For this study we use the three-dimensional chemical transport model (TOMCAT) (Chipperfield, 2006) coupled to the GLOMAP global aerosol microphysics scheme (Mann et al., 2010). The version of TOMCAT-GLOMAP used here is a further development of that described by Monks et al. (2017). Cloud fields are now provided from the European Centre for Medium-Range Weather Forecasts (ECMWF) reanalyses (Dee et al., 2011), replacing the climatological clouds fields used previously from the International Satellite Cloud Climatology Project (ISCCP) (Rossow and Schiffer, 1999), leading to improved representation of photolysis. Other developments include updated emission inventories, the inclusion of CERN Cosmic Leaving Outdoor Droplets (CLOUD)-based new particle formation and the introduction of Mårtensson sea spray emissions (Gordon et al., 2017; Monks et al., 2017). The model is run at 2.8∘×2.8∘ horizontal resolution with 31 vertical levels from the surface to 10 hPa, driven by 6-hourly ECMWF ERA-Interim reanalyses. The planetary boundary layer (PBL) scheme is based on Holtslag and Boville (1993) and sea surface temperatures are from ECMWF reanalyses. ECMWF ERA-Interim reanalyses have been shown to have good skill in capturing Madden–Julian Oscillation (MJO) events, which in turn impact the onset of ENSO events (Dee et al., 2011), giving confidence that the model competently simulates El Niño meteorological conditions.

The tropospheric chemistry scheme used is as described in Monks et al. (2017), with anthropogenic emissions from the Monitoring Atmospheric Composition and Climate (MACCity) emissions inventories (Lamarque et al., 2010). Annually varying emission inventories are included for all fire-emitted gas-species and aerosol emissions, such as black carbon (BC). The Global Fire Emissions Database (GFED) used by TOMCAT-GLOMAP has been updated to version 4, with CO, nitrogen oxides (NOx) and volatile organic compound (VOC) emissions from fires (Randerson et al., 2017; Reddington et al., 2018). Monthly varying biogenic VOC emissions are from the MEGAN-MACC emissions inventory for reference year 2000, calculated from the Model of Emissions of Gases and Aerosols from Nature (MEGAN version 2) (Sindelarova et al., 2014). The CH4 inventory was produced by McNorton et al. (2016b), with wetland emissions derived from the Joint UK Land Environment Simulator (JULES) and biomass burning emissions from GFEDv4 (Randerson et al., 2017). These are then combined with anthropogenic emissions from EDGAR version 3.2, paddy field emissions from Yan et al. (2009) and termite, wild animal, mud volcano, hydrate and ocean emissions from Matthews and Fung (1987) (McNorton et al., 2016b). The global mean surface CH4 mixing ratio is scaled in TOMCAT-GLOMAP to a best estimate based on observed global surface mean concentration (McNorton et al., 2016a; Dlugokencky, 2019).

Radiative effects of O3 changes are calculated using the O3 radiative kernel approach derived by Rap et al. (2015) using an offline version of the Edwards and Slingo (1996) radiative transfer model. This considers six bands in the shortwave (SW), nine bands in the longwave (LW) and uses a delta-Eddington two-stream scattering solver at all wavelengths (Rap et al., 2015). This version has been used extensively in conjunction with TOMCAT-GLOMAP for calculating radiative forcing from simulated distributions of several short-lived climate pollutants (SLCPs) including BC, O3 and CH4 (Spracklen et al., 2011; Riese et al., 2012; Rap et al., 2013, 2015; Richards et al., 2013).

All simulations are performed for 1997–2014 with a 4-year spin-up through 1993–1996. The control run (CTRL) allows all emissions and meteorology to vary throughout the modelled period. GFED biomass burning emission inventories began in 1997; therefore, the 1993–1996 spin-up simulation uses repeating 1999 emissions instead, as the closest year of “average” emissions, having excluded 1997 and 1998 due to the exceptionally high emissions in those years (Schultz et al., 2008).

To test the impact of El Niño events on atmospheric chemistry, we also performed three perturbed simulations (Table 1). Where model simulations used “fixed” parameters in Table 1, the year 2013 emissions or meteorology are specified as invariant throughout the simulation. This year is chosen as the ENSO-neutral case, due to it being the least active ENSO year during 1997–2014, with a maximum bimonthly multivariate ENSO index (MEI) magnitude of -0.4 and the only year without a single MEI value that could be considered an active El Niño or La Niña (Wolter and Timlin, 1993, 1998). Throughout this study, an El Niño event was taken to be ongoing if the MEI was greater than +1.0. We perform a factorial analysis based on perturbed simulations in which we fix global biomass burning emissions (FIREFIX) or global meteorology (METFIX) to the “ENSO-neutral” case. An additional perturbed simulation was performed in order to examine the secondary impact of CO on CH4 via oxidation changes, where only CO emissions from biomass burning were fixed (COFIX).

