Influence of Hurricane Wind Field Variability on Real‐Time Forecast Simulations of the Coastal Environment

Abstract

Dynamic conditions occur in the coastal ocean during severe storms. Forecasting these conditions is challenging, and large‐scale numerical models require significant computing power. In this paper, we describe a real‐time modeling system (DUNEX‐RT), developed in support of the During Nearshore Event experiment (DUNEX) off the coast of North Carolina, United States of America. The model is run with wave, current, and water level boundary conditions from larger‐scale models, and provides 36‐h forecasts of significant wave height, depth‐averaged velocity, and water levels every 6‐h using Delft3D‐SWAN. Observations and forecasts run at different times are compared and communicated via an interactive website to verify model performance in real‐time and to visualize uncertainty from changing inputs. Here, we evaluate model sensitivity to inputs from seven different atmospheric hindcasts and two atmospheric forecasts for Hurricane Dorian in September 2019. The results emphasize the importance of accurate wind forcing, with significant differences observed between the output model results for different input atmospheric forcing models and forecasts produced at different times. The best results were achieved using atmospheric forcing from the high resolution rapid refresh model, and overall, DUNEX‐RT had low errors at 33 wave, water level, and current sites across the system. The model results for water levels and significant wave heights were also accurate over a longer period of 49 days. Overall, the good forecast skill achieved for the wide range of conditions over this time results suggest that this high‐resolution regional approach could be applied to forecast conditions in other coastal areas.

Full text

1. Introduction
Tropical cyclones are a significant and increasing natural hazard for human life and infrastructure along
many coastlines throughout the world. Atlantic Ocean hurricanes deliver powerful conditions to the east
and Gulf coasts of North America annually and are the most destructive natural disaster in the United
States (Grinsted etal., 2019). The frequency and intensity of these storms are projected to increase with
future climate warming and longer storm formation periods (Knutson etal.,2010). During these storms,
large waves, high storm surge, and strong currents can combine to create a multihazard marine environ-
ment, making understanding the impacts of these events in coastal areas a vital research area (Mulligan &
Hanson,2016).
Since wind forcing acts as a critical driver for coastal hydrodynamic conditions, the selection of an atmos-
pheric forcing model represents a critical decision and several atmospheric models can be used to forecast
wind conditions during a storm. Moreover, large-scale ocean models can provide predictions of surface
waves, water levels, and currents. However, these forecasts lack the high resolution needed to resolve local
Abstract Dynamic conditions occur in the coastal ocean during severe storms. Forecasting these
conditions is challenging, and large-scale numerical models require significant computing power. In
this paper, we describe a real-time modeling system (DUNEX-RT), developed in support of the During
Nearshore Event experiment (DUNEX) off the coast of North Carolina, United States of America. The
model is run with wave, current, and water level boundary conditions from larger-scale models, and
provides 36-h forecasts of significant wave height, depth-averaged velocity, and water levels every 6-h
using Delft3D-SWAN. Observations and forecasts run at different times are compared and communicated
via an interactive website to verify model performance in real-time and to visualize uncertainty from
changing inputs. Here, we evaluate model sensitivity to inputs from seven different atmospheric hindcasts
and two atmospheric forecasts for Hurricane Dorian in September 2019. The results emphasize the
importance of accurate wind forcing, with significant differences observed between the output model
results for different input atmospheric forcing models and forecasts produced at different times. The best
results were achieved using atmospheric forcing from the high resolution rapid refresh model, and overall,
DUNEX-RT had low errors at 33 wave, water level, and current sites across the system. The model results
for water levels and significant wave heights were also accurate over a longer period of 49days. Overall,
the good forecast skill achieved for the wide range of conditions over this time results suggest that this
high-resolution regional approach could be applied to forecast conditions in other coastal areas.
Plain Language Summary Large waves and fast flowing currents occur in the coastal ocean
during severe storm events, including hurricanes. Forecasting these conditions is challenging, and existing
large-scale numerical models require significant computing power and can have limitations. In this paper,
we describe a real-time modeling system of coastal North Carolina, United States of America. This model
provides forecasts of the waves, currents, and water levels every 6-h. The model results are compared with
real-time observations and communicated on an interactive website to allow users to visualize differences
in results based on winds forecast at different times. Detailed results are presented for Hurricane Dorian
in September 2019, and the model had relatively low errors at many sites across the system. Overall, the
results suggest that this high-resolution regional modeling approach could be skillfully applied to forecast
conditions in other coastal areas.
REY AND MULLIGAN
© 2020. American Geophysical Union.
All Rights Reserved.
Influence of Hurricane Wind Field Variability on Real-
Time Forecast Simulations of the Coastal Environment
Alexander J. M. Rey1 and Ryan P. Mulligan1
1Department of Civil Engineering, Queen's University, Kingston, Ontario, Canada
Key Points:
• A high-resolution regional modeling
system for real-time coastal forecasts
of surface waves, currents, and water
levels is developed
• Forcing input from different
atmospheric model hindcasts and
forecasts are compared to assess the
accuracy of output results
• Model results are quantitatively
in very good agreement with
observations across coastal NC
during Hurricane Dorian in
September 2019
Supporting Information:
• Supporting Information S1
Correspondence to:
R. P. Mulligan,
ryan.mulligan@queensu.ca
Citation:
Rey, A. J. M. & Mulligan, R. P. (2021).
Influence of hurricane wind field
variability on real-time forecast
simulations of the coastal environment.
Journal of Geophysical Research:
Oceans, 126, e2020JC016489. https://
doi.org/10.1029/2020JC016489
Received 17 JUN 2020
Accepted 25 NOV 2020
10.1029/2020JC016489
RESEARCH ARTICLE
1 of 20

Journal of Geophysical Research: Oceans
conditions and smaller scale features, limiting their application in coastal and nearshore areas. Significant
computational resources are also typically required for simulations over large domains at high spatial reso-
lutions, further limiting their application (Bilskie etal.,2019).
Atmospheric modeling has progressed substantially over the last decade in conjunction with the availabil-
ity of high-performance computing resources; however, translating these advances into hurricane impacts
on coastal ocean environments remains an active research area. Most research has focused on hurricane
storm surge, for example Bennett etal.(2018) used a detailed wind hindcast with significant spatial var-
iability to simulate inundation and overwash in a back-barrier estuary during Hurricane Sandy. Thomas
etal. (2019) compared multiple wind hindcast models with a large observational data set to investigate
the effects of storm speed and timing on water levels. The importance of including wave effects on coastal
circulation during hurricanes has also been emphasized in several studies (e.g., Mulligan etal.2008). Sheng
etal.(2010) applied a wave-current model to the Outer Banks of North Carolina (NC) and Chesapeake
Bay during Hurricane Isabel in 2003 and found that including waves improved the results. This finding is
shared by Drost etal.(2017), who also highlighted that bottom friction is a priority area for research as a key
calibration parameter in coastal models.
While numerical models are commonly applied to help understand coastal processes during storms, most
existing models have been designed with a focus on assisting emergency management decision-making,
as opposed to the goal of aiding in the planning and deployment of the sensors in the field. The coastal
emergency risks assessment (Blanton etal., 2012) is a web-accessible portal showing seven-day forecasts
of water levels along the east coast of North America from the ADvanced CIRCulation model (ADCIRC)
Prediction System; however, currents are not reported in real time. Dresback etal.(2013) found good model
agreement with observations using this model but noted the importance of accurate atmospheric forcing,
a finding also emphasized by Cyriac etal.(2018) during Hurricane Arthur. This model was also applied
by Fleming etal.(2008) using an ensemble of Holland wind model simulations during the 2007 hurricane
season. This investigation highlighted the importance and challenge of communicating model uncertainty,
as well as the high (80–160 cores) computational requirements of the setup. The United States Geological
Survey (USGS) Total Water Level and Coastal Change Forecast Viewer provides inundation predictions
along selected coastlines but is limited to nearshore water levels (Aretxabaleta etal.,2019). While currents
are included in the Navy Coastal Ocean Model (NCOM), resolution is limited (>3.7km) and waves are not
included (Martin etal., 2009). Likewise, the National Oceanic and Atmospheric Administration (NOAA)
Real Time Ocean Forecast System provides an operational, global ocean forecast based on the Global Hy-
brid Coordinate Ocean Model, but does not include waves and operates at a 4–7km resolution (Mehra &
Rivin,2010).
Several investigations have considered various factors that are important for real-time coastal forecasting;
however, relatively few studies have analyzed the performance of coastal models in a real-time forecast
configuration using forcing from multiple wind fields. An early study by Robinson etal.(1996) applied the
Harvard ocean prediction system in a forecast configuration and highlighted the importance of effective
data assimilation for accurate model results. Paramygin etal.(2017) applied the CH3D model nested in a
large-scale ADCIRC (Luettich etal.,1992) grid and identified that enhanced resolution in coastal zones is
possible using this approach; however, the large-scale grid simulations require significant computation-
al time. Recently, Dietrich etal. (2018) found that atmospheric forecasts produce more accurate coastal
forecasts compared to parametric hurricane wind models. This was also identified by Garzon etal.(2018)
in Chesapeake Bay, with more accurate water levels when using the NOAA North American Mesoscale
Model (NAM) compared to parametric winds. An 84-h forecast for the northeast Atlantic is produced using
the Simulating WAves Nearshore (SWAN) model coupled toADCIRC; however, the large domain requires
significant computational resources (Ferreira,2017). Using the ADCIRC modeling system, an investigation
by Mattocks and Forbes(2008) found good agreement with observations during Hurricane Ophelia using
a parametric wind model combined with National Hurricane Center (NHC) best track forecasts, and sug-
gested that coupled waves would be a valuable addition to the operational model. Mulligan etal.(2011)
accurately predicted wave conditions in a small and semiprotected bay using the SWAN wave model (Boo-
ij et al., 1999) with boundary wave inputs from WaveWatch III (WW3, Chawla et al., 2013). Olabarrie-
ta etal. (2011) applied the Coupled Ocean-Atmosphere-Wave-Sediment Transport (COAWST) modeling
REY AND MULLIGAN
10.1029/2020JC016489
2 of 20

