Machine Learning Models For Multilevel Biodiversity Index Prediction In Apulia, Southern Italy: A Comparative Study

Environmental factors, notably air pollutants and meteorological conditions, significantly impact biodiversity distribution and ecosystem dynamics. Among these, PM10 is a major stressor, affecting plant physiological responses, reducing species abundance, and modifying soil microbial activity and nutrient cycling. These effects are often compounded when PM10 interacts with climatic conditions, particularly temperature and rainfall. Elevated temperatures contribute to habitat fragmentation and species loss, while changes in precipitation patterns influence ecological productivity and species composition. Furthermore, during extreme weather events, increased airborne particle deposition can lead to soil acidification and nutrient depletion, further compromising ecological integrity. Understanding these combined impacts is crucial for regions like Apulia, where biodiversity is shaped by both climate variability and localized human pressures. This study aims to provide biodiversity modeling for predicting Multilevel Biodiversity Index (Cazzolla Gatti and Notarnicola, 2018), taking into account the effect of PM10 concentrations and meteorological variables. A comparative analysis of the performance of Random Forests, Support Vector Machines, and Gradient Boosting algorithms, including Extreme Gradient Boosting with respect to multiple geographical regression analysis is discussed (Bayat et al, 2021; Chang, 2023). The empirical findings demonstrate a significant decline in biodiversity in areas with elevated PM10 levels, particularly in inland regions where agricultural and industrial activities are prominent.