Coastal Engineering 92 (2014) 1–11

Contents lists available at ScienceDirect

Coastal Engineering
journal homepage: www.elsevier.com/locate/coastaleng

Evaluation of wave runup predictions from numerical and

parametric models
H.F. Stockdon ⁎, D.M. Thompson, N.G. Plant, J.W. Long
St. Petersburg Coastal & Marine Science Center, U.S. Geological Survey, 600 4th Street S., St. Petersburg, FL, United States

a r t i c l e i n f o a b s t r a c t

Article history: Wave runup during storms is a primary driver of coastal evolution, including shoreline and dune erosion and bar-
Received 20 February 2014 rier island overwash. Runup and its components, setup and swash, can be predicted from a parameterized model
Received in revised form 6 June 2014 that was developed by comparing runup observations to offshore wave height, wave period, and local beach
Accepted 11 June 2014
slope. Because observations during extreme storms are often unavailable, a numerical model is used to simulate
Available online 17 July 2014
the storm-driven runup to compare to the parameterized model and then develop an approach to improve the
Keywords:
accuracy of the parameterization. Numerically simulated and parameterized runup were compared to observa-
Runup tions to evaluate model accuracies. The analysis demonstrated that setup was accurately predicted by both the
Swash parameterized model and numerical simulations. Infragravity swash heights were most accurately predicted
Setup by the parameterized model. The numerical model suffered from bias and gain errors that depended on whether
XBeach a one-dimensional or two-dimensional spatial domain was used. Nonetheless, all of the predictions were signif-
Observations icantly correlated to the observations, implying that the systematic errors can be corrected. The numerical
Storms simulations did not resolve the incident-band swash motions, as expected, and the parameterized model
performed best at predicting incident-band swash heights. An assimilated prediction using a weighted average
of the parameterized model and the numerical simulations resulted in a reduction in prediction error variance.
Finally, the numerical simulations were extended to include storm conditions that have not been previously ob-
served. These results indicated that the parameterized predictions of setup may need modification for extreme
conditions; numerical simulations can be used to extend the validity of the parameterized predictions of
infragravity swash; and numerical simulations systematically underpredict incident swash, which is relatively
unimportant under extreme conditions.
© 2014 Published by Elsevier B.V.

1. Introduction numerous observations of offshore wave conditions and synchronous

runup measurements, empirical parameterizations have been devel-
Hurricanes and other large storms can cause extensive changes to oped to predict the magnitude of runup and its components, setup
coastal topography, including shoreline erosion, destruction of protec- and swash (Holman, 1986; Nielsen and Hanslow, 1991; Ruessink
tive dunes, creation of large overwash deposits, and opening of new et al., 1998; Ruggiero et al., 2004; Stockdon et al., 2006). The Stockdon
inlets. These changes can have a profound impact on coastal environ- et al. (2006) parameterization (hereinafter referred to as S2006) in par-
ments and may increase coastal vulnerability to future storms. The ticular has been shown to support skillful predictions of coastal changes
type and magnitude of barrier island response to storms is dependent, in the vicinity of hurricane landfall (Plant and Stockdon, 2012; Stockdon
in part, on the interactions between beach morphology and the ocean- et al., 2007), despite not having been originally formulated using storm
ographic forces associated with waves and storm surge. The shoreline conditions. However, the accuracy of parameterized swash and setup
manifestation of these forces is wave runup, which can, in general, during extreme storm conditions has not been examined. Evaluating
be estimated from knowledge of offshore wave height and period the S2006 parameterization under extreme storm conditions, when
(or wave spectra) and nearshore topography, including the slope of observational data are typically unavailable, requires a new approach.
the intermittently wet and dry foreshore (Bowen et al., 1968; The S2006 parameterizations were determined by fitting a large
Kobayashi et al., 1990; Reniers et al., 2002). Using data sets with number of observations to a statistical model based on observed off-
shore significant wave height (H), dominant wave period, and foreshore
beach slope (β). The parameterizations are

⁎ Corresponding author. Tel.: +1 727 502 8074. pffiffiffiffiffiffi

E-mail address: [email protected] (H.F. Stockdon). η ¼ 0:35 β HL; ð1aÞ

http://dx.doi.org/10.1016/j.coastaleng.2014.06.004
0378-3839/© 2014 Published by Elsevier B.V.
2 H.F. Stockdon et al. / Coastal Engineering 92 (2014) 1–11

