Introduction

Extreme weather events have become more frequent, longer, and intense due to anthropogenic global warming1. These events can occur independently from each other but also simultaneously (compound events) or sequentially (cascading hazards)2 amplifying their overall effects3,4. Drought conditions, for example, can exacerbate heatwaves, increasing fire risk5. Vegetation fires, in turn, can contribute notably to local, regional6,7,8, and global air pollution. While single impacts of isolated weather-driven natural extreme events are well understood, a significant gap remains in our knowledge of how these impacts accumulate when combined so that considering such events independently can lead to misestimating the magnitude, intensity, and duration of the involved processes and their actual risk9. Therefore, understanding the complex relationships and feedback mechanisms between extreme events is crucial for risk assessment, which requires integrated approaches considering the interconnected nature of these hazards10.

In recent years, substantial efforts have been made towards understanding compound events and cascading hazards caused by the combination of several drivers. During Australia’s 2019/2020 fire season, for example, moderate rain and floods following extreme drought and subsequent wildfires had their impacts amplified due to increased surface runoff, leading to reductions in water quality and rising ash and soil erosion in the country’s east coast9. In the United States, the record-breaking 2021 heatwave was associated with a preexisting drought that exacerbated the escalation of temperatures11. In western Russia, land-atmosphere interactions intensified drought conditions, promoting the advection of warm and dry air, developing the heatwave downwind, and resulting in cascading impacts in 2010 due to a coupled lack of rainfall and enhanced evaporative demand12.

Climate-related extreme hazards pose severe threats to natural and human systems, including infrastructure, livelihoods, ecosystems, and human health, thereby contributing to societal and environmental risks13. For instance, in Moscow, over 2000 excess deaths were attributed to the interaction between high temperatures and air pollution from wildfires during the 2010 Russian heatwave14. In Brazil, studies have shown a positive association between exposure to drought15 and heat16 with increased mortality, particularly affecting females, children, and older adults. Regarding extreme heat, 37.0% of warm-season heat-related deaths globally can be attributed to current human-induced climate change, reaching 60% in Brazil17. In the Amazon region, drought and vegetation fires have increased respiratory disease hospitalizations, with compounded effects significantly impacting older adults and children18. Additionally, a projected increase in warm-season mortality from the 2000s to the 2090 s is expected across various climate zones under different climate change scenarios, shifting the mortality peak from cold to warm seasons in arid, temperate, and continental zones19.

Although the analysis of compound events and cascading hazards is poorly documented in South America (SA)20,21, recent global studies point out the continent as a current and future hotspot for these events22,23. A recent showcase was the severe 2019–2022 outstanding drought over central-east SA coincident with several heatwaves24,25. Notably, in 2020, compound drought and heatwave (CDHW) conditions triggered the worst fire season in the last two decades over the Pantanal (Fig. 1), the largest contiguous wetland in the world21. Land-atmosphere feedback contributed decisively to the unprecedented wildfires registered in the region, with approximately 70% of the burned area that year attributed to CDHW events26. The Pantanal 2020 fire season (P20F) accounted for more than 3.9 million ha burned, an area four times larger than the long-term average27,28, with several impacts on the ecosystem29,30, hydrological cycle31, the economy32, and on the COVID19-associated hospitalizations33,34. Despite the notable advance in understanding the local effects of the P20F-related CDHW-fires, multi-hazard evaluations still need to be addressed; therefore, the widespread knock-on effects still need to be explored.

Fig. 1: Map of the study area.
figure 1

a Geographical location of São Paulo state (SPS) and Brazilian Pantanal, (b) scars from the first and second peaks of burned area in 2020 over the Pantanal (BAP1 and BAP2), and (c) the location of CETESB monitoring stations. The SPS is divided into 15 mesoregions labeled according to their acronyms. Map created using the Free and Open Source QGIS.

Full size image

Here, we investigated the cascading chain of hazards associated with the CDHW-fire events during the P20F beyond the local perspective by linking the smoke transport, the worsening air quality, and the health impacts in the São Paulo state (SPS) in southeastern Brazil (Fig. 1), which includes the Metropolitan Area of São Paulo (MASP), the largest SA megacity located approximately 1.500 km from Pantanal with over 21 million inhabitants (according to Brazilian 2022 census). Moreover, local weather and environmental conditions over the SPS were investigated as interconnected processes that could amplify/reduce such impacts. We combine different data sets, satellite-derived fire products, and atmospheric models, providing multiple lines of evidence of the long-range cascading impacts of the CDHW-fire events from the P20F on the southeast region of SA.

Results

Cascade chain: a storyline for the unprecedented Pantanal Fires in 2020

The year 2020 was the driest in the Pantanal since at least 198035,36, which took place under an outstanding drought that lasted from 2019 to 202225 depicted by the most significant negative peaks in the monthly SPEI-6 time-series (Fig. S1). These prolonged drought conditions resulted from changes in the Walker and Hadley Cells’ circulation and the establishment of a Rossby wave extending from the west South Pacific towards South America25, associated with the rare consecutive three La Niña events37,38. In the months preceding the fire season, large precipitation deficits were observed, close to zero for most days (Fig. S2a), leading to record-breaking (the lowest value since 1951) negative soil moisture anomalies26. Notably, the 2020 drought was simultaneously marked by unprecedented extreme hot conditions, which favored land-atmosphere feedback, amplifying fire episodes21,26,39,40. Three consecutive heatwaves occurred between August and October, with temperature anomalies reaching 6 °C (Fig. S2b). According to Libonati et al. 21, two distinct mechanisms have fueled fire occurrence in a cascading effect during the short span of these CDHW events. First, unprecedented long-term deficits in precipitation and significant evaporation rates dry out the soil and vegetation, reducing flood pulse and fueling fires. Second, increased sensible heat surface-atmosphere flux and soil desiccation establish a water-limited regime, boosting the concurrence of extreme heat and rising flammability thresholds26. Although CDHW conditions last for only 37% of the fire season, they accounted for 71% of the burned area in the Brazilian Pantanal21.

