Introduction

Cerrado, located in the northeastern to central part of Brazil (Fig. 1a), is a savanna region with native flora that generates a sporadic mix of woodlands, shrublands, and forests. As Brazil’s second-largest biome, covering more than 20% of the territory1,2, Cerrado holds considerable socioeconomic importance. For example, it is the origin of some major South American rivers, which are crucial for hydropower generation3. Land use change across much of the Cerrado means that it serves as a pivotal region of food production and agriculture4,5. In addition, its unique landscape with abundant biodiversity, including numerous endemic species, underscores the ecological and environmental importance of the area6,7.

Fig. 1: Observed mean characteristics of fire activity and precipitation in the Cerrado region.
figure 1

a Geographical location of Brazilian biomes (Global Forest Watch. Retrieved from https://data.globalforestwatch.org/datasets/gfw::brazil-biomes/about). The gray rectangle represents the domain displayed in (b, c). b, c Mean patterns of annual total burned area (BA) (Kha) and precipitation (mm day−1) for the period of 1982–2018. The Cerrado region is outlined by the solid green line, with the dashed line indicating the latitude of 16°S. d, e Monthly climatology of BA (Kha) and precipitation (mm day-1) averaged over the northern Cerrado region (>16°S), with error bars indicating one standard deviation from the mean. The fire season (August–October) highlighted is highlighted by gray shading.

Full size image

Fires play a significant role in Cerrado’s ecology, impacting the carbon balance through enhanced turnover of living vegetation and dead organic pools, and by selecting for fire-adapted endemic species. The fire characteristics in this region have been shaped by the clear seasonality of precipitation, with a pronounced dry season from April to September (Fig. 1e). The timing, duration, and intensity of the wet season is governed by the migration of the South Atlantic Convergence Zone (SACZ) in conjunction with the South American Monsoon System (SAMS)8,9,10. Dry season fires are an important driver of Land-Use and Land-Cover Change (LULCC). Planned fires have been used to clear vegetation to create soybean farmland, particularly in northern Cerrado2,11. Natural fires have also actively occurred even before these agricultural activities12, ignited by lightning at the beginning of or during the wet season, which typically runs from October to March13,14,15. The combination of these factors causes Cerrado to exhibit a large and frequent fire regime16,17 with a strong seasonality, where the primary area of activity is north of 16°S (Fig. 1b). In this region, the annual peak of burned area (BA) occurs in September (Fig. 1d), where fires in September accounts for an average of ~40% of the annual total BA, with its year-to-year variations between 30% and 57% for 1982–2018. The fire season, when defined as the three-month period from August to October, accounts for ~81% of the annual total BA, with its year-to-year variations between 70% and 87%.

The interannual variation in fire activity is inherently linked to climate variability. While human-planned fires can be ignited regardless of climate conditions, when the vegetation and litter are dry, fires more readily escalate into large fires. Here, one might suspect anthropogenic land use conversions as the primary driver of interannual variability of fires, which involve vegetation clearing for agricultural purposes5,11,18, given the significant land cover changes in Cerrado over the past decades. However, a previous study indicates that fires in the Cerrado predominantly occur in native vegetation rather than on areas altered by anthropogenic land use18. Furthermore, the influence of land use change on fire probability appears to be limited in Cerrado, especially when compared to the Amazon19. Therefore, it becomes crucial to explore and understand the climatic factors that influence the interannual variability of fires in Cerrado.

To date, considerable efforts have been made to understand the influence of climate variabilities on fire activities in Brazil; however, many of these efforts have focused on the Amazon. Specifically, drought conditions induced by El Niño–Southern Oscillation (ENSO)20,21,22,23 have been associated with increased fire activity in Amazonia20,24,25,26. There is also evidence for a positive relationship between sea surface temperature anomalies (SSTAs) in the North Atlantic and the Amazonian fires20,25,27. This widely acknowledged impact of ENSO on Amazonia fire activity, coupled with the additional influence from the Atlantic SSTA, has shaped the narrative surrounding the climatic drivers of fire events in tropical South America. However, as Cerrado is clearly distinguished from the Amazon in terms of the landscape, climatology, and fire behavior, it might exhibit a different relationship with large-scale climate variability. Indeed, while the Amazon generally experiences small and infrequent fires due to rare anthropogenic activity and high moisture, Cerrado is exposed to a considerably drier environment throughout the year (Fig. 1c), hence, experiences more frequent and larger fires17. This strongly suggests a need for a separate examination of Cerrado fires from the Amazon rainforest.

