PhD Defence: On Coastal Hazards in the Baltic Sea

Kévin Dubois defends his doctoral thesis.

Main supervisor: Professor Anna Rutgersson

Opponent: Professor Eduardo Zorita-Calvo, Hereon

Abstract: Coastal regions face increasing risks from compound events, extreme sea levels, and wave hazards, all of which are expected to be influenced by ongoing climate change. In the Baltic Sea region, characterised by strong seasonal variability and semi-enclosed oceanographic conditions, projecting and managing these hazards is particularly challenging due to considerable climate-related uncertainties. This thesis explores how statistical and machine learning methods can support more accurate and flexible assessments of present and future coastal hazards. A key focus is on developing approaches to reconstruct and project high-resolution sea level and wave time series at local scales. The use of Random Forest, a machine learning algorithm, proved effective in extending historical records in data-scarce areas and enabled more robust estimates of extreme sea level probabilities by incorporating ensemble-based uncertainty quantification. The risk of flooding increases when compound events, caused by the interaction of river discharge and high sea levels, are considered at the river mouth in Halmstad, Sweden. However, results show that joint risk estimates are highly sensitive to the selection of input data and the structure of statistical models, underscoring the need for transparency and methodological care. Projections of future extremes reveal spatial variability in storm tide behaviour, and a slight decline in extreme wave heights in the Baltic Sea. However, these changes are often within the range of natural variability and differ substantially between climate models, indicating limited dependence on emissions scenarios in the case of waves. The thesis highlights the dominant role of internal atmospheric variability and underscores the importance of local, ensemble-based assessments. Overall, this work provides robust, transferable tools for coastal hazard assessment. It emphasises the value of statistical modelling to improve predictions in support of climate adaptation planning.

Read the thesis in Diva