Two significant peaks of BA were identified during the CDHW period (Fig. S2c). The first BA peak (hereafter BAP1) occurred between September 9th and 17th (9 days), accounting for 618.6 kha, and the second (hereafter BAP2) between September 25th and October 11th (17 days) and reaching 887.6 kha (Table 1). In this context of CDHW-fire conditions, the 2020 fire season was unprecedented, with monthly BA exceeding the historical average by up to 5 times (Fig. S2d). Notably, BAP1 and BAP2 presented a temporal match with intense heatwave periods (Fig. S2b) from 5 to 20 September and between 25 September and 15 October. Burned area during the BAP1 was concentrated in the Pantanal of the Mato Grosso state, while occurrences in BAP2 were concentrated in the west and central north of the biome (Fig. 1b). A brief description of BAP1 and BAP2 is presented in Table 1.

Table 1 Summary of relevant information about the peaks of burned area in the Brazilian Pantanal, transport episodes, and smoke-induced PM2.5 peaks in the SPS
Full size table

Large-scale cascading hazards: from local to long-range effects

We investigated the smoke transport from the CDHW-fire events during the P20F to the SPS through HYSPLIT trajectory simulations (Fig. 2) during the BAP1 and BAP2 periods. Overall, air mass trajectories from the Brazilian Pantanal were advected towards the south of the country during the analyzed periods. However, three noteworthy episodes of smoke transport were identified in the southeastern direction reaching SPS. The first episode (T1) occurred at the end of BAP1 (September 17), as confirmed by both forward (Fig. 2a) and backward (Fig. 2d) trajectories starting in the Pantanal and ending in the MASP. In particular, backward trajectories above the atmospheric boundary layer indicate a long-range transport pattern. The second episode (T2) occurred during the BAP2 (September 27), as observed for forward (Fig. 2b) and backward (Fig. 2e) trajectories. The third episode (T3), also during BAP2, occurred on October 7, reaching the SPS on October 9 (Fig. 2c, f). A brief description of T1, T2, and T3 is presented in Table 1.

Fig. 2: Airmass trajectories and long-range transport of P20F-related smoke.
figure 2

HYSPLIT 48-h air-mass trajectories during the three episodes of transport of smoke from the Brazilian Pantanal to the SPS (T1, T2, and T3), including forward trajectories starting in the Pantanal at 1800 UTC on (a) September 17, (b) September 27, (c) October 7, and backward trajectories ending in the Metropolitan Area of São Paulo (MASP) at 1800 UTC on (d) September 18, (e) September 29, and (f) October 9, respectively.

Full size image

During T1, T2, and T2 episodes, EURAD-IM model simulations (Fig. 3) showed significant contributions from the Pantanal emissions to the PM2.5 levels in large areas of the SPS both at the surface level and at 850 hPa. The role of the SA Low-Level Jet (SALLJ)41 in transporting the smoke plumes from the Pantanal is evident from the wind patterns at 850 hPa (Fig. 3d–f). EURAD-IM model quality indicators are shown in Tables S1 and S2.

Fig. 3: Contributions of the Pantanal fires to air pollution in the SPS.
figure 3

Increment from GFAS Pantanal in PM2.5 concentrations modeled with the EURAD-IM on September 19 (T1), September 29 (T2), and October 9 (T3) at surface level (a, b, and c, respectively) and 850hPa (d, e, and f, respectively).

Full size image

PM2.5 simulations showed high concentrations in regions with extensive fire activity during T1, T2, and T3 episodes (Figs. S3, S4, and S5), including the Brazilian Pantanal, Bolivia, northern Argentina, and SPS. Particularly at the SPS, it can be attributed to the high number of active fires in the state (Fig. S6) in August, September, and October, reaching 5591, 14,567, and 6879 fires, respectively (compared to <500 fire counts in the other 2020 months). Relative to the monthly climatology, the total fire count in September and October 2020 was approximately twice the average for these months from 2012 to 2023 (excluding 2020) (Fig. S6b). In particular, September 13th and 14th were the top fire days, with 2030 and 1335 fires, respectively (Fig. S6a). However, a sharp decrease in the fire counts was observed in the following days, so the contribution of local fires during the T1 is expected to be small. Despite being lower, peaks of local fire activity were also observed near T2 (1295 on October 1st) and T3 (1280 on October 7th). Therefore, although the significant influence of smoke from CDHW-fire events during the P20F on PM2.5 levels over SPS during transport episodes, the contribution of the local fire emissions cannot be neglected.

