Atmospheric and landscape controls on fire size in tropical dry forests: insights from the South American Gran Chaco

Fire is a dominant disturbance in tropical and subtropical dry forests and a major contributor to variability in carbon emissions, atmospheric composition, and land–atmosphere interactions. Despite their global extent and rapid transformation, the processes controlling fire size and extreme fire events in dry forest systems remain less understood than in savannas or humid tropical forests.

We investigated the controls on fire size using the South American Gran Chaco as a representative large-scale tropical dry forest system spanning strong climatic, ecological, and land-use gradients. We analyzed two decades (2001–2022) of satellite-derived fire patches from the FRY v2.0 burned-area database, combined with ERA5-Land meteorology and Fire Weather Index diagnostics, land-cover composition, landscape fragmentation metrics, topography, and anthropogenic pressure proxies. Our analysis focuses explicitly on fire size rather than fire occurrence, using statistical approaches and machine learning tools such as Random Forest models with SHAP-based interpretation to disentangle the relative and interacting roles of atmospheric forcing, landscape structure, and human-driven land transformation.

Our results show that fire size distributions are highly skewed across the region, with a small fraction of large and extreme events accounting for a disproportionately large share of total burned area. Wind and atmospheric dryness exert a strong influence on the final shape and size. At the same time, precipitation plays opposing roles by constraining fire spread through fuel moisture and enhancing fuel accumulation in fuel-limited environments. Landscape structure mediates the translation of meteorological extremes into large burned areas, with land-cover composition, fuel continuity, and fragmentation consistently ranking among the most influential predictors of burned area. Topography systematically emerges as the dominant predictor across subregions and seasons, acting not as a direct driver of fire spread but as an integrative proxy capturing hydrological gradients, vegetation structure, and human accessibility. Direct anthropogenic proxies show weaker importance at the event scale but exert strong indirect control through long-term land-use change and fuel reorganization, which in turn modulate fuel continuity and landscape configuration.

These results highlight tropical dry forests as a distinct fire domain where fire size emerges from coupled climate–biosphere–human interactions. By combining Earth observation fire products with explainable machine-learning approaches, this study advances understanding of fire–Earth system interactions and supports improved fire-risk assessment in rapidly transforming dry forest regions.