The Jaú National Park (JNP) was created in September 1980 (Federal Decree 85.200) and is located between the municipalities of Barcelos and Novo Airão of the state of Amazonas, 220 km distant in northwest direction from Manaus (Figure 1). JNP is one of the largest National Parks in Brazil, covering 2,272,000 hectares and is inserted in a mosaic of state and federal conservation units. The limits of the southern edge of the park extend to the Carabinani River, forming the border with the Rio Negro State Park North Section. The northern portion is limited by the Unini River forming the border with the Extractive Reserve (RESEX) Rio Unini. Toward the east the JNP reaches the Negro River and toward the west the Amanã Sustainable Development Reserve. The JNP, together with the Sustainable Development Reserves Amaña and Mamirauá, and the Anavilhanas National Park, form the Central Amazonian Conservation Complex, declared by UNESCO as World Heritage Site in 2000.

The lower Negro River up to the confluence with the Branco River (Figure 1) is strongly influenced by the backwater effect of the mainstem (Amazonas-Solimões) (Meade et al., 1991). The backwater effect also extends to the lower sections of the Unini, Jaú and Carabinani rivers until the first rapids in each river (Figure 2). Hydrographic conditions in this region are similar in the timing of maximum (June) and minimum (October/November) water levels and flood amplitude (about 10 m) in comparison to the water level record of Manaus (Schöngart and Junk, 2007). About 70% of the JNP is covered by dense upland forests (terra-firme) while igapó forests comprise about 12% and white-sand ecosystems (campinarana) another 10% with some palm swamps dominated by Mauritia flexuosa (buritizal) in interfluvial regions (Borges et al., 2001). The relief is flat with some hilly areas of up to 150 m above average sea level cut by large valleys with igapó floodplains. Dominating soil types in the JNP are orthic Acrisols and hydromorphic Laterites and Podzols (FVA, 1998). Almost 86% of the JNP is destinated for strict protection of biodiversity covering all main vegetation types, while about 14% is destinated for residence, resource extraction and small-scale agriculture of the human population mainly occurring in igapós and adjacent terra-firme forests along the middle and lower section of the Jaú and Uniní rivers (FVA, 1998).

With the creation of the RESEX Rio Unini in 2007 (Law No. 11,516) a management plan was implemented defining zones for different land-use intensities and biodiversity protection (ICMBio, 2014). Areas destined for intensive community use (about 3,000 ha) are located in the eastern part of the conservation unit at the river margins close to the confluence with the Negro River, allowing settlements, livestock production and small-scale agriculture, including the use of fire (ICMBio, 2014). The zone for management of natural resources (extraction of timber and non-timber forest products and fishing) comprises an area of approximately 340,000 ha and extend mainly along the lower and middle course of the Unini River, including igapó floodplains. Areas destined for strict biodiversity protection (about 524,000 ha) are located in the western part of the RESEX Rio Unini, integrating mainly campinarana, terra-firme and igapó in the headwaters of the Unini River and its tributaries (ICMBio, 2014).

Both, the JNP and RESEX Rio Unini are integrated with other 18 conservation units in the Ramsar Regional Site Rio Negro, created in March 2018, covering an area of more than 12 million hectares (Figure 1). The Ramsar Convention on Wetlands of International Importance provides a framework for the conservation and wise use of these ecosystems and their natural resources based on national inventories to establish specific conservation strategies (Wittmann et al., 2015). The Ramsar Regional Site Rio Negro extends along the middle and lower sections of the Negro River basin and interfluvial regions between the Negro and Japurá rivers. It integrates important oligotrophic wetlands such as igapó floodplains, among them the two largest freshwater archipelagos on Earth (Mariuá and Anavilhanas) (Latrubesse and Stevaux, 2015), campinarana and interfluvial palm swamps (buritizal).