In 2020, the highest PM2.5 concentrations in the SPS were observed during the dry months (May to October), surpassing both the Brazilian 24-hour air quality guidelines established by the Conselho Nacional do Meio Ambiente (CONAMA) (25 μg/m3) and the World Health Organisation (WHO) standards (15 μg/m3 in 24 h), on most days (Fig. 4). In particular, notable PM2.5 peaks concurrent with the T1, T2, and T3 were observed in most state mesoregions. The first PM2.5 peak (PMP1) occurred on September 18th–19th, encompassing all mesoregions. The sharp rise in concentrations simultaneously with the T1 indicates a substantial contribution of smoke transport from the Brazilian Pantanal during the BAP1 period to the PMP1, corroborating the results of the EURAD-model simulations. In particular, daily averages reached 70 μg/m3 in the MASP (in São Caetano do Sul station) and 115 μg/m3 in Ribeirão Preto. Despite the high PM2.5 levels, PMP1 was a short-term peak, with low concentrations (<10 μg/m3) from September 20th.

Fig. 4: Ground-level PM2.5 concentrations over the SPS in 2020.
figure 4

Daily PM2.5 mean concentration [μg/m3] from CETESB monitoring stations in the mesoregions of São José do Rio Preto (SJP), Ribeirão Preto (RP), Piracicaba (PI), Macro Metropolitana Paulista (MMP) and Metropolitana de São Paulo (MASP). For the PI and MASP mesoregions, the multi-site mean and standard deviation are presented. The red dotted lines represent a threshold of 30 μg/m3 (two times WHO air quality guidelines). The red bars represent the PMP1, PMP2, and PMP3 periods.

Full size image

The second PM2.5 peak (PMP2) extended from September 29th to October 3rd, with daily averages above 30 μg/m3 for most CETESB stations during the five days. Exceptionally, in the MASP, the PMP2 was later and shorter, from September 30th to October 2nd. In the MASP, the concentrations reached a maximum on October 2nd, ranging between 33 μg/m3 (Pico do Jaraguá station) and 50 μg/m3 (Osasco station), while for the other mesoregions, the peak occurred on October 3rd, with daily averages in the 40 to 50 μg/m3 range. The PMP2 period fits with T2 dates, indicating the contribution of smoke transported from P20F during the BAP2 to the high-PM2.5 episode. Overall, concentrations decreased on October 4th, except for Ribeirão Preto and São José do Rio Preto. Particularly in the MASP, were below 10 μg/m3 PM2.5 levels after PMP2.

The third PM2.5 peak (PMP3) covered the period between October 5th and 9th, with daily averages above 30 μg/m3 across the SPS, reaching 85 μg/m3 in São José do Rio Preto on October 9th. Although the maximum PM2.5 levels during the PMP3 fits the T3 dates in most mesoregions, suggesting the contribution of smoke transported from P20F, this air pollution peak appears to be associated with a combination of factors. First, the high PM2.5 levels over the SPS occurred a few days after the PMP2 and, therefore, can be related to the persistence of the smoke in the atmosphere. Moreover, it can also be attributed to local fire emissions since a high number of active fires was observed on October 7th in SPS (Fig. S6). In São José do Rio Preto and Ribeirão Preto, the PM2.5 levels remained high between PMP2 and PMP3. Therefore, in these mesoregions, PMP2 and PMP3 can be interpreted as a single smoke-induced PM2.5 peak.

Cascading risks: public health implications and multiple hazard synergy

The distribution of non-COVID-19 mortality in the SPS is presented in Fig. 5. A very different pattern was observed in 2020 (blue and red lines) compared to the previous ten years (black line). A substantial increase in daily non-COVID-19 deaths was observed from February to June 2020 (Fig. 5a), markedly in the first months of the pandemic, mainly due to an outbreak of respiratory-related deaths (Fig. S7). These findings are in close agreement with those observed in eight of the largest Brazilian urban centers, including São Paulo city42. According to the authors, from February 23rd to August 8th, the excess mortality from respiratory system diseases was 312% in the cities analyzed. Although São Paulo had the lowest increasing rate among the cities included in the referred study (174%, while in Manaus, in the north of the country, this number reached 758%), this is equivalent to 7780 excess deaths from respiratory diseases in the period. The authors attributed the high number of excess respiratory deaths to the high underreporting of deaths from COVID-19 at the beginning of the pandemic. Despite these issues, from July onwards, the distribution of mortality in 2020 tends to be closer to the expected. This normalization can be explained by improving the diagnosis of COVID-19 deaths, reducing underreporting, and, therefore, reducing the overestimation of respiratory deaths.

Fig. 5: Daily mortality over the SPS.
figure 5

a Daily non-COVID deaths over the SPS throughout 2020 (blue), including the corresponding 7-day moving average (red). The black line shows the expected mortality (15-day centered moving average considering the mean values of daily deaths for the 2008–2019 period). The dark gray dotted line on the secondary y-axis represents COVID-19 deaths. The gray bars represent the smoke-induced PM2.5 peaks (PMP1, PMP2 and PMP3). b Excess mortality across SPS mesoregions from October 1st to October 14th. c Excess deaths observed during control cases that occurred in the SPS, including fire-count spike days alone from 2012 to 2019 (light red boxplots) and heatwave days alone from 2008 to 2019 (orange boxplot). The red error bar shows the P20F-related mortality burden estimated in the present study.

