Spatio-temporal Analysis of Vegetation Disturbance and Recovery in the Cerrado-Amazon Transition Using Landsat Time Series and Deep Learning

Chuanze Li, Angela Harris, Beatriz Schwantes Marimon, Ben Hur Marimon Junior, Matthew Dennis, Ricardo Dalagnol, Polyanna da Conceição Bispo

The Cerrado-Amazon Transition (CAT) represents the world’s largest tropical ecotone, demarcating the boundary between the Brazilian Cerrado and Amazon biomes. Extensive deforestation and degradation within the CAT are driving irreversible ecological transformations and significant biodiversity loss. The escalating incidence of fire and agriculture-induced deforestation has rendered the CAT a dynamic ecological frontier, situated within the globally recognized ‘Arc of Deforestation’. However, our understanding of deforestation and ecosystem degradation in the CAT is limited due to insufficient knowledge regarding the spatial and temporal patterns of these disturbances. Here, we utilize satellite image time-series segmentation combined with a convolution neural network (CNN) to identify and quantify disturbances within the CAT over a 35-year period. Using the Landtrendr algorithm, Landsat time-series data, and a Residual Neural Network (ResNet), we categorized four distinct types of disturbances—Amazon forest clear-cutting, Cerrado clear-cutting, Amazon forest fire, and Cerrado fire—based on their temporal-spectral trajectories and disturbance-recovery patterns. We identified over 384,000 km² of land cover disturbance between 1986 and 2020, with Amazon forest clear-cutting accounting for the largest proportion (35%) of the detected changes. In the southern CAT, disturbances in the Cerrado vegetation were widespread, while in the northern CAT, Amazon forest disturbances indicated a gradual loss of native forest boundaries. Our findings also reveal that whilst fire caused less immediate damage to vegetation than clear-cutting, neither the Amazon forest nor the Cerrado vegetation fully recovered to their pre-fire conditions within a decade post fire. These findings emphasize the necessity of adopting targeted conservation strategies and restoration measures to mitigate the long-term ecosystem degradation of this critical ecological transition zone.