Positive feedback to regional climate enhances African wildfires

Summary Regional climate strongly regulates the occurrence of wildfires partly because drying of fuel load increases fires. The large amounts of aerosols released by wildfires can also strongly affect regional climate. Here we show positive feedback (a seasonal burned area enhancement of 7–17%) due to wildfire aerosol forcing in Africa found in the simulations using the interactive REgion-Specific ecosystem feedback Fire (RESFire) model in the Community Earth System Model (CESM). The positive feedback results partly from the transport of fire aerosols from burning (dry) to wet regions, reducing precipitation and drying fuel load to enhance fires toward the non-burning (wet) region. This internally self-enhanced burning is an important mechanism for the regulation of African ecosystems and for understanding African fire behaviors in a changing climate. A similar mechanism may also help sustain wildfires in other tropical regions.


INTRODUCTION
Wildfire initiated by natural or human factors has a profound impact on ecosystems, 1 carbon cycles, 2 climate change 3 and human society. 4lthough the increase of population and land use conversion tends to suppress fires over most of regions and potentially in the future, [5][6][7][8][9][10][11][12][13][14] fire weather season length in Africa has increased by up to 40% in Africa in the past four decades, potentially leading shifts in both the geographical distribution and variability of burned areas. 9,13,14Wildfires also exert a significant impact on regional and global climate systems, as well and ecosystems.This perturbation stems from a complex interplay of short-term and long-term changes in fire weather, terrestrial ecosystems, and human activities.6][17][18] While there remain divergent estimations regarding the trends in wildfire emissions across regions, [7][8][9][10][11][12][13] studies have indicated shifts in both the geographical distribution and variability of burned areas, including those in Africa. 9,13,19ires and regional climate interact through several means. 20Fires impact climate by reducing local precipitation 21,22 and changing the global radiative balance through the emissions of greenhouse gases (GHG) and aerosols. 3,235][26][27][28] Climate change has increased the occurrences of weather conditions conducive to fires over the past 40 years, 29 and the effect will potentially become larger in the future. 30ires are a major source of atmospheric aerosols, 31 which strongly affect the global radiative balance 2 and cloud processes. 32Aerosols can have both positive and negative impacts on precipitation. 33An increase in the number of cloud condensation nuclei (CCN) leads to competition for the available water vapor, resulting in smaller CCN sizes and making it more difficult to form raindrops. 34 The aerosol radiative impact may also suppress evaporation, which then affects cloud formation.On the other hand, while aerosols may reduce precipitation from shallow clouds, they can stimulate deep convection with a warm cloud base. 35A positive feedback among fires, cloud, and precipitation was previously hypothesized. 36,37However, atmospheric modeling studies often focused on the perturbations by fire emissions to regional and global weather and climate, 38 while fire modeling studies often considered only the impacts of weather and climate on fires. 16n this work, we use the interactive Region-Specific ecosystem feedback Fire (RESFire) model in the Community Earth System Model (CESM) 19,39 to investigate positive feedback through aerosol forcing that enhances wildfires in Africa, which is an ideal place to study fire and climate interactions.The area of highly flammable savanna-like vegetation dry conditions is vast, and the extensive vegetation cover and favorable weather conditions contribute to substantial fuel accumulation in Africa.As a result, about half of global biomass burning occurs in Africa and the continent has more fire aerosol emissions than any other continent. 40The alternation of dry seasons in northern and southern Africa also means that fires are persistent throughout the year in Africa.The environmental and socioeconomic challenges caused by fires are exacerbated by climate change. 41Understanding the systematic fire-climate feedback via aerosol forcing in Africa is therefore crucial for examining potential climate change caused vulnerabilities in the continent.