Full size image

A substantial rise in non-COVID-19 deaths was observed in the period from October 1st to October 14th, reaching 12,602 deaths across SPS. The period covers both PMP2 and PMP3. Compared to the expected mortality ratio, it represents 2150 (2095–2206) excess deaths in a 14-day window, with an observed-to-expected ratio of 1.21 (1.17–1.24). When looking at the regional distribution of the excess deaths, however, we observed that the mortality effects were not homogeneous throughout the mesoregions (Fig. 5b). In the western SPS (closer to the Brazilian Pantanal and more impacted by smoke-induced PM2.5 during PMP2 and PMP3), the mortality increase was higher (128.0% in Araçatuba, 112.4% in São José do Rio Preto, 78.4% in Marília). Conversely, in the eastern part of the SPS (mesoregions Itapetininga, Litoral Sul Paulista, Metropolitana de São Paulo, and Vale do Paraíba Paulista), mortality increase was not statistically significant (Table S4).

Comparing the P20F-related mortality burden with control cases (previous heatwaves and fire-count spike days that occurred in the SPS in September or October months) provides an assessment of the expected impact of such events if they occur alone (Fig. 5c). Considering fire-count spike days (state daily number of active fires higher than the 95th percentile between 2012 and 2019), the mortality increase over the state was below 5% (median), both on the fire day (D) and on subsequent days (D + 1, D + 2 or D + 3). For state-wide heatwave days (i.e., heatwaves that affected more than 90% of the SPS area and occurred between 2008 and 2019), the impact was estimated at 10% (median). Both estimations are significantly lower than the 21% mortality increase observed within the 14-day window of coexposure to extreme heat and P20F-related air pollution.

CDHW events exacerbated fire risk in the Pantanal, promoting two main periods of burned area peaks (BAP1 and BAP2). Three episodes of long-range smoke transport to the SPS (T1, T2, and T3) were identified during BAP1 and BAP2 periods. Accordingly, three PM2.5 peaks related to T1, T2, and T3 were observed throughout the state (PMP1, PMP2, and PMP3). Among these episodes, PMP2 and PMP3 were simultaneous to a large number of non-COVID-19 excess deaths in the SPS, especially in the northwestern SPS, closer to the Brazilian Pantanal, where large increments of PM2.5 were attributed to the P20F through EURAD simulations. In this context, quantitatively determining the isolated effect of high PM2.5 levels on mortality rates is quite complex. Some hypotheses can be formulated from a more in-depth analysis of meteorological conditions over the SPS.

Overall, we observed an inverse correlation between daily precipitation and the duration of the PM2.5 peaks (Fig. 6). The rainfall is expected to play a significant role in reducing PM2.5 levels via particle wet deposition processes. In particular, the PMP1 was a short-duration peak (2 days). The sharp decrease in concentrations can be attributed to precipitation in the days following the peak (3, 6, and 10 mm on September 20th, 21st, and 22nd). Conversely, the PMP2 occurred in the absence of precipitation (September 29th to October 3rd), which may have contributed to the extensive time interval with high PM2.5 concentrations during the PMP2. Regarding PMP3, a behavior similar to PMP1 was observed, with a sharp decrease in concentrations and an increase in rainfall (2 and 8 mm on October 8th and 9th).

Fig. 6: Dry-season daily surface temperature anomalies and precipitation over the SPS.
figure 6

Daily mean temperature (DMT) anomalies in the state mesoregions and daily total precipitation as depicted by the (a) time series and (b) maximum DMT anomaly in each grid cell between September 27th and October 7th. Anomalies were computed as the difference between the mean daily temperature in 2020 and the base period (1981–2010) for each mesoregion. The PMP1, PMP2, and PMP3 periods are represented by the gray bars in (a). Median and IQ intervals were calculated considering the average temperatures from each mesoregion.

Full size image

The distribution of daily mean temperatures (DMT) in the SPS also provides insights into the PM2.5 dynamics during the smoke-induced air pollution episodes. In particular, the days after PMP1 (20–21 September) are marked by a significant drop in DMT (Fig. 6a). Conversely, the PMP2 and PMP3 episodes occur in a period characterized by atypically high temperatures over the state. Compared to the respective climatological averages (1981–2010), all mesoregions presented positive temperature anomalies that began on September 24th, persisting for up to 16 days (encompassing PMP2 and PMP3). Over the state, the maximum anomalies in this period ranged from 4 to 12 °C, reaching higher values in the central part of the SPS (Fig. 6b). All mesoregions were under an HW regime (Fig. 6), from September 28th, with a duration varying from 7 days (in the mesoregions of Itapetininga, Litoral Sul Paulista, Macro Metropolitana Paulista, and Metropolitana de São Paulo) and 12 days (in the mesoregions Araçatuba, Araraquara, Presidente Prudente, Marília, Assis, Ribeirão Preto and São José do Rio Preto), with a significant coastal-to-inland gradient (lower duration in mesoregions closer to the ocean).

Notably, synoptic atmospheric blocking conditions were observed in the central region of Brazil during the 2020 dry season, which were widespread and affected southeastern Brazil, including the SPS24,25. The observed blocking patterns exhibited a large anticyclonic anomaly responsible for air subsidence, persistent clear sky conditions, low humidity levels, and episodes of absence of precipitation26. These specific weather conditions and the occurrence of extreme temperatures in the SPS between September 28th and October 9th may have been responsible for the worsening of the air pollution condition, maintaining high pollution levels between the PMP2 and PMP3 periods and amplifying the effect caused by the transport of smoke during T2 and T3. Conversely, the rapid improvement in air quality after PMP1 may be linked to a sudden decay of the blocking pattern followed by a return to transient waves and precipitation. Moreover, the co-occurrence of extremely hot weather and air pollution episodes is expected to amplify the impact on health beyond the sum of their effects individually43,44.