We compare annual mean simulated gas-phase species for 1999 against a climatological dataset of aircraft observations from 16 campaigns conducted with a broad spatial and temporal range from 1992 to 2001 (Emmons et al., 2010). While the comparison of observational data from intermittent aircraft campaigns does not offer a perfect comparison with the model simulated long-term mean concentrations, it allows evaluation of broad characteristics of a number of species over vertical profiles in many global regions. Figure 2 shows the comparison of simulated annual mean global concentrations of CO, CH4 and Peroxyacetyl nitrate (PAN), with aircraft observations at 0–2, 2–6 and 6–10 km. We have also calculated the normalized mean bias between the model and observations (Fig. S6). Full details of the aircraft measurement campaigns used can be found in Table S1 in the Supplement.

The model captures broad characteristics of spatial distribution for all species, simulating higher concentrations in polluted urban or biomass burning regions, with lower concentrations over ocean and in the SH. CO concentrations decrease with altitude but the largest values still occur around urban areas and burning regions, which can be seen in both model and aircraft concentrations. Consistent with the comparison with MOPITT satellite retrievals (Figs. S1 and S2), the model underestimates CO concentrations particularly near the surface, with a NMB of -11.1 %, -9.93 % and -0.25 % at 0–2, 2–6 and 6–10 km, respectively. Absolute concentrations of CH4 in TOMCAT simulations match aircraft data very well, although given the global mean surface concentration scaling, we expect the magnitude of CH4 to be well simulated. The latitudinal and vertical distributions are also well captured, giving confidence in the model transport and OH simulation. Aircraft observations show CH4 also decreases with altitude and the hemispheric disparity becomes more pronounced, with higher concentrations in the NH. For PAN concentrations, the simulated spatial distribution is broadly well captured, as is the increased concentration with altitude. There is a general low bias in absolute concentrations near the surface (NMB =-12.3 %), with a better comparison at 2–6 km (NMB =1.68 %) and overestimation at 6–10 km (NMB =18.17 %).

Due to its very short lifetime, it is challenging to evaluate model-simulated OH over representative spatial and temporal scales. Here we follow the evaluation methodology recommended by Lawrence et al. (2001) of dividing tropospheric OH into 12 subdomains, from the surface to a climatologically derived tropopause. This method was also used to evaluate a previous version of TOMCAT(version 1.76) by Monks et al. (2017), allowing direct comparison. The evaluation is performed for the year 2000. Figure 3 shows our simulated OH compared to Monks et al. (2017), the ACCMIP model mean (Naik et al., 2013) and the Spivakovsky et al. (2000) OH dataset estimated from methyl chloroform observations.

The models and observationally constrained distribution broadly agree in terms of the latitudinal spread of OH concentrations with a minimum in the SH and a maximum in the tropics; however, there is disagreement over the exact altitude of the maximum OH concentrations. In both versions of TOMCAT the highest concentration is between the surface and 750 hPa, while ACCMIP and Spivakovsky et al. (2000) find peak OH in the upper and mid-level troposphere, respectively. The updated cloud fields used in the current TOMCAT-GLOMAP version have slightly increased OH concentrations in the mid-level and upper domains compared to Monks et al. (2017) but concentrations remain significantly higher in the NH and surface domains than in other studies. In addition, our simulated NH:SH ratio of 1.48 in the current TOMCAT version remains substantially higher than in the ACCMIP models (1.28±0.1), indicating that TOMCAT photolysis rates and OH production in the NH are larger.

