Land-Cover Dynamics in a Miombo Woodland Landscape: A Multi-Scale Machine Learning Analysis of the Niassa Special Reserve (2000–2024)

Vollständiger Abstract

Worum geht es in dieser Arbeit?

Understanding land-cover change in savanna systems requires approaches that account for ecological gradients and human land use across multiple spatial scales. This study examines land-cover dynamics in Niassa Special Reserve (NSR), Mozambique, and its surrounding buffer from 2000 to 2024. Land cover was classified from satellite imagery using random forest, achieving ~98% overall accuracy, and change was quantified across landscape, management, and local scales. Model robustness was assessed across 1000 iterations, and classification uncertainty was propagated through 10,000 Monte Carlo simulations. Results show a shift toward denser vegetation within NSR, with increases in open and dense woodland and declines in grass-dominated areas. Closed-canopy forest increased by approximately 552 km2 within NSR (95% MCSI: 375–586 km2), but declined by approximately 1549 km2 in the surrounding buffer (95% MCSI: −1499 to −1226 km2). Community-use areas within NSR collectively lost approximately 969 km2 of closed-canopy forest (95% MCSI: −1014 to −954 km2). Dominant transitions occurred between adjacent vegetation classes, indicating structural change rather than discrete categorical conversion. These results indicate that protection is associated with greater persistence and expansion of woody vegetation at the landscape scale, while localized human use areas show contrasting trajectories.