Skill Assessment of a Real-Time Forecast System Utilizing a Coupled Hydrologic and Coastal Hydrodynamic Model During Hurricane Irene (2011)

Continental Shelf Research 71 (2013) 78–94

Contents lists available at ScienceDirect

Continental Shelf Research

journal homepage: www.elsevier.com/locate/csr

Research papers Skill assessment of a real-time forecast system utilizing a coupled hydrologic and coastal hydrodynamic model during Hurricane Irene (2011)

Kendra M. Dresback a,n, Jason G. Fleming b, Brian O. Blanton c, Carola Kaiser d, Jonathan J. Gourley e, Evan M. Tromble a, Richard A. Luettich Jr. f, Randall L. Kolar a, Yang Hong a, Suzanne Van Cooten g, Humberto J. Vergara a, Zac L. Flamig h, Howard M. Lander c, Kevin E. Kelleher e, Kodi L. Nemunaitis-Monroe e,h a School of Civil Engineering and Environmental Science, University of Oklahoma, Norman, OK, United States b Seahorse Coastal Consulting, Morehead City, NC, United States c RENCI – Renaissance Computing Institute, Raleigh, NC, United States d School of the Coast and Environment, Louisiana State University, Baton Rouge, LA, United States e NOAA National Severe Storms Laboratory, Norman, OK, United States f Institute of Marine Sciences, University of North Carolina, Morehead City, NC, United States g NOAA/National Weather Service, Lower Mississippi River Forecast Center, Slidell, LA, United States h CIMMS – Cooperative Institute for Mesoscale Meteorological Studies, University of Oklahoma, Norman, OK, United States article info abstract

Article history: Due to the devastating effects of recent hurricanes in the Gulf of Mexico (e.g., Katrina, Rita, Ike and Received 21 August 2012 Gustav), the development of a high-resolution, real-time, total water level prototype system has been Received in revised form accelerated. The fully coupled model system that includes hydrology is an extension of the ADCIRC Surge 28 September 2013 Guidance System (ASGS), and will henceforth be referred to as ASGS-STORM (Scalable, Terrestrial, Ocean, Accepted 8 October 2013 River, Meteorological) to emphasize the major processes that are represented by the system.The ASGS- Available online 24 October 2013 STORM system incorporates tides, waves, winds, rivers and surge to produce a total water level, which Keywords: provides a holistic representation of coastal flooding. ASGS-STORM was rigorously tested during Prototype total water level modeling system Hurricane Irene, which made landfall in late August 2011 in North Carolina. All results from ASGS- Hydrodynamics STORM for the advisories were produced in real-time, forced by forecast wind and pressure fields Hydrology computed using a parametric tropical cyclone model, and made available via the web. Herein, a skill Water waves fi Wind assessment, analyzing wind speed and direction, signi cant wave heights, and total water levels, is used Hurricanes to evaluate ASGS-STORM's performance during Irene for three advisories and the best track from the National Hurricane Center (NHC). ASGS-STORM showed slight over-prediction for two advisories (Advisory 23 and 25) due to the over-estimation of the storm intensity. However, ASGS-STORM shows notable skill in capturing total water levels, wind speed and direction, and significant wave heights in North Carolina when utilizing Advisory 28, which had a slight shift in the track but provided a more accurate estimation of the storm intensity, along with the best track from the NHC. Results from ASGS- STORM have shown that as the forecast of the advisories improves, so does the accuracy of the models used in the study; therefore, accurate input from the weather forecast is a necessary, but not sufficient, condition to ensure the accuracy of the guidance provided by the system. While Irene provided a real- time test of the viability of a total water level system, the relatively insignificant freshwater discharges precludes definitive conclusions about the role of freshwater discharges on total water levels in estuarine zones. Now that the system has been developed, on-going work will examine storms (e.g., Floyd) for which the freshwater discharge played a more meaningful role. & 2013 Elsevier Ltd. All rights reserved.

