Disentangling the drivers of deforestation and forest degradation in the Miombo landscape: A case study from Mozambique

Abstract

The fragmented and complex landscape in the Miombo landscape makes it a challenge to map and disentangle the various forest change drivers (FCD) associated with these changes and relate them to other underlying drivers. To overcome these challenges, we developed a method to spatially disentangle the drivers of deforestation (smallholder and commercial agriculture, mining, and clean-cutting charcoal), forest degradation (selective charcoal production, wildfires, logging), and forest growth (abandoned land, regrowth including plantations) in the Beira corridor, in central Mozambique. We identified ten potential FCD from the literature and created two land use and land cover (LULC) maps for 2000 and 2020 to identify areas of forest change. We used stratified random sampling based on the LULC change map, visually interpreted high-resolution satellite imagery, and NDVI time series to characterise and collected observation points of the FCD. We derived several potential underlying drivers as explanatory spatial variables. We used the random forest algorithm to evaluate their relative importance and generate a map of FCD. The forest loss due to deforestation and degradation accounts for 82.8 % (38,553.1 ha year−1) and 5.2 % (2,399.1 ha year−1), respectively, while the gain due to plantations accounts for 2.8 % (1,314.4 ha year−1) and regrowth for 9.2 % (4,297 ha year−1) of the total forest change area from 2000 to 2020. Smallholder agriculture (72.2 % of the total forest change), clear-cutting charcoal (9.1 %), abandoned land (5.4 %) and regrowth (5.7 %) were the main FCD in the study area. They are explained mainly by the intensity of change, altitude, population density and proximity to the main road. The results show a satisfactory accuracy for the LULC map (overall accuracy = 88 % and F1-score = 80 % for LULC 2000 and 90 % and 88 % for LULC 2020) and for the FCD map (overall accuracy = 79 % and F1-score = 73 %). This study provides a significant improvement in quantifying FCD by using spatially explicit data. The method could help decision-makers design land use policies better and monitor their impacts.

Keywords

• Beira Corridor
• Southern Africa
• Machine Learning
• Land-Use Change
• Underlying Drivers