A Statistical Approach to Detect Land Cover Changes in Mediterranean Ecosystems Using Multi-Temporal Landsat Data: The Case Study of Pianosa Island, Italy

The normalized difference vegetation index (NDVI) is commonly used to detect spatiotemporal changes of vegetation cover. This study modeled the spatiotemporal changes of land cover on Pianosa Island, Italy, in the period 1999–2015, using the multi-temporal Landsat images. Since the end of the 1990s, the natural vegetation has been re-colonizing an area of abandoned agricultural land and the island is undergoing a process of re-naturalization in harsh (drought and hot) environmental conditions. Hence, it is an ideal test site to monitor the effects of anthropogenic and climatic stressors on vegetation dynamics under Mediterranean climate. In this work, we proposed a new statistical approach based on a pixel-by-pixel analysis of multi-temporal Landsat images. Mean (µ) and standard deviation () values of the NDVI images taken in 2015 were used for the determination of the pixel thresholds (µ ± 3). The evaluation of land cover change was carried out by comparing the µ value of a single NDVI pixel for 2015 with the same pixel of different years of the study period. The results indicate that surface reflectance (SR) Landsat images are more suitable in detecting the vegetation dynamics on the island than the top of atmosphere (TOA) ones and highlight an increasing trend of vegetation cover on Pianosa Island, mainly during the early seven years following the land abandonment in all the main land cover classes: abandoned crops and pastures, Mediterranean macchia, and woodland. However, the abandoned agricultural and pasture areas showed a higher increase in the vegetation cover and a shift in the shape of the normalized frequency distribution of the SR NDVI data during the study period, suggesting that a colonization process from other vegetation classes is occurring (i.e., Mediterranean macchia and trees are colonizing the abandoned land, partly replacing herbaceous species). Our data highlight that the statistical approach applied in this study is suitable for detecting vegetation cover changes associated with anthropogenic and climatic drivers in a typical Mediterranean environment and could be proposed as a new methodological approach in several other land monitoring studies.