The total global tropospheric average OH in this version of TOMCAT is 1.04×106 molec. cm-3, a decrease from Monks et al. (2017) and within the range of other published values (Table 2). This is primarily due an updated treatment of clouds, in which climatological cloud fields have been replaced with cloud fraction from ECMWF reanalyses data, affecting photolysis rates. The tropospheric O3 burden of 342 Tg has increased relative to Monks et al. (2017) (331 Tg) and is within the range found in Wild (2007) (335±10 Tg) and ACCMIP models (337±23 Tg) (Young et al., 2013). Due to the simplified treatment of CH4, the scaling applied and its relatively long atmospheric lifetime, the total atmospheric lifetime cannot be determined from TOMCAT simulations. Instead a chemical lifetime due to reaction with OH is calculated from CH4 and OH burdens, disregarding stratospheric sinks and soil sinks (Fuglestvedt et al., 1999; Berntsen et al., 2005; Voulgarakis et al., 2013). The lifetime diagnosed from TOMCAT is 8.0 years, compared to the multi-model mean and range of 9.3±0.9 years from Voulgarakis et al. (2013). The shorter lifetime in TOMCAT is due to the overestimation of OH at the surface, particularly in the NH where CH4 concentrations are highest due to anthropogenic emissions.

First we examine the mechanisms controlling interannual variability of simulated tropospheric CO, O3 and mean OH. We use the difference between the control (CTRL) and the perturbed simulations with fixed fires (FIREFIX) and fixed meteorology (METFIX) to determine the driving cause of IAV. Of particular interest is the effect of the 1997–1998 El Niño event (henceforth referred to as 1997 El Niño) and how the prevailing mechanisms controlling IAV change during such events. To define El Niño events, we use the bimonthly multivariate ENSO index, which is calculated from six observed variables and standardized to accurately monitor ENSO occurrence (Wolter and Timlin, 1998, 2011).

Previous studies examining the dominant factor controlling global CO IAV have found contrasting results. Szopa et al. (2007) suggested that meteorology was the main driver, accounting for 50 %–90 % of IAV in the tropics. Conversely, a study by Monks et al. (2012) considered CO IAV in the Arctic, finding that biomass burning was the dominant driver with a strong correlation to El Niño. Voulgarakis et al. (2015) also suggested that biomass burning was the more important driver of IAV with only a small effect from meteorology. Some of these differences in results can be explained by Szopa et al. (2007) considering only surface CO, rather than the whole troposphere as in Voulgarakis et al. (2015). Here we also consider the whole of tropospheric CO and our results are in line with those from Voulgarakis et al. (2015). We find the dominant source of IAV across the entire period is emissions from biomass burning – indicated by the large difference between simulations CTRL and FIREFIX (Fig. 4a), with a small effect from meteorological changes (CTRL – METFIX). This effect was largest during the 1997 El Niño, where an increase in fire events increased CO concentrations by more than 40 %. Smaller increases of 5.8 % and 7.6 % occur during less extreme El Niño events of 2002/03 and 2006, respectively, with only a 1.8 % increase during the 2009/10 El Niño, indicating that El Niño only significantly impacts CO concentrations when there is an associated increase in global fire events.

Expanding on the work of Voulgarakis et al. (2015), we analysed IAV using the coefficient of variation (CV), calculated as the multi-year standard deviation normalized by the mean (Fig. 5). The global annual mean CO IAV over the whole period is 11.0 % for the whole troposphere and 14.3 % for surface concentrations. This is in very good agreement with Voulgarakis et al. (2015), who calculated 10 % IAV; in fact, the comparison is even better when we consider the same time period (2005–2009) with our corresponding IAV estimate at 9.7 %. The slightly lower estimate here may be a result of the fixed-year biogenic volatile organic compound (BVOC) emissions, removing the effect of the IAV of biogenic emissions on CO IAV. BVOC oxidation is estimated to contribute 15 % of the total source of CO (Duncan et al., 2007); however, the IAV of BVOC emissions has been found to be relatively small, ∼2–4 % (Naik et al., 2004; Lathière et al., 2005). Despite good global comparison with Voulgarakis et al. (2015), there are regional differences; CO IAV from TOMCAT is much larger in high-latitude boreal regions. This is likely due to the difference in the period studied, meaning this study includes additional extreme events including unusually large Russia boreal wildfires in 2010 and 2012 (Gorchakov et al., 2014; Kozlov et al., 2014). Infrequent and extreme events such as these significantly increase IAV.