Enhanced seasonal to annual variations of fires in equatorial Africa
Africa accounts for >50% of the total area burned by fires on average. 13,42,43It is also an area with the largest aerosol-cloud radiative forcing by aerosols emitted from fires. 19We conduct 10-year RESFire simulations for the present-day climate with and without coupled fire-climate interactions (hereafter referred to as the ''fire aerosol feedback'' and ''no-fire aerosol feedback'' simulations, respectively) to explore the impacts and mechanisms of the feedbacks between fire aerosols and regional climate.
Aerosol forcing in general is short-term because the lifetime of aerosols is on the order of 1 week due to wet scavenging. 44,45However, over the equatorial region of Africa (10 S -10 N, 10 E À 28 E), fire aerosol forcing can lead to variations of the burned area on time scales much longer than a week.Figure S1 compares the power spectral densities (PSDs) of monthly mean burned area over equatorial Africa with and without fire aerosol feedbacks.Aerosol feedbacks enhance the seasonal and annual variations (3, 4, 6 and 12 months) by up to 60%.This simulated fire aerosol feedback differs from those related to fuel load, climate patterns, 46 or human impact, 47 and contributes to both short-term (within a week) and long-term (seasonal to annual) fire variations.

Positive feedbacks of fire aerosols on burning
The main fire seasons occur in the Northern Hemisphere (NH) Africa in December, January, and February (DJF) and in the Southern Hemisphere (SH) Africa in June, July, and August (JJA).This seasonal cross-equator fire migration is a key factor for extending short-term aerosol forcing into seasonal and interannual time scales shown in Figure S1.An important clue is shown in Figure 1.The seasonal increase of burned area is 13-25% in equatorial Africa.Compared to NH burning in DJF and SH burning in JJA without fire aerosol feedbacks (Figure 1A), the enhancements of burned area extend from the equatorial region toward the other hemisphere (Figure 1B).The feedback enhancement extends from 0 -8 N to 0 -15 S in DJF and from 0 -30 S to 0 -10 N in JJA, implying the significance of cross-equatorial processes in fire aerosol feedback.
The regional shift of fire aerosol-induced burned area is related to the transport of fire aerosols.The features of wind transport are illustrated by the circulation patterns (Figure S2).While the surface wind pattern around the equator is affected by the location of the intertropical convergence zone (ITCZ), which is mostly to the north of the equator, the wind transport in the lower free troposphere (850 and 700 hPa, 1-3 km in altitude) is quite different.In DJF, the burning takes place in NH Africa (Figure S1).The lofted fire aerosols are transported southward across the equator in the lower free troposphere.In JJA, the burning takes place in the SH Africa.The southerly wind near the surface in part due to the convergence of ITCZ transports fires aerosols northward.Figure S3 shows that the fire aerosol feedback did not significantly change the location of ITCZ, but affected atmospheric circulation.Near the surface, the convergence north to the ITCZ is 10%-20% larger with fire feedback in DJF and MAM, but the increase is less than 10% in JJA and SON.Therefore, the aerosol effects are the major contributor to this feedback mechanism. 19s a result of fire-aerosol transport, the distribution of aerosol optical depth (AOD) from fire differs significantly from that of burned areas.The fire aerosol AOD distribution shows that the enhancements shifted more toward the non-burning hemisphere (Figure 1D).The simulation results show that fire aerosols reduce precipitation (Figure 1C).We compared the radiative effects due to direct aerosol-radiative interactions and indirect aerosol-cloud interactions (Figure S4) and find that that the aerosol radiative impact by aerosol-cloud interaction is much larger, in agreement with the previous study. 19he fire aerosol-cloud effect on precipitation is shown more clearly on the zonal mean distribution changes due fire aerosols (Figure 2).In DJF, the largest zonal-mean burned fraction increase occurs at 2 -8 N. The southward transport of fire aerosols at 1-3 km results in the largest zonal-mean increases of AOD and PM 2.5 and the largest decrease of precipitation at 0 -4 N, to the south of the largest burned fraction.In-between the former maxima lie the largest zonal-mean increase of burned fraction at 3 N.The largest fire aerosol effect on precipitation is shifted away from the burning region because of the increase of RH and cloud water content away from the burning region.The effect of fire aerosols on precipitation moves toward the burning region for two reasons. 