However, the impact of such large-scale climate variability on the Cerrado fire activity (namely, teleconnection patterns) has received far less attention, despite the Cerrado having a much larger fire-affected area than Amazonia. It is necessary to note here that tropical Pacific and Atlantic SST warming has been widely linked to drought conditions around the Cerrado region, through modulations of the subsiding branch of Walker or Hadley cell28,29,30,31,32,33. Nevertheless, few studies have directly explored whether or how such impacts of SSTA-induced drought extend to fire activity20,34. This lack of relevant research has contributed to a significant gap in our understanding of the interannual variability of Cerrado fires. In addition, previous studies linking SST variabilities and Cerrado fires have relied on a relatively short observation period (13 years at most), precluding the establishment of a robust relationship between Cerrado fires and large-scale climatic variability. In an effort to fill this knowledge gap, we examined large-scale climatic factors that affect fire activity in Brazilian Cerrado, utilizing the 36-year data of global BA products derived from satellite observations and meteorological observation/reanalysis data (1982–2018 except 1994; see Methods). Our results demonstrate that the SST variability in the North Tropical Atlantic (NTA) is an important factor affecting the interannual variability of Cerrado fires, distinguished from the effect of ENSO on the Amazon. We also highlight the role of atmospheric circulation patterns and land–atmosphere interactions in bridging the Atlantic SSTA variability and fire activity in Cerrado.

Results

NTA as a primary climatic factor linked to Cerrado fires

The regression analysis of SSTAs on the BA over the northern Cerrado during the August–October (ASO) season (BA_NC; Methods) shown in Fig. 2 revealed a significant and spatially organized signal over the tropical North Atlantic, particularly in the January–March (JFM) season. This region largely overlaps with the region recognized for its interannual SST variability, referred to as NTA variability (hereafter, ‘NTA’)35,36,37,38. NTA is often regarded as the northern component of the Atlantic Meridional Mode (AMM), which is characterized by the SSTA contrast between the northern and southern hemispheres near the intertropical convergence zone (ITCZ)39,40,41,42,43. In line with this, in the April–June (AMJ) season, a negative relationship also emerges south of the equator, near the coast of Namibia and Angola, leading to a dipole SSTA structure throughout the tropical North Atlantic. In the subsequent season [July–September (JAS)], the remnants of this SSTA in the Atlantic are still visible. Overall, it indicates that a positive phase of NTA is related to the higher fire activity during the fire season. Additional analysis using the annual total BA instead of that for the fire season showed qualitatively similar results (not shown).

Fig. 2: Regression maps of sea surface temperature anomalies on the fire-season Cerrado BA for each season.
figure 2

a January to March (JFM), b April to June (AMJ), and c July to September (JAS). The fire season is defined as August to October (ASO). Diagonal lines indicate regions that are statistically significant at the 95% confidence interval. The green shading over the South American continent represents Cerrado north of 16°S, over which the annual BA was aggregated and used as a predictor in the regression analysis. The Niño3.4 and NTA regions are indicated by green boxes in (a).

Full size image

Beyond the Atlantic, El Niño-like SSTA warming in the equatorial Pacific, as well as basin-wide warming in the Indian Ocean, tends to occur from January to June in large-fire years. The statistical significance of these signals appears weaker or absent compared to the Atlantic, implying the importance of the Atlantic’s role. However, complex interplays exist between SSTA variabilities in different ocean basins, necessitating further analysis to elucidate the underlying causal relationships. Particularly, it is well-known that strong El Niño events are often followed by warm NTA in the ensuing spring44,45,46,47,48,49,50 (Supplementary Fig. 1). This recurring sequence of El Niño and NTA warming complicates the understanding of the fundamental drivers influencing Cerrado fires. Therefore, it is necessary to separately examine the individual effects of El Niño and NTA in order to better discern their respective roles.

