ArticlePDF Available

Abstract and Figures

Many low-elevation, coral reef-lined, tropical coasts are vulnerable to the effects of climate change, sea-level rise, and wave-induced flooding. The considerable morphological diversity of these coasts and the variability of the hydrodynamic forcing that they are exposed to make predicting wave-induced flooding a challenge. A process-based wave-resolving hydrodynamic model (XBeach Non-Hydrostatic, ‘XBNH') was used to create a large synthetic database for use in a “Bayesian Estimator for Wave Attack in Reef Environments” (BEWARE), relating incident hydrodynamics and coral reef geomorphology to coastal flooding hazards on reef-lined coasts. Building on previous work, BEWARE improves system understanding of reef hydrodynamics by examining the intrinsic reef and extrinsic forcing factors controlling runup and flooding on reef-lined coasts. The Bayesian estimator has high predictive skill for the XBNH model outputs that are flooding indicators, and was validated for a number of available field cases. It was found that, in order to accurately predict flooding hazards, water depth over the reef flat, incident wave conditions, and reef flat width are the most essential factors, whereas other factors such as beach slope and bed friction due to the presence or absence of corals are less important. BEWARE is a potentially powerful tool for use in early warning systems or risk assessment studies, and can be used to make projections about how wave-induced flooding on coral reef-lined coasts may change due to climate change.
Content may be subject to copyright.
RESEARCH ARTICLE
10.1002/2017JC013204
A Bayesian-Based System to Assess Wave-Driven Flooding
Hazards on Coral Reef-Lined Coasts
S. G. Pearson
1,2
, C. D. Storlazzi
3
, A. R. van Dongeren
1
, M. F. S. Tissier
2
, and
A. J. H. M. Reniers
2
1
Department of Applied Morphodynamics, Unit of Marine and Coastal Systems, Deltares, Delft, the Netherlands,
2
Faculty
of Civil Engineering and Geosciences, Delft University of Technology, Delft, the Netherlands,
3
Pacific Coastal and Marine
Science Center, U.S. Geological Survey, Santa Cruz, CA, USA
Abstract Many low-elevation, coral reef-lined, tropical coasts are vulnerable to the effects of climate
change, sea level rise, and wave-induced flooding. The considerable morphological diversity of these coasts
and the variability of the hydrodynamic forcing that they are exposed to make predicting wave-induced
flooding a challenge. A process-based wave-resolving hydrodynamic model (XBeach Non-Hydrostatic,
‘‘XBNH’’) was used to create a large synthetic database for use in a ‘‘Bayesian Estimator for Wave Attack in
Reef Environments’’ (BEWARE), relating incident hydrodynamics and coral reef geomorphology to coastal
flooding hazards on reef-lined coasts. Building on previous work, BEWARE improves system understanding
of reef hydrodynamics by examining the intrinsic reef and extrinsic forcing factors controlling runup and
flooding on reef-lined coasts. The Bayesian estimator has high predictive skill for the XBNH model outputs
that are flooding indicators, and was validated for a number of available field cases. It was found that, in
order to accurately predict flooding hazards, water depth over the reef flat, incident wave conditions, and
reef flat width are the most essential factors, whereas other factors such as beach slope and bed friction
due to the presence or absence of corals are less important. BEWARE is a potentially powerful tool for use in
early warning systems or risk assessment studies, and can be used to make projections about how wave-
induced flooding on coral reef-lined coasts may change due to climate change.
Plain Language Summary Low-lying tropical coasts fronted by coral reefs are threatened by the
effects of climate change, sea level rise, and flooding caused by waves. However, the reefs on these coasts
differ widely in their shape, size, and physical characteristics; the wave and water level conditions affecting
these coastlines also vary in space and time. These factors make it difficult to predict flooding caused by
waves along coral reef-lined coasts. We created a system (‘‘BEWARE’’) that estimates how different wave,
water level, and reef combinations can lead to flooding. This tool tells us what information is needed to
make good predictions of flooding. We found that information on water levels and waves is most important,
followed by the width of the reef. BEWARE can be used to make short-term predictions of flooding in early
warning systems, or long-term predictions of how climate change will affect flooding caused by waves on
coral reef-lined coasts.
1. Introduction
Thousands of reef-lined tropical islands are threatened by climate change, sea level rise, and coral degrada-
tion (Ferrario et al., 2014). Many of these islands, such as atolls, have low (<4 m above MSL) maximum ele-
vations, making them particularly vulnerable to sea level rise and the impact of wave-driven flooding. For
instance, on the Gilbert, Marshall, Caroline, and Maldives island chains, over 90% of the population and land
area are located within 5 m of mean sea level (UN-Habitat, 2015). On many low-lying coral atolls, freshwater
is constrained to a relatively thin (<15 m) freshwater lens. These aquifers are susceptible to wave-driven
flooding that salinizes the freshwater lens, making the water unsuitable for human consumption (Gingerich
et al., 2017; Terry & Falkland, 2010). For example, large storm-driven wave events occurring in 2008, 2009,
and 2011 that coincided with high tides destroyed crops, demolished infrastructure, and contaminated
freshwater drinking supplies on numerous atolls in the Pacific Ocean (Fletcher & Richmond, 2010; Hoeke
Key Points:
Created large synthetic data set
relating hydrodynamics,
geomorphology, and runup on reef-
lined coasts
Offshore forcing and reef width are
the dominant controls of the
hydrodynamic response on the reef
Potential application in early warning
systems and climate change impact
assessments
Supporting Information:
Supporting Information S1
Figure S1
Figure S2
Figure S3
Figure S4
Correspondence to:
S. G. Pearson,
s.g.pearson@tudelft.nl
Citation:
Pearson, S. G., Storlazzi, C. D., van
Dongeren, A. R., Tissier, M. F. S., &
Reniers, A. J. H. M. (2017). A Bayesian-
based system to assess wave-driven
flooding hazards on coral reef-lined
coasts. Journal of Geophysical Research:
Oceans,122, 10,099–10,117. https://doi.
org/10.1002/2017JC013204
Received 20 JUN 2017
Accepted 22 OCT 2017
Accepted article online 2 NOV 2017
Published online 20 DEC 2017
The copyright line for this article was
changed on 24 JAN 2018 after original
online publication.
Published 2017. This article is a U.S.
Government work and is in the public
domain in the USA.
PEARSON ET AL. BEWARE: FLOODING ON REEF-LINED COASTS 10,099
Journal of Geophysical Research: Oceans
PUBLICATIONS
et al., 2013; U.S. Fish and Wildlife Service, 2015), underscoring the evident vulnerability of these reef-lined
island communities. Even on more mountainous tropical islands, the majority of housing, critical infrastruc-
ture, and agriculture are situated on narrow coastal plains close to sea level. The susceptibility of these trop-
ical islands to changing oceanic and climatic conditions represents a severe threat to food and water
security, public safety, and environmental health.
Wave-driven flooding hazards on coral reef-lined islands not only result from large tropical cyclones but
more commonly are the result of remotely generated (‘‘sunny-day’’) swell events (Hoeke et al., 2013).
When these waves or those from tropical cyclones encounter a reef-lined coast, they usually undergo
significant transformation due to the abrupt changes in bathymetry. Through wave breaking and bot-
tom friction dissipation, incident wave heights are reduced on the order of 90–99%, depending on the
tidal stage (Ferrario et al., 2014; P
equignet et al., 2011). However, not all of this incident sea-swell wave
energy (‘‘SS,’>0.04 Hz) is dissipated, and some is transferred to infragravity (‘‘IG,’’ 0.004–0.04 Hz) and
very low frequencies (‘‘VLF,’<0.004 Hz), generally via the breakpoint mechanism/dynamic setup on the
steep fore reef slope (Pomeroy et al., 2012b; Symonds et al., 1982). As a result, low-frequency wave
energy often dominates at the shoreline, which promotes higher runup and hence an increased chance
of flooding (Beetham et al., 2015; Cheriton et al., 2016; Gawehn et al., 2016; Roeber & Bricker, 2015; Shi-
mozono et al., 2015).
Resonant amplification occurs when the highly energetic IG and VLF wave frequencies on the reef flat coin-
cide with the nth natural frequency of the reef (f
n
):
fn52n11ðÞ
ffiffiffiffiffiffiffiffiffiffiffi
ghreef
p
4Wreef
(1)
where nis mode number, gis gravitational acceleration, h
reef
is the mean water depth on the reef flat, and
W
reef
is the width of the reef flat. This phenomenon is controlled by parameters characterizing the reef (e.g.,
morphology) and extrinsic hydrodynamic forcing (Cheriton et al., 2016; Gawehn et al., 2016; P
equignet
et al., 2009). The consequences of resonant amplification are dire because higher water levels can be
excited on the reef flat than might be expected for the incident wave conditions, resulting in flooding or
damage to coastal infrastructure (Nakaza et al., 1990; P
equignet et al., 2009; Roeber & Bricker, 2015; Tajima
et al., 2016).
Vulnerability to wave-induced flooding is spatially heterogeneous due to the highly variable morphology
of reef-lined coasts. Quataert et al. (2015) found that relatively smooth, deep, narrow reef flats fronted by
steep fore reefs are prone to higher runup and therefore increased flood risk than other coral reef mor-
phologies. Furthermore, Owen et al. (2016) noted that small variations in island topography and land use
also influence wave-driven flooding and associated impacts. Therefore, the timing and severity of wave-
induced impacts depend, in part, on island characteristics, but the uncertainty regarding the spatially var-
iable shoreline morphology contributes considerable uncertainty to the prediction of these impacts. The
vulnerability of Small Island Developing States to natural hazards is further enhanced by their small physi-
cal size, relative isolation, often limited resources (Meheux et al., 2007), and existing socioeconomic vul-
nerabilities (Ferrario et al., 2014). These vulnerabilities will likely increase in years to come due to
population growth and climate-change effects such as sea level rise (Hinkel et al., 2014; Nicholls & Caze-
nave, 2010). All these factors may affect the habitability of these islands in the next century, displacing
their people and causing internal migration or emigration. Indeed, Storlazzi et al. (2015) project that
many atolls may become uninhabitable within the next few decades, when the recurrence interval of cat-
astrophic floods becomes shorter than the recovery period for freshwater lenses, vegetation, wildlife pop-
ulations, and repair of critical infrastructure. To combat threats like these, United Nations-endorsed
Sendai Framework for Disaster Risk Reduction calls for improved access to early warning systems and
disaster risk assessments by 2030 (UNISDR, 2015).
To answer this call and plan suitable adaptations, there is a need to improve how we evaluate and predict
wave-driven flooding threats to these regions, which constitute a large, diverse set of islands and reef
morphologies, all subject to a range of offshore oceanographic conditions. Bayesian networks are proba-
bilistic models that have been successfully used to make predictions of hydrodynamics and morphology
in numerous coastal applications (Gutierrez et al., 2011, 2015; Plant & Holland, 2011; Poelhekke et al.,
Journal of Geophysical Research: Oceans 10.1002/2017JC013204
PEARSON ET AL. BEWARE: FLOODING ON REEF-LINED COASTS 10,100
2016). They do not require a detailed description of the physical pro-
cesses, as they only investigate the correlations between each vari-
able. Hirschberg et al. (2011) call for a probabilistic approach for
hydrometeorological forecasts, making Bayesian networks fit-for-
purpose, since they handle the uncertainty inherent to climate change
and short-term meteorologic-oceanographic processes. While process-
based model simulations can be time consuming, Bayesian networks
compile and provide probabilities nearly instantly once supplied with
data and trained. This speed makes Bayesian networks an ideal tool for
early warning systems, which require rapid decision making under uncer-
tainty. Furthermore, they have the potential to lower barriers to entry for
end users by presenting scientific output in a more accessible and interac-
tive way.
The main disadvantage of Bayesian networks is that they are data-
intensive, requiring sufficient input in order to derive the probabilis-
tic relationships used in their predictions. This may make their appli-
cation in data-poor environments (e.g., low-lying tropical islands) challenging. To overcome this
limitation, we used a process-based numerical wave and water level model to generate a synthetic data-
base of model results that captures a wide range of intrinsic coral reef properties and extrinsic hydrody-
namic conditions. The network then acts as an emulator or surrogate for the process-based model, as
also applied by Poelhekke et al. (2016). Given the numerous combinations of island morphologies and
physical forcing, Bayesian networks are a powerful tool for improving our prediction strategies for wave-
driven flooding threats.