Discussion

CDHW events triggered uncontrolled fires during the 2020 dry season. Significant deficits in precipitation and large land evaporation rates pre-fire season combined with soil moisture-temperature feedbacks have boosted the occurrence of extreme heat and the unprecedented P20F21. Natural and anthropogenic mechanisms coupling has been responsible for cascading effects elsewhere. In Portugal, for example, the 2017 record-breaking burned area was attributed to the combination of prolonged drought, extensive vegetation hydric stress, the passage of a hurricane, and human actions (negligent ignitions)45. The CDHW-fire events during the P20F were widespread, affecting environmental conservation areas, traditional communities, indigenous lands, and settlements46. The impacts included 17 million killed vertebrates29, decreasing post-fire vegetation productivity and increasing runoff30; national economic losses of ∼USD 3.6 billion47; and health effects, increasing COVID-19 hospitalizations among older people33. The wide range of local impacts reinforces the claims for developing integrated fire management (IFM) in the region21,26,27,28, strengthening prevention actions48. Despite recent advances in reducing burned areas, fire intensity, and associated emissions, IFM is still restricted to a few protected areas in Brazil49,50,51. Thus, expanding IFM is necessary, besides strengthening environmental protection laws and sustainable land management policies countrywide.

Going beyond the local level, CDHW-fire events in the Brazilian Pantanal resulted in significant long-range cascading effects. We identified three episodes of smoke transport to the SPS related to the CDHW-fire events during the P20F. The SALLJ was important in transporting smoke from the P20F to southeastern Brazil, similar to what was previously observed in central Brazil and the Amazon basin46,52. However, a deep understanding of long-range transport mechanisms in SA is still crucial to better predict and manage risks in smoke-impacted regions. It is worth highlighting that IFM strategies are generally designed based on traditional place-based governance (i.e., focusing on governance or management of individual places), such as parks, communities, municipalities, and states. Nevertheless, as our results suggest, such strategies and policies need to be improved by adopting a telecoupled governance perspective53, which opens a new perspective of linkages between systems over multiple scales, across distance, and through time in the policy, human health, education, communication, economic, social, and management strategies, including flows between them54. In particular, multisectoral and inter-ministerial strategies transcending state government spheres and international cooperation are needed.

Along with previous studies46,55, our work highlights the role of the Pantanal fires in disrupting air quality across the country, as already recognized for fires in the Amazon and Cerrado56,57,58. In particular, the smoke loading over the Brazilian Pantanal during the P20F was higher than over Amazonia46. Within the framework of national and international political articulations, integrating fire policies and management, including environment, health, and infrastructure sectors, involving different regions and biomes, could help fire emission monitoring and create alert systems to help society avoid the consequences of fires in SA.

The chain of extreme events in the Pantanal combined with long-range transport directly impacted air quality in the SPS. Three smoke-induced PM2.5 peaks were identified over the SPS related to the CDHW-fire events from the P20F, exceeding CONAMA and WHO standards. Overall, air pollution episodes in the SPS are attributed to low precipitation periods and poor vertical mixing and horizontal dispersion conditions44. In the last decades, restrictive and regulatory measures reduced primary emissions from vehicular and industrial sources in the state, although secondary pollutants remain at high levels59. In particular, vehicle-derived precursors may interact with compounds in the atmosphere, further elevating secondary pollutant concentrations60. Local biomass burning emissions (such as waste burning and wood stoves61) and the transport of smoke from regional fires also appear as significant emission sources, worsening air quality in the region62. In 2014, biomass burning emissions contributed to 18.3% of PM2.5 during the winter, mostly related to the transport of particles from areas affected by sugarcane burning63. The SPS is the leading sugarcane producer in Brazil, accounting for over 50% of the national production, with cultivation mainly located in the central, northwest, and western regions of the state64. Despite the pre-harvesting biomass burning for sugarcane crops being responsible for high pollutant emissions in the region65, the legislation established by the SPS law N°11.241/2002 to gradually end the burning practice by 2031 reflects the observed reduction in fire occurrences over the region, despite the increase of sugarcane cultivation area and production66. Although emission control policies represent an essential step in improving air quality in urban and densely populated areas such as the SPS, our findings reinforce the importance of compound events in pollution levels at the regional scale, highlighting the interconnection between CDHW-fires and air quality in SA. Previous studies have reported similar effects of long-range smoke transport worsening regional air quality in Southeastern Asia67, Australia68, and the US69. Therefore, our work underscores the need to align multisectoral and multicountry policies to promote air quality.

The coexposure to smoke-induced air pollution and extreme heat led to 2150 excess deaths throughout the SPS. The 14-day peak of non-COVID-19 deaths represented a 21% mortality increase statewide, being more pronounced in mesoregions closer to the Brazilian Pantanal. Two PM2.5 peaks (PMP2 and PMP3) occurred in the SPS during the period mentioned above, along with prolonged and state-wide heatwaves (7 to 12 days), with temperature anomalies ranging from 4 to 12 °C, potentially amplifying health effects throughout the SPS. In Brazil, short-term smoke exposure has been associated with significant mortality risks70,71, increased hospital admissions72, and substantial economic losses73. PM2.5 exposure has also been linked to significant productivity-adjusted life years lost (5.11% for every 10 µgm−3) and economic costs estimated at 268 billion dollars between 2000 and 201974. Regarding heat exposure, a recent study attributed approximately 50 thousand excess deaths to heatwaves in Brazilian urban areas between 2000 and 2018, 14,850 only in the MASP16. Besides health effects, significant socioeconomic impacts, such as supply-chain disruption effects, health costs, and labor productivity loss, are related to heat exposure, with most severe losses disproportionately concentrated in Central and Southern Africa, Southeast Asia, and Latin America75.

