Post-fire forest recovery in the Amazon region (2001–2023): A patch-level analysis of spatiotemporal patterns and compound drivers

Post-fire forest recovery in the Amazon is critical for ecosystem resilience and carbon dynamics. However, most existing assessments rely on pixel-level approaches that fail to represent fire disturbances as coherent events, limiting understanding of recovery. Here, we construct a Landsat-based, patch-scale fire loss inventory and analyze post-fire vegetation recovery using relative NDVI trajectories for over 110,000 fire-affected forest patches across the Amazon from 2001 to 2023. Recovery patterns are examined in relation to fire characteristics, climate and hydrological conditions, and anthropogenic pressure. We find that post-fire recovery is strongly constrained across the region: only 27.6% of burned forest patches recovered within the study period, with a mean recovery time of 3.81 years. Recovery likelihood and speed declined sharply with increasing burn severity (mean recovery time: 3.53 years for low severity, 5.15 years for moderate severity, and 5.79 years for high severity), while patch geometry exerted a secondary but consistent influence, with more compact fire patches recovering more slowly. Thermal and hydrological factors showed pronounced nonlinear effects, indicating that recovery is governed by interacting energy and water constraints. Anthropogenic pressure imposed a clear threshold effect, with markedly delayed recovery in heavily modified landscapes, particularly within the Arc of Deforestation. These results highlight widespread limitations in Amazon forest resilience and provide a patch-based framework for post-fire recovery monitoring and restoration planning.