pffiffiffiffiffiffi
Sin ¼ 0:75 β HL; and ð1bÞ NC, in October 1997. Runup data were collected over a 9-day period be-
tween October 16 and 24. These data have been presented elsewhere
pffiffiffiffiffiffi
Sig ¼ 0:06 HL; ð1cÞ (e.g. Stockdon et al. (2006)), hence, we provide only a brief summary
and then describe the XBeach model and runup extraction.
where η is the wave setup, defined as the time-average of the non-tidal
water level fluctuations at the shoreline. Sin and Sig are the significant 2.1. Observations
swash heights in the incident (frequency N 0.05 Hz) and infragravity
bands (frequency b 0.05 Hz), respectively, defined as four times the Daily beach surveys during the SandyDuck experiment provided the
standard deviation of the water levels within each frequency band. bathymetry for XBeach and foreshore beach slope in the S2006 param-
Wave length (L) is computed from local wave period. Local wave height eterizations (Fig. 1). Wave height from the FRF Waverider buoy, located
is reverse shoaled to deep water to obtain an estimate of offshore wave offshore in approximately 17 m water depth, was reverse shoaled to
height (H). The coefficients estimated as part of the parameterization deep water and used as input in the parameterized model (Eqs. 1a, 1b,
development were based on observations from a restricted range of 1c). Wave spectra collected at the FRF 8-m array (Fig. 1) provided the
conditions (Stockdon and Holman, 2011). Specifically, the maximum offshore wave-boundary condition data for XBeach (Fig. 2). Waves mea-
offshore wave height was 4 m; therefore, the parameterization does sured from a cross-shore array of pressure sensors in 0–5 m water
not include extreme conditions associated with major storms or hurri- depth, between the shoreline and the 8-m array, (Raubenheimer et al.,
canes when wave heights reach 7 m or more (Doran et al., 2013; 2001) were used to evaluate XBeach simulated surf zone wave transfor-
Stockdon et al., 2012, 2013). mation. A tide gauge located at the end of the FRF pier was used for de-
The importance of understanding and quantifying the accuracy of fining tide levels in XBeach (Fig. 2). Observed tide levels were removed
runup predictions under extreme conditions is twofold. First, the predic- from both the modeled and observed runup in order to focus on the
tions of sediment transport in detailed numerical models and of morpho- wave driven processes.
logic change in statistical approaches are based on calibration. We want to Observed runup time series were extracted at six alongshore loca-
know if these calibrations are correcting for underlying prediction errors tions (Fig. 1) from video images (Fig. 3). This analysis produced 50 17-
in hydrodynamic processes. Second, the predictions of extreme water minute runup time series over the study period. Collection times are
levels that include wave runup are required for more accurate assess- shown in Fig. 2. Each 17-minute time series was analyzed to extract
ments of coastal hazards (Stockdon et al., 2012). Wave-induced water setup and significant incident and infragravity swash. (See Stockdon
levels can be a direct threat to people, infrastructure, and ecosystems; et al. (2006) for more detail.)
however they are not routinely included in the analysis of coastal hazards
in, for instance, the weather forecasting community. Understanding the 2.2. Model simulations
accuracy of runup predictions during extreme wave events will help to in-
form and improve assessments of potential hazards to people and wildlife Water levels at the shoreline were modeled using XBeach (v18),
that build communities (e.g., roads, houses, nests) in dynamic coastal which solves coupled two-dimensional (2-d), depth-averaged equa-
environments that shift and change with each storm. tions for short-wave envelope propagation and flow for varying spectral
Wave runup processes are not easy to measure, particularly under wave and flow boundary conditions (Roelvink et al., 2009). The low-
extreme conditions. Powerful wave forces and significant beach change frequency wave motions interact and evolve to produce both low-
can damage observing equipment or introduce uncertainty in the frequency and, due to nonlinear behavior, some incident-frequency
underlying topographic elevations needed to understand the runup swash (Fig. 4b). Incident waves are dissipated due to breaking and are
processes. One approach to circumventing observational challenges is expected to vanish when the depth is zero. Sediment transport and
to numerically simulate runup. This has been done using the XBeach morphology changes were not included in the simulations. In our
model, which couples runup to sediment transport and dune erosion case, the 2-d model spanned 380 m in the alongshore and about 800
(Roelvink et al., 2009). The model does not resolve incident-frequency m in the cross-shore (Fig. 1). The alongshore resolution was 10 m, and
motions but directly computes setup and low-frequency wave motions the cross-shore resolution varied from 0.5 m in the swash region to 8
which tend to dominate the runup processes during dissipative storm m at the offshore boundary. Bathymetry was derived from daily survey
conditions (Roelvink et al., 2009; Ruggiero et al., 2004; Thornton and data which were interpolated to the XBeach domain using a smoothing
Guza, 1982). Model predictions have been compared to observed method that adapted to the grid resolution (Plant et al., 2002).
beach and dune changes to test the accuracy of the coupled runup and Direction-frequency wave spectra from the 8-m array were applied to
sediment transport formulations with skillful results (McCall et al., the offshore boundary of the model domain. Water levels from the
2010; Roelvink et al., 2009; Splinter and Palmsten, 2012). tide gauge at the end of the FRF pier were applied uniformly to the off-
In order to use XBeach to simulate storm waves and runup for the shore boundary. The lateral boundaries of the domain were treated as
purpose of extending runup parameterizations to more energetic Neumann or no-gradient boundaries. All Xbeach parameters were set
wave conditions, the accuracy of the modeled runup must be evaluated. to default values except for the wave breaking parameter γ, which
Here, we conduct a comparison and sensitivity study to assess the accu- was set to 0.42. Details of model sensitivity to wave breaking parame-
racy of XBeach runup predictions across a range of conditions that have ters are described in Section 4.1.
corresponding runup measurements. The objective is to evaluate the The XBeach model can also be implemented in a horizontally one-
model skill at predicting setup, incident swash, and infragravity dimensional (1-d) domain (i.e., along a single cross-shore transect)
swash. Then, using this information, we can test the application of the where alongshore uniformity is assumed. The 1-d approach has several
S2006 parameterizations to extreme conditions and compare them advantages, including faster simulation times and reduction of required
with numerical simulations. Finally, we present a methodology for alongshore bathymetric detail. Because the alongshore components of
improving statistical parameterizations based on assimilating model bathymetry, wave groups, and swash are not fully resolved, it is expect-
results and observations. ed that 1-d simulations will produce different swash levels than the 2-d
simulations. When implemented in 1-d, separate XBeach domains were
2. Methods defined along each of the six video-based runup measurement lines,
while using the same offshore wave and water level boundary condi-
XBeach runup predictions were evaluated using data from the tions as in the 2-d simulations. The sensitivity of wave runup to the
SandyDuck field experiment (Stockdon and Holman, 2011) at the U.S. choice of dimensional space used in the model will be evaluated in
Army Corps of Engineers Field Research Facility (FRF) located in Duck, later sections.
 H.F. Stockdon et al. / Coastal Engineering 92 (2014) 1–11 3

Fig. 1. Nearshore bathymetry (NGVD), measurement locations, and XBeach model domain in the vicinity of the U.S. Army Corps of Engineers Field Research Facility.