Journal of Geophysical Research: Oceans
system with WW3 results at the boundaries to examine a hurricane. High computational demands can be
necessary to simulate large domains at high resolution, for example, requiring 1,000–3,000 cores to com-
plete a five-day simulation within a 2-h forecast time frame (Bilskie etal.,2019).
Increasing our understanding of risks posed by major storms has been identified as an important area for
investigation by the nearshore research community (Elko etal.,2015). This includes improving numerical
models of the fundamental coastal ocean processes that contribute to damage during storms, including
waves (Bennett & Mulligan,2017; Drost etal., 2017), erosion (Gittman etal.,2014; Xie etal.,2018), and
storm surge (Dietrich etal.,2018; Powell etal.,2010). Despite these advances, the nearshore research com-
munity has recently united to determine remaining research gaps and modeling limitations in the response
of coastal environments to storms (Elko etal.,2019). The extreme spatial and temporal variability of hurri-
canes requires model validation at many observation points during a storm, and collaboration is necessary
to facilitate sensor deployment over a large area. The During Nearshore Event experiment (DUNEX) was
proposed by the United State Coastal Research Program (Cialone etal.,2019) to support this by providing
a platform for research collaboration. This project aims to simultaneously study many coastal processes,
including flooding, sediment transport, erosion, and swash zone hydrodynamics, at sites throughout the
Outer Banks region of NC. Accurate and high-resolution real-time coastal surface forecasts provide a useful
way of planning and optimizing deployment sites immediately before a major storm event.
Recognizing the need for coastal forecasts that incorporate relevant physical processes and communicate
uncertainty in the predictions, the objective of this study is to develop a real-time, high-resolution coastal
forecast where the results are interactively compared with observations and with the results of other fore-
casts. This paper describes the development, wind input sensitivity testing, validation, and real-time com-
munication of the results for Hurricane Dorian in 2019. This includes evaluating the accuracy of different
hindcast and forecast wind models on the accuracy of hydrodynamic predictions across a range of coastal
environments, including the continental shelf, barrier islands, inlets, and estuaries.
2. Methods
2.1. Hurricane Dorian
Hurricane (H) Dorian caused major destruction along the United State east coast from Florida though North
Carolina in September 2019. Dorian made landfall in the Bahamas on September 1 as a category 5 Hurricane
on the Saffir-Simpson scale, and was the strongest recorded storm to hit the island (Lixion & Cangialosi,2019;
Royal Meterological Society,2019). With 82m/s sustained wind speeds and a peak storm surge of 7m, Dorian
resulted in 69 fatalities and widespread devastation throughout the Bahamas (UNICEF,2019). After moving
along the United State southeast coast, Dorian made landfall again on September 6 at Cape Hatteras, NC, as
a category 1 storm with 33m/s sustained winds (Avila etal.,2020). Widespread wind damage, offshore waves
of over 6m, up to 200mm of rain, and significant flooding were reported, producing mandatory evacuations,
impacting 681 homes, and causing 3 deaths (FEMA,2019; National Weather Service,2019). The post-tropical
cyclone continued northward and impacted Nova Scotia, Canada, on September 7 (Avila etal.,2020). In this
study, we investigate the storm conditions as it impacted eastern NC.
2.2. Observations
Observations are obtained from 18 wind anemometers, 21 water level gauges, 8 wave buoys, and 4 current
sensors at sites shown in Figure1. Water level measurements are obtained from the NOAA, USGS, US Army
Corps of Engineers (USACE), and the National Weather Service (NWS) (Herzmann etal.,2004). Wave ob-
servations are sourced from the National Data Buoy Center (NDBC) and the Coastal Data Information Pro-
gram (CDIP) (Flick etal.,1993). Current velocity observations are collected by the USACE Field Research
Facility (FRF). Real time observations during Hurricane Dorian were saved every 6h and communicated
together with the model results via the web interface. A complete list of all observation sources is provided
in Table1. Observations from across the system are used to statistically quantify model errors and are dis-
cussed in Section3.
REY AND MULLIGAN
10.1029/2020JC016489
3 of 20

Journal of Geophysical Research: Oceans
2.3. Numerical Model Set Up
The DUNEX real-time (DUNEX-RT) model domain was selected to cover the DUNEX project area, max-
imize coverage of different coastal environments, and include relevant observation points. Shown in Fig-
ure1, the domain covers the Albemarle-Pamlico Estuarine System (APES), including back-barrier estuar-
ies, inlets, barrier islands, and the coastal ocean across the continental shelf. Delft3D (Lesser etal.,2004)
solves the Navier-Stokes horizontal momentum equations and is capable of simulating water levels and
currents forced by both spatially varying meteorology and boundary inputs (currents and water levels).
Waves, including wave-current interactions, are simulated using SWAN (Booij etal.,1999), a third genera-
tion shallow water spectral wave model which predicts wave generation, propagation, and dissipation, and
is coupled to Delft3D to account for wave-current interactions. Delft3D-SWAN has been applied successful-
ly to this environment, notably by Mulligan etal.(2015) to model Hurricane Irene, by Clunies etal.(2017)
to investigate waves and storm surge, and by Mulligan etal.(2019) to study long-term estuarine response
to changing morphology and sea-level rise. In the present study, a 2D structured grid is used, with the
flow grid resolution varying from 100m to 400m using approximately 1.40×106 cells. The wave grid has
a coarser resolution varying between 250m and 1,000 m using approximately 5.70 × 104 cells. Bathym-
REY AND MULLIGAN
10.1029/2020JC016489
4 of 20
Figure 1. Map of the DUNEX-RT model domain including bathymetry, model boundaries, selected validation sites,
the location of the minimum sea-level pressure from 36-h HRRR model runs every 6h between September 5, 12:00
UTC and September 7, 00:00 UTC for and National Hurricane Center 6-h best track for Hurricane Dorian. Site details
are shown in Table1, and a map with all sites labeled is shown in theSupporting Information. DUNEX-RT, During
Nearshore Event experiment-real time; HRRR, high resolution rapid refresh; UTC, Coordinated Universal Time.

Journal of Geophysical Research: Oceans
etry was obtained from the NOAA coastal relief model (CRM), with a resolution of approximately 30m
(NOAA,2016). The DUNEX-RT system operates every 6h by producing 36-h forecasts that are “hot-started”
using results from the previous 6-h forecast and the most recent atmospheric forecast. Since the regional
models used for forcing incorporate recent observations into their forecasts, they are included into the mod-
el domain at the boundaries. Computations are performed with a 15-s time step. DUNEX-RT is run on 16
Intel Xeon processors with 32GB of RAM. Simulations are automatically started when new atmospheric
forecasts are released, approximately 2h after initialization, and take approximately 3 h for simulation
and 0.5h for processing, producing results with a 6-h lag. All parameters are the model defaults except
for bed roughness, which was decreased by adjusting the Chezy bottom roughness parameter (inversely
REY AND MULLIGAN
10.1029/2020JC016489
5 of 20
ID Name Parameters Grid/depth Source
FP FRF Pier Water Level; Wind 6 m USACE
BF Beaufort Duke Marine Lab Water Level; Wind N/A NOAA/Duke
VB Virginia Beach Wave Wave; Wind 47m NDBC/USACE
DS Diamond Shoals Buoy Wave; Wind 59m NDBC
OB Oslow Bay Buoy Wave 30m CDIP/USACE
CH Cape Henry Buoy Wave 18m CDIP/USACE
CN Currituck Sound North Water Level; Wave; Current 2.3m UNC
CS Currituck Sound South Water Level; Wave; Current 2.6m UNC
F6 FRF AWAC Current 6 m USACE
F11 FRF AWAC Current 11m USACE
NH Nags Head Buoy Wave 21m NDBC/UNC
O18 Oregon Inlet Buoy Wave 18m NDBC/UNC
F17 FRF 17m Buoy Wave 17m USACE
F26 FRF 26m Buoy Wave 26m USACE
OC Coast Guard Station @ Ocracoke Water Level; Wind N/A ISU/HADS
OI Oregon Inlet Marina Water Level N/A NOAA
AB Bogue Sound @ Atlantic Beach Water Level N/A USGS
HT Hatteras Coast Guard Water Level N/A NOAA/USCG
AS Albemarle Sound @ Leonards Point Water Level N/A USGS
CC Currituck Sound @ Corolla Water Level N/A USGS
PH Currituck Sound @ Point Harbor Water Level N/A USGS
JC Jean Guite Creek Outlet Water Level N/A USACE
HD Kill Devil Hills @ Hayman Street Water Level N/A USACE
VD Villa Dunes Dock Water Level N/A USACE
PI Roanoke Sound @ Point Island Water Level N/A USGS
KH Albemarle Sound @ Kitty Hawk Water Level N/A ISU/HADS
WO Roanoke River @ Westover Water Level N/A ISU/HADS
BH Pungo River @ Belhaven Water Level N/A ISU/HADS
WH Pamlico River @ Washington Water Level N/A ISU/HADS
BI Pamlico Sound @ Bell Island Pier Water Level N/A ISU/HADS
RF Pamlico Sound @ Rodanthe Water Level N/A ISU/HADS
Ferry Terminal
CDIP, Coastal Data Information Program; FRF, Field Research Facility; NDBC, National Data Buoy Center; NOAA, National Oceanic and Atmospheric
Administration; USACE, US Army Corps of Engineers, USGS, United States Geological Survey.
Table 1
List of Data Sources