This paper aims to demonstrate the use of a physics-based, deterministic numerical wave and water
level model (XBeach Non-Hydrostatic) and probabilistic Bayesian network (Netica) for estimating
wave-induced flooding of reef-fronted coastlines. First, the methodology used to construct the syn-
thetic database and design the Bayesian network is reviewed, and then the results of the network,
here termed, ‘‘Bayesian Estimator for Wave Attack in Reef Environments’’ (BEWARE), are presented
and discussed. The findings focus on the most important parameters for estimating wave-driven
flooding of reef-lined coasts, and the implications of using this system for early warning systems, cli-
mate change impact assessments, or adaptive planning such as prioritizing reef restoration projects.
The resulting Bayesian estimator is powerful in that it will enable researchers and coastal managers
to assess wave-induced flood hazards on a coast even if only approximate information is available.
The paper is organized as follows. In section 2, we discuss the methods; in section 3, the results
and analysis; in section 4, we discuss the implications of our findings and future applications; in sec-
tion 5, we provide our conclusions. Additional material regarding the XBeach Non-Hydrostatic model
validation, runup decomposition calculation, and BEWARE database are included in supporting infor-
mation (S1).
2. Methods
To construct BEWARE, the results of a validated process-based numerical wave model were combined with
a probabilistic Bayesian network. There are five steps in the methodology:
1. Schematize the reef and forcing conditions, and formulate a range of input parameters based on field
measurements and typical values from the literature, as per Quataert et al. (2015).
2. Simulate nearshore hydrodynamics for the full range of parameters using the validated process-based
wave and water level XBeach Non-Hydrostatic (XBHN) model to create a synthetic database of hydrody-
namic responses to extrinsic forcing and intrinsic coral reef geomorphology.
3. Develop a Bayesian network and train with model results.
4. Validate the Bayesian network by comparing predictions to field observations.
5. Assess the performance of the Bayesian network using techniques such a log likelihood ratios and confu-
sion matrices.
Figure 1. The idealized reef profile modeled in XBeach-Non-Hydrostatic with
the relevant hydrodynamic and morphological parameters indicated: offshore
water level with respect to the reef flat (g
0
), offshore significant wave height
(H
0
), wave steepness (H
0
/L
0
), fore reef slope (b
f
), bed roughness (c
f
), reef width
(W
reef
), beach slope (b
f
), and beach crest elevation (z
beach
).
Journal of Geophysical Research: Oceans 10.1002/2017JC013204
PEARSON ET AL. BEWARE: FLOODING ON REEF-LINED COASTS 10,101
2.1. Model Reef and Forcing Schematization
Key parameters were chosen based on findings of previous studies, following the methodology outlined by
Quataert et al. (2015). Multiple extrinsic and intrinsic parameters were covaried. The extrinsic hydrodynamic
parameters were offshore water level (g
0
), wave height (H
0
), and wave steepness (H
0
/L
0
), while the intrinsic
reef morphologic parameters examined were fore reef slope (b
f
), reef flat width (W
reef
), beach slope (b
b
),
and bed roughness (c
f
), as shown in Figure 1. Values were chosen to represent typical conditions reported
in the literature and observed at field sites (Table 1). Beach crest elevation, or island height (z
beach
), was fixed
at a height of 30 m to focus on runup as a proxy for overtopping, as per (Matias et al., 2012). However, the
method can easily be extended to include lower values of z
beach
for direct computation of overtopping.
2.2. Simulation of Nearshore Hydrodynamics
To generate the synthetic database, the process-based XBeach Non-Hydrostatic (XBNH) model (version
1.22.4867) was used with varying reef morphology and hydrodynamic forcing based on the schematization
of section 2.1. XBNH is a depth-averaged, wave-resolving model that solves the shallow water equations
including nonhydrostatic pressure (McCall et al., 2014; Smit et al., 2014; Roelvink et al., 2015). The model
was first validated using data from a fringing reef hydrodynamics laboratory experiment (Demirbilek et al.,
2007) (see supporting information S1).
An idealized 1-D reef profile was created in XBNH and varied for a range of parameter values (Table 1). This
study extends the 57 XBNH simulations of Quataert et al. (2015) to 174,372 (seven parameters, 3–12 varia-
tions per parameter, and four 30 min simulation periods with random realizations of the surface elevation
time series at the offshore boundary). A variable spin-up time was implemented to account for the differ-
ences in the time to achieve stationary conditions.
The XBNH model complexity had to be balanced carefully with time constraints because computational
demand increases exponentially with the number of parameters and variations. The scope of this study was
thus limited to remotely generated swell with unimodal JONSWAP spectra and maximum significant wave
height of 5 m, rather than more extreme cyclone conditions. The idealized setup used here has several other
limitations, in that it was one-dimensional, had spatially uniform bed roughness, and greatly simplified the
complex bathymetry characteristic of most coral reefs. The application of a one-dimensional model along a
cross-shore profile neglects some of the dynamics that occur on natural reefs, such as lateral flow. It does, how-
ever, represent a conservative estimate for IG wave generation and runup, as the forcing is shore-normal.
2.3. Development and Training of the Bayesian Network
Bayesian networks such as Netica (Norsys, 2003), which was used here, are probabilistic graphical models
that rely on Bayesian probability to make predictions. By examining the statistical relationships between
each result in the database, the network develops conditional probabilistic relationships between each
parameter, which are updated as more data (here: model results) are added (see supporting information S2).
The first step in developing a Bayesian network is to define key parameters as nodes, and then to create
links between them based on their dependencies. The eight main parameters varied in XBNH served as the
input nodes. Output nodes or ‘‘hazard indicators’’ were chosen from model variables (Table 2) that either
indicate the potential for flooding (in this case the top 2% of runup, R
2%
) or provide insight into the
Table 1
Primary XBeach Non-Hydrostatic Model Input Parameters and Their Values
Parameter Symbol Units Values
Offshore water level g
0
m21.0, 20.5, 0, 0.5, 1.0, 1.5, 2.0, 2.5, 3.0
Offshore significant wave height H
0
m 1,2,3,4,5
Offshore wave length L
0
m–
Offshore wave steepness H
0
/L
0
0.005, 0.001, 0.050
Fore reef slope b
f
1/2, 1/10, 1/20
Reef flat width W
reef
m 0, 50, 100, 150, 200, 250, 300, 350, 400,
500, 1,000, 1,500
Beach slope b
b
1/5, 1/10, 1/20
Coefficient of friction c
f
0.01, 0.05, 0.10
Journal of Geophysical Research: Oceans 10.1002/2017JC013204
PEARSON ET AL. BEWARE: FLOODING ON REEF-LINED COASTS 10,102
hydrodynamic processes acting on reefs, such as mean wave period (T
m-1,0
). The main Bayesian network
configuration used here (Figure 2) assumes that all input parameters influenced each of the output
parameters.
In order to represent the model variables as probability distributions, the data set was divided into bins (Fig-
ure 2). Input parameters were discretized using the same parameter values as were tested in XBNH, result-
ing in uniform distributions. Discretization of output variables required more careful consideration, since
they were continuously distributed. Furthermore, because of nonlinear processes, a uniform input may yield
a nonuniform output distribution, which influences bin discretization. The chosen bins took into account
both the distribution of data and the desired precision of the estimates. Care was taken to prevent overfit-
ting by checking that increasing the number of bins per node did not increase validation error rates, as per
Gutierrez et al. (2015). The last step in constructing the Bayesian network was to train it using the synthetic
data set created with XBNH, resulting in the prior predictions (i.e., probability distributions of all model
results in the absence of additional information or constraints). Simulations that did not achieve stationary
conditions before the end of the spin-up period (30–120 min, depending on reef width) were excluded
from the Bayesian network.
2.4. Bayesian Network Validation
The Bayesian network was constructed using a synthetic data set, so it needed to be validated using field
data from case studies. However, there are limited suitable field measurements available, so at this time
only a field data set from Roi-Namur in the Republic of the Marshall Islands (Quataert et al., 2015; Cheriton
et al., 2016; Gawehn et al., 2016) and numerical model results from Funafuti, Tuvalu (Beetham et al., 2015),
were used for validation. The availability of additional runup time series recorded on fringing coral reef-
fronted beaches would provide more opportunities to test the network, although this is hampered by the
dearth of published R
2%
field measurements in such locations.
To test the network, input nodes were constrained based on the prescribed hydrodynamic boundary condi-
tions and given reef geomorphology. The posterior probability distributions of runup, wave height at the
toe of the beach (SS, IG, and VLF frequencies), and wave setup were then compared with their observed val-
ues. An ideal prediction would show a narrower posterior distribution (indicative of precision) that is cen-
tered on the observed value (representative of accuracy).
2.5. Assessment of Bayesian Network Performance
The performance of the Bayesian network (how often estimates are correct) was assessed by comparing the
predictive skill of different configurations, and by testing its accuracy in predicting a subset of the database
Table 2
Primary XBeach Non-Hydrostatic Output Parameters, Calculated at the Inner Reef Flat Unless Otherwise Noted
Parameter Symbol Units
Significant sea/swell wave height (0.04–1 Hz) H
m0,SS
m
Significant infragravity wave height (0.004–0.04 Hz) H
m0,IG
m
Significant very low frequency wave height (0.001–0.004 Hz) H
m0,VLF
m
Significant low-frequency wave height (0.001–0.04 Hz) H
m0,LF
m
Wave setup gsetup m
Mean water depth (averaged across entire reef flat) h
reef
m
Extreme water level (mean of values greater than
2% exceedance value)
g2%m
Sea/swell contribution to g2%(0.04–1 Hz) g2%;SS m
Infragravity contribution to g2%(0.004–0.04 Hz) g2%;IG m
Very low frequency contribution to g2%(0.001–0.004 Hz) g2%;VLF m
Runup (2% exceedance value) on beach slope R
2%
m
Runup (mean of values greater than 2% exceedance value) R2%m
Sea/swell contribution to R2%(0.04–1 Hz) R2%;SS m
Infragravity contribution to R2%(0.004–0.04 Hz) R2%;IG m
Very low frequency contribution to R2%(0.001–0.004 Hz) R2%;VLF m
Mean spectral period T
m21,0
s
Mean spectral frequency f
m21,0
Hz
Journal of Geophysical Research: Oceans 10.1002/2017JC013204
PEARSON ET AL. BEWARE: FLOODING ON REEF-LINED COASTS 10,103
Figure 2. Layout of the ‘‘Bayesian Estimator for Wave Attack in Reef Environments’’ (BEWARE) system, illustrating key parameters and the links between
them. Extrinsic hydrodynamic input parameters are shaded in blue, intrinsic reef morphologic input parameters in teal, and output variables (calculated at
the inner reef flat/beach toe) in yellow. Within the nodes are a histogram indicating the prior probability distributions, mean, and standard deviation for
each parameter (n5174,372). The key parameters are defined in Tables 1 and 2. Negative values of R
2%
can be explained by cases where g
0
<0and
runup on the fore reef does not exceed the reef crest. Negative values of g
setup
can be explained by cases where the W
reef
50 and set down is occurring
at the observation point.
Journal of Geophysical Research: Oceans 10.1002/2017JC013204
PEARSON ET AL. BEWARE: FLOODING ON REEF-LINED COASTS 10,104
that had been excluded from the training. The first means of evaluating the predictive skill of the Bayesian
network here was through the use of confusion matrices, which break down predictive error rates into over-
prediction and underprediction. A confusion matrix thus provides a ‘‘hit rate’’ for the Bayesian network,
identifying how often the network correctly predicts what was observed in reality, or, in this case, calculated
by the XBNH model. To compute the error rates and confusion matrices, a k-fold cross validation was per-
formed, as per Poelhekke et al. (2016) and Gutierrez et al. (2015). This entailed randomly dividing the data-
base into k(in this case 10) folds or subsets, excluding them one at a time from the training, and comparing
the network predictions of the excluded data with the actual values. Although there is no restriction on the
size of confusion matrices, their complexity increases greatly with the number of bins for a given output
node. Hence, only a binary confusion matrix (two bins) was considered here, such that a variable lying
below a set threshold was considered negative, and positive above the threshold.