21,36The maximum loading of fire aerosols cools the surface and reduces the vertical instability of the atmosphere.At the edge of the burning region, where RH is relatively low (compared to the rainy region), the influx of fire aerosols competes for the available atmospheric water, increasing the number but reducing the size of CCN, resulting in a reduction of precipitation. 36CCN accumulated in the cloud, increasing the amount of low to middle clouds above 1 km (Figure 2D).This could potentially enhance precipitation elsewhere downwind, such as on the East Coast in DJF (Figure 1C), due to the conversion of water vapor into cloud droplets and the convection invigoration mechanism. 33ire aerosol induced drying at the edges of fire regions subsequently increases fires in DJF (Figure 1).The maximum zonal mean burned area increase due to fire aerosols is >50% (Figure 2A), while the maximum relative burned area increase in this region (Figure 1) reached 9 times that without fire aerosol feedbacks.In the non-burning regions in the SH Africa, on the other hand, the reduction of precipitation causes a dryer condition in the subsequent burning season.The observation-based RESFire parameterization predicts larger fire spreads under dryer conditions in the subsequent burning season, 19,39 contributing to the increase of burning in March, April, and May (Figure S5).It is worth noting that this effect is inter-seasonal, but its precise quantification in a fully coupled climate model is beyond the scope of this work.The net increase of burned area is a result of both intra-seasonal and inter-seasonal feedbacks of fire aerosols.
Similar fire-aerosol feedback is simulated in JJA, but the seasonal burning takes place in the SH Africa and the positive feedback on burning extends from the northern burning area to the equatorial region with the largest relative zonal-mean enhancement of burned fraction in the equatorial region (Figures 1 and 2).Since the ITCZ, determined by the maximum and zero-gradient latitudes for the surface wind divergence, 48,49 is located north of the equator and the burning region, the near-surface flow is northward, bringing large amounts of fire aerosols to the northern equatorial region (Figure S2).At 3-4 km ($700 hPa), the south-ward flow crosses the ITCZ to 10 N. The flow pattern leads to large fire aerosol loading in the equatorial region, which decreases precipitation near the equator.Since the flow northward-southward direction switch altitude increases from the ITCZ to the equator, the vertical extent of fire aerosols maximizes in the equatorial region (Figure 2).The zonal-mean moisture has a maximum at 10 N, resulting in a larger effect of fire aerosols on cloud liquid amount and precipitation in the northern than southern equatorial region.The maximum increase of the zonal mean burned fraction is 12% of the maximum zonal mean burned fraction.Beyond the intra-seasonal effect, burning moves to the northern equatorial region in September, October, and November (SON).The drying of the northern equatorial region in JJA helps the inter-seasonal burning enhancement in SON.
In DJF and JJA, the largest increases of zonal mean burned fraction, AOD, and cloud liquid water content and the largest decrease of precipitation occur downwind from the largest burning region, where burning occurs in the subsequent season.The DJF and JJA total burned areas are comparable at 2.9 3 10 5 km/month, larger than 1.2 3 10 5 km/month in MAM and 1.8 3 10 5 km/month in SON.The burning in MAM and SON is, however, more widespread covering both NH and SH Africa (Figure S5).The seasonal NH-to-SH shift of precipitation occurs in MAM and the reverse is the case in SON. 50Burning occurs in the dry regions shifts seasonally following the shift of the precipitation regions, resulting in burning in both NH and SH Africa simulated in the model similar to the observations 40 (Figure S5).One consequence of the widespread burning in MAM and SON is that aerosol loading is enhanced by fires, leading to precipitation reduction and burning enhancement in both NH and SH Africa.The geographical spread of burning and aerosol effect result in a larger relative positive feedback by fire aerosols on burning in MAM and SON (15-17%) than DJF and JJA (7-9%), although the absolute burned area increase due to fire aerosols is more comparable among DJF, MAM, JJA, and SON.