Journal of Geophysical Research: Oceans
proportional to the bottom drag coefficient) from Cz=65 to 95m1/2s−1, similar to the approach used by
Drost etal.(2017). This adjustment increases predicted current velocities at two offshore stations (F6, F11),
reducing the Root-Mean-Square-Difference (RMSD) depth-averaged velocity during the 36-h crossing of
Hurricane Dorian by 10% and 19% at these two sites. This change has negligible impacts on the accuracy
of wave and water level results elsewhere in the domain, and the remainder of this paper focuses on the
sensitivity to different input wind conditions. Model accuracy statistics from all sensitivity tests are included
in the supporting information.
2.4. Forcing From Large-Scale Models
To minimize computational requirements and enable forecast runs to be completed in under 6h, the high
resolution grid is forced at the boundaries from large-scale ocean forecast models. Riemann type bound-
aries (Stelling,1983) are used in 183 segments at 5km intervals for depth-averaged currents and water
levels. Multiple sources (summarized in Table2) are used for the boundary conditions. Water level forecasts
at the offshore boundaries, including tides and storm surge, are provided by the Extratropical Surge and
Tide Operational Forecast System (ESTOFS), a North Atlantic tide and storm surge model based on AD-
CIRC (Funakoshi etal.,2012). NCOM provides large-scale ocean currents (Martin etal.,2009), which are
depth-averaged to approximate boundary flow following the method described by Edwards etal.(2012).
REY AND MULLIGAN
10.1029/2020JC016489
6 of 20
Type Atmospheric hindcasts
Abbreviation GFS NAM CFSv2 RAP ERA5
Name Global Forecast North American Climate Forecast Rapid European
System Mesoscale System v2 Refresh Reanalysis
Forecast
System
Reference Yang etal. Rogers etal. Saha etal. Benjamin etal. Hersbach and Dee
(2006) (2009) (2010) (2016) (2016)
Source NOAA NOAA NOAA NOAA ECMWRF
Domain Global North America Global CONUS Global
Horizontal res. 27km 12km 27km 13km 30km
Output time step 6 h 6 h 1 h 1 h 1 h
Type Atmospheric Hindcasts/forecasts Ocean boundary forecasts
Abbreviation RDPS HRRR ESTOFS NCOM WW3
Name Regional High Resolution Extratropical Navy Multigrid
Deterministic Rapid Refresh Surge and Tide Coastal WaveWatch III
Prediction Operational Ocean
System Forecast System Model
Reference Caron etal. Smith etal. Funakoshi etal. Martin etal. Chawla etal.
(2015) (2008) (2012) (2009) (2013)
Source Env. Can. NOAA NOAA NAVOCEANO NOAA
Domain North America CONUS North Atlantic Global Global
Horizontal res. 2.5km 3.5km 0.2km 3.7km 6.7km
Output time step 1 h 1 h 1 h 3 h 3 h
Abbreviations: CFS, Climate Forecast System; ERA, European Reanalysis; ESTOFS, Extratropical Surge and Tide
Operational Forecast System; GFS, Global Forecast System, HRRR, high resolution rapid refresh; NAM, North American
Mesoscale Model; NCOM, Navy Coastal Ocean Model; NOAA, National Oceanic and Atmospheric Administration;
RAP, Rapid Refresh; RDPS, Regional Deterministic Prediction System.
Table 2
Summary of Large-Scale Model Outputs Used as Inputs to DUNEX-RT

Journal of Geophysical Research: Oceans
The NOAA multi-grid WaveWatch III model (Chawla etal.,2013) forecasts significant wave height, peak
period, and mean wave directions, which are then applied to DUNEX-RT in 36 ocean boundary segments at
25km intervals. All large-scale ocean models are forced using the Global Forecast System (GFS) atmospher-
ic forecast, creating the potential for boundary instability from the change in atmospheric forcing between
the large-scale ocean models and DUNEX-RT. However, boundary instabilities were not observed during
the study period due to the low model sensitivity to atmospheric inputs at the model boundaries, farther
offshore. Hindcast simulations were also performed, using hindcast wind fields, to compare atmospheric
wind inputs and to provide a best-track result for comparison with the forecast predictions. These hindcast
simulations used observed wave and water level conditions at the boundaries to allow the specific impacts
from different atmospheric models to be emphasized. Water levels at the FRF and the Beaufort Marine Lab
(Figure1, FP and BF) were used at the north and south model boundaries and linearly interpolated across
the east boundary. Directional wave spectra from four wave buoys (Figure 1, OB, DS, VB, and CH) were
linearly interpolated between observation sites across the entire boundary.
Atmospheric forcing (pressure and winds) are input from several global and mesoscale models summarized
in Table2. Analysis products from the Global Forecast System (GFS), North American Mesoscale Forecast
System (NAM), and Rapid Refresh (RAP) models were used to hindcast the storm (Benjamin etal.,2016;
Rogers et al., 2009; Yang et al., 2006), in addition to reanalysis data from the Climate Forecast System
(CFSv2) and the European Reanalysis (ERA5; Copernicus Climate Change Service (C3S),2017; Hersbach
& Dee,2016; Saha etal.,2010). Atmospheric data is linearly interpolated to a 2.5km input grid. Forecasts
from the Regional Deterministic Prediction System (RDPS; from Mai etal.(2019)), and the High Resolution
Rapid Refresh Model (HRRR; from Blaylock etal.,2017; Agrawal etal., 2019) are used in both hindcast
(zero-hour initialization) and forecast configurations and are described in Caron etal. (2015) and Smith
etal.(2008), respectively. The atmospheric models vary in resolution, data assimilation sources, and in-
ternal physics, all of which can impact the accuracy of atmospheric forecasts (Garzon et al., 2018; Ruti
etal.,2008; Wedam etal.,2009).
3. Results
3.1. Input Wind Fields
A comparison of wind speeds for the wind fields described in Table2 at nine locations over a 36-h period is
shown in Figure2. All atmospheric models were effective in simulating the overall wind trends during H
Dorian, and in particular, both the HRRR and RDPS forecasts effectively predicted wind speeds throughout
the domain. Despite this broad alignment, differences between the model results and the observations are
evident. Storm timing varies between models, evidenced by the timing of the eye at station CPL (Figure2e)
near the southern boundary of the domain, with the RDPS forecast slightly lagging observations and most
hindcasts showing an arrival time ahead of observations. Variations in the track of the eye (Table3) and
more resolution are also apparent, particularly at site BF (Figure2e), with the lower resolution ERA5,
CFSv2, GFS, and NAM models including a sudden drop in wind speed which did not occur in higher reso-
lution models or observations.
The wind fields are described in Table2 and are shown in Figure3 at 18:00 Coordinated Universal Time
(UTC) on September 6. On this figure, wind magnitude observations are shown in colored circles, and the
bulk descriptive hurricane parameters, including maximum wind speed, minimum central pressure, and
radius to maximum winds are shown alongside each wind field. Notable differences exist between wind
fields, especially with respect to wind directions, causing significant changes in hydrodynamic predictions.
Overall hurricane shape and strength are similar; however, the size and location of the eye varies between
models, as well as between model runs. The difference in the modeled location of minimum pressure and
the National Hurricane Center Best Track (Avila etal.,2020) is quantified in Table3. Offshore at the Virgin-
ia Beach buoy (Figure1, VB), the HRRR and RAP winds were from the northeast (Figures3a and 3d), while
the RDPS winds were from the northwest (Figure3b) at the same time, and all models are different in wind
speed. Resolution differences between models are apparent, with the lower resolution CFSv2 (27km), GFS
(27km), and ERA5 (30km) models failing to resolve local variations in wind speeds compared to the high
resolution RAP (13km), HRRR (3.5km), and RDPS (2.5km) models during this storm with high spatial
variability in winds. Differences between wind forecasts at a specified time (e.g., 18:00 UTC) are evident
REY AND MULLIGAN
10.1029/2020JC016489
7 of 20