In this section, we provide a summary of the demographic and socioeconomic background of the study region, as this information is essential for the later discussion about the relationship of igapó fires and human presence in a historical context (for more detailed information, see Supplementary Material). Humans have been present in the study region since the late Holocene, reflected by hundreds of petroglyphs at the JNP and RESEX Rio Unini (Valle, 2012). The foundation of Santo Elias do Jaú by Carmelite missionaries in 1694 (named Airão in 1759), located close to the mouth of the Jaú River (Figure 1), was the beginning of a profound sociocultural transformation of the indigenous populations in the study region (Leonardi, 1999). During the 18th century, the native population was decimated by conflicts, slavery, and diseases, or used as labor for resource extraction. The first half of the 19th century is characterized by economic and demographic decline of the region and massive decimation of the native and descendant population (Leonardi, 1999; Sampaio, 2003). During the second half of the 19th century, Airão gained economic importance, especially with the beginning of the rubber boom, when thousands of immigrants from the Brazilian Northeast settled along the Jaú, Carabinani, and Unini rivers (Leonardi, 1999). The collapse of the rubber boom in the 1920s caused again a progressive demographic decline in the study region which continued after the establishment of the JNP in 1980 since people are not allowed to reside in this category of protected area (SNUC Law 9.985/2000). This conflict is still not resolved, as traditional populations lived in the region long before the establishment of the JNP. The demographic census of 2001 indicates around 377 residents in the JNP and about 669 residents along the lower section of the Unini River (Pinheiro and Macedo, 2004).

For the remote sensing analyses we considered the JNP and an 8-km wide buffer zone, integrating the igapós along the blackwater rivers forming the border of the protected area comprising in total 3,072,295 ha (Figure 2). Using the wetland mask of Hess et al. (2015), a total of 518,135 ha igapó floodplain forests were identified, corresponding to 17% of the area. Of those, about 70% (363,420 ha) were located inside the JNP and 30% (154,715 ha) in the buffer zone. The reconstruction of fire occurrences in the study region covered the period from 1984 to 2017. The fires were identified from images obtained by Thematic Mapper (TM), Enhanced Thematic Mapper Plus (ETM+), and Operational Land Imager (OLI) aboard the Landsat 5, 7 and 8 satellites, respectively, with a spatial resolution of 30 meters. Images were made available by the United States Geological Survey (USGS) of NASA by the website Earth Explorer¹ of collection 1 (Higher level science data – Level 2 – on demand) with pixel values already corrected for surface reflectance. The study area is included in scenes 232/61; 232/62; 233/61 and 233/62, Datum WGS84, designed in the Universal Transverse Mercator Projection (UTM) zone 20S. The detection and delimitation of fire scars were carried out following the method proposed by Alvarado et al. (2017). For this purpose, false color fusion compositions were generated using the mid-infrared (SWIR), near-infrared (NIR) and red (R) bands in the red, green, and blue (RGB) channels. The bands used were 5-4-3 for the TM and ETM+ sensors, and bands 6-5-4 for the OLI sensor, obtaining composite images in false RGB color with fixed contrast levels to minimize misinterpretation due to fluctuations in reflectance intensity. The delimitation of the burned areas was then performed by visual interpretation and manual vectorization of the fire scars, with a standardized cartographic scale of 1:25,000. For the composition of the images, the R software (R Core Team, 2018) was used through the “raster” (Hijmans and van Etten, 2013) and “rgdal” (Bivand et al., 2014) packages for the vectorization of scars. For the reprojection of the images, we applied ArcGIS 10.4 software. The fire scars mapped in the January 1984 image were attributed to fires that occurred in the period 1982/1983, as despite the lack of Landsat images prior to 1984, the mortality of post-fire trees in the igapó forests is usually visible for a few weeks or months after the fire. Fires were classified according to the size of the burned area (<4 ha; 4.1–40 ha, 40.1–100 ha; >100 ha) to assess the extent of the fire in relation to hydroclimatic conditions and human occupation. Areas which burnt at least twice during the monitored period were analyzed considering the location, burnt area, year and frequency of recurring fire events on previously burnt areas. Based on these data we calculated the mean and standard deviation of the return interval and the percentage of the originally burnt area. To evaluate the anthropic factor on fire occurrence in the JNP and its surroundings, we applied three concentric zones with 2.5 km (intensive land use), 5 km (extensive land use) and 10 km (remote land use) around the settlements (Heckenberger et al., 2008). Therefore, the geographic coordinates of each settlement in the study region were placed on the map of fire scars, considering the three radii to count the number of fire scars. For this process, ArcGIS 10.4 software was used.

