Habitat suitability modelling for the African buffaloes in Northern Kenya using an ensemble approach.

Wildlife conservation relies heavily on the identification and mapping of species' geographic distributions. Many species face habitat loss and even extinction due to an increasingly fragile climate and the threat of human-induced changes. This study aims to model suitable habitats for African buffaloes using an ensemble approach. The datasets used include bioclimatic variables (temperature and precipitation), land use and land cover, normalised difference vegetation index, normalised difference built-up index, slope, proximity to settlements, and proximity to water sources. Multicollinearity analysis was used to test the correlation among the variables. An ensemble model predicting suitable habitats for the African buffaloes for the wet season (MAM) and dry season (JAS) in 2009, 2016, and 2023 was created by taking the geometric mean of three independent models: maximum entropy, random forest, and support vector machine. Results indicated that temperature and precipitation contributed the most during the modelling process. The results further indicated that from 2009 to 2023, the areal percentage of high and moderate potential areas reduced by 4.3% (158.13 km) and 30.5% (1071.43 km), respectively, while low potential areas increased by 2.2% (1229.56 km) in MAM. Additionally, high and moderate potential areas decreased by 4.9% (177.88 km) and 25.8% (966.16 km), respectively, while the low potential areas increased by 2.0% (1144.04 km) in JAS. The models' accuracy was evaluated using the area under curve (AUC), yielding AUC scores ≥ 0.96 for both seasons in each epoch, which was good. These results are valuable in supporting the conservation of buffaloes under changing climatic and environmental conditions.