Multiple linear regression of detrended fire-season BA to SSTAs in the tropical Atlantic (the NTA index; linearly detrended SSTA averaged over 5°N–20°N, 20°–60°W for the JFM season)51 and Pacific (the Niño3.4 index; linearly detrended SSTA averaged over 5°S–5°N, 170°W–120°W for the DJF season; Methods) indicates that variability in fire activity in the Cerrado is more sensitive to tropical Atlantic SSTA than to ENSO, particularly in the north (>16°S), where fire activity is concentrated (Fig. 3a, b). This highlights the need to consider the role of NTA, in addition to the widely acknowledged impact of ENSO, in understanding the dynamics of fires in the Cerrado region.

Fig. 3: Comparison of NTA–Cerrado BA relationship and ENSO–Cerrado BA relationship.
figure 3

a, b Multiple regression coefficients of the JFM NTA and DJF Niño3.4 indices on the gridded annual BA. c, d Scatterplots of fire-season BA summed over northern Cerrado (BA_NC) against the JFM NTA index (c) and DJF Niño3.4 index (d), shown in anomalies. Also shown are Pearson’s correlation coefficients, with asterisk marks (**) indicating a statistically significant result at the 95% confidence level.

Full size image

The more pronounced relationship of Cerrado fire activity with tropical Atlantic SSTA than with ENSO is also qualitatively consistent when considering the spatially aggregated BA over the northern Cerrado region. In Fig. 3c, d, detrended anomalies for the northern Cerrado BA for the fire season are compared with the NTA and Niño3.4 index, respectively. The ASO Cerrado BA exhibits a statistically significant positive relationship with the JFM NTA index at the 95% confidence level (r = 0.37 for 36 years of data). That is, Cerrado tends to experience high fire activity in years with a positive NTA phase. However, there are also some noticeable cases where the above-normal BA in Cerrado coincided with years of the anomalously cool NTA region (the 2nd quadrant); They were identified as 1985, 2012, 2014, and 2015, respectively. Given the complexity of fire dynamics, it is not surprising that such cases might arise under the influence of diverse climatic and/or environmental factors. Considering the distribution of these instances and the significant correlation found in Fig. 3c, it suggests that fire activity in Cerrado is largely proportional to temperature anomalies of NTA, intermittently interrupted by other factors. On the other hand, the BA_NC and DJF Niño3.4 yield a correlation coefficient of r = 0.19, failing to reject the null hypothesis (H0: r = 0), and the distribution is more dispersed (Fig. 2d).

We further consider partial correlations to investigate the independent effects of each phenomenon. When linear effects from ENSO are removed, the Cerrado BA–NTA relationship yields a correlation of 0.33, which is still statistically significant at the 90% confidence level (p = 0.049). On the other hand, without the NTA effects, the correlation between BA_NC and Niño3.4 drops to 0.05 (p = 0.76). These findings robustly indicate NTA as the primary climate variability that has a closer relationship than ENSO for understanding the interannual variation of Cerrado fire activity.

Physical mechanism of the NTA–Cerrado fire relationship

To unravel the physical mechanism by which NTA is related to Cerrado fire activities, it is necessary to first explore the seasonality in local meteorological variables involved in fire dynamics. The fire activity in the fire season can be influenced by meteorological conditions that are established several months in advance. For instance, negative precipitation anomalies (‘Pr’ in Fig. 4a) in Cerrado during not only the peak fire month in September but also the dry season (JJA) correspond to enhanced fire activity. Other variables indicating aridity, such as increased potential evapotranspiration (‘PET’; the maximum possible evaporation, where its higher value is associated with high atmospheric demand for moisture52,53,54) or reduced relative humidity (‘RH’), also present significant relationships starting from July–August, preceding the fire season. That is, the formation of fire-favorable conditions during dry periods is important for shaping the overall annual fire activity.