In contrast to the non-flooded upland forests (terra-firme), where climatic water deficit is mainly related to rainfall seasonality, the Central Amazonian igapó presents a more complex regime. The regular and annual flood pulse in the Central Amazonian igapó generally extends from the middle of the rainy season (around March) to the middle of the dry season with a 3-months shift compared to the precipitation regime (Schöngart et al., 2002). This has strong implications for the climatic water deficit in this environment, which generally remains flooded during the first months of the dry season. For our research approach we therefore considered the period from September to February as terrestrial phase, which is representative for the majority of igapó forests in the study region (Carvalho, 2019). During this period igapó floodplains become susceptible to fire due to climatic water deficit, especially in periods of El Niño occurrences, which decrease precipitation during the early rainy season (Aragão et al., 2007), leading to an extension of the non-flooded period in Central Amazonian floodplains into the wet season (Schöngart and Junk, 2007).

Monthly data for mean (T_mean), minimum (T_min) and maximum (T_max) temperature were obtained from the Brazilian National Institute for Meteorology (INMET) for the stations Manaus (located about 200 km in SE direction) and Barcelos (located about 150 km in NE direction) (Figure 1) for the period 1982–2017 (Supplementary Table 1), as no ground-based data were available for the study region. Monthly and annual averages and standard deviations were calculated for T_mean, T_min and T_max based on mean values from the records of Manaus and Barcelos.

Based on the temperature record, the potential evapotranspiration (ET_pot) was estimated using the equation suggested by Thornthwaite (1948):

where, ET_pot is the monthly potential evapotranspiration (cm), T_mean is the monthly mean temperature (°C), I represents a heat index calculated by the sum of the 12 monthly mean temperatures i as follows:

The parameter a is computed by the following equation (Ribeiro and Villa Nova, 1979):

Monthly precipitation data were obtained from the Hidroweb platform of the National Water Resources Information System (SNIRH) operated by the Brazilian National Water Agency (ANA) and the Geological Survey of Brazil (CPRM) (for more details, see Supplementary Table 1). Three stations situated within (Baruri, Umanapana, and Seringalzinho) (Figure 2) or nearby (Moura, Novo Airão) the study region (Figure 1) were used to calculate average and standard deviation of monthly and annual precipitation from July 1982 to June 2017. The maximum cumulative water deficit (MCWD) was estimated to represent annual climatological drought intensity during the terrestrial phase of the igapó floodplains (September to February) from 1982/1983 to 2016/2017. For MCWD, monthly values with ET_pot > precipitation (water deficit) were accumulated during all months of the terrestrial phase (Aragão et al., 2007; Anderson et al., 2018).

As a measure of the El Niño-Southern Oscillation (ENSO) we used the Ocean Niño Index (ONI). The ONI is a 3-month running mean of SST anomalies (ERSST.v5; Huang et al., 2017) in the Niño 3.4 region (5^oN-5^oS/120^o-170^oW) based on a 30-year base period (1986–2015) (Supplementary Table 1). Based on the 3-month periods November-December-January (NDJ) and DJF, El Niño events were classified into weak (ONI ≥ 0.5 and < 1.0°C), intermediate (ONI ≥ 1.0 and < 1.5°C) and strong (ONI ≥ 1.5°C) episodes. Data were obtained by the Climate Prediction Center of the U.S. National Oceanic and Atmospheric Administration (NOAA)².

Data of daily water levels were obtained from four stations available on the Hidroweb platform (SNIRH/ANA) for the Jaú, Carabinani and Unini rivers within the study region, and for the Negro River at Moura, nearby the JNP (for more details, see Supplementary Table 1). The hydrological year was defined from 1st July to 30th June of the subsequent year comprising the period between the regularely occurring maximum water levels from 1982 to 2017. Missing data for period less than one month were interpolated. Applying the method of indicators of hydrologic alteration (Richter et al., 1996), baseflow index (BFI; 7-day minimum water level divided by the annual mean water level), and 90-day minimum water level (WL_min90) were calculated as a measure of the hydrological drought in the igapó floodplains. Therefore, we calculated average and standard deviation of the available data of daily water levels from the four hydrological stations, setting the absolute minimum water level of each time series to zero (Supplementary Figure 2).

