Modeling Climate‐Driven Vegetation Changes Under Contrasting Temperate and Arid Conditions in the Mediterranean Basin

Abstract

ABSTRACT This study investigates climate change impacts on spontaneous vegetation, focusing on the Mediterranean basin, a hotspot for climatic changes. Two case study areas, Monti Sibillini (central Italy, temperate) and Sidi Makhlouf (Southern Tunisia, arid), were selected for their contrasting climates and vegetation. Using WorldClim's CMCC‐ESM2 climate model, future vegetation distribution was predicted for 2050 and 2080 under SSP 245 (optimistic) and 585 (pessimistic) scenarios. Two spectral indices, NDVI (temperate area) and SAVI (arid area), served as vegetation proxies, classified into three classes using K‐means (NDVI: high = mainly associated with woodlands, medium = shrublands, continuous grasslands and crops, low = discontinuous grasslands, bare soil, and rocks; SAVI: high = mainly associated with woods, olive trees, and crops, medium = shrublands and sparse olive trees, low = bare soil and saline areas). Classes validated with ESA WorldCover 2020 data and sampled via 1390 presence‐only points. A set of 33 environmental variables (topography, soil, and bioclimatic) was screened using Pearson correlation matrices, and predictive models were built using four algorithms: MaxEnt, Random Forest, XG Boost, and Neural Network. Results revealed increasing temperatures and declining precipitation in both regions, confirming Mediterranean climate trends. Vegetation changes varied by area: in the temperate area, woodlands and shrublands were stable, but discontinuous grasslands expanded. In the arid area, woodlands gained suitable habitat, while bare soil declined under the pessimistic SSP 585 scenario. Both areas showed an upward shift for shrublands and grasslands. The models indicated significant shifts in areal distribution and environmental conditions, affecting habitat suitability and ecosystem dynamics. MaxEnt emerged as the most reliable algorithm for small presence‐only datasets, effectively predicting habitat suitability. The findings highlight significant vegetation redistribution and altered ecosystem dynamics due to climate change, underscoring the importance of these models in planning for future ecological challenges.

Keywords

• climate change
• machine learning
• pastoral systems
• predictive vegetation models
• rangelands