2.2.1. Storm scenarios existing simulations that were designed to provide hurricane forcing
In addition to using measured wave inputs from the 8-m array, nu- input to the runup S2006 parameterization (Stockdon et al., 2013). Cat-
merical simulations of swash and setup were conducted using hurricane egory 1–5 storms scenarios were constructed by applying hurricane
conditions as boundary conditions. To cover a larger range of parameter wind speeds and imposing several wind directions to a wave model
space, inputs for category 1–5 storm scenarios were extracted from (Simulating WAves Nearshore (Booij et al., 1999)). The maximum

Fig. 2. Observed conditions during the SandyDuck field experiment. Significant wave height (a), peak period (b), and peak direction (c) measured at the FRF 8-m array. Water-levels
measured at the end of the FRF pier (d). Vertical lines indicate times of runup observations.
4 H.F. Stockdon et al. / Coastal Engineering 92 (2014) 1–11

Fig. 3. Camera view from the FRF (a) and corresponding runup timestack (b). The cross-shore transect represents the location of video-derived runup observations. In the timestack of pixel
intensity along the transect, each vertical line is the cross-shore variability of intensity at a single time step. The leading edge of swash is digitized through time (green line) and then
converted into a time series of water-level elevations. (Modified from Stockdon et al. (2006).).

wave height and corresponding wave period from each hurricane cate- simulations were compared to each other, with observations, and with
gory were used as input at the offshore XBeach boundary in 1-d simula- the S2006 parameterizations.
tions, which, as will be shown later, have higher accuracy than the 2-d
implementation. Similarly, storm-induced water levels for these five 3. Results
scenarios came from SLOSH model simulations (Jelesnianski et al.,
1992) as described by Stockdon et al. (2013). The storm scenarios 3.1. Wave transformation
were run in 1-d mode with the offshore boundary extended out to the
FRF Waverider buoy, which was moored approximately 3.6 km offshore The accuracy of the XBeach simulated waves were evaluated using
in 17.4 m of water at the time of the study. This was necessary to adapt wave observations in the surf zone (for a description see Raubenheimer
to the relatively coarse resolution of the wave and water level inputs, et al. (2001)). Using the 2-d implementation, the simulated significant
1.5 km and 0.5–2.0 km, respectively. The bathymetry for these scenarios wave heights (Fig. 5) compared well to the surf zone observations,
was a temporal mean of all the 1-d spatially-averaged profiles used in with root-mean-square errors (rmse) ranging from 0.21 m at the
the previously described 1-d simulations. Observations of wave runup wave gauges located near x = 500 m (mean depth = 5.3 m) to 0.41
do not exist for these conditions, so the accuracy of simulated swash m at the shallowest locations near x = 160 m (mean depth = 1.1 m).
and setup could not be assessed. However, the simulated values were The mean difference error (μΔ) for the 2-d runs was 0.08 m, indicating
compared to calculations obtained using the runup parameterization a small positive bias in the modeled wave heights. The 1-d implementa-
(Stockdon et al., 2006) to determine if they were consistent with what tion produced similar results with rmse ranging from 0.28 m offshore to
would be expected based on observations from a range of sites and 0.36 m nearshore and μΔ of −0.04 m.
conditions.
3.2. Wave setup and swash
2.2.2. Extraction of simulated runup
The modeled shoreline water levels can be treated similarly to mea- XBeach runup simulations of setup, incident swash, and infragravity
surements obtained from video-based observations. The numerically swash were compared to both video-based observations and predic-
simulated swash location was extracted at each of the six cross-shore tions using the S2006 parameterization (Fig. 6). The 1-d simulations
profiles by detecting the shoreward-most wet point at a threshold produced the best predictions of setup with rmse of 0.13 m and a skill
depth δ (here, δ = 10 cm). The sensitivity of swash measurements to (R2) of 0.68, which is significant at the 95% confidence level. For this
choice of δ has been described by others (Holland et al., 1995; and the following analyses, the 95% significance threshold was about
Raubenheimer et al., 1995), and we will return to this issue in 0.02 based on 209 observations. The predictions using S2006, which
Section 4. In the 2-d simulations, where runup profile locations did was based in part on this dataset, had the poorest performance (μΔ =
not correspond exactly to model grid lines, simulated water levels 0.01 m; R2 = 0.41; rmse = 0.21 m, Table 1), reflecting site-specific
were interpolated in space. Swash and setup values from 1-d and 2-d uncertainty in the parameterization, which we revisit in Section 4.
 H.F. Stockdon et al. / Coastal Engineering 92 (2014) 1–11 5

Fig. 4. Plan view showing the spatial variability of XBeach water-level output at one time step Oct 19, 1997 (a). The timestack (b) shows the cross-shore and temporal variation of XBeach
water levels at a single longshore location (y = 815 m, green vertical line in a).