Journal of Geophysical Research: Oceans
between runs initialized at 18:00 UTC (18:00 UTC start, zero-hour (00Z) forecast), and runs that were ini-
tialized 18h prior (00:00 UTC start, 18-h (18Z) forecast), shown in Figures3a and 3h for the HRRR model
and Figures3b and 3i for the RDPS model. The radius to maximum winds decreased significantly (from
125km to 60km) between RDPS runs. Between HRRR runs, the eye moved 50km to the sound (Table3),
and a stronger northern wind was forecasted along the coast during the 00Z run compared to the 18Z run,
producing important differences in the forecasted currents. For example, currents at site F11 were predicted
to be 1.4m/s at 18:00 UTC from the HRRR wind field forecast initialized at 00:00 UTC, compared to 0.6m/s
from forecast started at 18:00 UTC.
3.2. Wave and Hydrodynamic Simulations
Example maps of DUNEX-RT results from the 00Z run are shown with observations for September 6 12:00
UTC in Figure4 and 18:00 UTC in Figure5. At 12:00 UTC (Figure4), the eye of H Dorian was located in the
southern half of the domain; however, the HRRR model had slightly offset eye location, with a location of
minimum sea-level pressure 22km east of the NHC best track (Table3). This is particularly visible at station
BF, with an observed speed much slower than predicted by the HRRR model. Model results were in good
agreement with observations at this time step. The wind-driven water level gradient, varying from 1.5m
to −0.5m over the 60km distance between OC and RF on the southern shore of Pamlico Sound, matched
observations (Figure4c), and the slower current speeds (<0.5m/s) observed in the north half of the domain
were predicted by the model. Significant wave heights were accurately predicted at the central O18 and DS
stations, but were overpredicted at the northern F17 and F26 stations due to the shape and timing of the
input wind field. Mean wave direction was generally well predicted throughout the domain. Model results
at these stations improved in subsequent model runs (Figure6).
At 18:00 UTC (Figure5), waves are directed from north to south, contrasting with the south to north pattern
at 12:00 UTC, with Hs=5–6m on the shelf and Hs= 1–2m in the APES, and model Hs and mean wave
REY AND MULLIGAN
10.1029/2020JC016489
8 of 20
Figure 2. Comparison between hindcast and forecasted wind speeds and observations at six selected sites shown in Figure1.

Journal of Geophysical Research: Oceans
direction results generally agree with observations (Figure5b). The strong northern winds drove water to-
ward the southern shores of the APES and produced up to 1.5m of surge in the large back-barrier estuary
(Figure5c). A strong (1.5m/s) southward alongshore current on the shelf, occurring in conjunction with a
strong southerly wind, occurred offshore of the Outer Banks, and is in agreement with the observations at
F6 and F11 (Figure5d), with measured and predicted currents of up to 0.9m/s in Currituck Sound. Com-
paring the water levels at different times during the passage of the storm in Figures4c and 5c, it is important
to note that the combination of physical processes produces the high spatial variability. For example, along
the Outer Banks barrier islands the wind-driven surge is the primary driver of sound-side water levels, and
tides dominate water levels on the ocean side.
The model results for different HRRR wind forecast start times are shown through time with observations
in Figure6. Earlier forecasts are shown in green, with later forecasts (“hot-started” from the 6-h time step)
in blue, which illustrates the impacts of differences in atmospheric forcing. Model accuracy and consistency
are visualized via overlapping forecasts, which helps identify areas with higher or lower errors. Overall wa-
ter levels forecasts are accurate and relatively consistent, particularly near inlets, with a RMSD of 0.13m at
Oregon Inlet (OI, Figure6c). Wave heights are also subject to variation from changes in boundary wave fore-
casts and winds; however, overall results were fairly accurate, with a RMSD of 0.77m at an offshore wave
buoy with a peak observed Hs of 4.5m (F17, Figure6e). While wave heights were over-predicted early in
the storm at this station, predictions improved with time, illustrated by the closer agreement between later
forecasts and observations. Current speed observations on the inner shelf are very strong (1–2m/s) during
the hurricane and thus closely depend on the input wind field, demonstrated by the very different model
predictions at F6 and F11 through time (Figures6h–6i). A more accurate wind field occurred in earlier
HRRR forecasts, and this is communicated through the overlapping curves that terminate in a vertical line
at the end of each forecast period. Despite this, depth-averaged velocity RMSDs remained low, with errors
of 0.18m/s and 0.20m/s at FRF sites F6 and F11. Displaying these changing results in real time intuitively
communicates differences between model results, forecast runs, and observations, without the additional
pre-event computational demands of a probabilistic model (e.g., Irish etal.2011).
Modified Taylor diagrams (Elvidge etal.,2014; Taylor,2001) are a useful way to visualize model performance
by comparing three statistics on a single plot. The results of DUNEX-RT using seven hindcasts and two fore-
casts as input over a 36-h period (from September 6 00:00 UTC to September 7 12:00 UTC) are shown at
nine selected sites across the system for three different parameters (η, Hs, |u|) in Figure7. These diagrams
display the correlation coefficients (R) along the azimuthal angle, the model standard deviations (σm) are
normalized against observed standard deviations (σo) and are shown along the radial axis (σ*=σm/σo). In
addition, the standard deviation normalized centered-root-mean-square-differences (CRMSD, bias correct-
ed RMSD) are radially distributed from the observation point at σ*=1 and R=1. Using this approach,
REY AND MULLIGAN
10.1029/2020JC016489
9 of 20
Track error (km) 06/09 00h 06/09 06h 06/09 12h 06/09 18h Mean
HRRR 09/06 00:00 UTC 16.7 37.1 22.2 91.8 41.9
RDPS 09/06 00:00 UTC 19.1 14.5 9.7 22.6 16.5
HRRR 16.7 28.8 20.6 42.3 27.1
RDPS 19.1 14.5 8.6 23.7 16.5
ERA5 17.8 21.6 12 42.8 23.6
RAP 3.2 20.4 23.4 35.5 20.6
CFSv2 14.5 14.4 21.3 28.6 19.7
GFS 14.5 14.4 21.3 37.8 22
NAM 6 9 6.9 8 7.5
Abbreviations: CFS, Climate Forecast System; ERA, European Reanalysis; GFS, Global Forecast System, HRRR, high resolution rapid refresh; NAM, North
American Mesoscale Model; RAP, Rapid Refresh; RDPS, Regional Deterministic Prediction System; UTC, Coordinated Universal Time.
Table 3
Mean Distance Between the Modeled Location of Minimum Sea-Level Pressure and the National Hurricane Center Best Track Over a 24-h Period on September 6,
2016, During the Crossing of H Dorian (Avila etal.,2020; Landsea & Franklin,2013)

Journal of Geophysical Research: Oceans
REY AND MULLIGAN
10.1029/2020JC016489
10 of 20
Figure 3. Comparison of hindcast and forecast wind fields on September 6 at 18:00 UTC: (a–g) 7 wind model hindcasts; (h–i) 2 wind model 18-h (18Z) forecast
products from simulations started on September 6 at 00:00 UTC. Bulk hurricane wind parameters including the peak wind speed (Umax), minimum sea-level
pressure (Pmin), and the radius to maximum winds (Rmax) are shown for each wind field. Observations are shown by colored circles on the same scale. UTC,
Coordinated Universal Time.

Journal of Geophysical Research: Oceans
model results with better agreement with observations are plotted closer to the location of the normalized
observation point. The overall statistics indicate that the zero-hour HRRR provided the best hindcast results
(RMSD=0.16m for η; 0.42m for Hs; and 0.23m/s for |u|), with the zero-hour RDPS model similarly accu-
rate (RMSD=0.21m for η; 0.61m for Hs; and 0.17m/s for |u|). Consequently, the HRRR and RDPS models
were evaluated in a forecast configuration, with slightly more accurate results from HRRR (RMSD=0.16m
for η; 0.56m for Hs; and 0.25m/s for |u|) than RDPS (RMSD=0.21m for η; 0.66m for Hs; and 0.20m/s for
|u|). Despite the overall higher accuracy of the HRRR forecast, more accurate southward winds at the FRF
sites in the RDPS forecasts produced improved depth-averaged velocity forecasts at the observed sites in the
coastal ocean (F6, F11). Results from all models and locations are available in supporting information. Over-
all statistics indicate that HRRR and RDPS provide the best description of the wind structure of Hurricane
Dorian. Notably, the range in model results occurred despite relatively small differences in the modeled
location of minimum pressure between models (Table3), emphasizing the spatial complexity of tropical
cyclones, and limiting the utility of comparisons based only on track location.
REY AND MULLIGAN
10.1029/2020JC016489
11 of 20
Figure 4. Example of model forcing and results on September 6, 2019, at 12:00 UTC: (a) winds forecasted from the September 6 00:00 UTC HRRR model run;
(b) significant wave height; (c) water levels; and (d) depth-averaged currents. Observations are shown by colored circles and magenta vectors and model results
are shown by the color contours on the same scale. Every 20th vector is shown in subplot (a), and every 12th vector is shown in subplot (b), and every 50th
vector is shown in subplot (d), and additional times are shown inSupporting Information. UTC, Coordinated Universal Time.