1. Introduction the development of a high-resolution, real-time, total water level prototype forecast system has been accelerated: the development Since the devastating effects of Hurricanes Katrina, Rita, Ike and and assessment of such a system is the subject of this manuscript. Gustav in the Louisiana and the northern Gulf of Mexico region, It produces a holistic representation of the coastal ﬂooding through development of a total water level product (deﬁned herein to be tidesþwavesþ riversþsurgeþprecipitation). n Corresponding author. Tel.: þ1 405 325 8529; fax: þ1 405 325 4217. Over the last several years, the ADCIRC Coastal Circulation and E-mail address: [email protected] (K.M. Dresback). Surge Model, which forms the backbone of the system described

0278-4343/$ - see front matter & 2013 Elsevier Ltd. All rights reserved. http://dx.doi.org/10.1016/j.csr.2013.10.007 K.M. Dresback et al. / Continental Shelf Research 71 (2013) 78–94 79 herein, has been used in many hurricane studies, in particular, levels on the Tar/Pamlico River) (Tromble et al., 2011). Hurricane the Interagency Performance Evaluation Task Force to study Hurri- Dennis preceded Floyd and alleviated the drought conditions in cane Katrina (Link et al., 2006), and evaluations of the hindcasts of the North Carolina coastal areas; however, it also set up the Hurricanes Gustav (Forbes et al., 2010)andRita(Dietrich et al., 2010). antecedent conditions that brought about massive flooding during As part of this process, a wave component (Dietrich et al., 2010)was Hurricane Floyd. Thus, in order to accurately represent the coastal added and there has been continued development of a parametric flood inundation caused by both the upstream flooding and storm tropical cyclone model following Holland (1980) and Mattocks et al. surge, a system is needed that incorporates not only the tides, (2006). In addition, a real-time system has been developed, the waves, winds and atmospheric pressure differences due to the ADCIRC Surge Guidance System (ASGS), to automate the use of storm but also the antecedent conditions, precipitation and ADCIRC in forecast applications (Fleming et al., 2008). The model riverine flows. Events such as these have spurred the development system described in this paper demonstrates a further extension of of a total water level product, which has been the focus of both ASGS to include a hydrologic model, capturing both coastal and DHS (Department of Homeland Security) and NOAA projects. Thus, inland total water levels associated with tropical and extratropical herein, we describe progress towards the development of a storms. This model system is herein referred to as ASGS-STORM modeling system to produce high-resolution, total water level (ASGS-Scalable, Terrestrial, Ocean, River, Meteorology). A parallel, forecasts for coastal, near-shore and inland areas. collaborative effort with NOAA (National Oceanic and Atmospheric Previous studies have looked at real-time systems; however, Administration) through their CI-FLOW (Coastal Inland-Flood none of these previous systems have coupled a hydrologic and Observation and Warning) project is reported in Van Cooten hydrodynamic model in a real-time framework. Blain et al. (2002) et al. (2011), which documents the history of the system's devel- documented a real-time tidal modeling system designed to inte- opment and its initial testing. grate meteorological information from different scale meteorolo- In some coastal storm events, the surge and wave components gical models. However, this system did not include waves, play the most significant role; while for others, the freshwater hydrological components, or the ability to develop the meteor- component from upstream river flow is the dominant component. ological forcing from tropical cyclone advisories. Mattocks et al. An event that illustrated the need to incorporate the upstream (2006) developed a system that provided real-time coastal mod- freshwater component in the coupled model system was Hurri- eling utilizing ADCIRC and a new parametric tropical cyclone cane Floyd. The North Carolina watersheds saw only a small surge model. This system included the tidal effects; however, it did not ð3mÞ during Hurricane Floyd, but inland main stem river include the waves and hydrological models. Modern cyberinfras- reaches and their tributaries experienced record precipitation that tructure tools were applied to ADCIRC by Ramakrishnan et al. produced historical flooding (comparable to the 500-year flood (2006) to develop a real-time modeling capability that could