Infragravity-band swash predictions varied substantially for the prediction of infragravity-band swash to the observations was also
three methods. Based on rmse, the most accurate prediction for Sig high, but included a regression slope error (b = 0.55, less than an
was from the S2006 model (R2 = 0.54, rmse = 0.26 m), and the least ideal regression slope of 1.0) such that low swash heights were
accurate was from the 2-d implementation of XBeach (R2 = 0.59, underpredicted and high swash heights were overpredicted (Fig. 6;
rmse = 0.66 m). The large rmse reflects a bias (μΔ = −61 cm), indicat- Table 1).
ing an underprediction of observed Sig. The skill of the 1-d XBeach The most accurate incident-band swash height predictions were
made by the S2006 model, based on high R2 value and low rmse, 0.42
and 0.36 m, respectively. The S2006 prediction had a higher R2 than
the XBeach predictions but also included a regression slope error such
that low swash heights were underestimated and high swash heights
were overestimated. Because of short-wave averaging in the modeled
processes, XBeach simulations underpredicted incident-band swash
heights, had low R2 values (0.14 and 0.12 for 1-d and 2-d, respectively),
and high rmse (0.57 m and 0.82 m for 1-d and 2-d, respectively). It is
clear from the scatter in the model predictions that XBeach modeled
incident-band swash does not offer an improvement over predicting
the mean value (about 1 m). The 1-d model offered only a minor
improvement over the 2-d model.

3.3. Simulations of hurricane scenarios

Because the simulations during hurricane conditions could not be

compared to observations, these results were compared to the predic-
tions made using the S2006 parameterization. Previously presented
comparisons show the extent to which XBeach simulations can be
used as surrogates for actual runup (setup and swash) observations
for the purpose of extending the parameterizations to extreme storm
conditions (Table 1). Using hurricane forcing as input, XBeach-
modeled setup for category 1 conditions lay near 1:1 trend line
Fig. 5. Comparison of modeled and observed wave height in the surf zone for both the (Fig. 7a), indicating that XBeach and S2006 produced nearly the same
1- and 2-dimensional XBeach implementations. result under these conditions. However, as storm intensity grew, and
6 H.F. Stockdon et al. / Coastal Engineering 92 (2014) 1–11

Fig. 6. Observed and modeled setup (a), infragravity swash (b), and incident swash (c). The S2006, 1-d XBeach, and 2-d XBeach models are represented by green, red, and blue dots,
respectively. Solid lines indicate linear-regressions fitting the model results to the data.

waves and surge interacted with the upper beach and/or dune face, 4. Discussion
XBeach modeled setup increased more than was predicted by the
S2006 parameterization. If these XBeach results during storm condi- The analysis and results presented here relied on making a number
tions can be verified using field observations, particularly when waves of assumptions, many of which were contained in model-parameter
are interacting with the dune, then the model simulations could be use- choices that gave the best comparisons of the XBeach model to the
ful in modifying the parameterized range of the S2006 model. For observational data. The danger of using numerical models is that the re-
infragravity swash, differences between 1-d XBeach simulations and sults may depend on poorly constrained coefficients. For instance,
S2006 were about the same as differences between observations and Apotsos et al. (2008) showed that the best parameter values for predict-
S2006 (Table 2). XBeach values for hurricane conditions are lower ed wave-heights varied with external conditions that changed from one
expected from simply extrapolating the 1-d model results from non- field experiment to another and even changed within a single field
hurricane conditions (Fig. 7b). The mean and rmse differences between experiment as the boundary conditions changed. This effect has been
the 1-d hurricane results and the S2006 model are very similar to the noted elsewhere (Plant et al., 2011; Ruessink et al., 2003). Site-specific
differences between S2006 and the observations, suggesting that both sensitivity to model parameter choice also affects the S2006 model.
models are equally valid (Table 2). Thus, XBeach can be used to extend Stockdon et al. (2006) showed that the model coefficients that best fit
the S2006 parameterization for infragravity swash during storms. As all the data resulted in systematic prediction errors for individual data
expected, XBeach-simulated Sin during hurricane conditions was sets, including the 1997 SandyDuck data set used here (e.g., Fig. 6c,
substantially smaller than that predicted using S2006 (Fig. 7c). When which exhibits a slope error). Similarly, XBeach is vulnerable to predic-
evaluating whether XBeach can be used to simulate swash during tion errors due to sensitivity to adjustable model coefficients.
hurricane conditions, it is important to note that swash processes
under highly dissipative conditions, such as during storms, are typically 4.1. XBeach parameter sensitivity
dominated by infragravity energy (Thornton and Guza, 1982). As such,
an underestimate of incident band swash in XBeach will have a relative- Energy dissipation due to depth-induced wave breaking of the
ly small influence on the total swash magnitude during hurricane incident band waves in XBeach is modeled as a dissipation rate times
conditions. a probability of wave breaking (Roelvink, 1993). This wave-breaking
formulation includes two free parameters; one that controls the magni-
tude of the dissipation rate, α, and the other dictates the fraction
of breaking waves as a function of the wave height to water depth
ratio, γ. The first parameter, α, is typically set at 1.0 (Battjes and
Table 1 Janssen, 1978; Roelvink, 1993; Roelvink and Brøker, 1993). Sensitivity
Statistics describing the fit between observations and parameterized, numerical, and
to this parameter was not explored here. The default value for γ is
assimilated results (Figs. 6 and 11). Prediction errors for setup (η), infragravity swash
(Sig), and incident swash (Sin) include the mean difference (μΔ), root-mean-square error 0.55 and is based on a limited number of tests, primarily laboratory.
(rmse), skill (R2), and slope (b) of best-fit linear regression. The rsme for assimilation Comparisons between field observations and model predictions of
results using the uncorrected XBeach model are shown in parentheses. wave height show large variation in the optimal value of γ depending
Parameter Model μΔ (m) rmse (m) R2 b
on both the wave conditions and the empirical relationship chosen for
γ (Apotsos et al., 2008).
η S2006 0.10 0.21 0.41 0.78
Using the 1-d XBeach domains for all runup observation times, we
1-d Xbeach −0.02 0.13 0.68 1.02
2-d Xbeach 0.00 0.17 0.46 0.82 explored the sensitivity of modeled wave heights, setup, and swash to
Assimilation 0.05 0.16 (0.16) 0.91 1.13 the choice of γ (Fig. 8; wave height comparisons are not shown). In ad-
Sig S2006 −0.06 0.26 0.54 0.91 dition to the default value for γ, 0.32 and 0.42 were tested. These values
1-d Xbeach −0.09 0.35 0.55 0.55
are consistent with field observations of wave height to water depth ra-
2-d Xbeach −0.61 0.66 0.59 1.08
Assimilation −0.03 0.23 (0.42) 0.96 1.04 tios observed at the field site in Duck (Guza and Thornton, 1981;
Sin S2006 0.19 0.36 0.42 0.51 Sallenger and Holman, 1985). For η and Sig, the value of γ = 0.42 provid-
1-d Xbeach −0.49 0.57 0.14 0.63 ed the best fit with the observations, as indicated by lower mean and
2-d Xbeach −0.76 0.82 0.12 0.98 rmse differences (Fig. 8). In an additional sensitivity test, XBeach was
Assimilation 0.28 0.28 (0.32) 0.94 0.71
implemented using an advective-deterministic breaking formulation
 H.F. Stockdon et al. / Coastal Engineering 92 (2014) 1–11 7

