Random forest classification of bush encroachment patterns in Lake Chivero Game Park, Zimbabwe

Savanna ecosystems, dominated by a continuous grass layer and scattered trees, face threats from bush encroachment. Remote sensing data offer a cost-effective way for park managers and ecologists to monitor bush encroachment. This study provides detailed insights into vegetation changes, greatly improving upon traditional ground surveys in identifying affected areas, tracking encroachment and evaluating management efforts. This enables faster, data-driven decisions for habitat restoration and biodiversity maintenance. This study aimed to i) map the spatial and temporal distribution of land cover in Lake Chivero Game Park (LCGP), Zimbabwe; ii) detect bush encroachment using a random forest algorithm; and iii) evaluate the rate of bush encroachment within LCGP from 1986 to 2022. Landsat images from Landsat 4 and 5 TM C2 Level 1 for 1986 and 2004 and Landsat 8 and 9 OLI/TIRS C2 Level 1 for 2013 and 2022 were analysed for classification and change detection. Landsat imagery was analysed using a random forest algorithm, achieving classification accuracy between 79.89% and 95.55%, with kappa coefficients from 0.61 to 0.91. The findings revealed that LCGP vegetation cover changed in space and time due to bush encroachment. Our study contributes towards adaptive and sustainable control of encroaching species in LCGP through providing a methodology for detecting and monitoring bush encroachment. We recommend further studies to identify possible drivers and solutions to bush encroachment within the LCGP.