CO IAV is significantly greater in September–October, with peaks in known fire regions such as tropical South America, Africa, Southeast Asia and boreal forests. This indicates a strong contribution of fire emissions to IAV, especially from Indonesia (Fig. 4a), as also suggested by previous studies (Monks et al., 2012; Huang et al., 2014; Voulgarakis et al., 2015). In the FIREFIX simulation IAV is ∼55 % of the CTRL value, showing a large reduction in variability when interannual variability in fire emissions is removed. The IAV in March–April is significantly smaller than September–October as this period is outside the primary fire season for South America and Eurasia, although hotspots remain in Southeast Asia and Africa where fires commonly occur in March–April (van der Werf et al., 2017). Meteorology and atmospheric transport changes are most important in Africa in September–October and Indonesia in March–April (Fig. 5c, d). Fire emissions occur in these regions but the meteorological effects are important sources of IAV. This is in good agreement with Voulgarakis et al. (2015), who found that with fixed biomass burning emissions, high IAV remained over Africa during December–January, and Huang et al. (2014), who found CO over Central Africa correlated more closely with ice water content than CO emissions due to increased convective transport. However, the overall effect of meteorology on global IAV found here is much smaller than the 50 %–90 % suggested by Szopa et al. (2007): when we consider only surface CO over the same period, fixing meteorology decreases the mean CO IAV by just 5 %.

The IAV of OH and O3 have more complex contributions from fire emissions and meteorology (Fig. 4b, c). For both species, meteorology is the dominant cause of variability for the majority of the period, indicated by, on average, greater deviation from CTRL in METFIX simulation than FIREFIX, including during El Niño events outside of the 1997 El Niño, such as in 2006. Our results compare well to Inness et al. (2015), who also found that changes to tropospheric O3 during El Niño were driven by a combination of emissions and atmospheric dynamics. This is also in agreement with Doherty et al. (2006), where a strong correlation was found between ENSO meteorology and global O3 burden, albeit with a lag period of several months. Various meteorological variables are known to affect OH and O3 variability, including humidity, clouds and temperature (Stevenson et al., 2005; Holmes et al., 2013; Nicely et al., 2018). OH variability is particularly sensitive to changes in lightning NOx production which decreases during El Niño conditions (Turner et al., 2018). Murray et al. (2014) also examined factors affecting OH variability since the last glacial maximum, finding tropospheric water vapour, overhead stratospheric O3 and lightning NOx to be key controlling factors. Furthermore, circulation changes during El Niño events have been linked to lower stratospheric O3 variability (Zhang et al., 2015; Manatsa and Mukwada, 2017), which in turn influences tropospheric OH and O3 concentrations (Holmes et al., 2013; Murray et al., 2014). Despite the importance of meteorological drivers, we find that fire emissions are the dominant cause of variation in both OH and O3 during the 1997 El Niño, increasing global tropospheric O3 burden by up to ∼7 % and decreasing tropospheric OH by up to ∼6 %. This result is supported by several other studies, which have found that during large fire events such as that caused by the 1997 El Niño, fire emissions substantially decrease tropospheric OH and increase tropospheric O3 (Hauglustaine et al., 1999; Sudo and Takahashi, 2001; Holmes et al., 2013). Our results indicate that while meteorology is generally the most important driver of IAV in global tropospheric OH and O3, fire emissions can also play a key role and become the dominant driver when there are particularly large fire emissions related to El Niño.

Figure 6 shows the IAV of O3, supporting the analysis of Fig. 4 that also suggests meteorology is the dominant process in controlling IAV. METFIX-simulated IAV differs substantially from the CTRL, with much lower IAV in September–October (33 % decrease) and in March–April (42 % decrease) when meteorology is repeated. However, in the METFIX run there remain peaks in variability in close proximity to regions with large biomass burning emissions, demonstrating the significant contribution from fire emissions. In the FIREFIX simulation the distribution of IAV is broadly similar to the CTRL simulation and only shows a small change in global mean CV, indicating that fire emissions have less control on O3 IAV. These results are again comparable to Voulgarakis et al. (2015) as the distribution of O3 IAV in both CTRL and FIREFIX simulations is similar, despite slightly larger values of variation due to differing time period.

The COFIX sensitivity experiment was conducted to determine the indirect influence of CO emissions on CH4 variability through changes in tropospheric OH concentrations. Figure 7a shows the difference in COFIX monthly mean OH concentrations from the control experiment, compared to that from the METFIX and FIREFIX simulations. When CO emissions from biomass burning are fixed, OH concentrations are consistently higher than in the CTRL simulation. This indicates that high CO emissions decrease global mean tropospheric OH. The greatest impact is during the 1997 El Niño, where CO emissions were abnormally large, suppressing mass weighted global monthly mean OH concentrations by up to 9 %. The mean effect on OH over the 1997 El Niño of -3.6 % is comparable to that simulated by Butler et al. (2005), who also found an increase in CO resulted in a change in OH of -2.2 %. Duncan et al. (2003) found a similar magnitude response in OH to the Indonesian wildfires in 1997 of between -2.1 % and -6.8 %. The suppression of OH concentrations due to CO emission is also simulated to a lesser degree in the 2003 and 2006 El Niño events but is absent in the 2010 El Niño as this event had little impact on global fire occurrence (Randerson et al., 2017). The effect of fixing only CO from fires is greater than the effect of fixing all fire emissions due to co-emitted species such as NOx, which act to increase OH concentrations.

