Unravelling the Importance of Uncertainties in Global-Scale Coastal Flood Risk Assessments under Sea Level Rise

Global scale assessments of coastal flood damage and adaptation costs under 21st century
sea-level rise are associated with a wide range of uncertainties, including those in future projections
of socioeconomic development (shared socioeconomic pathways (SSP) scenarios), of greenhouse
gas concentrations (RCP scenarios), and of sea-level rise at regional scale (RSLR), as well as structural
uncertainties related to the modelling of extreme sea levels, data on exposed population and
assets, and the costs of flood damages, etc. This raises the following questions: which sources of
uncertainty need to be considered in such assessments and what is the relative importance of each
source of uncertainty in the final results? Using the coastal flood module of the Dynamic Interactive
Vulnerability Assessment modelling framework, we extensively explore the impact of scenario,
data and model uncertainties in a global manner, i.e., by considering a large number (>2000) of
simulation results. The influence of the uncertainties on the two risk metrics of expected annual
damage (EAD), and adaptation costs (AC) related to coastal protection is assessed at global scale by
combining variance-based sensitivity indices with a regression-based machine learning technique.
On this basis, we show that the research priorities in terms of future data/knowledge acquisition to
reduce uncertainty on EAD and AC differ depending on the considered time horizon. In the short
term (before 2040), EAD uncertainty could be significantly decreased by 25 and 75% if the uncertainty
of the translation of physical damage into costs and of the modelling of extreme sea levels
could respectively be reduced. For AC, it is RSLR that primarily drives short-term uncertainty (with
a contribution ~50%). In the longer term (>2050), uncertainty in EAD could be largely reduced by
75% if the SSP scenario could be unambiguously identified. For AC, it is the RCP selection that helps
reducing uncertainty (up to 90% by the end of the century). Altogether, the uncertainty in future
human activities (SSP and RCP) are the dominant source of the uncertainty in future coastal flood
risk.