Journal of Geophysical Research: Oceans
A spatial visualization of the correlation coefficients (R) between model wave and water level results and
observations from the 36-h run starting on September 6, 00:00 UTC is shown in Figure8 for the HRRR and
RDPS forecasts. Overall agreement is good throughout the domain for both models; however, the HRRR
model produced slightly stronger agreement for both parameters. In particular, water level forecasts are
more accurate in the HRRR southern side of the domain compared to the RDPS model. Sound side waves
(CS N, CS S) are more accurate in the RDPS forecast than the HRRR forecast; however, this trend is reversed
for offshore wave stations due to differences in the wind fields in these areas.
4. Discussion
The modeling system designed in this study builds on previous research in the extensive body of literature
on real-time coastal forecasting to provide detailed coastal forecasts specifically to the nearshore research
community and provides a platform for the evaluation of the impact of wind field variability on the coastal
REY AND MULLIGAN
10.1029/2020JC016489
12 of 20
Figure 5. Example maps of model forcing and results on September 6, 2019, at 18:00 UTC: (a) winds forecasted from the September 6 00:00 UTC HRRR model
run, with a black box indicating zoom area for subsequent plots; (b) significant wave height; (c) water levels; and (d) depth-averaged currents. Observations
are shown by colored circles and magenta vectors and model results are shown by the color contours on the same scale. Every 20th vector is shown in subplot
(a), and every 12th vector is shown in subplot (b), and every 50th vector is shown in subplot (d), and additional times are shown in the supporting information.
UTC, Coordinated Universal Time.

Journal of Geophysical Research: Oceans
forecasts. While limitations of the present implementation of DUNEX-RT exist, the modeling system is
able to accurately predict waves, water levels, and currents at many sites across a coastal region during a
tropical storm. Existing operational models can provide accurate forecasts of coastal conditions; however,
they have been designed with a focus on emergency management decision-making, and consequently may
not meet the unique demands of interdisciplinary research. For example, while the ADCIRC Prediction Sys-
tem (Blanton etal.,2012; Cyriac etal.,2018; Dresback etal.,2013; Mattocks & Forbes,2008) provides high
REY AND MULLIGAN
10.1029/2020JC016489
13 of 20
Figure 6. Observations (black line) and six different 36-h HRRR forecast time-series results at selected sites across the system: (a–c) water levels; (d–f)
significant wave height; (g) depth-averaged currents; and (j) bathymetry and selected sites. Observations and model results for all sites are shown in supporting
information. HRRR, high-resolution rapid refresh.

Journal of Geophysical Research: Oceans
resolution water level forecasts, ocean current forecasts are not presently published in real time. Ocean cur-
rent forecasts are published by the NCOM model; however, the resolution is coarse (>3.7km) and coastal
areas, such as less than 1km wide inlets, are not well resolved (Martin etal.,2009). Previous investigations
have identified a need for high-resolution current data to improve the understanding of how coastal systems
behave and evolve (i.e., transport of materials such as sediments or contaminants), presenting a major gap
in the real-time information that is currently available (Velasquez-Montoya etal.,2020).
Recent research has emphasized the importance of wave effects on total water levels, circulation, and flood-
ing in coastal areas (Bunya etal.,2010; Drost etal.,2017; Hoeke etal.,2015; Mulligan etal.,2008; Niedoroda
REY AND MULLIGAN
10.1029/2020JC016489
14 of 20
Figure 7. Taylor diagrams showing three important statistics that quantify agreement between model results and observations (correlation coefficient (R: green
lines), standard deviation normalized centered-root-mean-square-difference (CRMSD: blue circles with origin at σ*=1 and R=1), and normalized standard
deviation (σ*=σm/σo: radially from black circles with origin at σ*=0) over 36h between September 6 00:00 and September 7 12:00 at nine selected sites for: (a)
water levels; (b) significant wave heights; and (c) depth-averaged currents. Black dots represent observations. Note scale differences between figures.
Figure 8. Visualization of model correlation coefficients (R) over a 36-h period between September 6 00:00 UTC and
September 7 12:00 UTC using the HRRR and RDPS forecasts for: (a,c) η and (b,d) Hs. HRRR, high-resolution rapid
refresh; UTC, Coordinated Universal Time.

Journal of Geophysical Research: Oceans
etal.,2010; Sheng etal.,2010). Accordingly, coupled hydrodynamic-wave models that simulate wave-cur-
rent interactions represent an important advance compared to models that do not include both waves and
currents in the ability to simulate wave-driven currents. This is shown in Figure9 through visualization
of the modeled depth-averaged-velocities at three subregions of the larger domain at the same time step
shown in Figure5. Complex current patterns are visible and the processes that are driving these flows are
different in each area. In Currituck Sound (Figure9b), a strong southward current generally occurs at this
time, in conjunction with large waves; however, local variations occur due to bathymetry. On the ocean side
of the sound, a strong alongshore current, driven by the combination of hurricane winds on the shelf and
the wave-driven flow in the surf zone, is evident in both DUNEX-RT results and observations. Contrasting
current patterns, with water flowing into the sound at Oregon Inlet (Figure9c) and out of the sound at Hat-
teras and Ocracoke Inlets (Figure9d), are a result of the wind and water level patterns at this time. Figure9
also highlights the importance of providing model results at varying levels of detail in order to visualize
small-scale variability in coastal flows.
In comparison with the high computational requirements (1,000–3,000 cores) that are often required by
coastal forecast models (Bilskie etal.,2019), DUNEX-RT completes 36-h forecasts with a grid resolution as
REY AND MULLIGAN
10.1029/2020JC016489
15 of 20
Figure 9. Depth averaged velocity results from the September 6, 00Z model run (a) at three subregions of the overall
domain: (b) Duck; and Currituck Sound; (c) Croatan Sound and Oregon Inlet; and (d) Hatteras Inlet and Okracoke
Inlet, on September 6, 18:00 UTC. Observations are shown in subplot (b) using colored circles and red arrows on the
same scale. Every 50th vector is shown in subplot (a), and every 6th vector is shown in subplot (b-d). UTC, Coordinated
Universal Time.

Journal of Geophysical Research: Oceans
low as 100m within 4h using only 16 cores. This approach, combining boundary conditions from multiple
large-scale models using Riemann boundary conditions, represents a unique computational optimization
to high-resolution coastal forecasting. This method is contrasted with existing larger-scale models, notably
the iFlood forecast by Ferreira(2017), which apply ocean-scale flexible mesh grids to reduce boundary con-
ditions but require complex grid refinement and may demand more computing resources. Relatively low
computational demand, a simplified orthogonal grid, and flexible boundary conditions of DUNEX-RT al-
lows for a wider application of similar coastal models to other high priority research areas in the future. The
use of a smaller domain with higher resolution also supports the application of ensemble forecast models
due to their significantly increased computing requirements (Fleming etal.,2008).
During H Dorian, the results presented here for many different atmospheric inputs emphasize the impor-
tance of accurate wind fields for coastal forecasting. Model accuracy varied significantly for different input
wind fields, with water level correlation coefficients ranging from 0.65 (CFSv2) to 0.76 (HRRR), and signifi-
cant wave height correlation coefficients ranging from 0.61 (GFS) to 0.88 (HRRR). This variation in accura-
cy occurred despite the relative consistency in modeled storm track (Table3) and suggests that mean track
error may not be an accurate way to compare atmospheric model accuracy compared to other hurricane
parameters, such as the radius to maximum winds and minimum central pressure (Figure3). Moreover,
changes in atmospheric forecasts through time had large effects on model outputs (Figure6), reiterating the
importance of assessing variability and uncertainty in model inputs and results (Elko etal.,2019).
Several recent investigations have highlighted the importance of and challenges associated with the com-
munication of hurricane forecast predictions (Broad etal.,2007; Morss etal.,2008; Ruginski etal.,2016).
In particular, Hyde(2017) emphasized the unique issues that arise when evaluating and communicating
uncertainty in numerical model results to forecast users in real time, and noted that present techniques may
not fully convey these concepts. The results presented here (Figure6), produced using different forecasts as
model input overlaid with observations, provide a new method to qualitatively and quickly communicate
forecast accuracy. While less comprehensive than a complete ensemble forecast, this simple technique has
significantly lower computational requirements and can facilitate straightforward comparisons with obser-
vations that are easily interpreted by users.
To assess the performance of the real-time modeling system over an extended period, a simulation was
completed during a 49-day period between September 20 to November 15, 2019. During this time, there
were nine wave events with Hs greater than 2m, and the wave and water level results at selected sites in the
center of the model domain near Oregon Inlet are shown with observations in Figure10. This simulation
was completed using zero-hour initialization atmospheric products from the HRRR model in conjunction
REY AND MULLIGAN
10.1029/2020JC016489
16 of 20
Figure 10. Observations (black line) and model results (red line) close to the center of the model domain near Oregon Inlet from a 49-days DUNEX-RT
simulation from September 20 to November 15, 2019: water level time series (a) and scatter plot (b) at OI; significant wave height time series (c) and scatter plot
(d) at O18. DUNEX-RT, During Nearshore Event experiment-real time.