As OH is also the primary sink of CH4 (∼90 %) (McNorton et al., 2016a), another effect of the decrease in OH due to CO emissions is to weaken the sink of CH4, increasing its atmospheric lifetime. The magnitude of this can be seen in Fig. 7b; the COFIX simulation indicates that CO emissions from fires extended CH4 atmospheric lifetime by more than 4 % during the 1997 El Niño. Fixing all fire emissions also enhances CH4 lifetime by around 2 %. Increasing the lifetime of a species increases its concentration in steady-state equilibrium. Due to the scaling applied to CH4 in TOMCAT we are unable to directly calculate the response in CH4 growth rate from TOMCAT, as simulated global mean surface CH4 concentrations are nudged to the observed value. Therefore, to determine the impact of the change to OH on CH4 concentrations, we used a simple global box model. This box model is similar to that described in McNorton et al. (2016a), which was found to compare well with other global and 12-box CH4 models (Rigby et al., 2013; McNorton et al., 2016a). In this case, the box model used monthly mean tropospheric OH concentrations and CH4 emissions for each simulation, while assuming constant temperature to calculate the effect of changing OH on global mean surface CH4. A fixed temperature was used as varying the temperature has been found to have a relatively small impact on derived CH4 concentrations (McNorton et al., 2016a). The impact of fire emissions on the CH4 growth rate is greatest in 1998, where all emissions from fires increased global CH4 by 10.5 ppb (Fig. 7c). Analysis of the COFIX simulation demonstrates that up to 7.5 ppb (72 %) of that change could have been caused by the release of CO alone and its role as a sink for OH. The effect on growth rate in the FIREFIX simulation is larger than in the COFIX despite a greater effect on CH4 lifetime from the COFIX, due to directly emitted CH4 varying with El Niño conditions in the COFIX simulation and not in FIREFIX. The influence of CO emissions on CH4 growth rate calculated here is smaller than in Butler et al. (2005), despite a much larger effect on tropospheric OH. The radiative effect of the change to CH4 from CO emitted from biomass burning alone in 1998 is 0.004 W m-2, calculated using updated expressions from Etminan et al. (2016).

In this section we examine trends and the impact of El Niño on the production of tropospheric O3. El Niño is known to have a large effect on tropospheric O3 precursors such as CO and NOx; therefore, examining O3 production regimes during El Niño can provide insights into the main mechanism responsible for the observed changes in tropospheric O3. The ratio between formaldehyde (HCHO) and nitrogen dioxide (NO2) concentrations can be used to indicate the limiting factor for tropospheric O3 production (Duncan et al., 2010). Ratios smaller than 1 indicate that removing VOCs will decrease tropospheric O3 formation (i.e. a VOC-limited regime), while ratios larger than 2 indicate that removing NOx will reduce O3 (i.e. a NOx-limited regime). Ratios of 1–2 indicate that both NOx and VOC reductions could decrease O3 (i.e. a “both-limited” regime). Here we apply this methodology to determine the changes to this ratio from 1997 to 2014 and dependence of O3 formation during the 1997 El Niño event. We compare the early period mean (1999–2003) to the end period mean (2010–2014) to determine whether significant changes have occurred over the 18-year period and compared mean El Niño conditions to both.

In general, the SH and tropical regions have very high ratios, meaning they are strongly NOx-limited (Fig. 8). The NH is also predominantly NOx-limited, although less robustly, and polluted regions tend to be either VOC-limited or both-limited regimes. The ratio is largely constant across the modelled period; however, there are some significant shifts, such as in India, which was once solely NOx-limited, becoming increasing VOC-limited due to increased NOx pollution (Hilboll et al., 2017). This shift in the spatial distribution of O3 precursor emissions to lower latitudes leads to increased tropospheric O3 production proportional to total emissions (Zhang et al., 2016).

During El Niño there are large changes, increasing the ratio and therefore the NOx limitation by more than 40 % in the tropical Pacific. Significant changes to the ratio were also found in biomass burning regions of South America and Southeast Asia. This is due to the increase in NOx emissions in larger fire seasons associated with El Niño. However, these regions are already very heavily NOx-limited due to high VOC emissions in forest regions, meaning that although the shift in HCHO/NO2 ratio during El Niño is large, it is not substantial enough to alter the limiting factor for formation of tropospheric O3 from one regime to another. Over India, El Niño conditions inhibit the trend towards a both-limited regime, as the NOx-limited regime continues to dominate throughout.

