From Mapping to Risk: Two Decades of Land-Use and Land-Cover Change in Tropical Dryland–Forest Mosaics

Tropical dryland–forest mosaics host hundreds of millions of people and are among the world’s most climate-sensitive landscapes. Yet these heterogeneous regions remain difficult to monitor consistently with Earth observation due to persistent cloud cover, sparse and variable vegetation with asynchronous phenological responses to irregular rainfall pulses, and limited ground reference data. These constraints have historically capped land-use and land-cover (LULC) mapping accuracy and hindered the detection of subtle but consequential transitions in dryland ecosystems. We developed a 21-year (2000–2020) time-series remote-sensing framework that overcomes these barriers by integrating climate-driven compositing optimized for rainfall gradients, region-specific classification, and machine learning. Our approach generates multiclass, annual, cloud-free mosaics with <0.5% pixel gaps from moderate-resolution satellite imagery and maps LULC using a Random Forest model with dozens of spectral, temporal, and fraction-based predictors. External validation demonstrates a step-change in performance for heterogeneous dryland–forest environments, achieving unprecedented >90% overall accuracy and enabling reliable tracking of both forest and non-forest formations at regional scale. Applying this dataset to Northeast Brazil reveals dramatic transformations over two decades: forest cover declined by 22%, grasslands by 68%, and agriculture expanded by 140%, equivalent to roughly 10 million soccer fields, while encroachment around protected Amazon areas intensified. Building on these maps, we apply interval-level intensity analysis and spatial driver diagnostics to examine how land transformation propagates through coupled human–environment systems. Results reveal sustained periods of rapid change in the early 2000s, followed by a partial slowdown after 2013, with distinct spatial pathways of expansion for farming and non-vegetated land. Vegetation losses are strongly correlated with demographic growth, economic activity, and energy use. Critically, severe multi-year droughts affecting ~60% of the study area amplify degradation in seasonally dry tropical forests. Over 70% of forest conversion occurs within 30 km of roads, with sharp decay beyond 50 km, highlighting infrastructure as a dominant organizing force of landscape change. By linking high-accuracy mapping with change intensity, climate stress, and accessibility gradients, this work moves beyond describing where change happens to explaining how and why it propagates. Our approach demonstrates significant improvements over existing datasets, showing 29–70% spatial concordance with alternative products while achieving superior class discrimination. This open-access product (http://www.dsr.inpe.br/DSR/laboratorios/LAF) provides a transferable blueprint for monitoring land transformation and assessing socio-environmental risk across dryland–forest mosaics worldwide.