Proximal sensing for scalable mapping of shallow coastal ecosystems

Abstract

Shallow coastal habitats, such as shallow coral reefs and mangrove forests, provide invaluable services to surrounding ecosystems and coastal human populations. In recent decades, they have experienced rapid decline and are under constant threat from direct and indirect anthropogenic stressors, such as: land run-offs, water pollution, over-fishing, coastal infrastructure development, sea-level rise and ocean acidification due to climate change. Thus, it is critical to understand how these fragile environments are fairing under present-day conditions, and how they can adapt, to be able to design and implement better regulations, protection plans and recovery efforts. Creating platforms fueled by qualitative and quantitative ecological information about the biotic communities is key in the task of posing and answering relevant questions. However, in past decades, survey efforts have struggled to capture thematically detailed, temporally frequent and spatially fine-grained information of ecosystems, partially failing to set concrete baselines. The current accelerated improvement of Artificial Intelligence (AI) algorithms and the increased accessibility of powerful imaging devices provide new tools to significantly reduce the costs of ecosystem monitoring, by automating and scaling up tedious processes. Furthermore, the increase in detail of environmental monitoring introduces the possibility of new discoveries to be made, previous beliefs to be challenged and concrete baselines to be set. This doctoral study identifies the shortcomings of traditional coral reef and mangrove forests surveying methods, on both the proximal sensing scale (e.g., underwater or on-ground sensors) and remote sensing scale (e.g., air- or spaceborne sensors). Furthermore, it shows that through well designed AI work- flows, more detail and new insights on these ecosystems can be drawn, while reducing uncertainty. Moving away from sparse sampling towards dense thematic mapping provides a closer view of the underlying biodiversity in shallow coastal ecosystems, captures intra-group composition and configuration patterns, without neglecting rare species or small specimens. An environmental correlation analysis shows that more detailed sampling helps unveil the mechanisms and drivers of shifts in community composition and configuration, as well as the co-occurrence of species and substrate classes. The modern capabilities of AI workflows also enable a shift from purely areal coverage percentage studies towards organism-focused analysis. This not only facilitates in-depth spatial and temporal investigations of individuals within populations, but also reduces the error in ecosystem accounting calculations. The subsequent studies explore the ecological applications of state-of-the-art imaging platforms and novel AI workflows to automatically create habitat maps with thematic detail and individual-organism resolution, as well as showing that these analyses are spatially and thematically scalable.