Assessing coastal erosion hazards at large spatial scales: insights and uncertainties
Panos Athanasiou is a PhD student in the research group Water Management. Supervisors are prof.dr. J.C.J. Kwadijk and prof.dr. R.W.M.R.J.B. Ranasinghe from the Faculty of Engineering Technology (ET). Co-supervisor is dr.ir. A.R. van Dongeren from IHE Delft / Deltares.
Sandy coasts provide numerous benefits that attract human settlement and economic developments. At the same time, they are dynamic coastal systems, constantly changing in response to atmospheric and marine forces at different spatial and temporal scales. Coastal erosion of sandy coasts due to extreme storms or long-term changes in marine forcing is one of the most serious natural hazards, since it can directly damage coastal investments and increase flooding in low-lying coastal areas. Assessing erosion hazard at large spatial scales (from global to regional) can provide useful information for the identification of general trends and vulnerable hotspots or to guide coastal zone management and adaptation priorities at a more regional scale. So far, continental and global assessments of coastal erosion use specific geophysical datasets or assumptions to describe the coastline, without looking into the effects of these choices on the assessments. Additionally, a large spatial scale of assessment can impose computational constrains even at regional scales, where extreme storm impacts on dunes are assessed mainly with simple qualitative models.
Using a merged product of available global elevation and bathymetric data together with global wave statistics, the distribution of nearshore slopes along the global coastline is analysed. The global map of nearshore slopes produced herein, gives a first indication of the spatial variability of the sandy and non sandy nearshore profile slope along the coastlines of the world. The variability of the estimated nearshore slopes is validated in a both qualitative and quantitative manner, using available coastal classifications and in-situ surveys, showing good agreement.
The uncertainties related to the use of available geophysical datasets to describe the sandy coastline in continental scale assessments are further studied at the European scale. The implications of using the aforementioned global spatially varying nearshore slope (SVNS) dataset are compared to the assumption of a 0.01 uniform slope at all sandy coasts (commonly employed in previous global and continental studies). These effects are studied in combination with two available datasets of sandy beach occurrence, using future coastal erosion due to SLR as the diagnostic. An uncertainty analysis is performed to compare the distribution of uncertainty sources for coastal erosion projections during the 21st century.
Next, data-driven statistical methods are combined with process-based modelling to produce a fast meta-model, able to predict dune erosion during extreme events for the entire Dutch coast (260 km). Since a large-enough set of observations of dune erosion are not readily available, synthetic cases (i.e., combinations of offshore storm conditions and morphological settings) are simulated using a calibrated version of the well-established XBeach dune impact model. In contrast to simple grid-based approaches commonly used to create the synthetic cases, which can lead to a computationally prohibitive number of simulations, without necessarily adding useful information, here a novel technique that re duces a dataset of 1,400 elevation profiles at the Dutch coast to 100 real-world representative profiles, called Typological Coastal Profiles (TCPs) is presented.
Using XBeach, the dune response of the 100 TCPs to 100 physically realistic offshore storms conditions is simulated, creating a synthetic dataset of 10,000 training cases of dune erosion. A meta-model based on artificial neural networks (ANNs) is created and trained with the previously described synthetic dataset. The model follows a 2-step approach where first a classification ANN is used to assess if there is dune erosion or not and then a regression ANN provides an estimate of dune erosion volume. The meta-model needs just 10 morphological and hydrodynamic local profile characteristics and 4 offshore storm parameters as input and can produce an estimate of dune erosion volumes for the whole Dutch coast in a matter of seconds.