For the reconstruction of past El Niño events we obtained SST anomalies for DJF of the Niño 3.4 region (>0.5°C) based on instrumental data (HadISST monthly SST data set from 1870 to 2016; Rayner et al., 2003) and the reconstruction of Cook et al. (2018) for the last 400 years, based on tree-ring chronology networks of the Eastern Australia New Zealand Drought Atlas (Palmer et al., 2015) and Mexican Drought Atlas (Stahle et al., 2016). We associated the occurrence of past El Niño events with instrumental (Henley, 2017) and reconstructed warm IPO phases for the past four centuries. The reconstructions of the IPO phases are based on tree-ring records from Vietnam and the North American Drought Atlas (Buckley et al., 2019) and an assemblage of four ice cores around the Pacific basin (Porter et al., 2021).

The total annual burnt igapó area and the hydroclimatic variables were tested for normal distribution by a Shapiro-Wilk test. With exception of the annual burnt area and the MCWD all hydroclimatic variables present a near normal distribution (p > 0.05). Hydroclimatic parameters were compared between years with fire occurrence and those without burning, as well as the hydroclimatic conditions between El Niño episodes and other years, using the non-parametric Mann-Whitney U test (R Core Team, 2018). Trends of hydroclimatic data during the study period were analyzed applying the non-parametric Mann-Kendell test (Pohlert, 2020), commonly used in detecting trends of meteorological and hydrological variables (Ahn and Merwade, 2014). The generated positive (negative) Z-values indicated increasing (decreasing) trends. For the terrestrial phase, average (T_max, ONI), cumulative (P, MCWD) and single values (BFI, WL_min90) were correlated to the annual burnt igapó area from 1982/1983 to 2016/2017 by non-linear regression models:

where, X represents the hydroclimatic parameters T_max, ONI, P, MCWD, BFI or WL_min90, and a, b and c the coefficients of the exponential equations with positive (negative) values for c turning them into growth (decay) functions (X-Act 7.0, SciLab).

The climate in the region during the analyzed period was characterized by average annual precipitation of 2,362 ± 273 mm (standard deviation) (Figure 3). Maximum monthly rainfall occurred in April (294 ± 68 mm) with only August demonstrating monthly precipitation of less than 100 mm (92 ± 32 mm). The estimates for monthly potential evapotranspiration (ET_pot) varied on average between 133 and 158 mm, indicating a period of climatic water deficit (ET_pot > precipitation) from July to November, characterizing a distinct dry season which occasionally extends until February. The mean annual air temperature was approximately 26.9 ± 0.6°C with maxima and minima of 32.2 ± 0.7°C and 23.4 ± 0.6°, respectively. Monthly maximum temperatures during the dry season were higher (32.0–33.5°C) than those of the rainy season (31.3–31.8°C). The hydrographic regime of the blackwater rivers (Negro, Unini, Jaú and Carabinani) in the study region was characterized by a monomodal flood pulse with maximum water levels peaking in the period end of June and beginning of July, while the minimum water levels occurred around October/November (Figure 3). The annual amplitude of the Negro River and the lower section of the Carabinani and Jaú rivers varied from 9.7 to 10.3 m. Upstream the rapids, the amplitude of the flood pulse declined and the hydro-regime became more irregular. Data from the Unini River upstream of the rapids indicates an annual amplitude of approximately 6.5 m with a more extended low-water period compared to the other analyzed river sections (Figure 3).

