Spatio-temporal variation in dry season determines the Amazonian fire calendar

The study area encompasses an area above 6.5 million km² of the Amazon basin, including seven countries in South America, with Brazil accounting for the largest area, covering 63% (figure 1). About 31% is concentrated in the Andean countries Peru (9%), Colombia (7%), Venezuela (7%), Bolivia (7%) and Ecuador (1%). The remaining area occurs in Guyana (3%), Suriname (2%) and French Guiana (1%). In addition to forest areas, savanna patches occur mainly in Bolivia, in the far north of Brazil and Venezuela (figure 1). The Amazon has one of the highest rainfall regimes globally (Villar et al 2009), reaching annual values above 5000 mm in some parts of the basin (figure 1).

Zoom In Zoom Out Reset image size

2.2.1. Stratifying the Amazon basin according to dry season variation

2.2.2. Fire seasonality

2.2.3. Critical fire period

The seasonal variation in rainfall was spatially distributed, resulting in 74 regions with different dry season timings (figures 2 and S1). There is no dry season in 21% of the Amazon, with the largest area concentrated close to the equator, northwest of the basin (figure 2 and table S1). For the remaining area, we observed the expected opposite seasonal patterns between the north and south hemispheres. The dry season between May and September represents 41% of the area and covers most of the basin (figures 2(a) and (b)). In the northern hemisphere, in the west part, the dry season occurs between November and April, while in the east, it is concentrated between August and December, encompassing 7% and 14% of the area, respectively (figures 2(a) and (b)). In 58% of the Amazon, the dry season length varied among three (20%), four (17%), and five (21%) months (figure 2(c) and table S1). The shortest dry season lengths, one and two months, covered 14% of the basin and were adjacent to the aseasonal region. Dry season length with more than five months represented only 6% of the area and was restricted mainly to the north and northeast of the basin (figure 2(c) and table S1).

Zoom In Zoom Out Reset image size

From 2003 to 2019, an annual average of 63 710 (±31 088) active fires were observed in an area that covered 42% of the Amazon (figure 3(a) and table 1). The annual mean was less than two active fires in 62% of cells, corresponding to 18% of observations (table 1). Although only 10% of the cells had more than six active fires, they accounted for 39% of the annual mean (24 752, table 1).

Zoom In Zoom Out Reset image size

Considering the distribution of fire activity throughout the year, more than half of the annual mean active fires (32 246) occurred in the peak month of each cell, (table S2). In 43% of the cells, 30% to 50% of the monthly percentage of fire of the year occurred in the peak month (figure 3(b) and table 2(a)). These contributions were greater in cells with up to six active fires in the peak month (table 2(b)). Cells with a monthly percentage of fire in the peak month between 50% and 70% added up to 28% of the cells, with the highest occurrence of these percentages in cells with more than six active fires in the month with the highest burning activity (figure 3(b) and table 2). Furthermore, the shorter the dry season length, the more concentrated the burning activity was in the peak month, with a predominance of the monthly percentage of fire above 70% in cells with a dry season varying between zero and three months (figure 4).

Zoom In Zoom Out Reset image size

We observed two opposite fire seasons in the northern and southern hemispheres, concentrated at the beginning and end of the year, respectively (figure S3). Furthermore, we observed a well-defined spatio-temporal relationship between the fire peak and the dry season end (figures 3(c) and (d)). Similar to dry season end, the fire peak month showed three clear patterns throughout the basin with different seasons among the south, north, and eastern Amazon (figures 3(c), (d) and table S3). In 52% of the cells, mainly covering the centre-south of the basin, the fire peak occurred in August (17%) and September (35%), (figure 3(c) and table S3(a)). Concentrated in the eastern Amazon, 28% of the cells presented the fire peak in the last trimester of the year. In 14% of them, the peak occurred in October, 11% in November and 4% in December (figure 3(c) and table S3(a)). Conversely, in the north of the basin, the fire peak occurred in the first three months of the year, encompassing 14% of the cells distributed in January (4%), February (6%) and March (4%), (figure 3(c) and table S3(a)).