DISCUSSION
Fire aerosols lead to both intra-seasonal and inter-seasonal burned fraction increases (Figures 1 and S5).These increases tend to occur in tropical Africa and can be attributed to the following process.Fire aerosols are transported from the burning (dry) to the non-burning (wet) hemisphere across the equator.A large amount of fire aerosols increases CCN, reduces precipitation, 51 and increases fire activity. 21he intra-seasonal effect occurs on the burning side toward the non-burning region, where a reduction of precipitation provides conducive conditions for more burning.Further down in the non-burning (wet) hemisphere across the equator, drying by fire aerosols in the current season enhances fires in the subsequent burning season.The coupling of the fire-aerosol feedback with the migration of precipitation is more apparent in MAM and SON, resulting in stronger fire-aerosol feedback than in DJF and JJA. Figure 3 shows a schematic diagram of the positive fire-aerosol feedback process.In DJF, the intra-seasonal burning enhancement due to fire aerosols is located at 3.3 N, which is 4.7 south to the peak of the maximum burned fraction.This enhancement from DJF continues to MAM and triggers the burning at the edge of the fire region toward the equator in MAM.The intra-seasonal fire enhancements from local burning in MAM are at 7 N and 16 S, respectively, which continue to JJA.In JJA, fire aerosols in SH Africa are transported near the surface northward and the tropical fire aerosols trigger an enhancement of burning across the equator, which also continues in SON.During the burning migration from the SH to NH in SON, fire aerosols are transported to the north and south of the fire regions, spreading out the burning feedback enhancements in both NH and SH Africa.

Limitations of the study
The overall effect of the fire aerosol feedback is to widen the region of burning.This positive feedback process simulated by the CESM-RESFire model is difficult to diagnose using the observations only.The good agreement between model simulations and the observations (see the STAR Methods section) lands confidence in the modeling results.However, the fire-aerosol-climate-ecosystem coupling is complex. 39As the interactive fire model and aerosol-climate parameterizations improve, additional feedback mechanisms may also exist.For instance, an increase in the intensity of high clouds and deep convection may result in more lightning events.In dry regions with a high potential for wildfires, lightning can play a crucial role in igniting fires. 52,53In addition, the global cooling effect of aerosols may lead to a negative feedback globally in a longer timescale due to a negative correlation between temperature and fire ignition. 535][56] This process and its effect on burning in Africa, particularly in the context of the seasonal migration of precipitation, 50 appear to be a key issue for future studies.Although the fire aerosol feedback through wind shift is insignificant in this study, analysis of the fire aerosol impacts to atmospheric circulation on different timescales is still necessary in the future.Furthermore, interactions among aerosols, lightning, and precipitation can also be important for fires in Africa. 57We did not consider the feedback from the ocean in this study.9][60][61] The timescale of the fire-ocean feedback can easily extend into the multi-decadal timescale.

Conclusions
The significant spatiotemporal burning and precipitation changes due to fire aerosols (Figures 1 and S6) indicate that the positive fire feedback is an important mechanism for the regulation of African ecosystems.The positive feedback results partly from the transport of fire aerosols from burning (dry) to wet regions, reducing precipitation and drying fuel load to enhance fires toward the non-burning (wet) region.The unique alternation of dry and wet season across the equator in Africa extends the short-term (days to weeks) aerosol forcing into longer interseasonal enhancements of burning (Figure 3).The spatiotemporal varying positive feedback of fire aerosols is manifested in the inter-seasonal and intra-annual variability of burning in Africa, increasing the intra-annual PSDs by up to 60% (Figure S1).The multi-year El Nin ˜o-Southern Oscillation (ENSO) is not simulated in the model.
The identification of the fire-aerosol positive feedback mechanism in Africa advances our knowledge of climate feedbacks related to wildfires globally.Studies have shown that in some coastal areas (such as the Mediterranean, Southeast Asia, and the western United States), fire smoke alters local fire weather, resulting in positive feedback. 37However, these regions have distinct fire seasons, and the local fire intensification caused by this feedback does not persist into the next fire season.In contrast, in Africa, the partial alignment of prevailing winds and shifting fire regions means that the positive feedback from large fire-aerosol events not only occurs within the current fire season but also amplifies burning in the subsequent season.This type of positive feedback implies that a warmer and dryer climate will likely lead to more persistent burning in Africa in the future.The systematic fire-climate feedback may also be present in other fire prone tropical regions and has significant ramifications for understanding the impacts of fires and climate change on humans and plant life.