Fig. 7. Observations and XBeach simulations of setup (a), infragravity swash (b), and incident swash (c) compared to values parameterized using the S2006 model. Comparisons to video-
based observations are shown in green. Comparisons to 1-d XBeach, 2-d XBeach, and XBeach during hurricane conditions are represented by the red, blue, and black dots, respectively.
Solid lines indicate linear regressions between the parameterized results and the Xbeach and observational data.

(Daly et al., 2012), where wave breaking is turned on and off using positively skewed (0.22), while the 2-d water level elevations were neg-
upper and lower values of γ, in this case 0.52 and 0.30, respectively. atively skewed (−0.30). The observed water level distributions had a
Comparison of 1-d model results to the observed data indicated that skewness of −0.05. The frequency spectra of water levels (Fig. 9c) indi-
this formulation did not offer an improvement over the static breaking cated that the swash elevation spectra were broader banded in the 1-d
formulation using γ = 0.42 (Fig. 8). case compared to the 2-d case.
Differences in swash heights between 1- and 2-d runs may result if
4.2. Comparison between 1-d and 2-d XBeach simulations cross-shore evolution of wave heights differs between the two simula-
tion approaches or if alongshore interactions dissipate swash differently
Estimates of infragravity and incident band significant swash through frictional or nonlinear processes (Cox et al., 2013; Guza and
differed between the 1-d and 2-d XBeach implementations. The 2-d im- Feddersen, 2012; Reniers et al., 2006, 2010). In this case, there were
plementation produced Sig that was consistently too low as compared to no substantial differences in the wave height modeling across the surf
the observations (Fig. 6). The 1-d implementation included both under zone between the two implementations (Fig. 5). A preliminary analysis
and over prediction of Sig, as indicated by the large slope error (b = 0.5). of the effects of longshore currents on swash magnitude variations be-
Model performance for the 1-d implementation was best during shore- tween 1- and 2-d implementations was inconclusive. Cross-spectral
normal wave conditions. When the wave approach was ± 15° analysis of modeled shoreline and surf zone water level time series
from shore-normal (approximately ~80% of the cases), the mean error began to reveal differences between 1- and 2-d swash modeling. The
in 1-d modeled Sig was 0.02 m compared to − 0.56 m when waves analysis showed standing-wave motions in the infragravity frequencies
were oblique to the shoreline. However, the errors in Sig for the 2-d and strong reflection and resonance in the 1-d domain, likely associated
implementation were high regardless of wave approach: − 0.59 m with the increased swash variance or magnitudes. Coherence between
and −0.71 m for shore-normal and oblique waves, respectively. Results the swash and surf zone water levels was lower in the 2-d model imple-
from a particular case with 3-m high waves, and a shore-normal ap- mentation, possibly suggesting that alongshore dissipation and/or non-
proach, illustrate the most extreme differences in the different spatial linear interactions in the swash zone may be decreasing water level
implementations (Fig. 9a). The 1-d case produced very high runup max- amplitudes. Additional investigations, which are outside the scope of
ima, while the 2-d case exhibited lower runup maxima but also lower this study, are required for a complete understanding of the swash
minima. The distribution (Fig. 9b) of 1-d water level elevations were differences between the two spatial implementations.