Journal of Geophysical Research: Oceans
with WW3 wave boundaries and observed water levels. Overall, the model is in good agreement for both
parameters, with a correlation coefficient (R) of 0.85 for water level and 0.83 for significant wave height.
The water level variability from both the daily ocean tides and the longer timescale changes driven by winds
over Pamlico Sound (e.g., October 8–13) are captured (Figure10a). All major wave events and times with
low waves are also generally accurately simulated (Figure10c), with a slight (∼0.1m) under-prediction of
wave heights during large events (>2.5m).
While the overall agreement between observations and the DUNEX-RT forecast results are best when the
HRRR model is used for atmospheric forcing, the accuracy is limited due to several important aspects. No-
tably, the overall performance of the modeling system is limited to the accuracy of forcing and boundary
conditions from the larger scale models that feed results into it. The results presented here emphasize the
model sensitivity to atmospheric forcing conditions, and consequently, errors in the wind, precipitation,
and boundary conditions reduce the model accuracy. The hydrodynamic computations in DUNEX-RT are
implemented in two-dimensions, limiting the ability of the model to simulate stratified conditions or buoy-
ancy driven processes in the ocean or estuaries. River discharges and other hydrological processes are also
not included in the model, which may introduce errors near river channels. While sufficiently detailed to
resolve coastal processes, the model resolution is not capable of simulating detailed nearshore hydrodynam-
ics, including the surf zone and overwash of the land surface. These processes are neglected in this system
to focus on the first-order processes of waves, currents, and surge for storm conditions, optimizing the use
of computational resources to achieve fast and efficient real-time model results. While the method present-
ed here is theoretically capable of being applied to a wide range of coastal areas, additional investigation is
required to further validate this application.
5. Summary and Conclusions
As most operational models have been designed to focus on the requirements of emergency management
decisions, existing modeling systems may not meet unique coastal research needs. Although existing mod-
eling systems provide accurate and effective coastal forecasts, limitations in resolution, real-time validation,
parameter inclusion, or interactive output results may present gaps. Notably, most existing operational mod-
els are currently published with 3–10km resolution, which does not allow for small-scale coastal features
to be resolved, or do not publicly publish wave effects or current forecasts. Moreover, important questions
remain about the impact of various atmospheric forcing models on coastal forecasts.
To address these challenges, a high-resolution real-time model application called DUNEX-RT was devel-
oped using Delft3D-SWAN for the Outer Banks region of NC, USA. This paper describes the performance
of the modeling system during Hurricane Dorian in September 2019. After evaluating seven atmospheric
hindcasts, considerable variability was observed between different atmospheric hindcasts, and model out-
puts were sensitive to the selection of atmospheric wind fields, particularly with respect to currents and
water levels. Based on hindcast results, the regional deterministic prediction system (RDPS) and the high
resolution rapid refresh model (HRRR) were selected for evaluation in a forecast configuration. Effective
coastal forecasts were obtained from both atmospheric forecast models, with lower errors from the HRRR
model for water levels and waves. Depth-averaged velocity forecasts were overall more accurate when using
the RDPS model.
Relying on accurately forecasted inputs from larger scale atmospheric, ocean, and wave models as boundary
conditions, the DUNEX-RT system provides high-resolution regional results with modest computational
resources. The application of accurate boundary condition forecasts from multiple large-scale models rep-
resents a method of optimizing computational resources to advance accurate forecasts of coastal conditions.
This produces useful predictions to assist in instrumentation deployment prior to storm events that are
communicated through an interactive web interface. The presentation of varying model outputs through
time together with observations intuitively conveys the impact of wind model accuracy and uncertainty in
real time. Research should continue to investigate differences in wind field models during future storms and
evaluate the impacts of 2D versus 3D models and enhanced grid resolution for simulating coastal processes.
Future work should also include analysis of results from additional atmospheric forecasts over a longer pe-
riod, including multiple storms, to characterize accuracy of these atmospheric forecasts. The relatively low
REY AND MULLIGAN
10.1029/2020JC016489
17 of 20

Journal of Geophysical Research: Oceans
computational requirements required by DUNEX-RT suggest that it could be effectively applied using forc-
ing from an ensemble of atmospheric forecasts to produce a probabilistic model, and future work should
investigate this possibility. The results presented here suggest that this method of developing a high-resolu-
tion regional forecast system and interactive real-time validation could be applied to forecast conditions in
other areas of the coastal ocean.
Data Availability Statement
The data used in this study are archived in the “Hydrodynamic Model Results for Hurricane Impacts in
Coastal North Carolina” Dataverse repository at Queen's University (https://dataverse.scholarsportal.info/
dataverse/CoastalModelling). The DUNEX-RT model results are available at http://coastlines.engineering.
queensu.ca/dunexrt.
References
Agrawal, S., Barrington, L., Bromberg, C., Burge, J., Gazen, C., & Hickey, J. (2019). Machine learning for precipitation nowcasting from
radar images. 33rd Conference on neural information processing systems (pp. 1–6). Vancouver: NeurIPS 2019. arXiv:1912.12132
Aretxabaleta, A. L., Doran, K. S., Long, J. W., Erikson, L. H., & Storlazzi, C. D. (2019). Toward a national Coastal hazard forecast of total
water levels. In J. D. Rosati, P. Wang, & M. Vallee (Eds.), Coastal sediments 2019 (pp. 1373–1384). St. Petersburg, Florida: WORLD SCI-
ENTIFIC. https://doi.org/10.1142/9789811204487_0120
Avila, L. A., Stewart, S. R., Berg, R., & Hagen, A. B. (2020). Hurricane Dorian. Technical report. National Hurricane Center, Miami, Florida,
United States Hurricane Dorian. Technical report AL052019. National Hurricane Center, Miami, Florida, United States. https://www.
nhc.noaa.gov/data/tcr/AL152017_Maria.pdf(1–74). https://www.nhc.noaa.gov/data/tcr/AL052019_Dorian.pdf
Benjamin, S. G., Weygandt, S. S., Brown, J. M., Hu, M., Alexander, C. R., Smirnova, T. G., etal. (2016). A North American hourly assimila-
tion and model forecast cycle: The rapid refresh. Monthly Weather Review, 144, 1669–1694. https://doi.org/10.1175/MWR-D-15-0242.1
Bennett, V. C., & Mulligan, R. P. (2017). Evaluation of surface wind fields for prediction of directional ocean wave spectra during Hurricane
Sandy. Coastal Engineering, 125, 1–15. https://doi.org/10.1016/j.coastaleng.2017.04.003
Bennett, V. C., Mulligan, R. P., & Hapke, C. J. (2018). A numerical model investigation of the impacts of Hurricane Sandy on water level
variability in Great South Bay, New York. Continental Shelf Research, 161, 1–11. https://doi.org/10.1016/j.csr.2018.04.003
Bilskie, M. V., Hagen, S. C., & Medeiros, S. C. (2019). Unstructured finite element mesh decimation for real-time Hurricane storm surge
forecasting. Coastal Engineering, 156, 103622. https://doi.org/10.1016/j.coastaleng.2019.103622
Blanton, B., McGee, J., Fleming, J., Kaiser, C., Kaiser, H., Lander, H., etal. (2012). Urgent computing of storm surge for North Carolina's
coast. Procedia Computer Science, 9, 1677–1686. https://doi.org/10.1016/j.procs.2012.04.185
Blaylock, B. K., Horel, J. D., & Liston, S. T. (2017). Cloud archiving and data mining of High-Resolution Rapid Refresh forecast model
output. Computers and Geosciences, 109, 43–50. https://doi.org/10.1016/j.cageo.2017.08.005
Booij, N., Ris, R. C., & Holthuijsen, L. H. (1999). A third-generation wave model for coastal regions 1. Model description and validation.
Journal of Geophysical Research, 104, 7649–7666. https://doi.org/10.1029/98JC02622
Broad, K., Leiserowitz, A., Weinkle, J., & Steketee, M. (2007). Misinterpretations of the “cone of uncertainty” in Florida during the 2004
hurricane season. Bulletin of the American Meteorological Society, 88, 651–667. https://doi.org/10.1175/BAMS-88-5-651
Bunya, S., Dietrich, J. C., Westerink, J. J., Ebersole, B. A., Smith, J. M., Atkinson, J. H., etal. (2010). A high-resolution coupled riverine
flow, tide, wind, wind wave, and storm surge model for southern Louisiana and Mississippi. Part I: Model development and validation.
Monthly Weather Review, 138, 345–377. https://doi.org/10.1175/2009MWR2906.1
Caron, J. F., Milewski, T., Buehner, M., Fillion, L., Reszka, M., Macpherson, S., etal. (2015). Implementation of deterministic weather fore-
casting systems based on ensemble-variational data assimilation at Environment Canada. Part II: The regional system. Monthly Weather
Review, 143, 2560–2580. https://doi.org/10.1175/MWR-D-14-00353.1
Chawla, A., Tolman, H. L., Gerald, V., Spindler, D., Spindler, T., Alves, J. H. G., etal. (2013). A multigrid wave forecasting model: A new
paradigm in operational wave forecasting. Weather and Forecasting, 28, 1057–1078. https://doi.org/10.1175/WAF-D-12-00007.1
Cialone, M., Elko, N., Lillycrop, J., Stockdon, H., Raubenheimer, B., & Rosati, J. (2019). During nearshore event experiment (DUNEX): A
collaborative community field data collection effort. In J. D. Rosati, P. Wang, & M. Vallee (Eds.), Coastal sediments 2019 (pp. 2958–2966).
St. Petersburg, Florida: WORLD SCIENTIFIC. https://doi.org/10.1142/9789811204487_0253
Clunies, G. J., Mulligan, R. P., Mallinson, D. J., & Walsh, J. P. (2017). Modeling hydrodynamics of large lagoons : Insights from the Albe-
marle-Pamlico Estuarine System. Estuarine, Coastal and Shelf Science, 189, 90–103. https://doi.org/10.1016/j.ecss.2017.03.012
Copernicus Climate Change Service (C3S) (2017). ERA5: Fifth generation of ECMWF atmospheric reanalyzes of the global climate. https://
cds.climate.copernicus.eu
Cyriac, R., Dietrich, J. C., Fleming, J. G., Blanton, B. O., Kaiser, C., Dawson, C. N., etal. (2018). Variability in coastal flooding predictions
due to forecast errors during hurricane Arthur. Coastal Engineering, 137, 59–78. https://doi.org/10.1016/j.coastaleng.2018.02.008
Dietrich, J. C., Muhammad, A., Curcic, M., Fathi, A., Dawson, C. N., Chen, S. S., etal. (2018). Sensitivity of storm surge predictions
to atmospheric forcing during Hurricane Isaac. Journal of Waterway, Port, Coastal, and Ocean Engineering, 144, 1–24. https://doi.
org/10.1061/(ASCE)WW.1943-5460.0000419
Dresback, K. M., Fleming, J. G., Blanton, B. O., Kaiser, C., Gourley, J. J., Tromble, E. M., etal. (2013). Skill assessment of a real-time forecast
system utilizing a coupled hydrologic and coastal hydrodynamic model during Hurricane Irene (2011). Continental Shelf Research, 71,
78–94. https://doi.org/10.1016/j.csr.2013.10.007
Drost, E. J., Lowe, R. J., Ivey, G. N., Jones, N. L., & Péquignet, C. A. (2017). The effects of tropical cyclone characteristics on the surface wave
fields in Australia's North West region. Continental Shelf Research, 139, 35–53. https://doi.org/10.1016/j.csr.2017.03.006
Edwards, K., Nguyen, T., & Sitton, D. (2012). Generating Delft3D boundary conditions using the Navy Coastal Ocean model. Technical report.
Naval research laboratory. Stennis Space Center, MS. (NRL/MR/7320--12-9448, pp. 1–16). Slidell, Louisiana: Naval Research Laboratory.
REY AND MULLIGAN
10.1029/2020JC016489
18 of 20
Acknowledgments
The authors thank the USCRP and
the DUNEX community, including
Britt Raubenheimer at the Woods Hole
Oceanographic Institution, Spicer Bak
and Ian Conery at the USACE Field
Research Facility, D. Reide Corbett at
the East Carolina University Coastal
Studies Institute (CSI), and Allison
Penko at the U.S. Naval Research
Laboratory (NRL). Research funding
for this project was provided by the US
Office of Naval Research (ONR) Global
science program with a Naval Inter-
national Cooperative Opportunities in
Science and Technology (NICOP) grant
awarded to R. Mulligan under award
number N62909-17-1-2169, and the
Queen's University Engineering Dean's
Graduate Research Award held by A.
Rey. R. P. Mulligan also acknowledges
support from the Natural Science and
Engineering Research Council of Can-
ada Discovery Grant Program under
award number RGPIN/04043-2018.
Computational support was provided by
SHARCNET (www.sharcnet.ca), Com-
pute Canada (http://computecanada.
ca), Cory Wyatt at Queen's University,
and Maria Aristizabal Vargas at Rutgers
University.