The second method for assessing the predictive capability of the network was using the log likelihood ratio
(LLR). The LLR is an indicator of predictive skill and model uncertainty that compares the prior predictions
of a network with the posterior predictions made using additional information (Plant & Holland, 2011). The
concept is explained in more detail by, van Verseveld et al. (2015), Gutierrez et al (2015), and Poelhekke
et al. (2016). When the LLR is calculated for key parameters, it makes it possible to consider which parame-
ters should be included in the Bayesian network, which parameter uncertainty should be constrained, and,
thus, which field measurements are most important to collect. By withholding parameters from the network
one at a time and comparing the resulting predictions with those of the full network, the relative impor-
tance of each parameter was assessed. For this study, the LLR score for each withheld parameter was nor-
malized by the LLR score of the full network.
3. Results and Analysis
3.1. XBNH Validation on Reefs
In order to validate XBNH and the parameter settings for wave transformation and runup on a fringing reef,
the model was tested against the Demirbilek et al. (2007) laboratory-derived experimental data set of cases
without wind, similar to the data sets used by Nwogu & Demirbilek (2010), Zijlema (2012), and Shimozono
et al. (2015) to validate their numerical models.
Modeled H
m0
at the inner reef flat shows good agreement with the laboratory data across the 29 tested
cases (Figure 3a), albeit with slight underestimation (R
2
50.786, bias 520.098). The model shows greater
skill at estimating wave setup at the inner reef flat, also with a slight negative bias (R
2
50.946,
bias 520.046), as shown in Figure 3b. The scatter in R
2%
predictions is wider but shows a positive correla-
tion and slight overestimation (R
2
50.642, bias 510.098), as shown in Figure 3c. The results of this valida-
tion suggest that XBNH can simulate reef hydrodynamics with reasonable accuracy, and give us the
Figure 3. Scatterplots of observed (Demirbilek et al., 2007) and computed properties at the inner reef flat (a) wave height
(H
m0
), (b) setup, and (c) runup (R
2%
) for all 15 tested cases.
Journal of Geophysical Research: Oceans 10.1002/2017JC013204
PEARSON ET AL. BEWARE: FLOODING ON REEF-LINED COASTS 10,105
confidence to use it in the subsequent analysis. More information regarding the validation can be found in
supporting information (S1).
3.2. XBNH Results
The nearshore hydrodynamic results of all XBNH simulations were aggregated to enable an examination of
general trends across the entire synthetic data set (Figure 4). Extreme water levels on the inner reef flat
(g2%) and runup (R2%) were defined as the mean of the highest 2% of the water level time series for each
simulation at the inner reef flat and waterline, respectively. We focus on gat the inner reef flat because it is
commonly measured in reef hydrodynamics studies (i.e., Cheriton et al., 2016; Merrifield et al., 2014), and on
R
2%
because it can be used as a proxy for overtopping and potential flooding (Matias et al., 2012). These
Figure 4. For the full set of XBeach Non-Hydrostatic simulations, variations in (a–g) extreme water levels, g2%, and (h–n) runup, R2%;as a function of the seven pri-
mary input parameters (Table 1). The different colors represent the mean relative contribution of water level (g
0
), setup, and each wave frequency band (VLF, very
low frequency, 0.004–0.001 Hz; IG, infragravity, 0.04–0.004 Hz; SS, sea swell, >0.04 Hz) to the total water level and runup. Results have been filtered to show only
cases with g00(n5136,032).
Journal of Geophysical Research: Oceans 10.1002/2017JC013204
PEARSON ET AL. BEWARE: FLOODING ON REEF-LINED COASTS 10,106
values were then decomposed into separate components (i.e., g0;gsetup ;g2%;SS ;g2%;IG ;g2%;VLF ) to yield
insight into the nature of wave transformation across the reef. This decomposition calculation is described
in supporting information (S3).
Some trends confirm a priori expectations: the extreme water level parameters g2%and R2%both increase
with increasing g
0
and H
0
(Figures 4a, 4b, 4h, and 4i), and both decrease for high H
0
/L
0
(Figures 4c and 4j).
In addition, the highest extreme water levels occur with low c
f
(Figures 4d and 4k), small W
reef
(Figures 4f
and 4m), steep b
f
(Figures 4e and 4l), and steep b
b
(Figures 4g and 4n), which concurs with the findings of
Quataert et al. (2015) and Shimozono et al. (2015). When runup (R
2%
) is evaluated as a function of reef width
and friction (Figure 5), it increases with reduced width (consistent with Shimozono et al., 2015) and reduced
friction (consistent with Quataert et al., 2015).
For a constant water level, as offshore wave heights increase, extreme water levels and runup at the shore-
line become primarily driven by setup and reef flat waves. With increasing H
0
, the combined contribution of
gsetup ,g2%;SS ,g2%;IG , and g2%;VLF (i.e., excluding g
0
) to total g2%increases from 20% to 60% (Figure 4b), and
similarly, the contribution of Rsetup ,R2%;SS ,R2%;IG , and R2%;VLF to total runup (R2%) rises from 23% to 67%
(Figure 4i). The large contribution of setup and reef flat waves to total extreme shoreline water levels and
runup during the occurrence of large offshore waves reinforces the importance of including these parame-
ters in predictions of flooding on reef-lined coasts, as opposed to simpler ‘‘bathtub’’ models that only
account for offshore water levels. Variations in H
0
/L
0
(Figures 4c and 4j) and b
f
(Figures 4e and 4l) have little
effect on the relative composition of g2%and R2%, but there is a proportionally larger R2%;IG component at
lower values of c
f
(23% of R2%at c
f
50.01, up from 11% at c
f
50.1; Figure 4k), indicating the importance of
frictional dissipation to resulting infragravity wave dynamics over reef flats. This relationship has been indi-
cated by field data (Cheriton et al., 2016) and physics-based models (Pomeroy et al., 2012a), particularly in
relation to resonance.
While offshore wave forcing is important, mean offshore water level (g
0
) was found to have the strongest
effect on the relative proportions of setup and reef flat waves to overall extreme shoreline water levels and
runup. Wave-driven setup makes up the largest proportion of g2%at lower values of g
0
(61% at 0 m; Figure
4a), but its influence decreases with increasing water depth on the reef flat (6% at 3 m); this inverse relation-
ship between water level setup and offshore water levels is well established (Becker et al., 2014; Beetham
et al., 2015; Vetter et al., 2010). Infragravity and VLF waves make a fairly constant contribution across the full
range of modeled water levels (g2%;IG :l50.14 m, r50.02 m; g2%;VLF :l50.25 m, r50.02 m); this rela-
tive insensitivity of low-frequency waves to offshore water level was also observed in the field (Beetham
et al., 2015; Merrifield et al., 2014). However, for the short-period waves, since they are depth-limited, as the
mean water level over the reef flat increases, the contribution of g2%;SS to the total g2%increases (from
0.15 m [11%] at 0 to 0.72 m [17%] at 3 m). However, while the magnitude of the contribution of SS waves to
runup, R2%;SS , increases with increasing g
0
, the relative proportion does not (from 0.16 m [10%] at 0 to
Figure 5. Runup (R2%Þas a function of reef width (W
reef
) and friction coefficient (c
f
), averaged across all tested XBeach
cases (n5174,372).
Journal of Geophysical Research: Oceans 10.1002/2017JC013204
PEARSON ET AL. BEWARE: FLOODING ON REEF-LINED COASTS 10,107
0.36 m [8%] at 3 m). This modulation of SS waves by offshore water level is consistent with previous findings
(Beetham et al., 2015; Merrifield et al., 2014; Storlazzi et al., 2011). Because increases in g
0
can be considered
a proxy for sea level rise, these findings imply that sea level rise may result in a greater contribution to
extreme shoreline water levels and runup from SS-band reef flat waves.
Although the composition of g2%is relatively insensitive to different beach slopes, b
b
, (Figure 4g), R2%
shows some variation (Figure 4n), with the contribution of R2%;SS increasing with steeper b
b
(from 2% to
15% of total). This trend was also noted by Shimozono et al. (2015), as runup is inversely proportional to dis-
sipation in the surf zone. Furthermore, higher-frequency waves break more readily on milder beach slopes
where they are depth limited (Brocchini & Baldock, 2008).
The overall trends for g2%and R2%are similar, although R2%is consistently 27% higher than g2%for the
same forcing conditions (Figure 6). This may have important implications for inferring R2%and flooding
characteristics from measurements on the reef flat in the absence of direct R2%measurements. Although
directly correlated (R
2
50.916), g2%by itself does not fully translate to R2%, as a result of continued wave
transformation on the beach slope.
In this section, we examined the average of all conditions in the data set, but there is considerable variation
around the mean. These variations can be attributed, in part, to particular combinations of parameters that
yield anomalously high runup, such as those that result in resonant amplification, which is explored in the
next section.
3.3. Reef Flat Resonance
As reef flat resonance may account for anomalously high runup (Gawehn et al., 2016; Nakaza et al., 1990;
Nwogu & Demirbilek, 2010; Shimozono et al., 2015), the model results were analyzed to determine if reso-
nant conditions were present. A peak in H
m0,VLF
wave height at the inner reef flat was identified for narrower
(50 to 250 m) reefs (Figure 7a). In order to verify whether this peak was related to resonance, cases of rela-
tively high (top 30%) IG and VLF waves, defined as Hm0;IG=H0

2>0:3 and Hm0;VLF=H0

2>0:1, respectively,
were isolated for additional analyses. The data from these cases reveal a distinct peak (Figure 7b) at the res-
onant frequency (f
m21
/f
n0
5160.1), indicating that resonance is likely occurring. Frictional and geometrical
effects can explain deviations from the theoretical resonant frequency (van Rijn, 2011). Furthermore, not all
high-energy VLF waves are necessarily resonant: some may also be standing or progressive waves (Gawehn
et al., 2016).
Figure 6. Extreme water levels at the inner reef flat (g2%) versus runup on the beach slope (R2%). Results have been fil-
tered to show only cases with g00 and z
beach
530 m (n5136,032).
Journal of Geophysical Research: Oceans 10.1002/2017JC013204
PEARSON ET AL. BEWARE: FLOODING ON REEF-LINED COASTS 10,108
To quantify the differences between mean conditions and resonant cases, the full set of model results was
filtered, selecting only those cases meeting the aforementioned resonance criteria, and the same analyses
of comparing the contributions to extreme shoreline water levels and runup were carried out for only these
resonance cases (Figure 8). In general, g2%and R2%were slightly higher for the resonant cases in Figure
8 when compared to the full set of simulations shown in Figure 4. The starkest difference between the
resonance-only and full set of simulations is found with the g2%and R2%trends as a function of reef width:
as W
reef
increases, g2%and R2%values from the full set of model results gradually decline (Figures 4f and
4m), while the g2%and R2%values for the resonance cases increase sharply (Figures 8f and 8m). The dis-
continuity at 500 m is due to the fact that none of the simulations with W
reef
>500 m met the aforemen-
tioned resonance criteria. Furthermore, the percent contribution of g2%;IG to g2%is larger than that of
g2%;VLF for W
reef
<300 m, whereas the percent contribution of g2%;VLF to g2%is larger at W
reef
>300 m (Fig-
ure 8f), likely due to resonance. When theoretical resonant frequencies are calculated for a mean reef flat
depth (h
reef
) of approximately 1.5 m across all simulations using Equation (1), W
reef
<240 m should be reso-
nant at IG frequencies, and resonant at VLF frequencies when W
reef
>240 m. A similar shift from IG to VLF
energy on wider reefs was also demonstrated by Shimozono et al. (2015).
Many of these resonant cases are associated with greater mean reef flat water depths (h
reef
), whether due
to higher g
0
or setup (Figure 8a). This coincides with the expected response, since greater water depth over
the reef increases the resonant frequency and reduces the effects of frictional dissipation (P
equignet et al.,
2009; Pomeroy et al., 2012a). Shimozono et al. (2015) also posited that the increase from extreme water lev-
els at the inner reef flat to runup on the shoreline can be partly attributed to resonant runup amplification
along the beach slope. However, this effect was found to be minor for the steep range of b
b
tested here.