Extreme heat conditions in the SPS likely amplified the health effects from smoke-induced air pollution transported from the Pantanal, in addition to the high number of active fires throughout the state, contributing to the worsening of air quality throughout the state. The simultaneous exposure to heat and air pollution has synergistic and cumulative effects on health since their impacts occur through similar pathophysiological pathways76. High temperatures and air pollution can cause inflammatory reactions in the respiratory system, increasing respiratory deaths76. In the cardiovascular system, short-term exposure to air pollution may disrupt compounds present in the blood. In contrast, heat exposure stresses thermoregulation ability and the immune system, increasing susceptibility to air pollution effects77. Nonetheless, several limitations are still associated with quantifying multi-hazard interrelationships in impact, as the interaction between different hazards can result in a more significant effect than the sum of their individual effects78. By comparing the 21% mortality burden associated with the P20F in the SPS with some single-event control cases (heatwaves and fire-count spike days alone), we assessed the exceptional nature of this burden and the potential individual effects of exposure to heat and air pollution from local fires across the state. During fire-count spike days, a mortality increase below 5% (median) was observed for events between 2012 and 2019.

Similarly, for state-wide heatwaves, the median excess mortality was 10% for events in the 2008–2019 period. Regarding heat exposure, the result was in close agreement with the observed by Monteiro dos Santos et al. 16, despite the authors considering only the MASP region (around half of the SPS’s population). This result suggests that neither extreme heat conditions in the SPS nor the high number of active fire outbreaks throughout the state can explain the 21% short-term (14-day) P20F-related mortality impact, reinforcing the substantial long-range health effects of CDHW-fire events in SA.

Although we considered only non-COVID-19 deaths, it is also worth mentioning that the pandemic context exerts additional strain on the healthcare system, potentially amplifying the mortality burden. Therefore, assessing health indicators in a pandemic is challenging, even for causes unrelated to the virus. In Europe, for example, the combined effect of COVID-19 and extreme temperatures resulted in an indirect amplification in heat-related mortality (>50% compared to previous years) as a consequence of the disruption of healthcare systems and the decrease in emergency room admissions due to fear of the population attending healthcare facilities79. PM2.5 exposure was identified as a risk factor, particularly in the Brazilian Pantanal, resulting in a 23% increase in hospitalizations by COVID-19 in the elderly33.

Like previous catastrophic fire events, the 2020 Pantanal wildfires resulted not only from severe drought and extreme heat conditions but also from a complex interplay among extreme climate conditions, lack of management, and human-induced ignitions27. It is worth mentioning that this event occurred during the COVID-19 pandemic, severely contributing to the lack of environmental law enforcement due to the COVID-19 lockdown28. The pandemic hindered Pantanal firefighting, reducing the number of firefighters and constraining their movements spatially and temporally. In particular, Indigenous firefighters, who represent a significant proportion of the workforce in the Pantanal, could only work in the Indigenous lands. Moreover, in the context of wide fire outbreaks in the Amazon, Atlantic Forest, and Cerrado biomes in 2020, several brigades had been working in other regions of Brazil under great pressure28. Equally important, land cover changes are pointed out to have contributed to homogenizing the landscape, favoring the occurrence of those mega-fires31. Furthermore, human ignition sources, either by accident, negligence, or arson, are responsible for most wildfires80.

The simultaneous occurrence of extreme events is projected to become more frequent under all future climate change scenarios, materializing as a chain of impacts that can trigger cascading hazards and systemic risks81,82,83. Multiple scale interactions between co-occurring drivers and cascading hazards amplify the impacts compared to the individual occurrence of such hazards82. Therefore, addressing and preparing for the effects of climate change demand a thorough examination of the intricate interplays between heat, droughts, air pollution, and other extreme events, integrating it into risk assessments and strategic planning, supported by robust monitoring systems to enhance readiness for potential future cascading hazards13. Integrating observational, satellite-based, and reanalysis data, atmospheric dispersion models and death records, we provided multiple lines of evidence of a long-range cascade chain of hazards connecting central and southeastern SA (Fig. 7). The simultaneous exposure to smoke-induced air pollution transported from the P20F and persistent state-wide heat conditions in the SPS, combined with high local fire incidence that further worsened air quality throughout the state, resulted in 2150 excess deaths. This 21% mortality increase attributable to a multi-hazard short-term exposure (14 days) was significantly higher than estimations for single-event control cases that occurred locally in previous years (heatwaves and fire-count spike days alone). Our results underscore the need for comprehensive risk management and public health strategies that address the long-range cascading impacts of compound drought-heatwave (CDHW)-fire events in SA. These impacts extend beyond the burned and immediate surrounding areas, affecting distant populations. This reinforces the necessity for countries to strengthen collaboration across global governance structures to develop successful adaptation strategies and provide scientific support for policy-making, addressing current scientific challenges and motivating future studies.