Fig. 1. Schematic of ASGS-STORM for predicting total water inundation in coastal North Carolina; upper boxes illustrate model components in the coupled system, lower boxes illustrate output products, and arrows and text boxes indicate data flow. 80 K.M. Dresback et al. / Continental Shelf Research 71 (2013) 78–94 ingest information from several different locations. The cyberin- 2.1.1. Hydrologic model and precipitation—HL-RDHM and QPE/QPF frastructure developed in this study was utilized in the initial test The hydrologic modeling component of the coupled model of the coupled model system for Hurricane Earl. Fleming et al. system, HL-RDHM (Koren et al., 2004), ingests precipitation (2008) describes an automated surge guidance system that was information from QPFs1 and/or QPEs2 to produce riverine flows. initially developed for use in Lake Pontchartrain in Louisiana; QPEs are used as model inputs for current and past precipitation however, as additional capabilities were added to the system, a (required for model states), and QPFs are used for future time more generalized format was added that allowed it to be utilized steps. Surface runoff generation is based on the original concepts outside of Louisiana. This system evolved into the ADCIRC Surge of the Sacramento soil moisture accounting model (SAC-SMA- Guidance System (ASGS). ASGS includes software automation, job Burnash, 1995). These concepts are applied to each 4-km grid cell control, fault tolerance and data flow services. The ASGS technol- in the domain, and water is routed downstream using the kine- ogy was implemented in ASGS-STORM for use after Hurricane Earl, matic wave equation. This model is the operational National to couple the hydrologic and hydrodynamic models. Lastly, Chu Weather Service (NWS) distributed hydrologic model and is fixed et al. (2009) developed a relocatable ADCIRC automation system on a 4-km grid based on the Hydrologic Rainfall Analysis Project that utilized a toolbox approach, which allowed the user to run the (HRAP—Reed and Maidment, 1999), which allows for the direct system in either an automated or manual fashion. It provided integration of the QPE/QPF information. There are a total of 19 five modules that included system set up, data acquisition, model parameters and 6 state variables that control the behavior of the configuration, model run status, and product generation. The system storages and fluxes in the model. A priori parameter settings for was designed to be relocatable to areas of interest in a short period of HL-RDHM have been estimated over the United States using timeandwasdevelopedfortheUnitedStatesNavy. readily available spatial maps of soil characteristics and land cover. NOAA utilizes several different models in operational settings in a The 19 parameters enable a great deal of flexibility for the model non-coupled fashion to simulate water quantity (e.g., SLOSH: Sea, to be applied in diverse geographic and hydroclimatological Lake, and Overland Surges from Hurricanes—Jelesnianski et al., 1992) settings. With this flexibility, however, comes the need to provide and HL-RDHM (National Weather Service/Hydrology Laboratory- accurate estimates of parameter settings, which may be difficult to Research Distributed Hydrologic Model—Koren et al., 2004). NOAA identify in order to optimize model performance for a broad range runs a variety of hurricane meteorological models, such as HWRF of scenarios (e.g., a land-falling hurricane with very dry initial (Hurricane Weather Research and Forecasting—Skamarock et al., conditions). Historical records of rainfall and runoff were used to 2008) and GFS (Global Forecast System—Environmental Modeling manually calibrate the HL-RDHM model parameters in each sub- Center, 2003). However, none of the models or systems that NOAA basin of the Tar/Pamlico and Neuse River basins. This process was runs in real-time currently couple the meteorological forcing, river completed efficiently by adjusting the routing parameters alone. inflows, tides, waves and storm surge into one complete system. Computational resources allowed for a total of 128 riverine The advantage of ASGS-STORM is that it can couple these compo- model simulations using HL-RDHM to be performed in real-time nents through physics-based models that produce a complete for each 6-hour update, corresponding to new QPFs issued by the picture of the total water level. HPC. These ensemble members were designed to encompass The paper is divided into six sections: Section 1 discusses the uncertainty in the rainfall forcing (estimated and forecast) and motivation for the project and background on other real-time model parameters. The 128 ensemble included four different systems; Section 2 describes the components of ASGS-STORM and parameter sets (designated event-based, automatic, multiple their development to produce the total water level response from basin, and a priori) with three of the parameter sets multiplied a tropical or extratropical event; Section 3 gives information about by 16 different rainfall multipliers that are uniformly distributed Hurricane Irene; Section 4 presents the results from ASGS-STORM between 0.8 and 1.2. The last parameter set (a priori) was multi- for Irene's winds, waves, and total water levels for three different plied by 5 different rainfall multipliers, along with 16 channel advisories (Advisories 23, 25 and 28) and the best track; Section 5 routing perturbations. Thus, with the first three parameter sets provides some general observations about the system's perfor- there