STAR+METHODS
Detailed methods are provided in the online version of this paper and include the following:
amount above 4 km, indicating a potential model bias of aerosol wet scavenging in deep convection. 91,92We also compared the zonal mean precipitation in Africa between the model simulations and Global Precipitation Climatology Project Version 2.3 (GPCP v2.3) produced 5 under the NOAA Climate Data Record (CDR) 93 from 2001 to 2010 (Figure S12).The model distribution of the zonal mean precipitation in Africa is similar to GPCP data.The equatorial precipitation from model is overestimated by 8% -15% in DJF, MAM and SON, but the bias is within the interannual variability.In DJF when the fire aerosol effect on precipitation reduction is large, the simulation with fire feedback agrees better with GPCP than the simulation without fire aerosols.More global evaluations for the CESM-RESFire model and the evaluations with ground based AOD measurements from level 2.0 Aerosol Robotic Network (AERONET) 94 and the carbon budget with MODIS primary production products 95,96 are described in Zou et al. (2019). 39

Figure 1 .
Figure 1.Effects of fire aerosol feedback in Africa in DJF and JJA (A) Burned fraction distribution without fire aerosol feedback.(B) Distribution of burned fraction change caused by fire aerosol feedbacks; the black dashed line denotes the location of ITCZ.(C) Precipitation changes due to fire aerosol feedback (unit: mm/day).(D) Distribution of fire aerosol AOD.In Panels (B), (C), and (D), a t-test was conducted on the seasonal mean values for each gridbox between experiments with and without fire aerosol feedback.Grid boxes with statistically significant differences (p < 0.05) are cross-shaded.

Figure 2 .
Figure 2. Zonal mean distributions of fire aerosol feedbacks averaged over the African land area in DJF and JJA (A) The burned fraction without fire aerosol feedback (in red lines) and the burned fraction change from no fire aerosol feedback simulation to fire aerosol feedback simulation (in blue lines).Colored contours represent averaged northward-southward wind speed (in unit of m s À1 , positive values are northward wind).(B) The 15 W -45 E zonal mean AOD (in blue lines) and precipitation (in black lines, in unit of mm/day) change from the no-fire aerosol feedback simulation to the fire aerosol feedback simulation.The zonal standard deviation from the ensemble mean is shaded.(C)The fire PM 2.5 distribution (color shaded, in units of mg m À3 ) and the RH change (contour, in units of percentage) from the no fire aerosol feedback simulation to the fire aerosol feedback simulation.(D) The cloud liquid amount (in units of kg kg À1 ) changes from the no fire aerosol feedback simulation to the fire aerosol feedback simulation.A Student's t test was performed to compare the seasonal mean values between experiments with and without fire aerosol feedback.Areas with statistically significant changes (p = 0.05) are cross-shaded.The dashed lines in red denote the latitude of peak burned area; dashed lines in blue denote the latitude of peak burned area change, and the black dashed lines denote the latitude of the most reduced precipitation.

Figure 3 .
Figure 3.A schematic diagram for the seasonal progression of fire aerosol feedback The colors of fire icons represent burning with no fire aerosol feedback (black), burning of fire aerosol feedback in the same season (red), and burning of fire feedback across the season (blue).The size of fire icons qualitatively represents the burned fractions.The yellow dashed lines point the locations of the interseasonal feedback.

TABLE
d RESOURCE AVAILABILITY B Lead contact B Materials availability B Data and code availability d METHOD DETAILS B Model description B Model simulation setup and evaluations