Abstract
This study investigated the application of Species Distribution Models (SDMs), based on Boosted Regression Tree (BRT) and Random Forest (RF), to predict the distribution of citrus crops in a Mediterranean climate by comparing climate data from WorldClim with those from the Regional Territorial Information System of Sicily (S.I.T.R.). To this aim, 19 bioclimatic variables were calculated from monthly temperature and precipitation data in the period 2003–2021 by using the biovars package in R software. Soil properties, terrain elevation, slope, and soil water retention capacity were considered to adequately simulate pedoclimatic conditions in the Syracuse area in Sicily (Italy). The SDM algorithms performed well (AUC: 0.84–0.93; TSS: 0.51–0.69), and Random Forest was selected to compare global and local outcomes. Using data from local meteorological stations increased the model’s reliability, resulting in a difference of approximately ~800 ha in the predicted citrus distribution compared to WorldClim data. This approach also provided a more accurate representation of precipitation patterns, for instance, in the municipality of Augusta, where WorldClim underestimated the average annual rainfall by 284 mm. These findings emphasise the importance of incorporating local environmental data into SDMs to improve prediction accuracy and inform future hybrid approaches to enhance model robustness in the context of climate change. Finally, the results contribute to expanding knowledge of citrus soil and climate conditions, with potential implications for land-use planning.