Fig. 4: Seasonal characteristics illustrating physical processes of NTA–Cerrado fire relationship.
figure 4

a Correlations between the fire-season BA_NC and various variables for each season. PET: evapotranspiration, Tsfc: surface temperature, RH: relative humidity, SM: soil moisture, Pr: precipitation. The y-axis of variables associated with humid conditions are reversed (green lines and the right axis) so that the upper part of the panel indicates a fire-favorable condition. b Correlations of BA_NC for each season with the JFM NTA index, and with DJF Niño3.4 index are shown respectively. In (a, b) the center month of each season is denoted on the x-axis, where each season is defined as the average of adjacent 3 months. The statistically significant values at 90% confidence are highlighted by thick lines and the gray shading highlights the peak fire season.

Full size image

The key factor that allows JJA drought to affect the fire season is the long memory of soil moisture across different seasons (Supplementary Fig. 2), further sustained through positive soil moisture–precipitation feedback55,56. When below-normal precipitation occurs during the dry season in JJA, the moisture deficit in soil layers inhibits surface evapotranspiration until the subsequent season. This reduces atmospheric moisture content, elongating the initial drought conditions (Supplementary Fig. 3). At the same time, the decreased latent heat flux associated with less evapotranspiration can maintain the surface temperature to be higher than normal conditions (‘Tsfc’ in Fig. 4a). This can further exacerbate the vapor pressure deficit due to higher atmospheric moisture demand according to the Clausius–Clapeyron relation. These coupled atmospheric–land–vegetation processes, persisting from the dry period to the fire season, enhance the warm and moisture-depleted (i.e., fire-favorable) conditions. This is portrayed in Fig. 4 by significant positive correlations of variables indicating dryness [reversed RH, Pr, soil moisture (‘SM’), and PET] and warmth (Tsfc) with the ASO BA.

Meanwhile, the seasonally-decomposed NTA SSTA exhibits its maximum correlation with ASO BA_NC during the JFM period (Fig. 4a). This indicates a delayed relationship between JFM NTA and fire-season fire activity. Indeed, when examining BA outside the peak fire season of ASO, JFM NTA is significantly correlated with BA in April to October, which encompasses the peak fire season (Fig. 4b). It is noteworthy that although the influence of JFM NTA extends into the fire season, there is no significant correlation with JFM BA_NC, despite being temporally closer (which will be further discussed later).

The processes through which JFM NTA leads precipitation and humidity deficits over the Cerrado region months later can be illustrated as follows. First, over the North Atlantic, through the oceanic thermal inertia and wind–evaporation–sea surface temperature (WES) feedback39,57,58, the SSTA associated with JFM NTA tends to last in the same sign until late in the year (Fig. 5a), including the boreal summer (JJA) during which the ITCZ typically resides in the northern hemisphere. Anomalous warming in the NTA intensifies this climatological pattern, with the ascending branch of the ITCZ anchored to the warm SSTA40. Accordingly, the downward motion in the southern hemisphere is strengthened over the Cerrado region (Fig. 5b). In alignment with these changes in the vertical motion, the local convective activity in Cerrado is reduced, as indicated by positive outgoing longwave radiation (OLR) anomalies (Fig. 5c). The suppressed convection brings a more arid condition there21,29, leading to higher fire danger around September through the atmospheric–land–vegetation coupled process illustrated above.

Fig. 5: Persistence of NTA SSTA and its effects on the JJA convection in Cerrado.
figure 5

a Composite of seasonal evolution of the NTA index. Individual years with positive NTA are displayed by thin pink lines, with their ensemble mean shown by a thick line. Dots indicate statistically significant results based on a bootstrap test (N = 5000). Similarly, negative NTA years are displayed in blue color. b, c Composite patterns related to the atmospheric convective activity for the JJA season in positive NTA years (pNTA). b Latitude-pressure cross-section (over 60°W-40°W) illustrating atmospheric meridional overturning circulation (MOC) by vectors. The x- and y-components of vectors correspond to atmospheric northward (‘v’, in m s-1) and vertical (‘\(\omega\)’, in Pa s-1) wind, respectively (obtained from ERA5). Originally, positive ω means a downward motion, but for visual clarity, it was rescaled and reversed by multiplying −100 so that positive indicates an upward motion. The gray bar at the bottom denotes the latitudinal range of Cerrado (darker color for northern Cerrado). c NOAA outgoing longwave radiation (in W m-2). d, e Same as (b, c) but for negative NTA years (nNTA). Statistically significant results identified from bootstrapping (N = 5000) are highlighted by colored arrows (blue for ascent, orange for descent) and hatches, respectively.