The 1997 El Niño significantly altered the vertical distribution of O3 in the troposphere, increasing O3 concentrations in the NH while decreasing in the SH and tropics with an overall decrease in tropospheric O3 of -0.82 % compared to the 1997–2014 mean (Fig. 9a). In the CTRL simulation there is decreased O3 in the tropical upper troposphere, possibly related to increased convection over the eastern Pacific (Oman et al., 2013; Neu et al., 2014). We also simulate large increases in the midlatitude upper troposphere of both hemispheres in the CTRL and FIREFIX simulations but not in METFIX, implying that this is produced by El Niño-associated meteorological processes which promote intrusion of stratospheric air into the troposphere. These positive anomalies were also observed in Oman et al. (2013) and Zeng and Pyle (2005), attributed to El Niño influence on circulation patterns and enhanced stratosphere–troposphere exchange.

In general, the METFIX run simulates higher O3 concentrations in the NH than the period mean and lower concentrations in the SH (Fig. 9b). This hemispherical shift is also present in the CTRL and FIREFIX simulations but with greater negative O3 anomalies in the SH. The simulated NH increases in the CTRL simulation correspond to other studies of the 1997 El Niño (Koumoutsaris et al., 2008), while Oman et al. (2013) similarly reported negative O3 anomalies in the SH during El Niño. Large increases in tropospheric O3 in the western Pacific, Indian Ocean and Europe contribute to the increase in O3 in the NH, despite decreased O3 in the eastern Pacific (Chandra et al., 1998; Koumoutsaris et al., 2008; Oman et al., 2011).

There is an overall increase in O3 (∼2 %) when meteorology was fixed to an ENSO-neutral year (i.e. 2013), meaning that meteorology during the 1997 El Niño caused a decrease in tropospheric O3 concentrations despite large increases in O3 in regions of the upper troposphere due to stratospheric intrusion. During the 1997 El Niño we find a 0.4 % increase in global tropospheric humidity compared to the period mean. This is likely partly responsible for the general decrease in O3 due to meteorology, as increased humidity enhances O3 loss (Stevenson et al., 2000; Isaksen et al., 2009; Kawase et al., 2011). Changes to transport and distribution of O3 will also impact how efficiently tropospheric O3 is produced and lost.

The similarities between the tropospheric O3 distribution in the CTRL and FIREFIX simulations show that fire emissions have a relatively small impact on the global distribution of O3 but do affect absolute values, as concentrations in the FIREFIX run are significantly lower in the tropics. This is likely because of the removal of large emissions of O3 precursors in that latitude band when fire emissions are fixed to a non-El Niño year, as several studies have found that enhanced fires in 1997 El Niño increased tropospheric O3 in the region (Chandra et al., 1998; Thompson et al., 2001; Doherty et al., 2006; Oman et al., 2013).

Figure 10 shows the tropospheric O3 radiative effect (RE) during the 1997 El Niño in each TOMCAT simulation, calculated using the Rap et al. (2015) tropospheric O3 radiative kernel. Consistent with the relative changes in O3 concentration, fire emissions and meteorology have contrasting effects on O3 RE. When isolated, these effects are opposite and almost equal: fire emissions increase O3 RE by 0.031 W m-2, while meteorology decreases by -0.030 W m-2. We performed an additional simulation to determine the effect of these factors occurring simultaneously (BOTHFIX) and found the increasing effect from fire emissions to be dominant over the decreasing effect from meteorology, leading to an overall increase in global mean O3 RE of 0.015 W m-2. The effect of fire emissions occurs almost entirely over Indonesia and the eastern Indian Ocean where the large influx of NOx, CO and CH4 from fire emissions during the 1997 El Niño causes large regional increases in tropospheric O3. This increase, also observed in Chandra et al. (1998), causes a regional RE of up to 0.17 W m-2. Meteorology has more varied impacts during El Niño, causing large decreases in O3 RE over the central Pacific Ocean (∼-0.36 W m-2) but also increases in the midlatitudes of the Pacific Ocean (∼0.33 W m-2). Globally the mean change to O3 RE is small, around 0.015 W m-2, but large regional changes have the potential to significantly alter atmospheric heating and dynamics.