The global burned area declined by nearly one-quarter between 1998 and 2015. Drylands contain a large proportion of these global fires but there are important differences within the drylands, e.g., savannas and tropical dry forests (TDF). Savannas, a biome fire-prone and fire-adapted, have reduced the burned area, while the fire in the TDF is one of the most critical factors impacting biodiversity and carbon emissions. Moreover, under climate change scenarios TDF is expected to increase its current extent and raise the risk of fires. Despite regional and global scale effects, and the influence of this ecosystem on the global carbon cycle, little effort has been dedicated to studying the influence of climate (seasonality and extreme events) and socioeconomic conditions of fire regimen in TDF. Here we use the Global Fire Emissions Database and, climate and socioeconomic metrics to better understand long-term factors explaining the variation in burned area and biomass in TDF at the Pantropical scale. On average, fires affected 1.4% of the total TDF’ area (60,208 km²) and burned 24.4% (259.6 Tg) of the global burned biomass annually at Pantropical scales. Climate modulators largely influence local and regional fire regimes. Inter-annual variation in fire regime is shaped by El Niño and La Niña. During El Niño and the forthcoming year of La Niña, there is an increment in extension (35.2 and 10.3%) and carbon emissions (42.9 and 10.6%). Socioeconomic indicators such as land management and population were modulators of the size of both, burned area and carbon emissions. Moreover, fires may reduce the capability to reach the target of “half protected species” in the globe, i.e., high-severity fires are recorded in ecoregions classified as nature could reach half protected. These observations may contribute to improving fire management.

We used the Global Fire Emissions Database, Version 4.1 (GFED4s) (Randerson et al., 2018). GFED4s includes the monthly and daily fire burned above-ground biomass from 1997 to 2020, for all fire sizes including "small fires”. Burned area covers the period 1997-2016. The information has a spatial resolution of 0.25 degrees. We included small fires to fully capture fire dynamics across the TDF in the Pantropic. To ensure that we dominantly evaluated the TDF ecosystem, we crossed the GFED4s database to the realms defined by Dinerstein et al. (2017), similar to others (Zubkova et al., 2019). The realms are redefined and updated from their previous version of the terrestrial ecoregions of the world (Olson et al., 2001). For this study, we included the Tropical & Subtropical Dry Broadleaf Forests and the TDF in Mato Grosso, Brazil (Biudes et al., 2022), and excluded the woody savannas from our analysis, such as the savannas of Africa, Australia, and South America (Lehmann et al., 2011; Moncrieff et al., 2016). A total of 52 different ecoregions were included for six Pantropical regions: (i) Caribbean Islands, Central America, and Mexico (CEAM); (ii) Northern Hemisphere South America (including Colombia and Venezuela) (NHSA); (iii) Southern Hemisphere South America (including Bolivia, Brazil, Ecuador, and Peru) (SHSA); (iv) Southern Hemisphere Africa (including Angola, Madagascar, and Zambia) (SAHF); (v) Southeast Asia (including Cambodia, India, Laos, Myanmar, Sri Lanka, Thailand, and Vietnam) (SEAS); and (vi) Equatorial Asia (EQAS).

To identify the main drivers of fires, we focused on characterizing the historical socio-ecological conditions (i.e climatic, biophysical, and socioeconomic drivers) in TDF regions for each grid-cell at 0.25° resolution. For this analysis, we used the most updated and improved biophysical and socio-economic data at the finest spatial resolution. All the selected databases have been used, either for modeling climate change or have been updated based on the most recent climatic modeling. Therefore, our results would be comparable to more recent and future studies. Also, all the spatial information has global coverage. Overall the spatial resolutions range from 90-m up to 10-km, and are interpolated to the common GFED4s grid-cell of 0.25 degrees.

