Preliminary evaluation of forest biodiversity and landscape classification using EnMAP data

Forest biodiversity and landscape classification are key issues for environmental management, particularly in the context of climate change. This thesis analyses the use of multispectral and hyperspectral remote sensing data for landscape classification and forest biodiversity assessment in the South Tyrol region. This study is divided into two main parts: the classification of land cover using the Random Forest algorithm with multispectral (Sentinel-2, Landsat-8) and hyperspectral (EnMAP) satellite imagery, and the assessment of forest biodiversity using the Spectral Variation Hypothesis (SVH). The first part focuses on evaluating the accuracy of land cover classification by comparing the performance of different satellite data, while the second part explores biodiversity estimation by relating field data to the spectral heterogeneity of images from Sentinel-2 and EnMAP. The Random Forest algorithm proved effective in identifying land cover types, including areas devastated by Storm Vaia, for all three satellites, demonstrating the value of remote sensing for monitoring environmental changes. However, the application of the SVH for biodiversity assessment has shown mixed results: while the multispectral data from Sentinel-2 have provided good results in estimating biodiversity, the hyperspectral data of EnMAP did not produce any significant correlations with field data. Despite the high spectral resolution of EnMAP, its application to the SVH has not met expectations in terms of biodiversity assessment. Rao’s Q index, used to quantify functional diversity, demonstrated its usefulness when combined with spectral data, although there were limitations in EnMAP data. This study represents the first attempt to test the SVH using EnMAP images, highlighting both the strengths and weaknesses of remote sensing technologies, with a particular focus on EnMAP's hyperspectral data, for monitoring forest ecosystems.