Drivers of forest loss and habitat degradation in Zambia’s Greater Kafue Ecosystem
- 1School of Graduate Studies, Copperbelt University
- 2Department of Plant and Environmental Sciences, School of Natural Resources, Copperbelt University
- 3Department of Zoology and Aquatic Sciences, School of Natural Resources, Copperbelt University
Abstract
Habitat conversion threatens the Greater Kafue Ecosystem (GKE), Zambia, one of southern Africa’s most important protected-area landscapes, yet its spatial predictors remain poorly quantified. We used a Random Forest classifier to map susceptibility to natural-habitat conversion across Kafue National Park (KNP) and its surrounding Game Management Areas (GMAs). The binary response was derived from Landsat classifications for 1984 and 2024, treating natural-to-anthropogenic cover transitions as conversion and fourteen socio-economic, climatic, topographic, proximity, and conservation predictors were evaluated. Because most predictors are contemporary, the model characterises spatial correlates of cumulative conversion rather than temporal drivers. On an independent, spatially blocked test set (n = 1,751), it achieved an accuracy of 0.77, AUC-ROC of 0.83, F1-score of 0.81, and Cohen’s Kappa of 0.53, remaining stable under ten-fold spatial cross-validation (accuracy = 0.84 ± 0.07). A higher false-positive rate (0.34) than false-negative rate (0.14) indicated a precautionary bias suited to screening. Percentage cultivated land was the strongest predictor (mean decrease in impurity = 33.6%; univariate AUC = 0.91). Model-derived cut-points near 10% cultivated cover, 0.036 persons ha⁻¹, 4 km from settlements, and 898 mm rainfall discriminated higher- from lower-susceptibility conditions, though these are screening values, not validated ecological thresholds. Susceptibility mapping identified Bilili, Namwala, and Mumbwa as the principal high-susceptibility GMAs, whereas KNP remained predominantly low. The results support targeted monitoring and conservation prioritisation across the GKE and comparable Miombo landscapes.
Related
- Spatiotemporal dynamics and drivers of forest ecosystem service value in China based on the equivalent factor method
- Unraveling nonlinear ecosystem water use efficiency dynamics and environmental drivers across Iran using remote sensing and XGBoost-SHAP
- Synergistic effects of climatic drivers and anthropogenic land degradation on cascading geomorphic hazards in Iran
- Spatiotemporal patterns of habitat quality and associated drivers in Anshun City, China, based on InVEST and Geodetector
- Fishery variability across distinct habitat groups in the Indian exclusive economic zone and its teleconnections to large-scale climate drivers
BibTeX
@article{mulenga2026drivers,
title = {{Drivers of forest loss and habitat degradation in Zambia’s Greater Kafue Ecosystem}},
author = {Gift Mulenga and Darius Phiri and Matamyo Simwanda and Vincent R. Nyirenda},
journal = {Discover Forests},
year = {2026},
month = oct,
doi = {10.1007/s44415-026-00143-3},
url = {https://doi.org/10.1007/s44415-026-00143-3},
note = {CC BY},
}
RIS
TY - JOUR
TI - Drivers of forest loss and habitat degradation in Zambia’s Greater Kafue Ecosystem
AU - Gift Mulenga
AU - Darius Phiri
AU - Matamyo Simwanda
AU - Vincent R. Nyirenda
PY - 2026
DA - 2026/10/25
JO - Discover Forests
DO - 10.1007/s44415-026-00143-3
UR - https://doi.org/10.1007/s44415-026-00143-3
AB - Habitat conversion threatens the Greater Kafue Ecosystem (GKE), Zambia, one of southern Africa’s most important protected-area landscapes, yet its spatial predictors remain poorly quantified. We used a Random Forest classifier to map susceptibility to natural-habitat conversion across Kafue National Park (KNP) and its surrounding Game Management Areas (GMAs). The binary response was derived from Landsat classifications for 1984 and 2024, treating natural-to-anthropogenic cover transitions as conversion and fourteen socio-economic, climatic, topographic, proximity, and conservation predictors were evaluated. Because most predictors are contemporary, the model characterises spatial correlates of cumulative conversion rather than temporal drivers. On an independent, spatially blocked test set (n = 1,751), it achieved an accuracy of 0.77, AUC-ROC of 0.83, F1-score of 0.81, and Cohen’s Kappa of 0.53, remaining stable under ten-fold spatial cross-validation (accuracy = 0.84 ± 0.07). A higher false-positive rate (0.34) than false-negative rate (0.14) indicated a precautionary bias suited to screening. Percentage cultivated land was the strongest predictor (mean decrease in impurity = 33.6%; univariate AUC = 0.91). Model-derived cut-points near 10% cultivated cover, 0.036 persons ha⁻¹, 4 km from settlements, and 898 mm rainfall discriminated higher- from lower-susceptibility conditions, though these are screening values, not validated ecological thresholds. Susceptibility mapping identified Bilili, Namwala, and Mumbwa as the principal high-susceptibility GMAs, whereas KNP remained predominantly low. The results support targeted monitoring and conservation prioritisation across the GKE and comparable Miombo landscapes.
SN - 3059-2666
ER -