are 48 members of the ensemble with the last parameter set mance; and Section 6 gives some concluding remarks on the providing 80 members of the ensemble. A full discussion of the coupled model system and directions for future work. results from these 128 ensemble members can be found in Section 4.2.1. Within ASGS-STORM, streamflow forecasts are produced every 6 hours in the Tar/Pamlico and Neuse River basins in North 2. Background Carolina and, subsequently, passed to the ADCIRC hydrodynamic model to serve as river boundary conditions (a discussion on these 2.1. Coupled model system—ASGS-STORM locations can be found in Section 2.2). All forecast and observed rainfall above the location of the river boundary conditions for Fig. 1 shows a schematic of ASGS-STORM, which can be utilized ADCIRC is accounted for by the hydrologic model and routed for both tropical and extratropical events (Tromble et al., 2011; downstream. Rainfall occurring directly on top of the estuaries, or Van Cooten et al., 2011). ASGS-STORM links precipitation, atmo- on the land surface below the boundary condition points, is spheric, hydrologic, wave, and hydrodynamic models into an neglected in the current configuration. The river boundary condi- integrated system to produce routine predictions of total water tions for ADCIRC were chosen to be well upstream of locations levels, along with the significant wave heights and other informa- where any tidal or storm surge effects would be experienced, as tion. The first test of ASGS-STORM occurred in the 2010 hurricane the routing component of HL-RDHM uses a kinematic wave season with results being produced for Hurricane Earl; however, equation, which cannot model backwater effects. Thus, any back- the system was not run in a real-time mode during the entire water effects caused by water moving up the estuaries must be event. In June 2011, ASGS-STORM was run in a prototype, real-time mode and was utilized during Hurricane Irene to provide total 1 water level information in the North Carolina area. A discussion of Quantitative Precipitation Forecast from NOAA Hydrometeorological Predic- tion Center (HPC). the system is given in Tromble et al. (2011) and Van Cooten et al. 2 Quantitative Precipitation Estimates from NOAA National Severe Storms (2011); a summary of that discussion is provided in the following Laboratory National Mosaic and Quantitative Precipitation Estimation System subsections for completeness. (NMQ/Q2) (Zhang et al., 2011). K.M. Dresback et al. / Continental Shelf Research 71 (2013) 78–94 81 captured by the hydrodynamic model, ADCIRC. Results from wind fields could not contain any information about future winds. HL-RDHM are produced at every 4-km grid point in the model NAM can provide gridded meteorological forcing, but only at a domain, but results from the entire ensemble are stored only at relatively coarse grid resolution (E 12 km grid resolution), which gauge locations (for validation purposes) and boundary points of may not capture the hurricane wind field in an accurate manner. the hydrodynamic model. In turn, this could cause inaccuracies in the QPFs. However, during normal meteorological conditions, NAM can provide a gridded meteorological forcing with relatively good accuracy. 2.1.2. Wind model—DAH In many cases, parametric wind models (Holland, 1980; Two different atmospheric models are utilized in ASGS-STORM Houston et al., 1999; Mattocks et al., 2006) produce wind fields based on the type of event occurring in the area of interest. For that capture storm surge in a reasonable manner (Forbes et al., 2010). tropical events, ASGS-STORM employs a parametric tropical These types of models require only a small amount of data in order to cyclone model that is based on Holland (1980) and Mattocks develop wind fields, produce information in a timely fashion as et al. (2006). This captures the asymmetric nature of the wind field required by real-time operational systems, and provide information and is referred to as the Dynamic Asymmetric Holland (DAH) on wind stresses and pressures at any location. Model, while for non-tropical events, the results from the North American Mesoscale (NAM—Janjic et al., 2001) model drive the wind fields. Meteorological forcing presents a challenge to an 2.1.3. Hydrodynamic model—ADCIRC operational modeling system that has limited computational ADCIRC (2D, depth-integrated mode) serves as the hydrody- resources due to the need for real-time, high-resolution spatial namic model in ASGS-STORM and ingests information from the data. Data-assimilated meteorological fields provide the most hydrologic model at riverine boundaries (Tromble et al., 2012), accurate representation of the meteorological forcing; however, while also taking information from the wind and wave models. these types of products (e.g., HnWind—Powell et al., 1998) typically ADCIRC utilizes the full nonlinear shallow water equations with are only produced after the event, sometimes days after the storm the hydrostatic assumption and Boussinesq approximation. The has already dissipated and the NOAA/NWS/National Hurricane model obtains the elevation changes from the Generalized Wave Center (NOAA/NWS/NHC) has stopped issuing advisories. There- Continuity (GWC) equation (Kinnmark, 1986) and the velocities fore, data-assimilated wind fields cannot be utilized in a real-time from the momentum equation. Spatially, it utilizes Galerkin finite environment because the data would not be available and the elements with equal-order interpolating functions (linear C0

