Use of remote sensing images for the assessment of the conservation status of biodiversity in protected areas

Remote sensing technologies play a vital role in monitoring forest ecosystems and assessing biodiversity. This thesis investigates the application of remote sensing technologies for forest ecosystem analysis, vegetation dynamics, and biodiversity assessment, addressing critical gaps in the field through innovative methodologies and interdisciplinary approaches. To achieve these objectives, we performed multiple studies utilizing multispectral data from Sentinel-2 and hyperspectral data from EnMAP and PRISMA sensors. The study began with a fundamental analysis of forest ecosystems using Sentinel-2 data to observe forest growth and loss patterns over time. This involved classifying deciduous and evergreen forests to understand basic forest composition. Building upon this foundation, EnMAP data, combined with Sentinel-2 temporal data, were used to classify tree species and generate biodiversity maps, employing various machine learning algorithms. The research then expanded to investigate the impact of both human-induced and natural disturbances on surrounding vegetation. A study on the effects of landfills on vegetation involved analyzing the phenological behavior of vegetation by dividing the study area into zones. Additionally, the impact of Cyclone Vaia on vegetation was assessed using the Normalized Difference Vegetation Index (NDVI), effectively identifying areas most severely affected. Two advanced studies were conducted, focusing on hyperspectral data. First, data fusion techniques were applied to enhance the spatial resolution of PRISMA data by integrating it with Sentinel-2 data. This process successfully enhanced PRISMA data to a 10-meter resolution for approximately 29 bands, using the 10-meter resolution bands from Sentinel-2 and employing various pansharpening methods, with the FIHS method demonstrating superior performance. Second, the transferability of machine learning models between PRISMA and EnMAP datasets was investigated. A Support Vector Machine (SVM) model trained with PRISMA spectral signatures and refined using 20% EnMAP data demonstrated a methodology for effectively utilizing these sensor data in conjunction, illustrating how model accuracies varied with different weighting of samples. These findings contribute to advancing remote sensing methodologies for forest monitoring and biodiversity conservation, providing actionable insights for sustainable ecosystem management, conservation planning, and disaster response.