrease photolysis rates (e.g. O3 + hν), which subsequently impacts [OH]. Fires are also a source of carbon monoxide and other VOCs (e.g. formaldehyde and furans), which may decrease local OH concentrations through their oxidation . Even though high VOC emission fluxes could decrease local [OH] in remote tropical regions, VOC oxidation in biomass burning plumes occurs in the presence of elevated NOx concentrations, which increases tropospheric and boundary layer ozone downwind of the plume and thus [OH]. In fact, biomass burning alone can contribute to a quarter of the boundary layer ozone in Africa, and some of this ozone can be transported globally, especially toward southeast Asia and South America . Enhanced O3 can lead to a net increase in local OH through higher HOx production.

We note that the lifetime of NOx is also shorter than CO’s lifetime, which results in larger decreases in NOx as the plume advects away from the fire and may result in different chemistry as the plume ages. Nevertheless, elevated NOx has still been observed in areas hundreds of kilometers downwind of biomass burning, and biomass burning in equatorial Africa also produces peroxyacetyl nitrates in the lower and mid-troposphere that can transport NOx species aloft over long distances . The advection of NOx and PANs over long distances, as well as downwind O3 production and transport, are potential ways in which biomass burning in African savannahs can affect isoprene chemistry in nearby tropical African forests.

Importantly, although high NOx emissions in NOx-saturated regimes (e.g. in some flaming wildfire plumes) can decrease [OH] through increased HNO3 and RONO2 formation, these increases in NOx emissions generally co-occur with increases in HOx production from species like HONO . HONO in particular is an important source of OH in early-stage (<3 h) plumes . Moreover, enhanced NO3 and O3 downwind of wildfires can provide alternative pathways for isoprene oxidation, which can become important in thick plumes with reduced photolysis or at nighttime . This positive correlation between biomass burning NOx and isoprene loss via OH or other oxidants is consistent with our GEOS-Chem simulations, which is further described in Sect. 6.

Thus, isoprene column variability over equatorial Africa is likely driven by sink (OH) variability outside of anomalously hot periods (displayed in Fig.  as 2015–2017). This relationship between biomass burning and isoprene would depend on plume chemistry, the fire's fuel type and characteristics, and the NOx lifetime within a plume. Unlike Amazonia, where isoprene column variability is emissions-driven due to the region's large dynamic range in temperature, over equatorial Africa the dynamic range in oxidant chemistry due to biomass burning NOx emissions generally exceeds the dynamic range in temperature outside of 2015–2017 and 2019. Therefore, equatorial Africa represents a region where isoprene column variability can be either emissions- or chemistry-driven, representing an intermediate regime between Amazonia and the Maritime Continent.

We observe different drivers of isoprene columns over Amazonia, the Maritime Continent, and equatorial Africa, with Amazonia representing an “emissions-controlled” regime, the Maritime Continent a “chemistry-controlled” regime, and equatorial Africa as an intermediate regime between the two. For the Maritime Continent, we described three hypotheses for the observed isoprene-precipitation relationship, which are in addition to established episodic contributions from biomass burning NOx: (1) soil NOx in oil palm plantations modulating [OH]; (2) satellite artifacts due to increased cloud cover or water vapor; and (3) convection and lightning affecting isoprene retrievals and chemistry. Although our radiative transfer simulations suggest that clouds and water vapor are not responsible for our observed correlation, the impact of convection/lightning on isoprene retrievals remains an important uncertainty on this observed isoprene-precipitation relationship. We emphasize that these remaining hypotheses are not mutually exclusive: it is possible that both soil NOx and convection/lightning modulate isoprene columns over the Maritime Continent, as well as in the other two regions. All three regions experience the same processes: temperature-dependent emissions, and variations in non-anthropogenic NOx emissions, and convection. What determines variability in isoprene columns is not whether a process occurs but rather the range of variability in that process relative to other controlling factors.

Given that all three non-anthropogenic NOx sources may impact isoprene columns over varying ways, we can quantify the sensitivity of isoprene columns to simulated changes in NOx emission fluxes using the chemical transport model GEOS-Chem. The sensitivity of isoprene columns to a NOx source depends on both the spatiotemporal colocation between isoprene and NOx emissions, as well as the altitude in which the NOx is emitted. If the NOx is emitted in regions with high isoprene emissions and closer to the ground, the NOx may be more likely to impact [OH] in areas with high rates of isoprene oxidation, potentially having a stronger impact on isoprene variability.

In our sensitivity studies, we independently decreased the soil, biomass burning (GFED4 and QFED2), and lightning NOx emission fluxes by 10 %. Since each NOx source has a different total flux, this scaling decreased biomass burning NOx in the Maritime Continent ten times more than soil or lightning NOx. To account for this unequal scaling, we normalized the resulting change in isoprene columns by the change in NOx flux to obtain the sensitivity of isoprene columns to each NOx source. We also take the absolute value, as a decrease in NOx always increases isoprene columns over these simulations. In this analysis, we are more interested in the magnitude of the change. The simulated 10 % decrease in soil NOx was lower than the monthly variability in BDSNP soil NOx by 1–2 orders of magnitude, representing a lower-bound on soil NOx impact on isoprene columns.

Over the Maritime Continent, lightning NOx had the smallest impact on isoprene columns on a per molecule basis (Fig. a). The same was true for equatorial Africa, but over Amazonia, the sensitivity of isoprene to lightning NOx was higher and comparable to the other NOx sources (Fig. b). Although lightning NOx plays an important role in the convective outflow's chemical regime, the modeled flux of isoprene that reaches the upper troposphere in GEOS-Chem was small relative to the total flux at the surface. This result may be sensitive to the choice of convection scheme. Additional work should be conducted on more high resolution models, e.g. large-eddy simulations, to compare the fraction of isoprene that gets lofted into the upper troposphere.