Climatic variables such as temperature, precipitation, and wind speed have been recognized as key drivers of moisture availability and fire propagation (Archibald et al., 2009). Some studies used weather conditions as drivers that modulate variations in ignition efficiency, fire spread rate, and fire size (Andela et al., 2017; van der Werf et al., 2017; Jiang et al., 2020), while others included eco-climatic zones (Chuvieco et al., 2008). Contrasting to previous studies, we used the “near current climate” (1970-2000) data from WorldClim version 2.1 released in 2020’ (Fick and Hijmans, 2017) with a spatial resolution of 1-km. We used the 19 current bioclimatic variables (Bio1 to Bio 19), mean solar radiation, water vapor pressure, and wind speed. All the bioclimatic variables are derived from the monthly temperature and rainfall values. The bioclimatic variables represent annual trends (e.g., mean annual temperature, annual precipitation), seasonality (e.g., annual range in temperature and precipitation), and extreme or limiting environmental factors (e.g., temperature of the coldest and warmest month, and precipitation of the wet and dry quarters) (Fick and Hijmans, 2017). Complementarily, from Trabucco and Zomer (2019) we used potential evapotranspiration, and Priestley-Taylor alpha coefficient (Lhomme, 1997). Precipitation and temperature alone have shown to be inadequate to measure the hydrological condition (Quan et al., 2013), we evaluated water deficiency in the TDF based on two different aridity indexes. The first index considers annual precipitation and temperature (Lang index). The second index further includes reference evapotranspiration (Trabucco and Zomer, 2019). Lang aridity index is the ratio of annual precipitation to the mean annual temperature (mm per °C) (Lang, 1920). This index suggests that the rise in temperature increases water deficiency and makes the air drier if not sufficiently recharged by precipitation and/or underground water (Quan et al., 2013). Trabucco and Zomers aridity index shows moisture availability/deficit for potential growth of reference vegetation excluding the impact of soil mediating water runoff events.

Fuel availability relates to biomass favoring burning. NDVI has been used as a proxy for the conditions of vegetation, particularly for fuel availability (Jiang et al., 2020). However, we propose to use above-ground biomass (AGB) in live plants as a fuel load proxy, similar to others (Corona-Núñez et al., 2020; Tang et al., 2021). The AGB dataset refers to the epoch of the years 2000’s with a spatial resolution of 1-km (Avitabile et al., 2016). The AGB was transformed to above-ground carbon stocks (AGC) assuming a mean C concentration of 47.2% (Corona-Núñez et al., 2018). To estimate the potential biomass losses from fires we assumed that the total fuel load refers to the AGB.

Socioeconomic factors have been shown to be important drivers of deforestation, forest degradation, and modulators of fire frequency and size. In this study, we evaluated population, richness, land accessibility, and land management as drivers of fires (Chuvieco et al., 2008; Archibald et al., 2009; Andela et al., 2017; Zubkova et al., 2019). We included population density (GPWv4) and gross domestic product (G-Econ v4) with a resolution of 30 arc seconds (CIESIN, 2018). We evaluated land accessibility by two means. Firstly, we used altitude. Altitude has been a significant driver to understand forest degradation and deforestation in the tropics (Mendoza-Ponce et al., 2018; Corona-Núñez et al., 2021). Altitude comes from a digital terrain model with a spatial resolution of 90-m from the SRTM (Shuttle Radar Topography Mission - V.2.1, NASA). Secondly, we used road density at a resolution of 5 arc-minutes (Meijer et al., 2018). Roads have been shown to be important drivers to explain burned areas (Archibald and Roy, 2009; Zubkova et al., 2019) and TDF degradation (Corona-Núñez et al., 2021). To integrate land-management practices we included the proportion of croplands, irrigated agriculture (Ramankutty et al., 2010a), and pastures (Ramankutty et al., 2010b) within 5 arc-minutes grid-cell, a similar approach undertaken by others (Chuvieco et al., 2008; Archibald et al., 2009; Andela et al., 2017; Zubkova et al., 2019). The global croplands, irrigated agriculture, and pastures data set represent the proportion of land areas used in the year 2000. This data was estimated from MODIS and SPOT sensors combined with inventory data (Ramankutty et al., 2008).