From 1982/1983 to 2016/2017, a total of 17,524 ha of burned floodplain forests have been mapped in the study region, distributed among 18 years and corresponding to 3.4% of the total studied igapó area. About 11,159 ha (63.7%) of burned igapós burned were located along the Unini River, 3,860 ha (22.0%) were identified along the Jaú River, 1,883 ha (10.7%) along the Negro River section and 628 ha (3.6%) along the Carabinani River (Figure 4). The majority of burned floodplain forest (61.5%) occurred in the buffer zone of the JNP, mainly along the Unini River and less extensively along the Negro River section. Interestingly, no fire scars were observed in the adjacent upland (terra-firme) forests during the 35-year study period. These fires comprised 254 fire scars, concentrated on the eastern and northern sections of the study region (Figure 4). The majority (78.7%) of the burnt areas was observed within a 10-km radius around human settlements, with 24.8% at 5–10 km, 24.8% at 2.5–5 km and 29.1% closer than 2.5 km distance. The detected fire scars were mainly concentrated in the size class of 4.1–40 ha (52.8%) (Table 1). Small-scale fires (<4.0 ha), large burnt areas (40.1–100 ha), and large-scale fires (>100 ha) shared 11.4, 15.7, and 20.1% of the fire scars, respectively. Large-scale fires accounted for 71.6% of the total burnt area of igapó in the study region, while the size classes < 4.0, 4.1–40, and 40.1–100 ha shared 0.4, 12.1, and 15.9%, respectively. Although large-scale fires comprise only 20% of the total number of detected fire scars, they jointly represent over 70% of the total burnt area during the study period and occurred mainly during strong El Niño episodes.

From the 254 detected fires scares, 37 (14.6%) burnt at least twice during the monitored period, most of them in the igapós of the Uniní (n = 29) and to a less extent along the Jaú (n = 6) and Negro (n = 2) rivers. Recurring fires on the same area were observed twice (31 areas), three (5 areas) or even four times (1 area) during the monitored 35-year period with an average return period of 13.2 ± 8.5 years. In general, the repeatedly occurring fires affected on average 57.0% of the original burnt area, however, with a huge variation (standard deviation of 40.7%). About 60% of the recurring fires occurred on areas < 10 ha and only 5% represented large-scale fires (>100 ha). The burnt area by recurring fires accounted for 8.1% (1,412.4 ha) of the total observed burnt area and was mainly concentrated along the Uniní River (95.5%). Most recurring fires were observed during strong El Niño episodes such as during 1997/1998 (5 scars with 312.7 ha) and 2015/2016 (18 scars with a total area of 315.3 ha). However, also during years without El Niño conditions repeated fires were detected such as in 2001/2002 and 2003/2004 with a total burnt area of even 493.2 ha, all concentrated in igapós along the Unini River.

Significant trends of hydroclimatic parameters during the terrestrial phase (September to February) of the igapó floodplains were observed for T_max (Z = 3.81, p < 0.001) and a negative trend for MCWD (Z = −2.95, p < 0.01) along the 35 years (Figure 5). The remaining hydroclimatic parameters showed no significant trends during the study period but a large interannual variation. The comparison of hydroclimatic conditions of the terrestrial phase between periods with and without fire occurrence indicated significant differences for all analyzed parameters (p < 0.05) (Table 2). Periods with fire occurrence had, on average, an elevated T_max by 0.9°C during the terrestrial phase, while P and MCWD were reduced by 230 mm and 120 mm on average, respectively. The WL_min90 was significantly lowered by 1.2 m, and BFI was decreased by 0.13 on average during periods with fire occurrence. Overall, the average ONI indicated for years with fire occurrence El Niño conditions (0.51 ± 1.24°C), while for years without burning ONI was negative (−0.48 ± 0.76°C).