Full size image

For years characterized by negative NTA, the results generally depict mirror images, with a weakening of the climatological MOC and increased convection in Cerrado (Fig. 5d, e). However, detailed features are not symmetric, as shown by the larger magnitude of changes in the Hadley cell for negative NTA years, especially in its southern branch, which lies close to Cerrado (Fig. 5b, d). Considering that positive NTA events tend to occur with a larger amplitude (Fig. 3c), this may seem counterintuitive. A detailed investigation into this nonlinear feature is left for future research topics. However, it could be related to inherent asymmetry in the decay timescale of NTA SSTA (i.e., how persistently the JFM SSTAs are maintained until the JJA period); In the composite anomalies depicting the time evolution of the NTA index, negative NTA years exhibit a larger amplitude in the JJA season, contrary to the JFM period (Fig. 5a).

Discussions

Before drawing final conclusions, there are a few points that may require further discussion and consideration. First, it should be mentioned that our findings are largely based on linear analysis methods, whereas fire dynamics inherently involve nonlinear processes. Although the monthly and spatial averaging used to produce the BA_NC index effectively reduces the nonlinearity in its distribution (e.g., the scatter of BA_NC in Fig. 3 shows no pronounced outliers), allowing for the application of linear analysis, it fails to capture the nonlinear interactions between fire and climate phenomena. Nevertheless, the observed consistency among the various physical variables (throughout Figs. 4, 5) supports a robust NTA–Cerrado BA relationship within the scope of linear processes.

It is important to clarify why the influence of the NTA SSTA on the northern Cerrado BA is not observed during the JFM season but only emerges in the JJA season, as shown by the lack of a significant simultaneous relationship in Fig. 4b. This can be attributed to the subtle interaction between interannual variability in SSTA and the seasonally-evolving South American monsoonal system. The anomalous atmospheric circulation, such as the strengthening of the upward (downward) motion at the northern (southern) side of the ITCZ, is induced by NTA warming not only during the JJA season, but also during the JFM season (not shown). However, during this season, an intense rainband associated with the SACZ extends from the northwest to southeast of the Cerrado (ref. 2 and Supplementary Fig. 4). The presence of such a strong monsoon system makes it more favorable for the sinking flow to occur over the ocean surface rather than over the South American continent (Supplementary Fig. 5). In other words, during the JFM season, the well-established convective system over Cerrado regulates fire weather in Cerrado more dominantly than the remote influences from NTA. Once the SACZ diminishes over the Cerrado region after a few months, the impact of NTA on BA_NC becomes evident.

Another aspect that needs to be considered is the role of other major SSTA variabilities, including ENSO and the Atlantic Multidecadal Oscillation (AMO)59, which may influence annual fire activity in Cerrado. While the impact of ENSO on Cerrado fires is not as pronounced as that of NTA (Fig. 3), ENSO can exert an indirect effect. It is well-known that strong El Niño events are often followed by warm NTA in the ensuing spring. This phenomenon is attributed to the weakening of entrainment and surface heat losses in the tropical North Atlantic caused by ENSO-induced tropical (e.g., displacement of the Walker circulation, excitement of the Gill-type response, and vertical stability change through tropospheric warming) and extratropical (e.g., the Pacific-North America teleconnection pattern) processes44,45,46,47,48,49,50. Therefore, the influence of ENSO, which triggers NTA events, may implicitly contribute to the annual total BA.

For AMO, which is characterized by long-term fluctuations in SSTA across the entire North Atlantic (0°–70°N), its positive phase has been associated with enhanced fire activity in the wide area of northeast South America20. Narrowing the scope to our target region, the annual mean AMO index indeed exhibited a positive relationship with the annual BA; however, the p-value of its correlation coefficient did not exceed the 90% confidence threshold (r = 0.26). Despite the fact that both AMO and NTA are indicators of North Atlantic SSTA variability, a difference is observed in their impacts on Cerrado fires. This could be attributed to the geographic distance from the Cerrado region, as well as the much longer characteristic timescale of AMO, which makes it less suitable for explaining interannual variability in Cerrado BA. Taken together, NTA is considered the primary factor that affects fire activity in Cerrado, with ENSO potentially playing an indirect role by modulating the NTA.