Regarding the time-lag to the dry season end, fire peaks were concentrated in the last two months of the dry season and the first month of the subsequent rainy season. In 47% of the cells that had a dry season defined, the fire peak and dry season end occurred in the same month (figures 3(c), (d) and table S4). Forty per cent of the cells had a one-month lag between the dry season end and the fire peak, of which 20% had a fire peak one month before the dry season end, and 20% had a fire peak in the month after the end of the dry season (figures 3(c), (d) and table S4). Considering the dry season length of the cells, the time-lag revealed a unimodal distribution for most lengths (figure 5). The greatest correspondence between the fire peak and the dry season end occurred in the dry season length of five months, accounting for 56% of the total cells (figure 5 and table S5). We also observed that the longer the dry season length, the greater the number of cells with the fire peak in the months before the end of the dry season (figure 5). Conversely, fire peaks after the dry season end were more frequent in cells with a dry season length less than three months, occurring mainly in the first month of the rainy season and representing 36% to 48% of the total of cells (figure 5 and table S5).

Zoom In Zoom Out Reset image size

We found four clusters with clear spatial patterns distributed among the northwest (C1), north (C2), south (C3), and northeast (C4), (figures 6(a) and (b)). C1 covered regions with a dry season length of less than three months and occurring at the beginning of the year. This cluster concentrated 2.38% (1519) of the annual mean active fires, with most critical periods start in January and last two to three months, which can encompass the entire dry season, as well as the subsequent onset of the rainy season (figure 6(c)). The longest critical periods were observed in C2, located in the savanna areas of northern Brazil and Venezuela, ranging from four to seven months (figures 6(b) and (c)). C2 concentrated 4.68% (2984) of the annual mean active fires, and the regions with the highest burning activity had critical periods ending in March/April (figure 6(c)). These regions showed a dry season length of four to seven months and had a more equitable distribution of the monthly percentage of fire during the critical period, with values mainly ranging between 10% and 30% (figure 6(c)). C3 covered the basin area with the highest burning activity, with critical periods mainly beginning in August and lasting for two to three months in the six regions that added up to 67% (42 837) of the annual mean active fires (figure 6(c)). In general, this period included the dry season end and the rainy season onset, with mean monthly percentages of fire for August, September, and October of 31%, 41% and 18%, respectively (figure 6(c)). The critical periods in C4 mainly included three to four months, ending in November/December. In the eight regions that concentrated 20% (12 766) of the annual mean active fires, the critical period was concentrated in the last two months of the dry season and the first month of the rainy season subsequent, accounting for about 62% and 23% of the monthly percentage of fire of the year, respectively.

Zoom In Zoom Out Reset image size

Our analysis clearly shows that fire analyses that disregards the seasonality of the dry season will misrepresent fire dynamics in several parts of the Amazon (Berenguer et al 2021). For example, an approach that uses the epicentres of deforestation to fix a single dry season for the entire basin, would misinterpret the fire dynamics in about 48% of the Amazon (table S3(a)). This includes some notable fire hotspots such as Roraima, the location of the 1997–98 megafires (Elvidge et al 2001); Colombia, where deforestation-related fires are expanding in recent years (Armenteras-Pascual et al 2011); and the eastern Amazon, including some of the deforested regions where both indigenous peoples and endemic species are highly threatened (Silva Junior et al 2020). The importance of these regions demonstrates the fact that Amazon-wide assessments of fire must consider seasonality.