4.3. Runup sensor depth choice

Table 2
The choice of threshold depth, δ, used to extract shoreline water
Statistics describing the fit between S2006 parameterizations and data/simulations
(Fig. 7). Prediction errors for setup (η), infragravity swash (Sig), and incident swash (Sin) levels (runup) from the XBeach model affected the estimates of setup
include the mean difference (μΔ), root-mean-square error (rmse), skill (R2), and slope and swash. Runup was extracted from all of the XBeach simulations
(b) of best-fit linear regression. using different values of δ, analogous to previous tests of runup sensitiv-
Parameter Data/simulations μΔ (m) rmse (m) R2 b
ity to the height of runup wires used in field experiments (Holland et al.,
1995). Setup elevation decreased more-or-less linearly with increasing
η Video 0.10 0.21 0.41 0.78
1-d 0.12 0.19 0.49 0.70
δ, with about a 2-cm loss in setup elevation for each 1-cm increase
2-d 0.10 0.20 0.32 0.58 in δ (Fig. 10). The sensitivity of setup to δ did not differ substantially
1-d hurricane −0.30 0.39 0.96 4.65 between the 1-d and 2-d implementations of XBeach. Analysis of the
Sig Video −0.06 0.26 0.54 0.91 sensitivity of field observations documents a similar dependence:
1-d 0.03 0.28 0.77 1.46
increasing δ results in a loss in setup magnitude, ranging between
2-d 0.55 0.57 0.69 0.73
1-d hurricane −0.18 0.27 0.01 0.16 about 0.5 and 1.0 cm (Holland et al., 1995). In the XBeach simulations,
Sin Video 0.19 0.36 0.42 0.51 significant swash was not as sensitive to δ. Swash gradually
1-d 0.68 0.75 0.44 0.31 increased as δ increased, with a maximum swash height occurring
2-d 0.95 1.01 0.50 0.20 at δ = 5 cm. Swash then decreased slightly with further increases
1-d hurricane 1.94 1.95 0.12 0.12
in δ (Fig. 10).
8 H.F. Stockdon et al. / Coastal Engineering 92 (2014) 1–11

Fig. 8. Sensitivity of setup and swash predictions to variation in the breaking parameter (γ) at six locations across the surf zone. The default formulation was used except for the case where
γ = 0.52–0.30, indicating that the Daly et al. (2012) formulation was used.

4.4. Improved parameterization utilizing XBeach output If the prediction errors from the different modeling approaches are
not correlated to each other, then the results from XBeach (1-d and/or
Our analysis tested the accuracy of XBeach model predictions of 2-d) and S2006 can be combined to reduce the total error by exploiting
setup and swash and systematic errors in both the 1-d and 2-d advantages of each approach. Because the S2006 parameterization
implementations. Systematic errors are also present in the S2006 provides a good fit to a broad range of conditions, we use it as a prior
parameterization. For example, at the Duck field site alone, rmse for condition that can be updated through assimilation of the XBeach
the S2006 runup parameterization varied between 34 to 69 cm, simulations.
depending on the specific field experiment (Stockdon et al., 2006).
 0 
The S2006 parameterization omits many surf zone process details, in-
Sassim ¼ Sparam þ w SXBeach −Sparam ; ð2aÞ
cluding alongshore and cross-shore variability of the shoreline and
sandbars. It has been shown in simulated tests that some infragravity
swash variation can be explained by including additional bathymetric where Sassim is the assimilated value, S'XBeach is the XBeach prediction,
details (Cox et al., 2013). It is possible to combine the observation- and Sparam is the value based on the S2006 parameterization for either
based parameterization and the numerical model results in a way that setup, incident swash, or infragravity swash. Bias and gain (the regres-
results in less overall error than using one model alone. sion slope error) corrections have been applied to XBeach values, as
 H.F. Stockdon et al. / Coastal Engineering 92 (2014) 1–11 9

indicated by the prime notation, to ensure that systematic errors are not
reintroduced to the assimilation. The weighting factor, w, is a function of
the expected errors of both the S2006 and the XBeach simulations:

 0 
w ¼ eparam = eparam þ eXBeach ; ð2bÞ

where eparam is the variance of the S2006 prediction errors and e'XBeach is
the variance of the XBeach prediction errors. The error terms are esti-
mates based on prior experience, such as from the results presented in
Table 1. In this formulation, if the errors of the XBeach model are very
small compared to the S2006 model, the value for w approaches 1,
and the assimilated result is equal to the XBeach prediction. Conversely,
if the XBeach errors are relatively large, the weight approaches 0, and
the assimilated result returns the S2006 prediction.
Using the 1-d XBeach implementation, assimilated setup and
infragravity and incident swash yielded errors (rmse) that were either
the same as or smaller than the individual model inputs (Table 1). For
instance, the assimilation of XBeach and parameterized predictions for
setup yielded errors that were equal to the XBeach errors, even though
the errors in S2006 alone were higher. The systematic over-prediction
of setup by the S2006 model (Fig. 11a) was reduced by giving higher
weight to the XBeach model. The roles were reversed for infragravity
swash. The S2006 model was more accurate than XBeach, and the as-
similation provided an improvement by (1) correcting the XBeach
bias and gain and (2) canceling errors in the cases where one model
overpredicted and the other underpredicted the swash height. In the
case of the incident swash, the XBeach model severely underpredicted
the observations, and higher weight was given to the S2006 model.
The value of including the XBeach incident swash in the assimilation
was to reduce scatter (Table 1).
Overall, the assimilation results reduced the predicted error variance
(total rmse = 0.40 m) by 19% compared to the S2006 parameterization
(total rmse = 0.49 m) and by 63% compared to the XBeach result (total
rmse = 1.07 m). The assimilation weights were, on average, 0.5 indicat-
ing that XBeach and S2006 contributed equally to the assimilation. Ad-
Fig. 9. Sample runup time series (a), histograms (b), and spectra (c) from the 1-d (red)
ditionally, the assimilation result was not overly sensitive to the precise
and 2-d (blue) Xbeach simulations. Time series were normalized to have zero mean and value of the assimilation weights. For instance, the results changed little
unit variance. if w = 0.5 was used in all cases (i.e., an average of XBeach and S2006
output), rather than allowing the weight to vary for setup and swash
components. Thus, prior knowledge of the prediction errors, which de-
termine the weights, need not be perfect. Furthermore, the assimilation
results were similar for setup and incident swash if the XBeach model
was not corrected for systematic errors (e.g., gain and bias errors,
Table 1). Correction of the systematic bias associated with the 2-d
XBeach simulations of infragravity swash improved the assimilation
results in this case (Table 1). Thus, XBeach simulations, corrected for
bias if possible, can be used to correct the S2006 parameterization by
either (1) updating the S2006 parameters to fit these additional data
or (2) assimilating the two, as we have done here.