Fig. 7: From local scale to long-range cascading effects:a schematic multi-hazard risk assessment framework connecting Central and Southeastern South America.
figure 7

Drought-heatwave coupling triggered fire risk and large burned areas in the Pantanal during the 2020 dry season. Long-range transport led to smoke-induced air pollution episodes in the São Paulo state, also influenced by local meteorological conditions and emission sources. Synergism and co-exposure to high levels of PM2.5 and extreme heat have resulted in a large burden of premature deaths. Map created using the Free and Open Source QGIS.

Full size image

Methods

Study area

The Pantanal biome is the largest continuous wetland in the world82. The Brazilian Pantanal (Fig. 1), located in central-western Brazil (Fig. 1), covers an area of approximately 150,000 km2, with a population of around 474,000 inhabitants (https:/ibge.gov.br/apps/biomas/). The geomorphology and rainfall regime determine the flood pulse, beginning in the northern region during spring and moving toward the southern region84. Fire in the Pantanal is mainly restricted to the dry season, from July to November80, and occurs more frequently in the grasslands. Still, the 2020 fires have preferentially burned forests28,85. Natural fires are rare in the Pantanal biome, accounting for less than 5% of the total fire scars80. Thus, fire is generally associated with anthropogenic activities, particularly land use, land cover change, and management86.

The São Paulo state (SPS), located in southeastern Brazil (Fig. 1), is the most populated in the country, with a population of 46.6 million inhabitants and an area of 248,219 km2. The Metropolitan Area of São Paulo (MASP), comprising 39 municipalities, approximately 1.200 km from the Brazilian Pantanal, is the largest South American megacity with over 21 million inhabitants (https://ibge.gov.br/cidades-e-estados.html). The state is characterized by high pollution rates, especially in the MASP. However, successful public policies implemented in the region for controlling and monitoring the emissions from industries and a fleet of more than 8 million vehicles have led to a significant reduction in the emissions of primary atmospheric pollutants over the last decades59. Additionally, local biomass burning emissions, such as waste burning and wood stoves61, as well as the transport of smoke from forest fires upwind SPS, contribute significantly to worsening air quality throughout the state62,86. The SPS is the leading sugarcane producer in Brazil, accounting for 54% of the national production (https://conab.gov.br/info-agro/safras/cana). Sugarcane cultivation is mainly located in the central, northwest, and west regions of the state64, and the pre-harvesting biomass burning for sugarcane crops is responsible for high pollutant emissions in the area65. In the MASP, Pereira et al. 63 estimated contributions of biomass burning for PM2.5 mass loadings ranging from 11.6% (annual average in 2014) to 18.3% (during the winter in 2014), mostly related to the long-range transport of particles from areas affected by sugarcane burning87.

Satellite-derived burned area and fire data

The Burned Area (BA) over the Brazilian Pantanal was obtained from the first version of the ALerta de ÁRea queimada com Monitoramento Estimado por Satélite - https://alarmes.lasa.ufrj.br - (ALARMES) satellite-derived annual dataset with 500 m spatial resolution for the main Brazilian biomes (Amazônia, Cerrado, and Pantanal). The algorithm combines daily images and active fires from the Visible Infrared Imaging Radiometer Suite (VIIRS) sensors and deep learning techniques to identify fire-affected areas88. The product provides a layer with the approximate burn date and a confidence level layer representing the classification quality. Moreover, daily information on active fire counts over the SPS was obtained from NASA/NOAA’s Visible Infrared Imaging Radiometer Suite (VIIRS) at 375 m (VNP14IMGTML)89.

Weather data and heatwave definition

Data on air temperature at 2 m above the surface and total precipitation were gathered from the ECMWF reanalysis version 5 (ERA5), with a spatial resolution of 31 km and 1-h temporal resolution90. The daily precipitation in 2020 over the entire SPS domain was computed from hourly data obtained from ERA5. The hourly temperature data were used to get each grid cell’s daily mean temperature (DMT). Furthermore, DMT in each mesoregion was computed as the spatial average over the grid cells.

Following the methodology proposed by Nairn & Fawcett91, we used the Excess Heat Factor (EHF) to identify the occurrence of heatwaves in the mesoregions over the SPS. As a health-relevant heat index, the EHF has been widely used92, including recent studies in Brazil16,93. The EHF is calculated as the product of two indices: (1) the significance index (EHIsig), computed as the difference between the three-day-averaged DMT and the 95th percentile of the DMT calculated across the 1981–2010 base period, and (2) acclimatization index (EHIaccl), which is calculated as the difference between the three-day-averaged DMT and the average DMT over the previous 30 days. Therefore, the EHF considers both the effect of short-term temperature anomalies and the human physical ability to adapt to them. If the EHF is positive, the three days are considered under heatwave conditions. More details about the methodology used here for heatwave identification can be found in ref. 16.

The drought severity was analyzed using the Standardized Precipitation Evapotranspiration Index at 6-month timescale (SPEI-6), derived from the ERA5 reanalysis product94. The dataset was obtained directly through the SPEI Crop Drought Monitor (https://global-drought-crops.csic.es/), and drought severity was categorized according to the US Drought Monitor system (https://droughtmonitor.unl.edu/About/AbouttheData/DroughtClassification.aspx). The SPEI-6 provides insights into agricultural droughts related to shorter-length timescales95 and has been preferred for assessing the dynamic water fluctuations in the Pantanal region. Its use is underscored by the established correlation between agricultural drought conditions and the incidence of wildfires96.

