From MODIS active-fire detections to dynamic fire regimes: an indicator-based spatiotemporal framework for fire-regime monitoring and modeling

Understanding how Fire Regimes evolve through time is critical for assessing ecosystem change. However, temporal and spatial fire analyses are typically conducted separately, limiting the ability to link temporal changes in fire activity with their spatial Fire-Regime patterns. This study addresses this gap by developing an indicator-based framework that integrates Fire Frequency and Fire Radiative Power (FRP) as complementary indicators of fire activity, linking their temporal trajectories to spatial Fire-Regime patterns, and evaluating the sensitivity of Fire-Regime delineation and environmental associations to detection-type composition. MODIS Active-Fire detections were analyzed for both Temporal Fire Evidence and spatial–temporal Fire-Regime patterns. Temporal Fire Evidence was evaluated for both 2000–2024 and 2003–2024 using seasonal Fire Frequency, relative seasonal Fire Frequency, and the mean FRP of detections exceeding the 95th percentile. Temporal trends in these indicators were evaluated using the modified Mann–Kendall (MK) test and Sen's slope. For spatial–temporal Fire-Regime analysis, annual Fire Density and FRP for 2003–2024 were independently evaluated using the MK test, and the resulting MK Z values were classified using ±1.96 as the thresholds for significant increasing and decreasing trends, respectively, before cross-classification to derive nine Fire Regimes. The resulting Fire Regimes were modeled using machine-learning classifiers. Two modeling strategies were considered, namely Regime 1, contrasting declining versus increasing Fire Regimes, and Regime 2, contrasting Increasing Fire Frequency versus Intensifying Fire Regimes. Both strategies were modeled independently for the complete retained MODIS detections (All-Type) and for Type 0 detections representing presumed vegetation fires. Spatial Block Cross-Validation was applied to the All-Type models to assess geographic transferability, while the Type 0-only analysis provided a sensitivity assessment of the influence of detection-type composition on Fire-Regime delineation and environmental associations. Temporal Fire Evidence revealed heterogeneous temporal robustness, with the most persistent increase in Fire Frequency occurring in spring and an increase in absolute Fire Frequency also retained in winter. Spatial analysis identified nine Fire Regimes and revealed contrasting trajectories of fire occurrence and energetic intensity, with the Increasing Fire Frequency Regime occupying 17.32% of the mapped All-Type area compared with 0.03% for the Intensifying Fire Regime. Across the All-Type and Type 0-only models, the four principal predictors of Regime 1, namely elevation, Wind speed, population proximity, and rural accessibility, were common to both configurations, although their relative importance was reordered. For Regime 2, topographic position and Wind speed remained prominent in both configurations, while the relative importance of secondary predictors varied. Overall, the results demonstrate that integrating temporal indicators with spatial Fire-Regime trajectories can reveal distinct and spatially structured pathways of changing fire activity, while sensitivity to detection-type composition provides an important basis for interpreting the environmental context of these Fire Regimes. The framework therefore offers a reproducible indicator-based approach for monitoring heterogeneous Fire-Regime dynamics and identifying areas requiring further ecological and environmental assessment.