Machine Learning and Geospatial Modeling of Climate Change Impacts on Ethiopian Honeybees for Conservation and...

Global climate change is negatively impacting honeybee production and productivity, threatening survival, health, and pollination functions which are vital for agriculture and biodiversity. Thus, this study employed integrated machine learning and geospatial modeling (Random Forest, Support Vector Machine, XGBoost, and LightGBM) to predict current and future habitat suitability in Ethiopia under SSP2-4.5 and SSP5-8.5 (2041-2080), therefore promoting conservation and climate-resilient agriculture. Variable importance analysis revealed that agro-ecological zones were the most influential predictors, accounting for 14%-22% of the variance across models. Among bioclimatic factors, Bio19 (coldest quarter precipitation) emerged as a prominent driver (14.1% in RF; 10.3% in XGBoost), indicating the importance of dry-season water availability. Model performance varied: Random Forest had the best predictive precision (specificity = 0.93); however, XGBoost better identified spatial clustering patterns. Under present conditions, Random Forest predicted 30.02% of the study area as highly suitable, especially in the Western Highlands, whereas LightGBM predicted 18.62%, showing increased habitat fragmentation. Forecasts for the future (considering only climate and static topography) indicate a significant reduction in highly suitable habitats, with a 46.2% decline under SSP5-8.5 by the 2070s. Landscape-level measurements indicated increased fragmentation, including a reduction in Shannon diversity (1.48-1.29) and a 19.2% increase in fractal dimension, indicating more complex patch topology. These findings recommended the need to restore pollinator corridors in highland refugia, promoting drought-tolerant plants like Vachellia abyssinica, and integrating adaptive apiculture approaches.