Living by the coast with accelerating sea-level rise

Abstract

This thesis addresses the increasing threat of coastal erosion under accelerating sea-level rise (SLR) by developing a global framework that integrates Earth-observation satellite data, cloud computing, and artificial intelligence (AI). Coastal erosion, a natural adjustment process of shorelines, can become a hazard when it coincides with human settlements, infrastructure, or ecosystems that cannot migrate landwards. A comprehensive understanding of erosion impacts under SLR is essential to inform climate adaptation policies. The research presents a three-part methodology, progressing from technical development to impact assessment at the scale of individual buildings worldwide.

The first part of the research focuses on creating a scalable, cloud-native framework capable of processing petabyte-scale satellite archives. Leveraging deep learning techniques, this framework classifies the global coastline at a 100-meter resolution, producing a high-resolution coastal typology that distinguishes geomorphological features such as sandy beaches, dunes, cliffs, and gravel or shingle coasts. This typology provides the necessary context to model coastal processes accurately and to constrain future shoreline projections to sediment-dominated coasts. The framework’s scalability demonstrates the feasibility of combining large-scale satellite data and AI for global coastal monitoring.

Using the derived typology, the second part of the study produces probabilistic future shoreline projections through an equilibrium-profile model. These projections estimate changes under multiple SLR scenarios, focusing on sandy sediment plains and dune coasts. By intersecting projected shorelines with global building datasets, the study quantifies exposure and identifies potentially impacted assets. The analysis reveals that 40% of the global coast is fronted by sandy, gravel, or shingle beaches—higher than prior estimates—and that 20% of these dynamic coasts are backed by cliffs, restricting natural landward migration. Approximately a quarter of the global coastline has buildings within the first kilometer of land, with around 76 million buildings globally. The “Empirical Setback Zone” metric developed here quantifies the de facto setback maintained by coastal communities, indicating that 10% of developed coasts have first buildings less than 37 meters from the shoreline.

Finally, the study provides a first-order global impact assessment. Under low- and high-emission scenarios (SSP1-2.6 and SSP5-8.5), between 1.6 and 2.4 million buildings are projected to be affected by 2100. These figures are conservative due to model constraints, which limit application to sandy and dune coasts with sufficient nearshore data. Limitations also arise from uncertainties in coastal typology and equilibrium-profile model assumptions, as well as the exclusion of full vulnerability and adaptive capacity assessments.

Scientifically, the thesis contributes both methodologically and thematically. It demonstrates a scalable AI-based workflow for analyzing large satellite datasets and provides globally consistent coastal data products, including a new transect system, high-resolution typology, exposure metrics, and probabilistic future shorelines. These outputs bridge local-scale studies with global analyses, enabling first-order asset-level impact assessments. The findings inform coastal management, supporting prioritization of high-risk areas, and advance coastal science by promoting open datasets, AI-based classification, and community-driven CoastalAI development. Continued support for open data, software, and cloud infrastructure is essential to sustain progress in intelligent coastal management and adaptation planning.