Conclusion

Our results robustly support a physical relationship between fires in northeastern Brazil and SSTA over the tropical North Atlantic Ocean. Specifically, warm SSTA maturing early in the year is linked to above-normal fire activity in Brazilian Cerrado during the fire season, which predominantly shapes the annual total BA (Supplementary Fig. 6). This relationship involves both spatial (the tropical Atlantic to Cerrado) and temporal gaps (JFM SSTA to ASO fires). The former, the remote influence from the North Atlantic to South America, is enabled by atmospheric meridional overturning circulation that involves cross-equatorial flows. The temporal lag can be understood as the interaction between NTA SSTA and the seasonal evolution of the ITCZ and SACZ. The warm NTA SSTA pulls the ITCZ northward during the JJA season, augmenting its climatological migration pattern. At the same time, the SACZ over the Cerrado region diminishes, allowing the intrusion of the subsiding branch into the Cerrado region. This, in turn, causes anomalous drought conditions, which are prolonged until the fire season through the positive feedback between atmospheric, land, and vegetation moisture. Consequently, fire activity in Cerrado tends to be higher than usual during years with a warm phase of NTA.

The findings of this study offer valuable insights with practical implications. Given that the NTA-associated SSTA peaks months ahead of the typical fire season in the Cerrado region, monitoring NTA SSTA can serve as a useful indicator of fire activity in the ensuing fire season. This will improve preparedness for fire risks and aid in minimizing socio-environmental losses in the region. Furthermore, elucidating the physical mechanisms, involving the dynamics of the ITCZ, SACZ, and land–atmosphere feedback processes, contributes to a deeper scientific understanding of how large-scale climate variabilities influence regional fire activities. This underscores the role of the teleconnections from climate variabilities in fire dynamics, which could potentially benefit understanding fire variability in other fire-prone regions.

As the next step to enhance predictability, it is necessary to elaborate on the physical understanding by considering the following factors. First, climate phenomena inherently exhibit oscillations on multi-decadal timescales. For example, it has been suggested that since 1990, the internal variability of the Atlantic has enhanced, affecting the interaction between the ENSO and NTA (e.g., their lead–lag relationship)43. Such changes can further influence direct and indirect pathways through which SST variabilities modulate Cerrado fire activity. In addition, anthropogenic forcing has modified/will modify the internal variability of the climate system in both the Atlantic60,61,62 and Pacific63,64,65. Therefore, it is essential to consider the climate system’s non-stationarity to improve the understanding of Cerrado fire dynamics. Continuous efforts to accumulate observational data and the utilization of climate models will greatly benefit this.

Methods

Observational datasets

We used the global long-term BA products from Advanced Very High Resolution Radiometer (AVHRR) images66, spanning from 1982 to 2018. Since the year 1994 was omitted due to insufficient input data for generating the BA product that year, the available data length is 36 years in total. This dataset represents the longest available record compared to similar datasets, such as the Moderate Resolution Imaging Spectroradiometer (MODIS) Fire_cci product, which began in 2000. Given that our study requires a sufficient number of ENSO and NTA events during the analysis period for more robust results, we opted for the AVHRR product. The spatial resolution of the data is 0.05°.

For precipitation in the main results, we used the data from the Climate Research Unit, University of East Anglia (CRU)67. The CRU precipitation is gridded data at 0.5° resolution that covers the global land domains (except Antarctica) by interpolating station observations. In Supplementary Figs. 4, 5, precipitation patterns are presented using the 2.5°\(\times\)2.5° CPC Merged Analysis of Precipitation (CMAP) dataset68, which covers both land and ocean. The CMAP precipitation is produced from gauge observations and satellite products. The Hadley Centre Global Sea Ice and Sea Surface Temperature (HadISST)69 data was used to investigate SST patterns correlated with the Cerrado BA. The Niño3.4 index was downloaded from NOAA (https://psl.noaa.gov/gcos_wgsp/Timeseries/Nino34/). The data from the fifth-generation European Centre for Medium-Range Weather Forecasts (ECMWF) Atmospheric Reanalysis (ERA5)70 were used for relative humidity, surface temperature, and atmospheric wind velocities at the 1°\(\times\)1° resolution. For soil moisture, a satellite-based product from the European Space Agency’s (ESA) Soil Moisture Climate Change Initiative (CCI) project was utilized71,72,73. The OLR data are from the National Oceanic and Atmospheric Administration (NOAA) satellites, provided at the 2.5°\(\times\)2.5° resolution74.