The trends in percent contribution of setup and the different wave frequency components to total extreme
water levels and runup are relatively similar between the resonant cases (Figure 8) and the full set of simula-
tions (Figure 4).
Though large R2%;VLF values are seen for W
reef
250–500 m in resonance cases (Figure 8m), a similar
increase in runup as a function of reef width is not seen for the full suite of cases (Figure 4m) because reso-
nant cases are rare (n57,608) relative to all others in the data set (n5136,032). However, given that the
synthetic data set is based on uniform input distributions, and T
p
values that are a function of H
0
(via the
steepness parameter), it is possible that resonant cases are underrepresented here compared to what might
be expected in the field.
3.4. Bayesian Network Validation
In the next step, the Bayesian network was trained using the results from XBNH, and validated against field
data from the 18 November 2013 runup event on Roi-Namur in the Republic of the Marshall Islands (Cheri-
ton et al., 2016; Quataert et al., 2015), and model simulations of the 23 June 2013 runup event on Funafuti,
Figure 7. (a) VLF wave height as a function of reef width (n5174,372). The red line indicates the mean, blue box indi-
cates 25th and 75th percentiles, and black whiskers denote the 5th and 95th percentiles, with red dots indicating upper-
range outliers. (b) Normalized, squared VLF wave height (H
m0,VLF
/H
0
)
2
as a function of the ratio between mean spectral
frequency at the inner reef flat (f
m21,0
) and the reef’s zeroth resonant frequency (f
n,0
). Points close to f
m21,0
/f
n,0
51 (108)
are near resonance.
Journal of Geophysical Research: Oceans 10.1002/2017JC013204
PEARSON ET AL. BEWARE: FLOODING ON REEF-LINED COASTS 10,109
Tuvalu (Beetham et al., 2015). The prior distributions in Figure 2 (also presented in Figure 9 in a different for-
mat) indicate the default prediction for each hazard indicator without any additional information (dark blue
bars in Figure 9). When the observed hydrodynamic forcing and reef characteristics are introduced to the
Bayesian network, the hazard predictions are constrained, yielding the posterior distributions. The network’s
predictive skill is judged based on whether the peak of the posterior prediction matches the observed value
for each hazard variable. In the Funafuti cases, results are presented as maximum runup (R
max
) and contri-
butions to R
max
(R
max,SS
and R
max,LF
), so we used R2%,R2%;SS , and R2%;LF to make comparisons rather than
R
2%
,H
m0,SS
, and H
m0,LF
as with Roi Namur.
Figure 8. For the XBeach Non-Hydrostatic (XBNH) model resonance cases, variations in (a–g) extreme water level, g2%, and (h–n) runup, R2%;as a function of the
seven primary input parameters (Table 1). The different colors represent the mean relative contribution of water level (g
0
), setup, and each wave frequency band
(VLF, very low frequency, 0.004–0.001 Hz; IG, infragravity, 0.04–0.004 Hz; SS, sea swell, >0.04 Hz) to the total water level. Resonant cases were filtered from the full
set of XBNH simulations presented in Figure 4 by selecting only simulations where f
m21,0
/f
n,0
5160.1 (n57,608).
Journal of Geophysical Research: Oceans 10.1002/2017JC013204
PEARSON ET AL. BEWARE: FLOODING ON REEF-LINED COASTS 10,110
The network underestimates R
2%
for Roi-Namur (Figure 9a), and overestimates R2%for the Funafuti high tide
case (Figure 9m), albeit by a single bin (20 cm elevation difference). However, it correctly predicts runup for
the low and mid tide Funafuti cases (Figures 9e and 9i). Although the network overpredicts H
m0,SS
f>0:04 HzðÞon Roi-Namur (Figure 9b), it successfully estimates R2%;SS for the Funafuti cases (Figures 8f, 8j,
and 8n). H
m0,LF
and R2%;LF f<0:04 HzðÞare confidently and correctly predicted by the network in three of
the tested cases (Figures 9c, 9g, and 9k), although the bimodal posterior probability distribution for Funafuti
at high tide (Figure 9o) suggests lower confidence in that prediction. Setup is overestimated for the Funafuti
cases (Figures 9h, 9l, and 9p) but correctly predicted for Roi-Namur and Funafuti at high tide (Figure 9d).
Figure 9. The ‘‘Bayesian Estimator for Wave Attack in Reef Environments’’ (BEWARE) system validation against case stud-
ies. Figures 9a–9d correspond to the runup event on 18 November 2013 in Roi Namur, Republic of the Marshall Islands
(Cheriton et al., 2016; Quataert et al., 2015). Figures 9e–9p correspond to a runup event on Funafuti, Tuvalu modeled by
Beetham et al. (2015). Please see Table 2 for parameter definitions. Note that for these test cases, H
m0,LF
and R2%;LF
encompass the full range of low-frequency waves from 0.001 to 0.04 Hz (i.e., including infragravity waves), as per the con-
vention used in the data sources. The dark blue bars represent the prior probability distribution for all cases in the net-
work, and the lighter blue bars represent the posterior probability distributions, based on the hydrodynamic forcing and
reef characteristics of each test case. The yellow triangles indicate the observed values of each variable from the case
study.
Journal of Geophysical Research: Oceans 10.1002/2017JC013204
PEARSON ET AL. BEWARE: FLOODING ON REEF-LINED COASTS 10,111
These results suggest that the Bayesian network is capable of predicting wave transformation processes
across the reef for the majority of the limited cases that were accessible, but that it is less able to capture
the final transformation from the inner reef flat across the beach. Discrepancies could be explained in part
by the schematized nature of the XBNH model when compared to the real conditions of the reefs, or to dif-
ferences in the runup/wave height decomposition calculation used by the validation cases. Underestima-
tion of the Roi-Namur case may also be explained by the available measurements—the observed runup
value is a single wave event that was captured on camera and may not correspond well with a statistical
value like R
2%
. The limited published data available for validation of flooding hazards on reefs underscores
the great need for further field measurements.
3.5. Bayesian Network Predictive Skill
After validating the Bayesian network on field and model data, its prediction of the XBNH results was tested,
making it possible to draw on a much larger pool of data for comparison. This provides some insight into
how often the Bayesian network overpredict or underpredict certain hazard indicators. The confusion matri-
ces in Table 3 indicate that the network has high positive prediction rates for large values of R
2%
,g
setup
,
T
m21,0
,H
m0,SS
, and H
m0,IG
(>96% correct) but is slightly less skilled at predicting large H
m0,VLF
(85%). The
high predictive skill for most of the hazard indicators (particularly runup) suggests that the Bayesian net-
work acts here as a suitable proxy for XBNH, similarly to the findings of Poelhekke et al. (2016) for the
XBeach Surf Beat model.
3.6. Bayesian Network Log Likelihood Ratio
After establishing the validity and performance of the network, the LLR test was used to investigate the rela-
tive importance of each parameter within the Bayesian network. When the normalized LLR was calculated
for R
2%
by withholding each input parameter successively from the network, it scored much lower when H
0
,
g
0
, and W
reef
were not taken into account (Figure 10a). This indicates that those parameters are more impor-
tant for making successful predictions of runup (and by extension flooding) than parameters such as H
0
/L
0,
c
f,
b
f
,orb
b
. Compared to the runup, setup is less sensitive to c
f
(Figure 10b), as the setup is dominated by a
balance between the radiation stress gradients and the pressure gradient with the friction force being a
smaller component. The T
m21,0
(Figure 10c) is sensitive mostly to W
reef
and g
0
as these parameters control
the degree to which the energy shifts to lower frequencies. H
m0,SS
(Figure 10d) shows similar behavior as
the runup. Compared to the short waves, H
m0,IG
(Figure 10e) is less dependent on g
0
because it is not satu-
rated and more determined by the breakpoint-generation and frictional dissipation processes as evidenced
the sensitivity to H
0
,H
0
/L
0
and c
f
.H
m0,VLF
(Figure 10f) shows poor predictive skill without H
0
,H
0
/L
0
,b
f
, and
W
reef
, indicating both offshore forcing and geometry are important to this response. This may reflect the
highly nonlinear nature of processes controlling H
m0,VLF
(i.e., resonance), and may also explain the lower pre-
dictive accuracy observed for this variable in Table 3.
Table 3
Confusion Matrices Depicting the Accuracy of the Bayesian Network in Predicting the XBeach Non-Hydrostatic (XBNH) Model
Output Parameters (Table 2) for a Given Set of Input Conditions (i.e., Validation Error Rates)
R2% (m)
Hm0,SS
(m)
Hm0,IG
(m)
ηsetup
(m)
Hm0,VLF
(m)
Tm-1,0
(s)
Predicted
Predicted
Predicted
Predicted
Predicted
Predicted
Predicted
Observed
Observed
Observed
Observed
Variable
Observed
Observed
Observed
Note. Values in the tables indicate the percentage of observed cases falling into a given prediction bin. Green values
along the main diagonal indicate correct predictions, whereas the bottom left corner indicates the false negative rate
(underpredictions) and the top left indicates the false positive rate (overpredictions).
Journal of Geophysical Research: Oceans 10.1002/2017JC013204
PEARSON ET AL. BEWARE: FLOODING ON REEF-LINED COASTS 10,112
4. Discussion
This study presents a proof-of-concept for the use of a process-based model and Bayesian network to esti-
mate flood hazard indicators such as runup on low-lying coral reef-lined coasts. This section examines the
sensitivity of key parameters and potential future applications in early warning systems and climate change
impact assessments.
4.1. Relevance of Key Parameters
The results of the LLR tests (Figure 10) across the entire data set indicate that hydrodynamic forcing (e.g.,
H
0
,g
0
,andH
0
/L
0
) is most important to an accurate prediction of the hydrodynamic response on the reef,
followed by morphological characteristics. Of these, reef geometry parameters (e.g., b
f
and W
reef
)are
more influential than the frictional characteristics (c
f
) and beach slope b
b
. The LLR tests imply that reef
properties that can be more easily obtained via remote sensing (e.g., W
reef
and b
f
)mayprovide(tofirst
order) more useful information than detailed, labor-intensive field surveys to measure parameters such as
bed roughness (c
f
). Similarly, although measurements of beach slope b
b
are unavailable for most low-
lying tropical islands, the LLR tests suggest that it is not critical for effective prediction of flooding
hazards.
4.2. Bayesian Network Improvements
The network’s validation could be improved by training it with additional XBNH simulations that include
higher resolution of input parameters (e.g., H
0
50.5, 1.5, 2.5, 3.5, 4.5 m). Furthermore, the predictive skill of
a Bayesian network improves when it has multiple cases from which to learn and gain experience (Poel-
hekke et al., 2016). At present, four cases were simulated with random wave forcing for each combination of
input parameters; the network’s experience could be improved by increasing this number. Lastly, the
Figure 10. Log likelihood ratio (LLR) comparisons of key output variables (Table 2) for withheld parameters. The dashed line at y 51 shows the normalized total
LLR for the full network with all parameters included. Each of the circles represents the total LLR for a network where that parameter has been withheld from the
prediction, normalized by the total LLR for the full network. A value of 1 would indicate that removing a given parameter does not affect the network’s predictive
skill, whereas a value of 0 means that the parameter is essential to making predictions of a particular output variable.
Journal of Geophysical Research: Oceans 10.1002/2017JC013204
PEARSON ET AL. BEWARE: FLOODING ON REEF-LINED COASTS 10,113
discretization of the Bayesian network’s output bins directly influences the accuracy and precision of its pre-
dictions, so sensitivity to alternate configurations could be carried out. For instance, if the observed value
was 0.51 and the predicted bin was 0.00–0.50, it would still be regarded as an incorrect prediction, even
though it was very close.