5. Conclusions

Runup and its components, setup and swash, can be predicted from
a parameterized model that was developed by comparing observations
of runup to offshore wave height, wave period, and local beach
slope (Stockdon et al., 2006). This parameterization can suffer from
systematic errors due to site-specific characteristics that were not
included in the model. Additionally, parameterization skill is unknown
under extreme conditions where observations are lacking. To address
the parameterization deficiencies, numerical models can be used to
simulate the storm-driven runup to improve and extend the parameter-
ized approach.
Runup was numerically simulated using XBeach and compared to
Fig. 10. Sensitivity of setup (solid lines) and swash (dashed lines) to threshold depth (δ) observations to investigate the simulation accuracy. XBeach simulated
used to extract runup from 1-d (circles) and 2-d (squares) XBeach simulations. runup was also compared to the S2006 parameterization to determine
10 H.F. Stockdon et al. / Coastal Engineering 92 (2014) 1–11

Fig. 11. Comparison of observed setup (a), infragravity swash (b), and incident swash (c) to that modeled using S2006, Xbeach, and an assimilated version of the two. Statistics are
presented in Table 1.

if predictions from the two are consistent. The analyses demonstrated Booij, N., Ris, R.C., Holthuijsen, L.H., 1999. A third generation model for coastal regions, 1.
Model description and validation. J. Geophys. Res. 104 (C4), 7649–7666.
that setup was accurately predicted by both the numerical simulations Bowen, A.J., Inman, D.L., Simmons, V.P., 1968. Wave ‘set-down’ and ‘set-up’. J. Geophys.
and parameterized model. Infragravity swash was most accurately pre- Res. 73 (8), 2569–2577.
dicted by the parameterized model. The numerical model predictions of Cox, N., Dunkin, L.M., Irish, J.L., 2013. An empirical model for infragravity swash on barred
beaches. Coast. Eng. 81, 44–50.
infragravity swash suffered from substantial bias or gain errors that Daly, C., Roelvink, D., van Dongeren, A., van Thiel de Vries, J., McCall, R., 2012. Validation of
were dependent on whether a 1-d or 2-d spatial domain was used. an advective-deterministic approach to short wave breaking in a surf-beat model.
Nonetheless, all of the predictions of infragravity swash were well cor- Coast. Eng. 60, 69–83.
Doran, K.S., Stockdon, H.F., Sopkin, K.L., Thompson, D.M., Plant, N.G., 2013. National
related to the observations if these systematic errors were corrected.
Assessment of Hurricane-induced Coastal Erosion Hazards: Mid-Atlantic coast. U.S.
The numerical simulations did not resolve the incident swash motions, Geological Survey, (2013-1131).
as expected, and the S2006 model performed best at predicting incident Guza, R., Feddersen, F., 2012. Effect of wave frequency and directional spread on shoreline
runup. Geophys. Res. Lett. 39 (11).
swash heights. With the systematic errors corrected, an assimilated
Guza, R.T., Thornton, E.B., 1981. Wave set-up on a natural beach. J. Geophys. Res. 86 (C5),
prediction using a weighted average of the S2006 model and the 4133–4137.
numerical simulations resulted in an error reduction of 19% compared Holland, K.T., Raubenheimer, B., Guza, R.T., Holman, R.A., 1995. Runup kinematics on a
to the parameterization and of 63% compared to XBeach. natural beach. J. Geophys. Res. 100 (C3), 4985–4993.
Holman, R.A., 1986. Extreme value statistics for wave run-up on a natural beach. Coast.
XBeach simulations of hurricane wave conditions were used to test Eng. 9, 527–544.
the parameterized runup to storm conditions that have not been previ- Jelesnianski, C.P., Chen, J., Shaffer, W.A., 1992. Slosh: sea, lake, and overland surges from
ously observed. The extreme-storm simulations produced infragravity hurricanes. Technical Report NWS 48. NOAA.
Kobayashi, N., Cox, D.T., Wurjanto, A., 1990. Irregular wave reflection and run-up on
swash results that were consistent with the observationally constrained rough impermeable slopes. J. Waterw. Port Coast. Ocean Eng. 116, 708–728.
simulations of swash. The simulations of setup under extreme condi- McCall, R., et al., 2010. Two-dimensional time dependent hurricane overwash and erosion
tions were consistent with the parameterized estimates under the modeling at Santa Rosa Island. Coast. Eng. 57 (7), 668–683.
Nielsen, P., Hanslow, D.J., 1991. Wave runup distributions on natural beaches. J. Coast. Res.
category-1 case, but for more extreme storms, simulated setup was 7 (4), 1139–1152.