Journal of Geophysical Research: Oceans
Elko, N., Dietrich, J., Cialone, M., Stockdon, H., Bilskie, M., Boyd, B., etal. (2019). Advancing the understanding of storm processes and
impacts. Shore and Beach, 87, 41–55.
Elko, N., Feddersen, F., Foster, D., Hapke, C., Mcninch, J., & Mulligan, R. P. (2015). The future of nearshore processes research. Shore and
Beach, 83, 13–38.
Elvidge, S., Angling, M. J., & Nava, B. (2014). On the use of modified Taylor diagrams to compare ionospheric assimilation models. 2014
XXXIth URSI general assembly and scientific symposium (URSI GASS) (pp. 1–4). IEEE. https://doi.org/10.1109/URSIGASS.2014.6929835
FEMA (2019). Preliminary damage assessment report: North Carolina–Hurricane Dorian. Technical report. Federal emergency management
agency (FEMA-4465-DR, pp. 1–2). Hyattsville, MD: Federal Emergency Management Agency. Retrieved from https://www.fema.gov/
sites/default/files/2020-03/FEMA4465DRNC.pdf
Ferreira, C. (2017). iFLOOD: A real time flood forecast system for total water modeling in the national capital region. AGU fall meeting
abstracts, New Orleans, Louisiana (pp. NH41A-0153).
Fleming, J. G., Fulcher, C. W., Luettich, R. A., Estrade, B. D., Allen, G. D., & Winer, H. S., (2008). A real time storm surge forecasting
system using ADCIRC. In Proceedings of the International Conference on Estuarine and Coastal Modeling (pp. 893–912). https://doi.
org/10.1061/40990(324)48
Flick, R. E., McGehee, D. D., Seymour, R. J., & Guza, R. T. (1993). The coastal data information program—A successful federal, state and
university cooperation. Coastal engineering considerations in coastal zone management. ASCE (pp. 245–249).
Funakoshi, Y., Feyen, J., Aikman, F., Tolman, H., van der Westhuysen, A., Chawla, A., etal. (2012). Development of extratropical surge
and tide operational forecast system (ESTOFS). Estuarine and coastal modeling (pp. 201–212). Reston, VA: American Society of Civil
Engineers. https://doi.org/10.1061/9780784412411.00012
Garzon, J. L., Ferreira, C. M., & Padilla-Hernandez, R. (2018). Evaluation of weather forecast systems for storm surge modeling in the
Chesapeake Bay. Ocean Dynamics, 68, 91–107. https://doi.org/10.1007/s10236-017-1120-x
Gittman, R. K., Popowich, A. M., Bruno, J. F., & Peterson, C. H. (2014). Marshes with and without sills protect estuarine shorelines from
erosion better than bulkheads during a Category 1 hurricane. Ocean and Coastal Management, 102, 94–102. https://doi.org/10.1016/j.
ocecoaman.2014.09.016
Grinsted, A., Ditlevsen, P., & Christensen, J. H. (2019). Normalized US hurricane damage estimates using area of total destruction 1900-
2018. Proceedings of the National Academy of Sciences, 116, 23942–23946. https://doi.org/10.1073/pnas.1912277116
Hersbach, H., & Dee, D. (2016). ERA5 reanalysis is in production, ECMWF Newsletter 147, ECMWF. ECMWF Newsletter, 147, 7.
Herzmann, D., Arritt, R., & Todey, D. (2004). Iowa environmental mesonet. Available at http://mesonet.agron.iastate.edu/request/coop/
fe.phtml (verified 27 September 2005). Ames, IA: Iowa State University, Department of Agronomy.
Hoeke, R. K., McInnes, K. L., & O'Grady, J. G. (2015). Wind and wave setup contributions to extreme sea levels at a tropical high island: A
stochastic cyclone simulation study for Apia, Samoa. Journal of Marine Science and Engineering, 3, 1117–1135. https://doi.org/10.3390/
jmse3031117
Hyde, J. T. (2017). Avoiding the windshield wiper effect: A survey of operational meteorologists on the uncertainty in hurricane track
forecasts and communication (PhD thesis).
Irish, J. L., Song, Y. K., & Chang, K. A. (2011). Probabilistic hurricane surge forecasting using parameterized surge response functions.
Geophysical Research Letters, 38, 1–5. https://doi.org/10.1029/2010GL046347
Knutson, T. R., McBride, J. L., Chan, J., Emanuel, K., Holland, G., Landsea, C., etal. (2010). Tropical cyclones and climate change. Nature
Geoscience, 3, 157–163. https://doi.org/10.1038/ngeo779
Landsea, C. W., & Franklin, J. L. (2013). Atlantic hurricane database uncertainty and presentation of a new database format. Monthly
Weather Review, 141(10), 3576–3592. https://doi.org/10.1175/MWR-D-12-00254.1
Lesser, G. R., Roelvink, J. A., van Kester, J. A., & Stelling, G. S. (2004). Development and validation of a three-dimensional morphological
model. Coastal Engineering, 51, 883–915. https://doi.org/10.1016/j.coastaleng.2004.07.014
Lixion, A., & Cangialosi, J. (2019). Hurricane Dorian advisory number 34. https://www.nhc.noaa.gov/archive/2019/al05/al052019.pub-
lic.034.shtml?
Luettich, R., Westerink, J., & Scheffner, N. (1992). ADCIRC: An advanced three-dimensional circulation model for shelves coasts and estuar-
ies, report 1: Theory and methodology of ADCIRC-2DDI and ADCIRC-3DL. Technical Report 32466, Vicksburg, MS, USA.
Mai, J., Kornelsen, K. C., Tolson, B. A., Fortin, V., Gasset, N., Bouhemhem, D., et al. (2019). The Canadian surface prediction archive
(CaSPAr): A platform to enhance environmental modeling in Canada and globally. Bulletin of the American Meteorological Society,
101(3), E341–E356. https://doi.org/10.1175/bams-d-19-0143.1
Martin, P. J., Barron, C. N., Smedstad, L. F., Campbell, T. J., Wallcraft, A. J., Rhodes, R. C., etal. (2009). User's manual for the Navy Coastal
Ocean model (NCOM) version 4.0. Technical report. Naval Research Lab Stennis Space Center Ms Ocean dynamics and prediction branch.
https://apps.dtic.mil/docs/citations/ADA508063
Mattocks, C., & Forbes, C. (2008). A real-time, event-triggered storm surge forecasting system for the state of North Carolina. Ocean Mod-
elling, 25, 95–119. https://doi.org/10.1016/j.ocemod.2008.06.008
Mehra, A., & Rivin, I. (2010). A real time ocean forecast system for the north Atlantic ocean. Terrestrial, Atmospheric and Oceanic Sciences,
21, 211–228. https://doi.org/10.3319/TAO.2009.04.16.01(IWNOP)
Morss, R. E., Demuth, J. L., & Lazo, J. K. (2008). Communicating uncertainty in weather forecasts: A survey of the U.S. public. Weather and
Forecasting, 23, 974–991. https://doi.org/10.1175/2008WAF2007088.1
Mulligan, R. P., & Hanson, J. L. (2016). Alongshore momentum transfer to the nearshore zone from energetic ocean waves generated by
passing hurricanes. Journal of Geophysical Research: Oceans, 121, 4178–4193. https://doi.org/10.1002/2016JC011706
Mulligan, R. P., Hay, A. E., & Bowen, A. J. (2008). Wave-driven circulation in a coastal bay during the landfall of a hurricane. Journal of
Geophysical Research, 113, 1–10. https://doi.org/10.1029/2007JC004500
Mulligan, R. P., Mallinson, D. J., Clunies, G. J., Rey, A., Culver, S. J., Zaremba, N., etal. (2019). Estuarine responses to long-term changes in
inlets, morphology, and sea level rise. Journal of Geophysical Research: Oceans, 124, 9235–9257. https://doi.org/10.1029/2018JC014732
Mulligan, R. P., Perrie, W., Toulany, B., Smith, P. C., Hay, A. E., & Bowen, A. J. (2011). Performance of nowcast and forecast wave models
for Lunenburg Bay, Nova Scotia. Atmosphere–Ocean, 49, 1–7. https://doi.org/10.1080/07055900.2011.558468
Mulligan, R. P., Walsh, J. P., & Wadman, H. M. (2015). Storm surge and surface waves in a Shallow Lagoonal Estuary during the crossing of a
hurricane. Journal of Waterway, Port, Coastal, and Ocean Engineering, 141, 1–11. https://doi.org/10.1061/(ASCE)WW.1943-5460.0000260
National Weather Service. (2019). Post tropical cyclone report: Hurricane Dorian. Technical report. National Weather Service, Raleigh, NC.
https://www.weather.gov/media/rah/HurricaneDorian/DorianPSH.pdf
Niedoroda, A. W., Resio, D. T., Toro, G. R., Divoky, D., Das, H. S., & Reed, C. W. (2010). Analysis of the coastal Mississippi storm surge
hazard. Ocean Engineering, 37, 82–90. https://doi.org/10.1016/j.oceaneng.2009.08.019
REY AND MULLIGAN
10.1029/2020JC016489
19 of 20