4.3. Early Warning Systems
The Bayesian network presented in this study can be used in an early warning system (EWS) to predict
flooding. Currently, most operational EWSs are capable of predicting offshore wave heights, tides and
surges, but not onshore hazards such as runup and flooding because these are computationally expensive
to predict. Following Poelhekke et al. (2016), one solution is to precompute the range of offshore forcing
and onshore hazards, compile the results in a Bayesian network, and then couple this network to an EWS.
Predictions of offshore forcing can be then used to obtain constrained posterior probability distributions of
the onshore hazard at negligible computational expense in operational mode. The alternative method of
incorporating a wave transformation model such as XBNH directly into an EWS (as was done in Bosserelle
et al., 2015) does not have these advantages of speed and capability to quantify the hazard uncertainty.
In the case of an EWS for low-elevation, reef-lined islands, the offshore forcing can be obtained as follows:
tides can be computed from a deterministic prediction, surges may be predicted using a hydrodynamic
model such as Delft3D (although they are usually less important on steep-sloped coasts like atolls), and off-
shore wave predictions can be obtained from existing operational models such as WAVEWATCH-III (Tolman,
2009).
If coupled with 2-D inundation models, the EWS can be extended to predict flooding occurrence, timing
and extent, which can be used for land use planning and evacuation purposes. Furthermore, if building
characteristics are known, damage to structures could also be estimated using simple stage-damage rela-
tionships or more sophisticated approaches where sufficient data is available. van Verseveld et al. (2015)
and J
ager et al. (2015, 2017) have used Bayesian networks to predict direct economic damage to houses
and infrastructure resulting from surge and wave-induced flooding on sandy, urbanized coastlines.
4.4. Climate Change Impact Assessments
In addition to EWS, the BEWARE system can be used to investigate hypothetical climate change scenarios,
such as changes to sea level, wave climate, or reef roughness due to coral degradation or restoration. Shope
et al. (2016) used the formulation of Stockdon et al. (2006) (developed on the basis of runup data obtained
on sandy sloping beaches under nonextreme offshore forcing) to estimate Pacific island runup under future
climate change scenarios. The BEWARE system developed for this study could provide a more comprehen-
sive estimate than those based on the Stockdon et al. (2006) equations by accounting for input uncertainty
and considering the full suite of processes involved in reef hydrodynamics (including resonance) and the
resulting wave-driven flooding.
The reaction to climate change does not have to be passive—mitigating measures can be taken by affected
island communities to improve resilience to flooding. The value of coral reefs as nature-based flood
defenses can also be analyzed with this model and used to prioritize conservation or restoration efforts. Fer-
rario et al. (2014) demonstrated that reef restoration is a more cost-effective solution for coastal risk reduc-
tion on coral reef-lined islands than the construction of artificial breakwaters. Given the scarce resources
available for such projects, the BEWARE system can be used to understand which coral reef-lined areas are
most vulnerable and where coral restoration can provide the largest return in terms of coastal hazard risk
reduction.
Since reef roughness is correlated to its coral health (Baldock et al., 2014), degradation of coral due to
bleaching or ocean acidification may reduce its ability to effectively dissipate wave energy (Sheppard et al.,
2005; Quataert et al., 2015). Conversely, restoration efforts that improve coral ecosystem quality (Fox et al.,
2005; Haisfield et al., 2010) may increase roughness and thus provide more effective wave attenuation. The
health of reef ecosystems under different climate change or restoration scenarios could be accounted for
by examining a given location’s sensitivity to c
f
in the model. It is thus possible that significant increases in
roughness brought on by reef restoration could help offset some of the effects of sea level rise on wave-
induced flooding.
Journal of Geophysical Research: Oceans 10.1002/2017JC013204
PEARSON ET AL. BEWARE: FLOODING ON REEF-LINED COASTS 10,114
Although the LLR analysis (Figure 10) suggests that c
f
is a less important parameter than W
reef
for determin-
ing wave transformation and the resulting water levels, it should be noted that it is easier to influence c
f
by
coral restoration than it is to change W
reef
. Thus, restoration is a viable strategy for flood risk reduction on
coral reef-lined islands.
5. Conclusions
The ‘‘Bayesian Estimator for Wave Attack in Reef Environments’’ (BEWARE) system for estimating flooding
hazards on coral reef-lined coasts was developed by training a Bayesian network with a synthetic database
generated by XBeach Non-Hydrostatic (XBNH) model simulations. The XBNH process-based numerical wave
and water level model is shown to be capable of reproducing wave transformation processes on fringing
reefs, including resonant reef flat amplification. BEWARE improves system understanding of reef hydrody-
namics, building on previous work by examining the intrinsic and extrinsic factors controlling runup on
reef-lined coasts.
BEWARE shows high predictive skill for flooding conditions from the XBNH model, and was validated for a
limited number of case studies. Using the log likelihood ratio test, it was found that offshore wave condi-
tions, water level, and reef width are the most important parameters required to estimate extreme water
levels and runup on reef-fronted coasts using a Bayesian network, whereas having knowledge of the reef
roughness or beach slope appears less important.
BEWARE has the potential to form the basis for early warning systems and scenario assessment applications
on reef-lined coasts. The applicability of the BEWARE system can be further enhanced if supplemented by
key parameters (e.g., reef flat width) obtained from remote sensing platforms as well as field measurements
of reef hydrodynamics.
References
Baldock, T. E., Golshani, A., Callaghan, D. P., Saunders, M. I., & Mumby, P. J. (2014). Impact of sea-level rise and coral mortality on the wave
dynamics and wave forces on barrier reefs. Marine Pollution Bulletin,83(1), 155–164. https://doi.org/10.1016/j.marpolbul.2014.03.058
Bayes, T., & Price, M. (1763). An essay towards solving a problem in the doctrine of chances. Philosophical Transactions of the Royal Society
of London,53, 370–418. https://doi.org/10.1093/biomet/45.3-4.293
Becker, J. M., Merrifield, M. A., & Ford, M. (2014). Water level effects on breaking wave setup for Pacific Island fringing reefs. Journal of Geo-
physical Research: Ocean,119, 914–932. https://doi.org/10.1002/2013JC009373
Beetham, E. P., Kench, P. S., O’callaghan, J., & Popinet, S. (2015). Wave transformation and shoreline water level on Funafuti Atoll, Tuvalu.
Journal of Geophysical Research: Ocean,120, 1–16. https://doi.org/10.1002/2014JC010472
Bosserelle, C., Kruger, J., Movono, M., & Reddy, S. (2015). Wave inundation on the Coral Coast of Fiji. In Australasian coasts & ports conference
2015 (pp. 96–101). Auckland, New Zealand: Engineers Australia and IPENZ.
Brocchini, M., & Baldock, T. E. (2008). Recent advances in modeling swash zone dynamics: Influence of surf-swash interaction on nearshore
hydrodynamics and morphodynamics. Reviews of Geophysics,46, RG3003. https://doi.org/10.1029/2006RG000215
Cheriton, O. M., Storlazzi, C. D., & Rosenberger, K. J. (2016). Observations of wave transformation over a fringing coral reef and the impor-
tance of low-frequency waves and offshore water levels to runup, overwash, and coastal flooding. Journal of Geophysical Research:
Ocean,121, 3121–3140. https://doi.org/10.1002/2015JC011231
Demirbilek, Z., Nwogu, O., & Ward, D. (2007). Laboratory study of wind effect on runup over fringing reefs (Tech. Rep. ERDC/CHL TR-07–8).
Washington, DC: U.S. Army Corps of Engineers.
Ferrario, F., Beck, M. W., Storlazzi, C. D., Micheli, F., Shepard, C. C., & Airoldi, L. (2014). The effectiveness of coral reefs for coastal hazard risk
reduction and adaptation. Nature Communications,5(May), 1–9. https://doi.org/10.1038/ncomms4794
Fletcher, C. H., & Richmond, B. M. (2010). Climate change in the Federated States of Micronesia: Food and water security, climate risk manage-
ment, and adaptive strategies (report). Honolulu: University of Hawai‘i.
Fox, H. E., Mous, P. J., Pet, J. S., Muljadi, A. H., & Caldwell, R. L. (2005). Experimental assessment of coral reef rehabilitation following blast
fishing. Conservation Biology,19(1), 98–107. https://doi.org/10.1111/j.1523-1739.2005.00261.x
Gawehn, M., van Dongeren, A. R., van Rooijen, A., Storlazzi, C. D., Cheriton, O. M., & Reniers, A. J. H. M. (2016). Identification and classification
of very low frequency waves on a coral reef flat. Journal of Geophysical Research: Ocean,121, 7560–7574. https://doi.org/10.1002/
2016JC011834
Gingerich, S. B., Voss, C. I., & Johnson, A. G. (2017). Seawater-flooding events and impact on freshwater lenses of low-lying islands: Control-
ling factors, basic management and mitigation. Journal of Hydrology,551, 676–688. https://doi.org/10.1016/j.jhydrol.2017.03.001
Gutierrez, B. T., Plant, N. G., & Thieler, E. R. (2011). A Bayesian network to predict coastal vulnerability to sea level rise. Journal of Geophysical
Research,116, F02009. https://doi.org/10.1029/2010JF001891
Gutierrez, B. T., Plant, N. G., Thieler, E. R., & Turecek, A. (2015). Using a Bayesian network to predict barrier island geomorphologic character-
istics. Journal of Geophysical Research: Earth Surface,120, 2452–2475. https://doi.org/10.1002/2014JF003270
Haisfield, K. M., Fox, H. E., Yen, S., Mangubhai, S., & Mous, P. J. (2010). An ounce of prevention: Cost-effectiveness of coral reef rehabilitation
relative to enforcement. Conservation Letters,3(4), 243–250. https://doi.org/10.1111/j.1755-263X.2010.00104.x
Hinkel, J., Lincke, D., Vafeidis, A. T., Perrette, M., Nicholls, R. J., Tol, R. S. J., ... Levermann, A. (2014). Coastal flood damage and adaptation
costs under 21st century sea-level rise. Proceedings of the National Academy of Sciences of the United States of America,111(9), 3292–
3297. https://doi.org/10.1073/pnas.1222469111
Acknowledgments
This work was funded by the U.S.
Department of Defense’s Strategic
Environmental Research and
Development Program (RC-2334), the
U.S.G.S. Coastal and Marine Geology
Program, by the European
Community’s 7th Framework
Programme through the grant to the
budget of RISC-KIT, contract 603458,
and Deltares Strategic Research in the
‘‘Hydro- and morphodynamics during
extreme events’’ program (1230002).
We thank the two anonymous
reviewers for their positive and
constructive feedback, which we feel
has improved the quality of our
manuscript. Thanks also to Maarten
van Ormondt (Deltares) for sharing his
XBNH preprocessing MATLAB code,
Robert McCall (Deltares) for his
support in the early stages of the
project, Wiebke J
ager (TU Delft) and
Nathaniel Plant (USGS) for their
invaluable guidance on Bayesian
networks, and to Olivia Cheriton
(USGS) for her thoughtful suggestions
and comments. Use of trademark
names does not imply USGS
endorsement of products. The
BEWARE database (174,372 line table
of input parameters and output
variables presented in this report) is
available as a NetCDF (*.nc) file in
supporting information Data Set S1.
Data are hosted at the following
location: https://doi.org/10.5066/
F7T43S20.
Journal of Geophysical Research: Oceans 10.1002/2017JC013204
PEARSON ET AL. BEWARE: FLOODING ON REEF-LINED COASTS 10,115
Hirschberg, P. A., Abrams, E., Bleistein, A., Bua, W., Monache, L. D., Dulong, T. W., ... Stuart, N. (2011). A weather and climate enterprise stra-
tegic implementation plan for generating and communicating forecast uncertainty information. Bulletin of the American Meteorological
Society,92(12), 1651–1666. https://doi.org/10.1175/BAMS-D-11-00073.1
Hoeke, R. K., McInnes, K., Kruger, J., McNaught, R. J., Hunter, J. R., & Smithers, S. G. (2013). Widespread inundation of Pacific islands triggered
by distant-source wind-waves. Global and Planetary Change,108, 128–138. https://doi.org/10.1016/j.gloplacha.2013.06.006
J
ager, W. S., Christie, E. K., Hanea, A., Den Heijer, M. C., & Spencer, T. (2017). A Bayesian network approach for coastal risk analysis and deci-
sion making. Coastal Engineering, https://doi.org/10.1016/j.coastaleng.2017.05.004. Retrieved from htt p://www.sciencedirect.com/sci-
ence/article/pii/S0378383917300625.
