A Frontier Metrics Framework for Monitoring Short-Term Deforestation Dynamics Using Multi-Sensor Satellite Data in Central Angola

Miombo woodlands represent one of the largest tropical dry forest ecosystems in sub-Saharan Africa, yet they are increasingly affected by anthropogenic land conversion in Central Angola. This study develops a multi-sensor remote sensing and machine learning framework, designed for operational scalability and reproducibility, to quantify deforestation frontier dynamics between 2015 and 2025. Annual cloud-free composites derived from Landsat-8 and Sentinel-2 imagery, harmonised to ensure temporal consistency across sensors, were classified using a Random Forest algorithm implemented in Google Earth Engine, producing consistent land use/land cover maps across the study period (overall accuracy: 0.68–0.91; Kappa: 0.63–0.90), reflecting inherent historical cloud-cover constraints and interannual variability in training conditions. Forest cover dynamics were further analysed using spatially aggregated fractional forest-cover trajectories at 90 m resolution, enabling the derivation of annual deforestation frontier maps. A Python-based frontier metrics framework was then applied to characterise spatial–temporal patterns of forest loss and landscape transition. Results indicate a substantial reduction in open Miombo woodland extent, declining from 42,330 km² in 2015 to 30,719 km² in 2025, corresponding to a 27.4% decrease and an average loss rate of 1,007 km² yr⁻¹. In contrast, grassland expanded by 166.4%, while smallholder agriculture increased by approximately 19%, indicating progressive land conversion. Frontier analysis shows an average onset year of 2018 and a mean expansion speed of 36.4 ha yr⁻¹. Spatial pattern analysis reveals predominantly diffusion-driven expansion (98.4%), with limited evidence of leapfrogging processes. Post-deforestation transitions indicate that 52.3% of converted areas still contain transitional vegetation, while cropland and pasture account for 23.4% and 24.3%, respectively. Frontier stage assessment shows that suspended frontiers dominate (51.0%), followed by emerging (30.5%) and active frontiers (18.6%). This framework provides a scalable and reproducible approach for monitoring tropical dry forest fragmentation and informing restoration strategies aligned with SDGs 13 and 15.