Data-driven tools for assessing ecosystem health

Abstract

This thesis investigates how in situ and satellite remote sensing data, combined via statistical and data-driven approaches, can be used to monitor coastal and terrestrial ecosystems in a scalable, cost-efficient, and scientifically robust way. The main objective of this work was to develop tools supporting the assessment and understanding of ecosystem health by exploiting the growing availability of Earth observation data. This thesis work revolves around two main facets: (i) the development of cost-efficient spatially scalable tools and (ii) the investigation of data integration of different data sources.
The research builds on satellite remote sensing data from the Copernicus mission (Sentinel-1 and Sentinel-2 data), complemented by in situ measurements from other open source repositories (such as Integrated Carbon Observation Systems (ICOS) and the European Fluxes Database Cluster) and additional remotely sensed data. All the models and algorithms used or developed during the research are published and available as open source.
The thesis starts by demonstrating the potential of satellite data as a complementary alternative to traditional in situ measurements. This was done by constructing a modeling framework for the retrieval of the shoreline position from Sentinel-1 data. The model is based on the Otsu method, a global thresholding method optimal for the recognition of the water/land interface. The resulting shorelines were validated against video monitoring systems-derived shorelines, showing sub-pixel accuracy. The results highlighted that satellite data may represent a cost-effective and low-maintenance complementary alternative to in situ measurements, especially in areas lacking dense ground-based instrumentation....