J
ager, W. S., den Heijer, C., Bolle, A., & Hanea, A. (2015). A Bayesian network approach to coastal storm impact modeling. Paper presented at 12th
international conference on applications of st atistics and probability in civil e ngineering, University of British Columbia, Vancou ver, BC.
Matias, A., Williams, J. J., Masselink, G., & Ferreira, O. (2012). Overwash threshold for gravel barriers. Coastal Engineering,63, 48–61. https://
doi.org/10.1016/j.coastaleng.2011.12.006
McCall, R. T., Masselink, G., Poate, T. G., Roelvink, J. A., Almeida, L. P., Davidson, M., & Russell, P. E. (2014). Modelling storm hydrodynamics
on gravel beaches with XBeach-G. Coastal Engineering,91, 231–250. https://doi.org/10.1016/j.coastaleng.2014.06.007
Meheux, K., Dominey-Howes, D., & Lloyd, K. (2007). Natural hazard impacts in small island developing states: A review of current knowl-
edge and future research needs, Natural Hazards,40(2), 429–446. https://doi.org/10.1007/s11069-006-9001-5
Merrifield, M. A., Becker, J. M., Ford, M., & Yao, Y. (2014). Observations and estimates of wave-driven water level extremes at the Marshall
Islands. Geophysical Research Letters,41, 7245–7253. https://doi.org/10.1002/2014GL061005
Nakaza, E., Tsukayama, S., & Hino, M. (1990). Bore-like surf beat on reef coasts. In Proceedings of 22nd international conference on coastal
engineering (pp. 743–756). Reston, VA: American Society of Civil Engineers.
Nicholls, R. J., & Cazenave, A. (2010). Sea level rise and its impact on coastal zones, Science,328(2010), 1517–1520. https://doi.org/10.1126/
science.1185782
Norsys. (2003). Netica—Application for belief networks and influence diagrams: User’s guide. Vancouver, BC: Author.
Nwogu, O., & Demirbilek, Z. (2010). Infragravity wave motions and runup over shallow fringing reefs. Journal of Waterway, Port, Coastal,
and Ocean Engineering,136(December), 295–305. https://doi.org/10.1061/(ASCE)WW.1943-5460.0000050
Owen, S. D., Kench, P. S., & Ford, M. (2016). Improving understanding of the spatial dimensions of biophysical change in atoll island coun-
tries and implications for island communities: A Marshall Islands’ case study. Applied Geography,72, 55–64. https://doi.org/10.1016/
j.apgeog.2016.05.004
P
equignet, A.-C., Becker, J. M., Merrifield, M. A., & Aucan, J. (2009). Forcing of resonant modes on a fringing reef during tropical storm Man-
Yi. Geophysical Research Letters,36, L03607. https://doi.org/10.1029/2008GL036259
P
equignet, A.-C., Becker, J. M., Merrifield, M. A., & Boc, S. J. (2011). The dissipation of wind wave energy across a fringing reef at Ipan, Guam.
Coral Reefs,30(Suppl. 1), 71–82. https://doi.org/10.1007/s00338-011-0719-5
Plant, N. G., & Holland, K. T. (2011). Prediction and assimilation of surf-zone processes using a Bayesian network. Part I: Forward models.
Coastal Engineering,58(1), 119–130. https://doi.org/10.1016/j.coastaleng.2010.09.003
Poelhekke, L., J
ager, W. S., van Dongeren, A. R., Plomaritis, T. A., McCall, R. T., & Ferreira, O. (2016). Predicting coastal hazards for sandy
coasts with a Bayesian network. Coastal Engineering,118, 21–34. https://doi.org/10.1016/j.coastaleng.2016. 08.011
Pomeroy, A., Lowe, R. J., Symonds, G., van Dongeren, A. R., & Moore, C. (2012b). The dynamics of infragravity wave transformation over a
fringing reef. Journal of Geophysical Research,117, C11022. https://doi.org/10.1029/2012JC008310
Pomeroy, A., van Dongeren, A. R., Lowe, R. J., van Thiel de Vries, J. S. M., & Roelvink, J. A. (2012a). Low-frequency wave resonance in fringing
reef environments. Coastal Engineering Proceedings,2012, 1–10.
Quataert, E., Storlazzi, C. D., van Rooijen, A., Cheriton, O. M., & van Dongeren, A. R. (2015). The influence of coral reefs and climate change
on wave-driven flooding of tropical coastlines. Geophysical Research Letters,42, 6407–6415. https://doi.org/10.1002/2015GL064861
Roeber, V., & Bricker, J. D. (2015). Destructive tsunami-like wave generated by surf beat over a coral reef during Typhoon Haiyan. Nature
Communications,6, 7854. https://doi.org/10.1038/ncomms8854
Roelvink, J. A., van Dongeren, A. R., McCall, R., Hoonhout, T., van Rooijen, B., van Geer, A., ...Quataert, E. (2015). XBeach technical reference:
Kingsday release (technical report). Delft, the Netherlands: Deltares.
Sheppard, C., Dixon, D. J., Gourlay, M. R., Sheppard, A., Payet, R., & Sheppard, C. (2005). Coral mortality increases wave energy reaching
shores protected by reef flats: Examples from the Seychelles. Estuarine, Coastal and Shelf Science,64(2–3), 223–234. https://doi.org/10.
1016/j.ecss.2005.02.016
Shimozono, T., Tajima, Y., Kennedy, A. B., Nobuoka, H., Sasaki, J., & Sato, S. (2015). Combined infragravity wave and sea-swell runup over
fringing reefs by super typhoon Haiyan. Journal of Geophysical Research: Ocean,120, 4463–4486. https://doi.org/10.1002/2015JC010760
Shope, J. B., Storlazzi, C. D., Erikson, L. H., & Hegermiller, C. A. (2016). Changes to extreme wave climates of islands within the Western Trop-
ical Pacific throughout the 21st century under RCP 4.5 and RCP 8.5, with implications for island vulnerability and sustainability. Global
and Planetary Change,141, 25–38. https://doi.org/10.1016/j.gloplacha.2016.03.009
Smit, P. B., Roelvink, J. A., van Thiel de Vries, J. S. M., McCall, R. T., van Dongeren, A. R., & Zwinkels, J. R. (2014). XBeach: Non-hydrostatic
model. Delft, The Netherlands: Delft University of Technology.
Stockdon, H. F., Holman, R. A., Howd, P. A., & Sallenger, A. H. (2006). Empirical parameterization of setup, swash, and runup. Coastal Engi-
neering,53(7), 573–588. https://doi.org/10.1016/j.coastaleng.2005.12.005
Storlazzi, C. D., Elias, E. P. L., & Berkowitz, P. (2015). Many atolls may be uninhabitable within decades due to climate change. Scientific
Reports,5, 14546. https://doi.org/10.1038/srep14546
Storlazzi, C. D., Elias, E. P. L., Field, M. E., & Presto, M. K. (2011). Numerical modeling of the impact of sea-level rise on fringing coral reef
hydrodynamics and sediment transport. Coral Reefs,30(Suppl. 1), 83–96. https://doi.org/10.1007/s00338-011-0723-9
Symonds, G., Huntley, D. A., & Bowen, A. J. (1982). Two-dimensional surf beat: Long wave generation by a time-varying breakpoint. Journal
of Geophysical Research,87(C1), 492–498. https://doi.org/10.1029/JC087iC01p00492
Tajima, Y., Shimozono, T., Gunasekara, K. H., & Cruz, E. C. (2016). Study on locally varying inundation characteristics induced by super
typhoon Haiyan part 2: Deformation of storm waves on the beach with fringing reef along the east coast of Eastern Samar. Coastal Engi-
neering Journal,58(1), 1640003. https://doi.org/10.1142/S0578563416400039
Terry, J. P., & Falkland, A. C. (2010). Responses of atoll freshwater lenses to storm-surge overwash in the Northern Cook Islands. Hydrogeol-
ogy Journal,18(3), 749–759. https://doi.org/10.1007/s10040-009-0544-x
Tolman, H. L. (2009). User manual and system documentation of WAVEWATCH-IIITM version 3.14 (technical note). Camp Springs, MD: U. S.
Department of Commerce.
UN-Habitat. (2015). Urbanization and climate change in small island developing states. Nairobi, Kenya: Author.
Journal of Geophysical Research: Oceans 10.1002/2017JC013204
PEARSON ET AL. BEWARE: FLOODING ON REEF-LINED COASTS 10,116
UNISDR. (2015). Sendai framework for disaster risk reduction 2015–2030. Sendai, Japan: Author.
U.S. Fish and Wildlife Service. (2015). U.S. Fish and Wildlife Service pacific region. Retrieved from https://www.fws.gov/pacific/
van Rijn, L. C. (2011). Principles of fluid flow and surface waves in rivers, estuaries, seas and oceans. Amsterdam, The Netherlands: Aqua
Publications.
van Verseveld, H. C. W., van Dongeren, A. R., Plant, N. G., J
ager, W. S., & den Heijer, C. (2015). Modelling multi-hazard hurricane damages on
an urbanized coast with a Bayesian Network approach. Coastal Engineering,103, 1–14. https://doi.o rg/10.1016/j.coastaleng.2015.05.006
Vetter, O., Becker, J. M., Merrifield, M. A., P
equignet, A.-C., Aucan, J., Boc, S. J., & Pollock, C. E. (2010). Wave setup over a Pacific Island fring-
ing reef. Journal of Geophysical Research,115, C12066. https://doi.org/10.1029/2010JC006455
Zijlema, M. (2012). Modelling wave transformation across a fringing reef using SWASH. Coastal Engineering,2012, 1–12. https://doi.org/10.
9753/icce.v33.currents.26
Journal of Geophysical Research: Oceans 10.1002/2017JC013204
PEARSON ET AL. BEWARE: FLOODING ON REEF-LINED COASTS 10,117
... As both surf zone IG wave generation and wave setup are enhanced by SS wave dissipation, the intense SS wave breaking observed on many coral reefs can lead to large IG waves and wave setup (Becker et al., 2016;Buckley et al., 2018a;Pomeroy et al., 2012). Certain reef geometries can also trap and resonantly amplify specific IG frequencies leading to enhanced coastal flooding (Becker et al., 2016;Buckley et al., 2018a;Demirbilek et al., 2009;Gawehn et al., 2016;Merrifield et al., 2014;Pearson et al., 2017;Péquignet et al., 2009;Roeber & Bricker, 2015;Shimozono et al., 2015). ...
... To efficiently simulate IG waves in nearshore applications, surfbeat models, which couple SS wave models with flow models on the time scale of wave groups, were developed (Quataert et al., 2015;Roelvink et al., 2009;van Dongeren et al., 2013). Whilst the wave runup from wave setup and IG swash is computed by these surfbeat models, SS swash motions are neglected or parameterized, which can result in substantial inaccuracies in the prediction of runup (Buckley et al., 2018a;Lashley et al., 2018;Pearson et al., 2017;Quataert et al., 2020). Finally, phase-resolving wave-flow models that treat waves and flow simultaneously have also been applied (Buckley et al., 2014;Lowe et al., 2019;Ma et al., 2014;Rijnsdorp et al., 2021;Roeber & Cheung, 2012;Yao et al., 2012Yao et al., , 2020aYao et al., , 2021Zijlema, 2012). ...
... 3 of 27 A number of laboratory, field, and numerical model studies have been undertaken to quantify the effectiveness of reefs at reducing wave runup (Baldock et al., 2014;Ferrario et al., 2014;Pearson et al., 2017;Quataert et al., 2015). However, a major limitation of many of these studies has been the lack of a detailed accounting of the effects of large bottom roughness on wave-driven hydrodynamics. ...