The horizontal resolution varies among different datasets used in this study. Nevertheless, we note that the qualitative results appear to be largely insensitive to the resolution. For example, when comparing precipitation from CRU (0.5°), ERA5 (1°), and CMAP (2.5°) datasets, they all exhibited negative precipitation anomalies over the Cerrado region associated with increased fire-season fire activity in northern Cerrado (Supplementary Fig. 7). Similarly, a comparison was made for the OLR composite anomalies during positive and negative NTA years, using NOAA (2°) and ERA5 (1°) datasets. Both datasets exhibited a clear contrast in OLR anomalies depending on the phase of NTA, with a reduced convective activity in positive NTA years and enhanced activity in negative NTA years (Supplementary Fig. 8).

To focus on the role of internal climatic variability, linear detrending was applied to anomalies of the variables.

Index definitions

The burned area index for the northern Cerrado region (BA_NC) is defined as the sum of BA over the northern Cerrado region (>16°S). The region was selected based on its role as the epicenter of fire activity, accounting for on average 87% of the annual total BA in the Cerrado region, with its year-to-year variations from 70% to 94% during the satellite observation period (Fig. 1b). The time series of the fire-season total BA_NC index is shown in Supplementary Fig. 9.

The tropical Atlantic SSTA variability was quantified by the NTA index (linearly detrended SSTA averaged over 5°N–20°N, 20°–60°W)34. The reason why NTA was preferred over other SSTA variabilities in the Atlantic [such as Benguela Niño (the south tropical Atlantic region that exhibits significant negative signals in Fig. 2b) and AMM] was because it exhibits a longer lead time against BA_NC, making it more useful as a predictor (Supplementary Fig. 10). The NTA index was defined for the January-to-March season, during which SSTA in the NTA region exhibits pronounced positive correlations with the fire-season Cerrado BA (Fig. 4a and Supplementary Fig. 11). This season also roughly corresponds to the peak season of the NTA SSTA, which is known as February to April38,75.

For the tropical Pacific SSTA variability, the Niño3.4 index (5°S–5°N, 170°W–120°W) was used, with SSTA linearly detrended before areal-averaging. The Niño3.4 index was defined for the ENSO mature season (D[-1]JF, where [-1] denotes the previous year).

Multiple linear regression analysis

$${{\rm{BA}}}_{{ij}}(t)={\alpha }_{{ij}}\cdot {\rm{NTA}}(t)+{\beta }_{{ij}}\cdot {\rm{Ni}}\widetilde{{\rm{n}}}{\rm{o}}3.4(t)+{\varepsilon }_{{ij}}$$
(1)

Multiple regression coefficients, \(\alpha\) and \(\beta\), are obtained from Eq. (1) and presented in Fig. 3. In the equation, ij denotes each grid point. BA is the annual sum, while the NTA and Niño3.4 indices are for the JFM and DJF seasons, with their center month falling in the same year as the BA. The NTA and Niño3.4 indices are normalized by their respective standard deviations so that the regression parameters have a unit of BA (Kha). This is to mitigate the potential bias where the parameter fitting is primarily influenced by the large amplitude of the Niño3.4 index.

There may be a concern that this normalization process could artificially exaggerate the NTA’s effect while undermining that of ENSO. To address this uncertainty, we re-examined the independent impacts of NTA and ENSO based on composite analysis. The results were qualitatively consistent (Supplementary Fig. 12): The BA in the Cerrado region tends to be large during years with positive NTA, even without El Niño events. On the other hand, the fire activity is less pronounced during the years of El \({\rm{Ni}}\widetilde{{\rm{n}}}{\rm{o}}\) only.