Two decades-long NDVI anomalies to understand Mediterranean forests ecosystem shifts

Abstract Mediterranean forest ecosystems are increasingly exposed to climate change, land-use transformations and altered disturbance regimes. Understanding vegetation shifts across these complex landscapes requires scalable, high-resolution tools. In this study, we analyzed a 24-year NDVI time series (2000–2023) from Landsat imagery over the Gargano National Park (Southern Italy) to identify anomalous vegetation dynamics and assess their potential drivers. We applied the Isolation Forest algorithm to detect extreme NDVI anomalies, followed by spatial cross-analysis with climate data, wildfire records, land cover change, and forest management information. Topographic variables (elevation, slope, aspect) were also included to evaluate their role in modulating anomaly occurrence. Results revealed that approximately 4% of the forested area exhibited significant anomalies, with the highest incidence observed in conifer and broadleaf plantations (up to 39%). Native forests such as beech and mixed oak stands accounted for the largest anomalous surfaces in absolute terms. Machine learning models (XGBoost, Random Forest) and SHAP analysis highlighted maximum temperature as the most influential climatic factor ( R 2 ≈ 0.37). A considerable portion of anomalies remained unexplained by the analyzed drivers, particularly in native forests, and is interpreted here as an exploratory hypothesis of vegetation changes not captured by the available attribution framework. Topographic analysis further revealed that anomalies tend to cluster on higher elevations, steeper slopes and south-facing aspects. Our findings underscore the value of combining remote sensing time series with machine learning and topographic analysis to support adaptive forest management under changing Mediterranean conditions.