Article
Full-text available
Wave breaking on the steep fore‐reef slopes of shallow fringing reefs can be effective at dissipating incident sea‐swell waves prior to reaching reef shorelines. However, wave setup and free infragravity waves generated during the sea‐swell breaking process are often the largest contributors to wave‐driven water levels (wave runup) at the shoreline. Laboratory flume experiments and a two‐dimensional vertical phase‐resolving nonhydrostatic wave‐flow model, which includes a canopy model to predict drag forces generated by roughness elements, were used to investigate wave‐driven water levels for along‐shore uniform fringing reefs. In contrast to many previous studies, both the laboratory experiment and the numerical model account for the effects of large bottom roughness. The numerical model reproduced the observations of the wave transformation and runup over both smooth and rough reef profiles. The numerical model was then extended to quantify the influence of reef geometry and compared to simulations of plane beaches lacking a reef. For a fixed offshore forcing condition, the fore‐reef slope controlled wave runup on reef‐fronted beaches, whereas the beach slope controlled wave runup on plane beaches. As a result, the coastal protection utility of reefs is dependent on these slopes. For our examples, with a fore‐reef slope of 1/5 and a 500 m prototype reef flat length, a beach slope of ∼1/30 marked the transition between the reef providing runup reduction for steeper beach slopes and enhancing wave runup for milder slopes. Roughness coverage, spacing, dimensions, and drag coefficient were investigated, with results indicating the greatest runup reductions were due to tall roughness elements on the reef flat.
... As both surf zone IG wave generation and wave setup are enhanced by SS wave dissipation, the intense SS wave breaking observed on many coral reefs can lead to large IG waves and wave setup (Becker et al., 2016;Buckley et al., 2018a;Pomeroy et al., 2012). Certain reef geometries can also trap and resonantly amplify specific IG frequencies leading to enhanced coastal flooding (Becker et al., 2016;Buckley et al., 2018a;Demirbilek et al., 2009;Gawehn et al., 2016;Merrifield et al., 2014;Pearson et al., 2017;Péquignet et al., 2009;Roeber & Bricker, 2015;Shimozono et al., 2015). ...
... To efficiently simulate IG waves in nearshore applications, surfbeat models, which couple SS wave models with flow models on the time scale of wave groups, were developed (Quataert et al., 2015;Roelvink et al., 2009;van Dongeren et al., 2013). Whilst the wave runup from wave setup and IG swash is computed by these surfbeat models, SS swash motions are neglected or parameterized, which can result in substantial inaccuracies in the prediction of runup (Buckley et al., 2018a;Lashley et al., 2018;Pearson et al., 2017;Quataert et al., 2020). Finally, phase-resolving wave-flow models that treat waves and flow simultaneously have also been applied (Buckley et al., 2014;Lowe et al., 2019;Ma et al., 2014;Rijnsdorp et al., 2021;Roeber & Cheung, 2012;Yao et al., 2012Yao et al., , 2020aYao et al., , 2021Zijlema, 2012). ...
... 3 of 27 A number of laboratory, field, and numerical model studies have been undertaken to quantify the effectiveness of reefs at reducing wave runup (Baldock et al., 2014;Ferrario et al., 2014;Pearson et al., 2017;Quataert et al., 2015). However, a major limitation of many of these studies has been the lack of a detailed accounting of the effects of large bottom roughness on wave-driven hydrodynamics. ...
Article
Full-text available
Engineered and natural submerged coastal structures (e.g., submerged breakwaters and reefs) modify incident wave fields and thus can alter hydrodynamic processes adjacent to coastlines. Although submerged structures are generally assumed to promote beach protection by dissipating waves offshore and creating sheltered conditions in their lee, their interaction with waves can result in mean wave-driven circulation patterns that may either promote shoreline accretion or erosion. Here, we analyse the mean flow patterns and shoreline water levels (wave runup) in the lee of idealised impermeable submerged structures with a phase-resolved nonhydrostatic numerical model. Waves propagating over submerged structures can drive either a 2-cell mean (wave-averaged) circulation, which is characterised by diverging flows behind the structure and at the shoreline, or 4-cell circulation, with converging flows at the shoreline and diverging flows in the immediate lee of the structure. The numerical results show that the mode of circulation can be predicted with a set of relationships depending on the incoming wave heights, the structure crest level, and distance to the shoreline (or structure depth). Qualitative agreement between the mean flow and proxies for the sediment transport using an energetics approach suggest that the mean flow can be a robust proxy for inferring sediment transport patterns. For the cases considered, the submerged structures had a minimal influence on shoreline wave setup and wave runup despite the wave energy dissipation by the structures due to alongshore wave energy fluxes in the lee. Consequently, these results suggest that the coastal protection provided by the range of impermeable submerged structures we modelled is primarily due to their capacity to promote beach accretion.
... Furthermore, Owen et al. (2016) noted that small variations in island topography and land use also influence wave-driven flooding and associated impacts. Responding to this need for flood predictions over reefs of widely varying shape and size but lacking sufficient field measurements, Pearson et al. (2017) numerically simulated over 174,000 combinations of different reef morphology and physical forcing. They showed that waves, water levels and reef width are the most important parameters to consider when predicting reef wave heights and runup. ...
... They showed that waves, water levels and reef width are the most important parameters to consider when predicting reef wave heights and runup. However, there is still a need for more detailed field observations of such sites (Pearson et al., 2017). Fringing reefs are particularly interesting as their narrow and shallow characteristics can result in the highest setup and runup at the shore front. ...
... The wave characteristics are described by the measured maximum values of the significant wave height H S (m) and peak period T P (s), and the mean peak direction D P (°) with its standard deviation. and for steep reef bathymetry has been widely tested and validated in the literature (Harris et al., 2018;Pearson et al., 2017;Quataert et al., 2020;Rueda et al., 2019;Storlazzi et al., 2018;van Dongeren et al., 2017). XBeach can be run in short wave-averaged mode (surfbeat) or short wave-resolving mode (non-hydrostatic). ...
Article
Full-text available
Coral reefs represent an efficient natural mechanical coastal defense against ocean waves. The focus of this study is La Saline fringing coral reef, located in the microtidal West of La Réunion Island in the Indian Ocean, frequently exposed to Southern Ocean swell and cyclonic events. The aim is to provide a better understanding of the reef's coastal defense characteristics for several Southern Ocean swell events. Pressure sensors were placed across the reef to measure water level fluctuations and to study wave transformation. A numerical model (XBeach surfbeat), validated using field observations, was used to deepen understanding of wave transformation, wave setup and runup. Field measurements and model outputs show that as gravity waves dissipate over the reef, and frequency‐dependent dissipation of infragravity waves by bottom‐friction occurs, the reef acts as a low‐pass filter. Wave‐induced setup is found to be the dominant hydrodynamic component. Setup and runup are each 98% and 79% driven by the offshore significant wave height, and 2% and 21% driven by the tide. The modulation of the water level by setup is the main contributor to runup in the fringing reef. At semidiurnal timescales, setup and runup are in antiphase with tidal variations as lower water levels result in higher gravity wave energy dissipation, setup and runup. Simple‐to‐use transfer functions relating incident wave characteristics to these hydrodynamic components are proposed. The effects of bottom friction and water level on the defensive capacity of the coral reef highlight future implications of structural damage and sea level rise.
... Even on high islands, the presence of steep terrain causes the majority of population, infrastructure (housing, businesses, ports, airports, power plants, sewage treatment plants, etc.), and economic activity to be concentrated on a narrow strip along the coastline susceptible to marine flooding. Locations where the reef is bisected by a shore-normal channel extending from the fore reef to the shoreline (Figure 1) are especially attractive for population centers, as channels provide a natural harbor, easy access to deeper water for vessels, and drinking Research is ongoing on this subject to understand which hydrodynamic processes lead to these high runup levels, and some valuable insights have been gained [7][8][9]: the recent increases in oceanic flooding along reef-lined coasts are primarily due to high offshore water levels coinciding with high energy wave events [3,9,10], circumstances which will become more frequent with sea-level rise [11,12]. Increasing water depth over the reef changes the hydrodynamics across the reef in such a way that larger waves reach the shoreline [13], causing high runup levels and flooding of the land behind the shoreline [3,4]. ...
... To address this knowledge gap, we used the physics-based numerical model, XBeach, which has previously been calibrated for fringing coral reefs, to conduct a parametric investigation of how variations in the reef and shore-normal channel morphology and oceanographic forcing influence waves, wave-driven water levels, and the resulting runup on fringing reef coasts. Research is ongoing on this subject to understand which hydrodynamic processes lead to these high runup levels, and some valuable insights have been gained [7][8][9]: the recent increases in oceanic flooding along reef-lined coasts are primarily due to high offshore water levels coinciding with high energy wave events [3,9,10], circumstances which will become more frequent with sea-level rise [11,12]. Increasing water depth over the reef changes the hydrodynamics across the reef in such a way that larger waves reach the shoreline [13], causing high runup levels and flooding of the land behind the shoreline [3,4]. ...
... However, most previous research on runup has been based on one-dimensional (1-D) schematization of coral reefs [3,7,10], representing a uniform and straight coastline, whereas most reefs have significant alongshore bathymetric variability. Alongshore variations on coral reef morphology lead to mean currents and circulation that affect nutrient transport [14,15] and hence coral development. ...
Article
Full-text available
Many populated, tropical coastlines fronted by fringing coral reefs are exposed to wave-driven marine flooding that is exacerbated by sea-level rise. Most fringing coral reefs are not alongshore uniform, but bisected by shore-normal channels; however, little is known about the influence of such channels on alongshore variations on runup and flooding of the adjacent coastline. We conducted a parametric study using the numeric model XBeach that demonstrates that a shore-normal channel results in substantial alongshore variations in waves, wave-driven water levels, and the resulting runup. Depending on the geometry and forcing, runup is greater either on the coastline adjacent to the channel terminus or at locations near the alongshore extent of the channel. The impact of channels on runup increases for higher incident waves, lower incident wave steepness, wider channels, a narrower reef, and shorter channel spacing. Alongshore variation of infragravity waves is predominantly responsible for large-scale variations in runup outside the channel, whereas setup, sea-swell waves, and very-low frequency waves mainly increase runup inside the channel. These results provide insight into which coastal locations adjacent to shore-normal channels are most vulnerable to high runup events, using only widely available data such as reef geometry and offshore wave conditions.
... XBeach is widely used and validated. These validations include both reef (Pearson et al., 2017) and beach (Elsayed and Oumeraci 2017) coastal environments. XBGPU has previously been tested against direct XBeach solutions and in reef environments (Bosserelle et al., 2021). ...
Article
Numerical prediction of coastal inundation can be complex due to the multiple physical processes involved and typically requires two-dimensional numerical model extents, particularly in areas with complex along-shore morphology. Such model domains often incur relatively high computational expense. Recent extreme inundation studies for Wellington, New Zealand, were executed using the numerical tool, XBeach. Here, the two-dimensional physical dynamics associated with multiple small embayments and both reef and sandy beach substrates, require large, high spatial resolution numerical model extents, informed by a multi-source elevation surface. XBGPU, is a translation of key XBeach features into code that permits GPU-based acceleration. The present study presents a comparison between XBeach and XBGPU for the same numerical model configuration and extents. Three model resolutions were employed, ranging from a typical desktop CPU based XBeach model resolution, to the highest resolution model that will require High Performance Computing (HPC) scale resources. Two CPU HPC facilities were used and five GPUs to investigate the scalability of both XBeach and XBGPU. The latter ranged from desktop grade units to GPUs associated with professional computing facilities. XBeach scalability is investigated by mean of the speed-up ratio, the time saving ratio and the computational efficiency. These are in reference to the computational speed of a model running on one CPU core. The XBGPU speed-up ratio is presented as a function of the slowest GPU. Direct comparisons between XBeach and XBGPU were achieved by using computational capacity as a metric. The results indicate that even a desktop grade GPU can compete with the computational efficiency of HPC-scale CPU facilities. Small model resolutions presented inefficient scaling on both high-performance CPUs and GPUs while the high-resolution model presented near linear scalability for the high-performance GPUs. Nevertheless, HPCs can be efficient to solve large computational problems if enough CPUs are employed.