During the studied period, fires in the igapó floodplain occurred mainly during El Niño events, which accounted for about 92.4% of the total burnt area in the igapó floodplains, while 7.6% of the burnt areas were attributed to eight further non-El Niño years (Table 1 and Figure 4). From the 12 El Niño episodes on record during the monitored period, only two events (1986/1987 and 2014/2015) were not related to igapó fires (Figure 5). Overall, El Niño affected mainly the climatic conditions in the study region leading to significantly increased T_max, while P and MCWD were significantly reduced (Table 2). Hydrological variables (WL_min90 and BFI) did not significantly differ between El Niño and other years. El Niño events of intermediate (n = 4) and weak intensities (n = 3) contributed only to 9.4% and 1.5% of the total burnt igapó area, respectively. Hydroclimatic conditions during weak and intermediate El Niño intensities did not show significant differences compared to other years. The five El Niño episodes with strong intensities (ONI > 1.5°C; 1982/1983, 1991/1992, 1997/1998, 2009/2010, and 2015/2016) were of remarkable impact, corresponding to 81.5% of the total burnt area (Table 1). Hydroclimatic conditions of the terrestrial phase during strong El Niño events were characterized by elevated T_max (33.6 ± 0.9°C) and extreme hydrometeorological droughts reflected by low P (742 ± 171 mm), MCWD (−389 ± 180 mm), WL_min90 (148 ± 77 cm) and BFI (0.16 ± 0.09) (Table 2). These severe drought episodes comprise more than 80% of the large-scale fires (> 100 ha). More than 7,506 ha (42.8%) of the total burnt igapó floodplain forests in the study region were associated with the strong El Niño 2015/2016 (Table 1). At the Jaú River, more than 2,428 ha igapó burnt during this period, corresponding to 62.9% of the total burnt area along this river. Also other river sections of the study area accounted in this period for the largest observed burnt areas with 67.5% (Carabinani River), 70.4% (Negro River) and 29.8% (Unini River), referring to the total of mapped burnt igapó during the study period (Figure 4). During this El Niño event, the highest T_max (34.8°C), and lowest values for P (507 mm), MCWD (−635 mm), WL_min90 (84 cm) and BFI (0.07) were observed for the terrestrial phase during the entire studied period (Figure 5). Hydroclimatic parameters explained a large portion of the annual burnt igapó forests in the study region as shown in Figure 6 by non-linear (exponential) regression models with adjusted R² varying from 0.61 (WL_min90) to 0.90 (MCWD). The standard deviation of the residuals (observed-estimated burnt area) ranged from 846 ha (WL_min90) to 436 ha (MCWD).

During our remotely monitored period, the IPO experienced a warm phase until 1998 and since 2014, with a cold phase in-between (1999–2013) (Buckley et al., 2019). The strongest El Niño events on record (1982/1983, 1997/1998, and 2015/2016) coincided with warm phases of the IPO, resulting in significantly increased air temperature and severe hydrometeorological conditions during the terrestrial phase of the studied igapó. These three events alone accounted for 62.5% of the burnt igapó detected along the 35-year period. The years with burnt igapó forests exceeding 1,000 ha in the study region occurred all during El Niño events throughout warm IPO phases with strong (1982/1983, 1991/1992, 1997/1998, and 2015/2016) or intermediate (1994/1995) intensities.

Generally, the humid conditions of Amazonian rainforests do not allow for the propagation of wildfires caused by natural ignition, such as lightning, and human presence is the leading cause of fires in the Amazon basin (Bowman et al., 2011; Pivello, 2011). During recent decades increasing human pressure resulted in spatiotemporal increases of fire, mainly in Amazonian terra-firme (Aragão et al., 2018). Igapós in our study region, however, are seasonally flooded for several months, have a relatively low human population density (< 0.04 persons/km²; Pinheiro and Macedo, 2004), do not suffer significant impacts from land-use changes, and are mostly integrated in protected areas (Figure 1). Even though, large-scale fires are evident, especially during El Niño-induced severe droughts. The fire spots detected in this study do not follow a homogeneous pattern of spread and frequency in igapó floodplains. About 79% of the mapped fire scars were detected in regions close to human settlements and communities (<10 km distance). This suggests that the origin of igapó fires is mainly caused by humans. More than 54% of the detected fire scars occurred close to human settlements (<5 km distance) suggesting that fires commonly applied during slash-and-burn activities in neighboring areas such as terra-firme forests, accidentally escape and are the main source of fire ignition in igapó forests. Most of the fires in igapós that occurred outside the community boundaries (>10 km radius) were located between settlements and communities, mainly along the lower section of the Unini River, destined for land-use (ICMBio, 2014). In this section we also observed most recurring fires events which generally occur at small-scales (<10 ha) and were detected also during non-El Niño conditions, suggesting anthropogenic ignition. Other common fire ignition sources in the igapó are not extinguished campfires on the river margins which might also occur at further distances to settlements (Ritter et al., 2012). Overall, fire ignition caused by humans in the study region seems to be accidental. Settlements of Pre-Columbian and modern populations until the late 1970s in northern Amazonia also showed a much higher amount of pyrogenic soil carbon in the region close to human occupation (< 3 km distance) than in more distant regions (Carvalho et al., 2018). These findings suggest that igapós in remote areas have a low probability of burning.

