Antecedent Rainfall as a Predictor of Biomass Accumulation and Lagged Fire Activity Across Botswana’s Savanna Gradient

Fire regimes in southern African savannas are strongly influenced by antecedent rainfall via vegetation-driven fuel accumulation. However, the temporal structure and spatial variability of this relationship remain poorly resolved at fine spatial scales across rainfall gradients. This study quantifies how rainfall anomalies propagate through vegetation dynamics to influence fire activity across Botswana’s savanna gradient (arid <350 mm yr⁻¹; semi-arid 350–500 mm yr⁻¹; mesic >500 mm yr⁻¹).

Monthly deseasonalized anomalies (z-scores) were derived for CHIRPS rainfall, MODIS burned area (MCD64A1), fire radiative power (MOD14A1), NDVI, and EVI from 2000 to 2024. Pixel-wise cross-correlation functions (CCFs) were computed at 5 km resolution across 25,553 pixels, with significance assessed using the Bretherton et al. (1999) effective sample size correction. Distributed Lag Nonlinear Models (DLNMs) were used to distinguish short-term moisture suppression from longer-term fuel accumulation effects. Zone-level bootstrapping and empirical orthogonal function (EOF) analysis were applied to characterize uncertainty and dominant modes of variability.

Rainfall anomalies translate into vegetation greenness within one month across all zones (median lag = 1 month; 89–99% of pixels significant). The strength of rainfall–burned area coupling declines from arid to mesic zones (median R = 0.833, 0.713, 0.647), with corresponding peak lags of 8, 10, and 13 months. A pronounced southwest–northeast lag gradient is observed in the burned area but not vegetation, indicating that rainfall gradients primarily regulate the vegetation-to-fire conversion process rather than vegetation response itself. DLNM results show initial moisture-driven fire suppression (lags 0–4 months) followed by fuel accumulation effects (lags 6–14 months).

Antecedent rainfall is a strong predictor of burned area across Botswana (94–99% of pixels significant), with distinct lag structures across climatic zones. These findings support a spatially stratified fire early warning framework, emphasizing same-season rainfall monitoring in arid regions and multi-year rainfall accumulation in semi-arid systems.