... Some other works of note in this area include: predicting coastal erosion related to storm events (Beuzen et al. (2018), Wilson et al. (2015), Hapke and Plant (2010), Beuzen et al. (2019), Beuzen et al. (2017)), or testing the relationship between erosion and sand nourishments (Giardino et al. 2019). Other studies that used Bayesian-based decision support systems include: Ferreira et al. (2019), which reported measures aimed at reducing risks associated with the occurrence of extreme events in southern Portugal; Pearson et al. (2017), which assessed wave-driven flooding hazards; Wright and Short (1984), which statistically determined beach state classification and Loureiro et al. (2013), which was based on sedimentological and hydrodynamic data from the Portuguese and Irish coastlines. ...
Article
Coastal areas are one of the most threatened natural systems in the world. Environmental beach indicators, such as erosion and deposition rates of exposed beaches in Andalusia (640 km), were calculated using the upper limit of the active beach profile and detailed orthophotos (1:2500) for the periods 1956–1977, 1977–2001 and 2001–2011. A hybrid classification method, both supervised and unsupervised, based on machine-learning (ML) techniques was then applied to model beach response and dynamics for this 55-year period. The use of a K-means technique allowed stratification into four beach groups that have responded similarly in terms of coastline mobility and erosion/deposition patterns. Furthermore, the application of a classification and regression tree (CART) based on the K-means results helped to identify the threshold values for erosional and depositional rates and the period that characterises each cluster or stratum, enabling correct classification of 1415 out of 1509 beaches (93.77%)
... Other predictive models for wave setup in coral reef islands, such as the Bayesian network model by 55 (that also provides runup and overtopping), already exist. However, as far as we know, our wave setup parameterisation is the simplest available way to estimate the wave setup in these islands that will avoid the use of adaptations of other wave setup parameterisations (such as 38 ) created for other environments different that coral reef islands. ...
Article
Full-text available
Atoll islands are among the places most vulnerable to climate change due to their low elevation above mean sea level. Even today, some of these islands suffer from severe flooding generated by wind-waves, that will be exacerbated with mean sea-level rise. Wave-induced flooding is a complex physical process that requires computationally-expensive numerical models to be reliably estimated, thus limiting its application to single island case studies. Here we present a new model-based parameterisation for wave setup and a set of numerical simulations for the wave-induced flooding in coral reef islands as a function of their morphology, the Manning friction coefficient, wave characteristics and projected mean sea level that can be used for rapid, broad scale (e.g. entire atoll island nations) flood risk assessments. We apply this new approach to the Maldives to compute the increase in wave hazard due to mean sea-level rise, as well as the change in island elevation or coastal protection required to keep wave-induced flooding constant. While future flooding in the Maldives is projected to increase drastically due to sea-level rise, we show that similar impacts in nearby islands can occur decades apart depending on the exposure to waves and the topobathymetry of each island. Such assessment can be useful to determine on which islands adaptation is most urgently needed.
Article
An extensive new set of laboratory measurements have been performed with random and monochromatic waves to study runup on fringing reef-fronted beaches. The experiments were conducted with an idealised fringing reef profile and tested with different forcing conditions (waves and water levels) and reef geometries. Experimental physical data are compared with empirical models developed on open coasts to predict runup. The results show that runup predictions are scattered and are overestimated by existing empirical models when using the off-reef wave conditions. Predictions based on the wave and water level conditions at the beach toe show much less scatter, with the run-up scaling remaining consistent with that proposed by Hunt (1959). Two revised optimizations of Hunt (1959) formula were derived from the experimental data. The results showed that these formulations and the existing runup scaling laws are useful to characterize runup on fringing reef-fronted beaches if the appropriate wave and water level conditions are adopted. Consequently, existing runup models can be combined with the results from phase-averaging wave models to estimate runup on beaches landward of fringing reefs.
Thesis
Full-text available
In an era of rising seas and other challenges posed by climate change, coastal regions like the Netherlands are facing ever graver threats. Strategic sand nourishments could mitigate the threat of coastal erosion and sea level rise on barrier island coasts while limiting ecological impacts. However, insufficient knowledge of sediment transport pathways at tidal inlets and ebb-tidal deltas prevents an informed response in these areas. The main goal of this project was to describe and quantify the pathways that sediment takes on an ebb-tidal delta. To reach this goal, we focused our analyses on Ameland ebb-tidal delta in the Netherlands. Before we could begin to tackle this challenge, we needed to develop new tools and techniques for analyzing a combination of field measurements and numerical models. These include a method for analyzing the stratigraphy and mapping the morphodynamic evolution of ebb-tidal deltas, a new metric for characterizing suspended sediment composition, and innovative use of sediment tracers. We also established a quantitative approach for looking at and thinking about sediment pathways via the sediment connectivity framework, and developed a Lagrangian model to visualize and predict these pathways efficiently. The techniques developed here are useful in a wider range of coastal settings beyond Ameland, and are already being applied in practice. We foresee that the main impacts of this project will be to improve nourishment strategies, numerical modelling, and field data analysis. This dissertation also points forward to numerous opportunities for further investigation, including the continued development of the connectivity framework and SedTRAILS. By managing our coastal sediment more effectively, we will set the stage for a more sustainable future, in spite of the challenges that lie ahead.
Conference Paper
Full-text available
Reef coasts, because the wave height of incoming stormy high waves are extremely dampened by coral reef, have been believed as calm sea areas. However, many coastal structures have actually been damaged by stormy waves dampened by the reef. Surf beat with large wave height was discovered through field observations. The surf beat has been named as "Bore-like Surf Beat". The wave height and the wave velocity of the Bore-like surf beat is larger than that of individual waves. Numerical simulation taking into account of non-linear effect of surf beat phenomenon on the reef shows good agreement with laboratory data. The main reason of the disasters of coastal structures on the reef coasts is existence of the Bore-like surf beat resonantly excited incoming wave groups.
Article
Full-text available
Very-low frequency (VLF, 0.001-0.005 Hz) waves are important drivers of flooding of low-lying coral reef-islands. In particular, VLF wave resonance is known to drive large wave runup and subsequent overwash. Using a five-month dataset of water levels and waves collected along a cross-reef transect on Roi-Namur Island in the Republic of the Marshall Islands, the observed VLF motions were categorized into four different classes: (1) resonant, (2) (non-resonant) standing, (3) progressive-growing and (4) progressive-dissipative waves. Each VLF class is set by the reef flat water depth and, in the case of resonance, the incident-band offshore wave period. Using an improved method to identify VLF wave resonance, we find that VLF wave resonance caused prolonged (∼0.5 – 6.0 hr), large-amplitude water surface oscillations at the inner reef flat ranging in wave height from 0.14 to 0.83 m. It was induced by relatively long-period, grouped, incident-band waves, and occurred under both storm and non-storm conditions. Moreover, observed resonant VLF waves had non-linear, bore-like wave shapes, which likely have a larger impact on the shoreline than regular, sinusoidal waveforms. As an alternative technique to the commonly used Fast Fourier Transformation, we propose the Hilbert-Huang Transformation that is more computationally expensive but can capture the wave shape more accurately. This research demonstrates that understanding VLF waves on reef flats is important for evaluating coastal flooding hazards. This article is protected by copyright. All rights reserved.
Technical Report
Full-text available
XBeach is an open-source numerical model which originally was developed to simulate hydrodynamic and morphodynamic processes and impacts on sandy coasts with a domain size of kilometers and on the time scale of storms. Since then, the model has been applied to other types of coasts and purposes. This reference guide provides an overview of the model formulations and many functionalities, options and parameters in the model. This technical reference applies to XBeach version 1.22 (revision 4567) also known as the ‘Kingsday’ release.
Article
Emergency management and long-term planning in coastal areas depend on detailed assessments (meter scale) of flood and erosion risks. Typically, models of the risk chain are fragmented into smaller parts, because the physical processes involved are very complex and consequences can be diverse. We developed a Bayesian network (BN) approach to integrate the separate models. An important contribution is the learning algorithm for the BN. As input data, we used hindcast and synthetic extreme event scenarios, information on land use and vulnerability relationships (e.g., depth-damage curves). As part of the RISC-KIT (Resilience-Increasing Strategies for Coasts toolKIT) project, we successfully tested the approach and algorithm in a range of morphological settings. We also showed that it is possible to include hazards from different origins, such as marine and riverine sources. In this article, we describe the application to the town of Wells-next-the-Sea, Norfolk, UK, which is vulnerable to storm surges. For any storm input scenario, the BN estimated the percentage of affected receptors in different zones of the site by predicting their hazards and damages. As receptor types, we considered people, residential and commercial properties, and a saltmarsh ecosystem. Additionally, the BN displays the outcome of different disaster risk reduction (DRR) measures. Because the model integrates the entire risk chain with DRR measures and predicts in real-time, it is useful for decision support in risk management of coastal areas. - See more at: https://doi.org/10.1016/j.coastaleng.2017.05.004
Article
An unprecedented set of hydrologic observations was collected after the Dec 2008 seawater-flooding event on Roi-Namur, Kwajalein Atoll, Republic of the Marshall Islands. By two days after the seawater flooding that occurred at the beginning of dry season, the observed salinity of water withdrawn by the island’s main skimming well increased to 100% seawater concentration, but by ten days later already decreased to only 10%–20% of seawater fraction. However, the damaging impact on the potability of the groundwater supply (when pumped water had concentrations above 1% seawater fraction) lasted 22 months longer. The data collected make possible analyses of the hydrologic factors that control recovery and management of the groundwater-supply quality on Roi-Namur and on similar low-lying islands.
Article
Low frequency, high impact storm events can have large impacts on sandy coasts. The physical processes governing these impacts are complex because of the feedback between the hydrodynamics of surges and waves, sediment transport and morphological change. Predicting these coastal changes using a numerical model requires a large amount of computational time, which in the case of an operational prediction for the purpose of Early Warning is not available. For this reason morphodynamic predictions are not commonly included in Early Warning Systems (EWSs). However, omitting these physical processes in an EWS may lead to potential under or over estimation of the impact of a storm event.
Article
Using coarse-scale approaches, existing national assessments of vulnerability and adaptation highlight physical land instability as a major threat to atoll island nationhood. However, such evaluations are bereft of detailed, ground-truthed analyses of the physical impacts of climatic change on reef islands, treating islands as homogenous in both biophysical and social characteristics. The distinct geomorphic context of two proximate reef islands (Jeh and Jabat) in the Marshall Islands was examined through conventional land survey techniques. A template documenting the nuances in island topography was used to evaluate simple inundation scenarios, reflecting current and future sea-level changes under storm surge conditions. The variations in local scale community exposure to inundation were discernible. The study highlights the importance of treating coarse-scale assessments with caution and underscores the need for continued commitment to resolving variations in community experiences to environmental change. Notions of risk and exposure are complex and embedded in both the biophysical and social contexts of each island community. Despite a number of targeted urban vulnerability studies in the Pacific there remains a need for efforts to document localised differences in experience to better inform contemporary adaptation efforts.
Article
Many low-lying tropical islands are susceptible to sea level rise and often subjected to overwash and flooding during large wave events. To quantify wave dynamics and wave-driven water levels on fringing coral reefs, a 5 month deployment of wave gauges and a current meter was conducted across two shore-normal transects on Roi-Namur Island in the Republic of the Marshall Islands. These observations captured two large wave events that had waves with maximum heights greater than 6 m with peak periods of 16 s over the fore reef. The larger event coincided with a peak spring tide, leading to energetic, highly skewed infragravity (0.04–0.004 Hz) and very low frequency (0.004–0.001 Hz) waves at the shoreline, which reached heights of 1.0 and 0.7 m, respectively. Water surface elevations, combined with wave runup, reached 3.7 m above the reef bed at the innermost reef flat adjacent to the toe of the beach, resulting in flooding of inland areas. This overwash occurred during a 3 h time window that coincided with high tide and maximum low-frequency reef flat wave heights. The relatively low-relief characteristics of this narrow reef flat may further drive shoreline amplification of low-frequency waves due to resonance modes. These results (1) demonstrate how the coupling of high offshore water levels with low-frequency reef flat wave energetics can lead to large impacts along fringing reef-lined shorelines, such as island overwash, and (2) lend support to the hypothesis that predicted higher sea levels will lead to more frequent occurrences of these extreme events, negatively impacting coastal resources and infrastructure.