Quantifying coastline change uncertainty using a multi-model aggregation approach

Abstract

Evolution
of coastline position under the influence of natural and anthropogenic
processes is directly linked to the development of seaside societies. In the
context of coastal zone management, process-based morphodynamic models are often
used topredict coastline evolution and support the decision-making process
for adaptation/mitigation strategies. Frequently, the processes driving
themorphodynamic evolution transcend the applicability limits of a single
model. In those cases, model ensembles can be used to estimate coastline change
under the joint effect of the relevant processes. However, model output and in
extent the aggregated result are characterised by uncertainty originating among
others from forcing variability and parameter imprecision. The increasing
exposure ofc oastal societies to coastal recession risks and emergence of
risk-informed coastal zone management create the need for aggregated coastal
recessionestimates with quantified uncertainty. This study
investigates different statistical methods for forcing and parameter uncertainty
quantification around coastline change estimates from process-based morphodynamic
models. Subsequently a numerical convolution approach for the aggregation of the
probabilistic coastal recession estimates from multiple models was formulated to
account for the combined uncertain effect of processes acting on different timescales. The methods
of this study were applied on Anmok beach, South Korea, a coastal stretch
experiencing erosion caused by long, intermediate and short timescale processes.
Available UNIBEST-CL+ and Delft3D model schematizations from theCoMIDAS
research program, capable of simulating the relevant processes, were utilised.
Following a literature review, two methods were considered applicable for
process-based morphodynamic models: Standard Monte Carlo (SMC) and
Latin Hypercube Sampling (LHS). The application of both methods on the
UNIBEST-CL+model schematisation enabled the evaluation of their relative
performance based on the precision of the different coastline change estimates
achieved for the different sample sizes. Only LHS was applied on the Delft3D model schematisation
due to computational demands limitations. Subsequently,the
scenario-based approach currently used for the aggregation of
multi-modelcoastline change outputs was extended to explicitly account for
the uncertainties quantified in the individual model outputs. A
numerical convolution approach, using Monte Carlo sampling, was suggested for
linear super position of the contributing probability distribution functions.
The advantages of this approach include speed, ease of
implementation,comprehensibility and high resolution even at the tails of the
aggregated distributions. Utilising this approach, the effect of alternative
interventions (combinations of various breakwater designs with a small-scale
nourishment) on the coastline change probabilities was quantified. The results
showed that both methods (i.e., SMC and LHS) with adequate sampling can produce
probability distribution outputs for coastline change when applied to the
process-based models. SMC remains the most suitable method for coastline change
uncertainty quantification for models with small simulation durations. The
method gives quantified estimates of the precision, enabling the achievement
specific target precisions, with the respective computational cost. For the
smaller sample sizes used, LHS gave better precision results, proving more
suitable for models with longer computational time. On the downside, without
extra iterations of the procedure only upper estimates of the achieved precision
for a specific sample size can be obtained. The probabilistic
aggregation framework presented in this thesis has several advantages compared
to the scenario-based approach currently used. It allowsfor quantified
coastline change uncertainty estimates with the respective precision estimates
and provides the distribution of the uncertainty across its range, information
that could not be derived using the scenario-based approach. Different coastline
change percentile estimates or confidence intervals with practical use to
decision makers and the likelihood of any coastline change realisation of
interest can be evaluated. The probabilistic uncertainty quantification and
aggregation framework is believed to be useful for intervention assessment and
comparison. It allows for the assessment of uncertainty around the morphological
response under the combined effect of the processes acting on the coast
with/without the intervention and thus the evaluation of the probabilistic
intervention impact. Different interventions can be compared in terms of the
probability of inducing desired/undesired morphodynamic realisations as well as
in terms of the uncertainty range in the coastline change estimates.