No fire scars have been detected during the analyzed 35-year period in the remote areas of the western part of the study region, lacking human settlements. An exception is the igapó along the Carabinani River. Although a small settlement exists in the lower section of the river downstream of the first rapid (Figure 2), the upstream areas with various rapids make access difficult and consequently reduce anthropogenic disturbance. However, some fire scars were also detected in the remote igapós. It might be possible that some of these fires were caused by natural events as observed by Moran (1990), such as frequent and intense lightning during the dry season in the study region (Christian et al., 2003). Amazonian floodplain trees represent the largest known non-ebullitive emission of methane (CH₄), produced under anaerobic conditions in the flooded soil mainly emitted by the trunk to the surface (Pangala et al., 2017). During field expeditions in igapó floodplains during the dry season, we frequently observed the enclosure of CH₄ in large trunk hollows (see Video in Supplementary Material) when wood cores from large trees were extracted by increment borers. If these trees are hit by lightning, they may act as natural fire ignition source during a severe drought period.

We showed that fire disturbances in the igapós floodplains of the JNP and its surroundings occur mainly during severe El Niño events, especially during warm IPO phases, evidencing the role of human presence as the main source of fire ignition. Based on these findings we compile in Figure 7 instrumental (Rayner et al., 2003) and reconstructed El Niño events (Cook et al., 2018) for the last 400 years. Compared to the analyzed period, past climate conditions were characterized by pluriannual El Niño episodes such as in 1606–1609, 1678–1681, 1717–1721, 1744–1746, 1790–1794, 1813–1816, 1826–1828, and 1831–1833 (Figure 7) suggesting that the study region was likely impacted by drought events with much stronger magnitudes and longer duration than any observed during instrumental periods (Parsons et al., 2018). We further argue that severe hydroclimatic droughts associated with El Niño events and large-scale fires in the igapó are intensified during warm IPO phases. Reconstructed warm IPO phases (Buckley et al., 2019; Porter et al., 2021) indicate congruence for the periods of 1610–1640, 1740–1765, at the end of the 18th century, and between 1900 and 1940. However, some periods, as 1680–1705 and during the 19th century, do not show temporal matches, probably because the proxies used for the reconstruction are related to different regions of the IPO (Buckley et al., 2019). The IPO reconstruction of Vance et al. (2015), based on a salt record from an Antarctic ice core, also suggests a warm phase for the period 1685–1720.

In the background of past El Niños and warm IPO phases, we highlight the historical background of the study region for the last 400 years (Figure 7), considering political, economic and sociocultural aspects mainly based on information from Leonardi (1999) (see Supplementary Material for more detailed information). We assume that the indigenous population inhabiting the study region before the arrival of the Portuguese colonizers made use of fire for small-scale agriculture or to develop ‘terra-preta’ soils (Bush et al., 2015). These anthropogenic soils have a large amount of pyrogenic charcoal and occur in the study region adjacent to the rivers (Clement et al., 2015). Denevan (1996) suggests that most of the ancient human populations in Amazonia settled on areas adjacent to river floodplains, preventing the populations from periodic flooding and supporting them by multiple subsistence systems of a year-around agriculture in floodplains and adjacent terra-firme areas, hunting and fishing. Pre-Columbian pyrogenic carbon derived from soil charcoal is closely related to the distance of the next major river (Sanford et al., 1985; McMichael et al., 2012; Bush et al., 2015). Soil profiles from river banks of the nearby Anavilhanas Archipelago often present charcoal fragments, some associated with archeological artifacts and logs (Latrubesse and Franzinelli, 2005). These findings suggest that fires might have affected the igapó forests of the study region during the 17th century, especially during frequent and prolonged El Niño episodes of warm IPO phases. However, anthropologic studies indicate that indigenous populations of the Negro River had a cautios and controlled use of fire for agriculture purposes (Ribeiro, 1995).