Journal of Geophysical Research: Oceans
NOAA. (2016). U.S. Coastal relief model. http://www.ngdc.noaa.gov/mgg/coastal/crm.html; https://doi.org/10.7289/V53R0QR5
Olabarrieta, M., Warner, J. C., & Kumar, N. (2011). Wave-current interaction in Willapa Bay. Journal of Geophysical Research: Oceans, 116,
1–27. https://doi.org/10.1029/2011JC007387
Paramygin, V., Sheng, Y., & Davis, J. (2017). Towards the development of an operational forecast system for the Florida Coast. Journal of
Marine Science and Engineering, 5, 8. https://doi.org/10.3390/jmse5010008
Powell, M. D., Murillo, S., Dodge, P., Uhlhorn, E., Gamache, J., Cardone, V., etal. (2010). Reconstruction of Hurricane Katrina's wind fields
for storm surge and wave hindcasting. Ocean Engineering, 37, 26–36. https://doi.org/10.1016/j.oceaneng.2009.08.014
Robinson, A. R., Arango, H. G., Warn-Varnas, A., Leslie, W. G., Miller, A. J., Haley, P. J., & Lozano, C. J. (1996). Real-time regional forecast-
ing. Elsevier oceanography series, 61, 377–410. https://doi.org/10.1016/S0422-9894(96)80017-1
Rogers, E., DiMego, G., Black, T., Ek, M., Ferrier, B., Gayno, G., etal. (2009). The NCEP North American mesoscale modeling system:
Recent changes and future plans. Preprints, 23rd Conference on weather analysis and forecasting/19th conference on numerical weather
prediction.
Royal Meteorological Society. (2019). Weather news. Weather, 74, 330–331. https://doi.org/10.1002/wea.3352
Ruginski, I. T., Boone, A. P., Padilla, L. M., Liu, L., Heydari, N., Kramer, H. S., etal. (2016). Non-expert interpretations of hurricane forecast
uncertainty visualizations. Spatial Cognition and Computation, 16, 154–172. https://doi.org/10.1080/13875868.2015.1137577
Ruti, P. M., Marullo, S., D'Ortenzio, F., & Tremant, M. (2008). Comparison of analyzed and measured wind speeds in the perspective of
oceanic simulations over the Mediterranean basin: Analyses, QuikSCAT and buoy data. Journal of Marine Systems, 70, 33–48. https://
doi.org/10.1016/j.jmarsys.2007.02.026
Saha, S., Moorthi, S., Pan, H. L., Wu, X., Wang, J., Nadiga, S., etal. (2010). The NCEP climate forecast system reanalysis. Bulletin of the
American Meteorological Society, 91, 1015–1058. https://doi.org/10.1175/2010BAMS3001.1
Sheng, Y. P., Alymov, V., & Paramygin, V. A. (2010). Simulation of storm surge, wave, currents, and inundation in the outer banks and
Chesapeake bay during Hurricane Isabel in 2003: The importance of waves. Journal of Geophysical Research, 115, (27 pp.) C04008.
https://doi.org/10.1029/2009JC005402
Smith, T. L., Benjamin, S. G., Brown, J. M., Weygandt, S., Smirnova, T., & Schwartz, B. (2008). Convection forecasts from the hourly updated,
3-km high resolution rapid refresh (HRRR) model. https://ams.confex.com/ams/24SLS/techprogram/paper_142055.htm
Stelling, G. S. (1983). On the construction of computational methods for shallow water flow problems (PhD thesis). TU Delft, Delft. http://
resolver.tudelft.nl/uuid:d3b818cb-9f91-4369-a03e-d90c8c175a96
Taylor, K. E. (2001). Summarizing multiple aspects of model performance in a single diagram. Journal of Geophysical Research, 106,
7183–7192. https://doi.org/10.1029/2000JD900719
Thomas, A., Dietrich, J. C., Asher, T. G., Bell, M., Blanton, B. O., Copeland, J. H., etal. (2019). Influence of storm timing and forward speed
on tides and storm surge during Hurricane Matthew. Ocean Modelling, 137, 1–19. https://doi.org/10.1016/j.ocemod.2019.03.004
UNICEF. (2019). Hurricane dorian the Bahamas humanitarian situation in numbers situation overview & humanitarian needs. Technical
Report 4. United Nations. https://reliefweb.int/report/bahamas/hurricane-dorian-bahamas-humanitarian-situation-report-no-4-re-
porting-period-25
Velasquez-Montoya, L., Overton, M. F., & Sciaudone, E. J. (2020). Natural and anthropogenic-induced changes in a tidal inlet: Morpholog-
ical evolution of Oregon Inlet. Geomorphology, 350, 106871. https://doi.org/10.1016/j.geomorph.2019.106871
Wedam, G. B., McMurdie, L. A., & Mass, C. F. (2009). Comparison of model forecast skill of sea level pressure along the east and west
coasts of the United States. Weather and Forecasting, 24, 843–854. https://doi.org/10.1175/2008WAF2222161.1
Xie, X., Li, M., & Ni, W. (2018). Roles of wind-driven currents and surface waves in sediment resuspension and transport during a tropical
storm. Journal of Geophysical Research: Oceans, 123, 8638–8654. https://doi.org/10.1029/2018JC014104
Yang, F., Pan, H. L., Krueger, S. K., Moorthi, S., & Lord, S. J. (2006). Evaluation of the NCEP global forecast system at the ARM SGP site.
Monthly Weather Review, 134, 3668–3690. https://doi.org/10.1175/MWR3264.1
REY AND MULLIGAN
10.1029/2020JC016489
20 of 20