Disturbance and incomplete recovery in the Cerrado-Amazon transition: Implications for conservation of a critical tropical ecotone

The Cerrado-Amazon Transition (CAT) is the world's largest tropical forest–savanna ecotone and a key component of Brazil's “Arc of Deforestation”. It harbours high biodiversity but remains weakly protected and increasingly exposed to agricultural expansion, deforestation, and fire. Using Landsat time series (1986–2020), we developed a disturbance–recovery framework to track how repeated and interacting disturbances reshape vegetation dynamics across the CAT. By combining LandTrendr to reconstruct disturbance histories, a residual neural network to distinguish disturbance pathways, and NBR-based proxy indicators to quantify disturbance-related spectral losses and post-disturbance trajectories, the framework captures both compound disturbance patterns and signals of resilience erosion. We identified four major disturbance trajectories associated with clear-cutting and fire. Over the past 35 years, approximately 493,000 km 2 of the CAT experienced at least one disturbance event, with clear-cutting dominating in Amazon forest and recurrent fire affecting both Amazon forest and Cerrado vegetation. Disturbance hotspots were concentrated near farming and ranching landscapes, while large areas of repeatedly disturbed vegetation remained outside the current protected-area network. Most fire-affected trajectories in both Amazon forest and Cerrado did not fully recover within 10 years, indicating persistent post-disturbance spectral legacies; although proxy indicator captures spectral recovery rather than all dimensions of ecosystem recovery, this pattern is consistent with widespread erosion of vegetation resilience under repeated disturbance pressure. By identifying hotspots of recurrent disturbance and incomplete recovery, and revealing major protection gaps, this study provides a practical basis for conservation planning, enforcement, and restoration across high-pressure frontier landscapes. • Developed a LandTrendr–1D ResNet framework for disturbance mapping in tropical ecotones. • Mapped 35-year disturbance and recovery dynamics across the Cerrado–Amazon Transition. • Identified hotspots of recurrent fire and clear-cutting linked to agricultural expansion. • Quantified incomplete post-fire recovery in Amazon forest and Cerrado vegetation types. • Supports conservation planning, restoration prioritisation and enforcement in high-pressure frontiers.