The colonization of the study region, first by the Carmelite missionaries (1694–1750) and later by the Portuguese Empire regime resulted in increasing extraction of forest products in the study region during the 18th century. Likely sporadic fires occurred in the igapós during prolonged and intense El Niño conditions, especially when warm IPO phases predominated, such as during the 1790s. This period is well-known in other tropical regions, such as the Settlement Drought in Eastern Australia (1791–1792; Palmer et al., 2015), corresponding to the East India Drought (1790–1796; Cook et al., 2010) and multiannual pluvials in subtropical regions of North (1791–1796) and South (1792/1793) America (Stahle et al., 2016; Morales et al., 2020), associated with the mega-El Niño in the period 1789–1793 (Grove, 2007). During the first half of the 19th century, the region suffered economic and demographic decay and heavy decimation of the indigenous and descendants, first by the Portuguese oppressors and later by the Brazilian Empire (Figure 7). Even if prolonged El Niño conditions occurred during the first half of the 19th century, the frequency and extension of fires in the igapós was probably reduced.

Since the creation of the Amazonas Province in 1850 with Manaus as capital (1856), this scenario changed profoundly. Steamboat navigation started in 1854 along the Negro River between Manaus and Sta. Isabel do Rio Negro and Airão became a strategic location to supply firewood for steamboats and harvests of igapó forests occurred in the lower Jaú River basin (Leonardi, 1999). The highest positive index in the IPO reconstruction based on tree rings is the year 1865 (Buckley et al., 2019), which is the driest year of a 250-year rainfall reconstruction in Eastern Amazonia within seven consecutive drought years (1864–1870) (Granato-Souza et al., 2019) associated with another short warm phase of the IPO, as suggested by Buckley et al. (2019). Another well-documented El Niño event occurred in 1877–1878, inducing severe drought hazards in many tropical regions, including the Amazon basin and the semi-arid Brazilian Northeast (Aceituno et al., 2009). The year 1878 appears as an outstanding positive IPO index in the 700-year reconstruction (Buckley et al., 2019) and is also evident in a tree ring-based reconstruction of teleconnections between El Niño and the hydrological regime at the middle Solimões River (Schöngart et al., 2004). These extreme drought events in a period of increasing harvest and human occupation in the study region likely caused frequent large-scale fires in igapó forests. After the strong El Niño episode in 1877–1878, which caused massive socioeconomic decay in the Brazilian Northeast, thousands of immigrants from these regions settled during the rubber boom distributed over a hundred locations in the lower sections of the Jaú, Carabinani and Unini rivers, but also upstream of the rapids, performing extraction of latex (Hevea spp.) and other forest products (Leonardi, 1999). This induced a profound sociocultural transformation of the study region. The use of fire in the predominantly savanna and semi-desert regions of Northeast Brazil is a common practice of these populations (Pivello, 2011), who had no intrinsic relationship with the humid tropical forests of the Amazon region (Leonardi, 1999). Intensified harvest occurred in igapó forests to extract firewood for the vulcanization of rubber and to sustain the intensifying steamboat navigation along the Negro River (Leonardi, 1999). Over the 40-year period of the rubber boom, frequent El Niño episodes occurred during predominant warm IPO phases as suggested from the reconstructed and instrumental data (Figure 7), also evident in the 200-year long floodplain tree-ring record (Schöngart et al., 2004). Strong hydroclimatic droughts induced by El Niño events occurred, for instance, in 1906 and 1916 (Marengo and Espinoza, 2016), leading to extreme low water levels of the lower Negro River (Schöngart and Junk, 2020). The strong El Niño event in the period of 1925/1926 already took place during the collapse of the rubber boom, but induced extreme hydrological drought conditions over an entire year in the Negro River basin (Williams et al., 2005), reflected by the lowest maximum water level on record. Most of the igapó forests during this event were not flooded (Piedade et al., 2013) and large-scale fires occurred in the Negro River basin as reported by a Salesian bishop from Barcelos (Sternberg, 1987; Sombroek, 2001). During the period 1950–1970 we assume again a low frequency and magnitude of igapó fires due to the predominating cold IPO phase, low El Niño frequency, declining population density and socioeconomic decay in the study region.

Overall, the provided insights suggest that fire disturbances in the igapó floodplains is not a recent disturbance vector. During periods in which intense human occupation in the study region coincided with frequent El Niño events and warm IPO phases, igapó floodplains were subjected to severe and frequent fire disturbances. This holds true especially for the period of the rubber boom (1880–1920). On the other side, periods characterized with low frequency and magnitude of El Niño events, especially during cold IPO phases, with simultaneously decline of human occupation in the study region, suggest minor fire disturbances in the floodplain landscape.