Defining the present-day fire calendar could also help track how climate change could modify fire seasons, which is important given fire's key role in any large-scale forest dieback (Nobre et al 2016). Extreme droughts are affecting the Amazon with greater frequency, extension and severity (Panisset et al 2017). Furthermore, the dry season is becoming longer (Anderson et al 2018), with some regions in the Southern Amazon extending by one week per decade (Fu et al 2013). Changes in the timing and length of the dry season can modify the fire-prone period throughout the year, and the calendar presented could help to track changes in fire seasonality more clearly. In the last 40 years, some parts of the Amazon have seen a 34% reduction in rainfall between August and October, in addition to temperature increases of up to 2.5 °C (Gatti et al 2021). As our results show that 65% of the Amazon has a fire peak between August and October, these changes may be relevant to the fire calendar, particularly in the south and east regions (table S3(a)). In a trend towards a warmer dry season and even drier months, we can also expect a change in the fire type occurring, mainly due to increase of forest fires, as these conditions boost the flammability of the forest, making it a fire-prone system (Brando et al 2020). Linking fire seasonality to fire type is challenging (Chen et al 2013), but remains an important research priority to track the impacts of fire (Barlow et al 2020).

Fire prevention/fighting actions are more successful when a fire calendar is available (Anderson et al 2021, McKemey et al 2021), and its lack has led to actions that do not fully contemplate fire planning, prevention and mitigation. For example, in August 2019, the Brazilian Amazon was exposed to a large increase in fire that drew the world's attention (Barlow et al 2020). To curb this environmental crisis, the Brazilian government enacted decrees prohibiting the use of fire until the end of October. However, the control did not last long, and, as soon as the law validity expired, fire rose once again in November (Silveira et al 2020). Our analysis shows that territorial boundaries are not the most suitable limits for defining seasonally-dependent environmental policies, especially in countries like Brazil with continental dimensions. We defined three different fire periods in the Brazilian Amazon, and these should be considered for more specific policy making (figure 6 and table S6). This is especially important for a Bill to develop a National Integrated Fire Management Policy in Brazil, that has been in progress since 2018 (Brasil 2018); our results would provide science-based guidelines to incorporate the seasonality and spatio-temporal variation of the fire in the region. The fire calendar can also improve policy at smaller scales: for example, Pará, one of the Brazilian states with the highest burning activity, has two distinct fire seasons (figure 6).

The fire calendar can also help define when prevention/firefighting actions need to be implemented in relation to the dry season onset. In general, the shorter the dry season length, the faster these actions must be taken to ensure its effectiveness. This is particularly important in countries such as Colombia, where favourable conditions for fire have a shorter duration, and for this reason, the burning activity begins as soon as the dry season starts. Importantly, we found that longer dry seasons do not imply a longer Critical Fire Period, except for savanna areas where dry season length can extend beyond six months; these savannas have evolved with fire and active fires were distributed more evenly during the dry season. However, in most of the basin, if fire containment interventions were concentrated for three to four months, the reduction could reach more than 80%. Furthermore, it is interesting to note that Critical Fire Period can also cover the rainy season onset. As the fuel load at the end of the dry season is in its driest condition, the fire can still be sustained at the onset of the subsequent rainy season, stressing the need to monitor this period.

Our research did not differentiate fire types, which is an important element of determining fire containment measures. In many cases, fire is just a signal of deforestation (Armenteras-Pascual et al 2011); here, efforts should go on detecting and combating forest loss, not fire per se. Fire is often used by those who depend on it for their livelihood, consisting in these cases the only viable tool to guarantee food production. However, it is necessary to implement a fire calendar indicating allowed burning periods and how the practice should be carried out to reduce the risk of fires escaping into the remaining forests. At the same time, environmental enforcement must focus on the Critical Fire Periods, ensuring greater success in actions to curb illegal use of fire in agriculture. By defining these periods it is possible to make the best use of the scarce resources across Amazon countries (Suarez et al 2018, Levis et al 2020, Hope 2021), and would help them meet their commitments to reduce carbon emissions from forest degradation. Finally, to support decisions and policy makers, we provide a user-friendly interface with our results: https://amazonianfirecalendar.shinyapps.io/fire_amazon/.