For the other two non-anthropogenic NOx sources, the sensitivity of isoprene to biomass-burning and soil NOx were similar in magnitude within each region, although the magnitude across all NOx source sensitivities was lowest in equatorial Africa relative to the other two regions. Isoprene over the Maritime Continent was most sensitive to soil NOx on a per-molecule basis in the beginning of the year and to biomass burning NOx at the end of the year. As noted before, biomass burning changes are more episodic than changes in soil NOx, and thus changes in soil NOx would better explain the consistently negative correlation between isoprene column anomalies and precipitation/soil moisture previously observed in Fig. . Nevertheless, both sources likely work in tandem to affect isoprene and [OH] in the Maritime Continent, with biomass burning likely contributing more to large, episodic changes in isoprene, OH, and NOx during biomass burning season, and soil NOx contributing more to gradual, continuous variations in all three species over time due to changes in temperature and soil moisture.

In general, decreasing NOx fluxes in GEOS-Chem decreases modeled formaldehyde columns (Fig. S18). By decreasing NOx fluxes and thus [OH], isoprene oxidation – and formaldehyde production from isoprene oxidation – slows down, with additional minor impacts from changes in RO2 branching at different NOx concentrations . Therefore, a negative correlation between isoprene and formaldehyde may indicate NOx-driven isoprene changes, while a positive correlation indicates isoprene emission-driven variability.

These modeled relationships between formaldehyde, isoprene, and NOx are consistent with observed daily L3 satellite retrievals from the Ozone Monitoring Instrument (OMI) that were bilinearly interpolated to the CrIS 0.5°×0.625° grid and resampled to monthly values. Outliers were removed using the 1.5× IQR threshold. Over Amazonia, isoprene and formaldehyde have a positive correlation (p<0.05), which indicates that changes in isoprene are likely due to isoprene emissions. On the other hand, formaldehyde and isoprene from the Maritime Continent have a consistently negative correlation, indicating that another driver (e.g. NOx) is responsible for changes in isoprene. These correlations provide observational evidence for the “emissions-controlled” regime over Amazonia and the “chemistry-controlled” regime over the Maritime Continent throughout the entire 8-year CrIS record (Fig. ).

In this paper, we show that the three tropical regions have different controls on isoprene column variability. Amazonia represents the most traditional regime: where temperature-dependent isoprene emissions control most of the isoprene column variability. Isoprene anomalies over the Maritime Continent, on the other hand, are controlled by a combination of non-anthropogenic NOx sources. Finally, equatorial Africa represents an intermediate regime, where isoprene emissions control isoprene columns during hot periods, while biomass burning NOx controls isoprene columns during cooler periods. These three regions span the spectrum between “emissions-controlled” and “chemistry-controlled” regimes.

The existence of these regimes is due to the dynamic range in temperature and the variability of NOx sources within each tropical region. Although isoprene over Amazonia is more sensitive to all three non-anthropogenic NOx sources than the other two regions (Fig. 8), Amazonia also has the highest variability in temperature and isoprene emissions (Fig. S11). Consequently, isoprene column variability caused by NOx sources (e.g. soils) is likely masked by the larger variability in temperature and isoprene emissions, resulting in the observed “emissions-controlled” regime. As mentioned previously, there is large uncertainty in soil NOx emissions, with soil NOx emission inventories potentially underestimating fluxes by an order of magnitude in tropical areas . The magnitude of these soil NOx emission fluxes is important in determining total isoprene columns and local chemistry, but for soil NOx to become a significant driver of isoprene column variability in Amazonia, the dynamic range in emission fluxes must be comparable or greater than the dynamic range in temperature. Regardless, the uncertainty in soil NOx fluxes highlights the need for more observations in remote tropical regions to better constrain soil NOx and isoprene emissions and chemistry.

Equatorial Africa represents a smaller region than Amazonia and also has the highest biomass burning NOx fluxes of the three regions by a factor of 3, particularly during boreal summer. Although most of these fires occur south of the areas with highest isoprene emissions, seasonally-averaged winds are oriented to transport plumes from regions with high biomass burning toward forested areas. Biomass burning plumes with elevated NOx originating from these fire hotspots have been detected in the mid- and upper-troposphere as far as the western Africa coast . Thus, the magnitude of biomass burning NOx and its transport may thus influence isoprene column variability.

In this paper, we suggest that isoprene variability over the Maritime Continent is largely driven by non-anthropogenic NOx sources. These sources include soil NOx from fertilized soils and episodic contributions from biomass burning. If convection is strong enough, then lightning NOx may also be an important driver of isoprene columns, and future work is required to determine the impact of convection and lightning NOx on CrIS isoprene retrievals. We note that certain chemical reactions, such as the rapid hydrolysis of 1,2-isoprene hydroxynitrate into nitric acid, and nitrogen deposition onto leaves may result in nonlinear relationships between the two species . Future modeling studies should simulate canopy effects (e.g., changes in turbulence, radiation, and deposition) and isoprene-NOx chemistry with a variety of different chemical mechanisms to determine the impact of these processes .

These potential drivers described above highlight the heterogeneity seen throughout the tropics, as well as how a combination of dynamics, chemistry, and biology influence the chemical composition of the remote atmosphere. Understanding these regional differences is critical for predicting future changes in atmospheric oxidants and methane lifetime as vegetation, fire regimes, and land use evolve in response to climate and human activity.