higher than predicted by the S2006 model. These results suggest that Plant, N.G., Stockdon, H.F., 2012. Probabilistic prediction of barrier-island response to
numerically simulated runup, with bias and gain errors corrected, may hurricanes. J. Geophys. Res. Earth Surf. 117 (F3).
Plant, N.G., Holland, K.T., Puleo, J.A., 2002. Analysis of the scale of errors in nearshore
be used to modify or expand field-based parameterizations of setup bathymetric data. Mar. Geol. 191 (1–2), 71–86.
and swash to more energetic storm conditions that have not been Plant, N.G., Thompson, D.M., Elias, E., 2011. Processed-based Model Predictions of
previously observed. Hurricane induced Morphodynamic Change on Low-lying Barrier Islands. Coastal
Sediments, Miami, FL, pp. 1299–1303.
Raubenheimer, B., Guza, R.T., Elgar, S., Kobayashi, N., 1995. Swash on a gently sloping
Acknowledgments beach. J. Geophys. Res. 100 (C5), 8751–8760.
Raubenheimer, B., Guza, R.T., Elgar, S., 2001. Field observations of wave-driven setdown
and setup. J. Geophys. Res. 106 (C3), 4629–4638.
This work benefited from many discussions with colleagues familiar Reniers, A., Van Dongeren, A., Battjes, J., Thornton, E., 2002. Linear modeling of
with runup observations and with the XBeach model, including Ad infragravity waves during Delilah. J. Geophys. Res. 107 (C10), 3137.
Reniers, A.J., MacMahan, J., Thornton, E., Stanton, T., 2006. Modelling infragravity motions
Reniers, Robert McCall, Jaap van Thiel de Vries, Ian Thompson, and on a rip-channel beach. Coast. Eng. 53 (2), 209–222.
Anouk de Bakker. These discussions helped to focus the work presented Reniers, A., et al., 2010. Estimation of infragravity waves at intermediate water depth.
here and exposed many more topics to explore in the future. We made Coast. Eng. 57 (1), 52–61.
Roelvink, J.A., 1993. Dissipation in random wave groups incident on a beach. Coast. Eng.
use of data collected and/or archived by the staff at the FRF and are in- 19, 127–150.
debted to their long-term investment in coastal research. We are grate- Roelvink, J., Brøker, I., 1993. Cross-shore profile models. Coast. Eng. 21 (1), 163–191.
ful to the individual researchers responsible for collecting data during Roelvink, D., et al., 2009. Modelling storm impacts on beaches, dunes and barrier islands.
Coast. Eng. 56 (11–12), 1133–1152.
this period including Rob Holman for the video imagery and Britt Ruessink, B.G., Kleinhaus, M.G., van den Beukel, P.G.L., 1998. Observations of swash under
Raubenheimer, Steve Elgar and Bob Guza for the surf zone water levels. highly dissipative conditions. J. Geophys. Res. 103 (C2), 3111–3118.
Ruessink, B., Walstra, D., Southgate, H., 2003. Calibration and verification of a parametric
wave model on barred beaches. Coast. Eng. 48 (3), 139–149.
References Ruggiero, P., Holman, R.A., Beach, R.A., 2004. Wave run-up on a high-energy dissipative
beach. J. Geophys. Res. 109 (C6).
Apotsos, A., Raubenheimer, B., Elgar, S., Guza, R., 2008. Testing and calibrating parametric Sallenger Jr., A.H., Holman, R.A., 1985. Wave-energy saturation on a natural beach of
wave transformation models on natural beaches. Coast. Eng. 55 (3), 224–235. variable slope. J. Geophys. Res. 90 (C6), 11939–11945.
Battjes, J.A., Janssen, J.P.F.M., 1978. Energy loss and set-up due to breaking of random Splinter, K.D., Palmsten, M.L., 2012. Modeling dune response to an east coast low. Mar.
waves. Proceedings of the 16th Conference on Coastal Engr. ASCE, pp. 569–587. Geol. 329, 46–57.
 H.F. Stockdon et al. / Coastal Engineering 92 (2014) 1–11 11

Stockdon, H.F., Holman, R.A., 2011. Observations of wave runup, setup, and swash on Stockdon, H.F., Doran, K.S., Thompson, D.M., Sopkin, K.L., Plant, N.G., 2013. National
natural beaches. Data Series 602U.S. Geological Survey. Assessment of Hurricane-induced Coastal Erosion Hazards: Southeast Atlantic
Stockdon, H.F., Holman, R.A., Howd, P.A., Sallenger, A.H., 2006. Empirical parameterization Coast. U.S. Geological Survey, (2013-1130).
of setup, swash, and runup. Coast. Eng. 53 (7), 573–588. Thornton, E.B., Guza, R.T., 1982. Energy saturation and phase speeds measured on a
Stockdon, H.F., Sallenger, A.H., Holman, R.A., Howd, P.A., 2007. A simple model for the natural beach. J. Geophys. Res. 87 (C12), 9499–9508.
spatially-variable coastal response to hurricanes. Mar. Geol. 238, 1–20.
Stockdon, H.F., et al., 2012. National Assessment of Hurricane-induced Coastal Erosion
Hazards: Gulf of Mexico. U.S. Geological Survey, (2012-1084).