Air quality data

Hourly PM2.5 [μg/m3] data were provided by the São Paulo State Environmental Company (CETESB, Companhia Ambiental do Estado de São Paulo, https://cetesb.sp.gov.br/ar/qualar) from 19 air quality monitoring stations across the SPS (Fig. 1) according to data availability. Data from CETESB monitoring stations have been largely used in air quality studies in the SPS59. For most analysis, the CETESB stations were grouped according to the mesoregions where they are located, namely: RP (Ribeirão Preto), SJP (São José do Rio Preto), PI (Piracicaba, Limeira, and Rio Claro), MMP (Jundiaí), and MASP (Santana, Pico do Jaraguá, Perus, Itaim Paulista, Ibirapuera, Grajaú-Parelheiros, Congonhas, USP-IPEN, São Caetano do Sul, São Bernardo-Centro, Osasco, Mauá, and Guarulhos-Pimentas).

Mortality data and estimations of excess deaths

Daily mortality data from the Brazilian Health Informatics Department (DATASUS) were provided by the Data Science Platform applied to Health (PCDaS) of the Oswaldo Cruz Foundation (https://pcdas.icict.fiocruz.br/, accessed on February 06, 2022). Deaths related to external causes (accidents, suicides, and homicide) were not included in the analysis; moreover, confirmed deaths corresponding to code B34.2 (coronavirus infection disease) were not considered. Total daily deaths from respiratory system diseases (cataloged in chapter X, according to the 10th revision of the International Statistical Classification of Diseases and Related Health Problems, ICD-10) for all municipalities within SPS were analyzed. Annual estimates of the population in the SPS provided by the Brazilian Institute of Geography and Statistics were used to normalize mortality data, reducing the effects of population growth on the estimations of the expected daily number of deaths. The analysis of mortality data was based on the observed-to-expected deaths (O/E) ratio16. This method allows for a percentage analysis, correlating the total daily deaths observed during 2020 with the daily deaths expected (15-day centered moving average of mortality between 2008 and 2019). Excess mortality was obtained by the difference between observed and expected daily deaths (O-E).

The observed-to-expected ratio was used to estimate excess deaths associated with long-range impacts of the CDHW-fire events from the P20F in the SPS. Moreover, some control cases that occurred in the SPS were compared with the P20F-related multi-hazard event. For that, the observed-to-expected ratio was applied for (1) all heatwaves days that have occurred within the SPS and reached all the state mesorregions simultaneously, filtered for September and October, and (2) all fire-count spike days, when daily fire counts in the SPS surpassed the 90th percentile of the 2012-2019 period (277 fire counts), filtered for September and October. We also included an analysis of the lag effects in the mortality for fire-count spike days, considering 1, 2, and 3 days after the fire spike.

Atmospheric chemistry and transport modeling

The HYSPLIT atmospheric dispersion model from NOAA’s Air Resources Laboratory was used to track wildfire smoke using forward and backward trajectory simulations97. Meteorological data for the 3D trajectories were obtained from the NCEP Global Forecast System (GFS) dataset (http://www.ready.noaa.gov/archives.php), which provides 0.25° × 0.25° spatial resolution, 55 vertical levels, and a 3-h temporal resolution. Simulation of 48-h forward trajectories at 1500 m above ground level (a.g.l.) from the Brazilian Pantanal (17.72°S, 56.03°W) and 48-h backward trajectories from São Paulo (23.51°S, 46.70°W) at 1800 UTC facilitated understanding of pollutant transport over different spatial and temporal scales. This approach allows for the visualization of airflow patterns and facilitates understanding pollutant transport over multiple spatial and temporal scales.

Moreover, the European Air Pollution Dispersion–Inverse Model (EURAD-IM) was used for mesoscale atmospheric chemistry and transport modeling98,99. It utilizes the Weather Research and Forecast (WRF) model100 as an offline meteorological model driver. The meteorological initial and boundary conditions are sourced from the United States National Center for Environmental Prediction (NCEP) Global Forecast System (GFS) at a horizontal resolution of 0.25° × 0.25° (available at https://rda.ucar.edu/datasets/ds083.3/). The EDGAR v4.3.2 anthropogenic emission inventory101 is applied within the model domain for simulations. EURAD-IM employs data from the global Copernicus Atmosphere Monitoring Service (CAMS) Integrated Forecasting System (IFS) model as chemical initial and boundary conditions. Fire emissions are sourced from the Global Fire Assimilation System Version 1.2 (GFASv1.2) products, with a horizontal resolution of 0.1° × 0.1° (available at http://apps.ecmwf.int/datasets/data/cams-gfas). The off-line model encompasses the south of Latina America focused on São Paulo with a 25 km horizontal resolution grid with 110 × 101 points covering an area of 2750 × 2525 km², including crucial regions in southern and southeastern Brazil adjacent to São Paulo. Other model configurations and parameterizations are detailed in Table S3. Four EURAD-IM simulations were conducted to evaluate biomass-burning contributions during the P20F. The first simulation (NO_GFAS) excluded GFAS fire emissions. The second (GFAS_SP) included GFAS fire emissions over São Paulo State, Rio de Janeiro State, and part of Minas Gerais. The third (GFAS_PANT) focused exclusively on GFAS fire emissions over the Pantanal region. Finally, the fourth EURAD-IM simulation (GFAS_BR) was performed with GFAS fire emissions over the entire model domain.