Fig. 2. Example of the CERA website during Hurricane Irene. Results are shown for Advisory 28 for the maximum water elevation. The track of the hurricane appears on the ﬁgure, along with the location of the hurricane at the time of the advisory, denoted by the black hurricane symbol. 82 K.M. Dresback et al. / Continental Shelf Research 71 (2013) 78–94

Fig. 3. Study domain for HL-RDHM showing the Tar/Pamlico and Neuse River basins. The black triangles correspond to USGS stream gauge stations. The “hand-off points” are locations where the ensemble mean of the forecasted river discharges are passed to ADCIRC. (For interpretation of the references to color in this figure caption, the reader is referred to the web version of this article.) triangular element). Temporally, ADCIRC uses a semi-implicit, the coupling of the wave interaction in the model (Dietrich et al., three-time level approximation (centered at k) for the GWC 2011b), which is further discussed in Section 2.1.4. equation and a lumped, two time-level approximation (centered þ 1 at k 2 ) for the momentum equation; the nonlinear terms are evaluated explicitly. Conversely, the equations can be evaluated in 2.1.4. Wave model—SWAN fully explicit mode, thus improving the scalability of the model The wave model utilized in ASGS-STORM is the Simulating (Tanaka et al., 2010), which is utilized within this system. The WAves Nearshore (SWAN) model, which was recently expanded to solution utilizes exact quadrature rules, but the advective terms utilize unstructured meshes in both the nearshore areas and in are approximated with L2 interpolation. The full development of deep water (Zijlema, 2010). The SWAN model employs the wave the model and equations can be found in Kinnmark (1986), action balance equation given in Booij et al. (1999) and includes Luettich et al. (1992) and Kolar et al. (1994). wave propagation in space and time, wave refraction and diffrac- ADCIRC has been used extensively for many applications tion, as well as shifting of the wave frequency with the change in over its 25-year history, such as hurricane storm surge modeling water depth or current. It takes into consideration bottom friction, (Blain et al., 1994, 1998; Westerink et al., 2008; Bunya et al., 2010; whitecapping of waves and their loss of energy, nonlinear effects Dietrich et al., 2010), Lagrangian particle transport, for both due to changes from deep to shallow regions, and surf breaking. chemical and biological applications (Luettich et al., 1999; Reyns Specifics on the model and its solution scheme can be found in et al., 2006; Oliveira et al., 2006; Dietrich et al., 2012), tidal and Booij et al. (1999). wind-driven circulation (Westerink et al., 1994; Blain et al., 2002), The SWAN model has been dynamically coupled to ADCIRC in and density-driven circulation (Blain et al., 2010, 2012). Recent recent years (Dietrich et al., 2011b), thus allowing the two models to advances have included a more accurate representation of the land pass information during run time. The tight coupling of the two use/land cover by incorporating Manning's resistance formula, the models is facilitated by using the same domain and grid in their reduction of the wind drag coefficient due to changes in the land solution. During a simulation, the ADCIRC model passes the wind cover, inclusion of the air–sea interaction (Bunya et al., 2010), and forcing, water levels and currents to the SWAN model, which in K.M. Dresback et al. / Continental Shelf Research 71 (2013) 78–94 83

Fig. 4. ADCIRC model domain: (a) the finite element grid, (b) mesh resolution (m) in the North Carolina area, and (c) bathymetry and topography (m), relative to MSL, for the North Carolina area. turns uses that information to calculate the wave radiation stresses (RENCI) in North Carolina. Third, the Coastal Emergency Risks over the entire domain. The wave radiation stresses are then used Assessment (CERA) website is an interactive interface, which was to calculate their gradients, which are subsequently passed back to developed by Louisiana State University. The CERA website (http:// the ADCIRC model. The models use an explicit “leap frog” algorithm nc-cera.renci.org) provides several different layers of results (e.g., to pass information between each other; the time between stag- water elevation, water inundation, significant wave heights, wave gered steps is a user-defined parameter, set at 20 minutes for results periods, and wind speed). Fig. 2 illustrates an example CERA product herein. This coupling interval is similar to what has been used in for the water elevation. other studies with the tightly coupled ADCIRC and SWAN model (e.g., 10 min intervals in Dietrich et al. (2011b) and Dietrich et al. 2.2. Domain and area of interest (2011a)). For other loosely coupled hydrodynamic-wave models, the coupling interval has been set to 15 or 30 min, respectively (Slinn, The study domain for HL-RDHM is shown in Fig. 3. The region 2008; Bunya et al., 2010). In addition, the convergence behavior of falls completely within the area of responsibility of the NWS SWAN for the advisories analyzed herein shows that it converged Southeast River Forecast Center (SERFC). The total basin area within 5 or 6 iterations, which is well before the 20 iterations modeled by HL-RDHM is 14,426 km2. HL-RDHM produces outputs specified, indicating that the wave results from the SWAN model at every 4-km grid cell, but time series and statistical analyses are did not change more than the convergence criteria between the produced only at USGS stations, where there are observed dis- iterations during the coupling interval. Thus, a coupling interval of charges. HL-RDHM also records output at four locations that 20 min seems appropriate for this application. provide the upper river boundary conditions to ADCIRC. These four locations are shown in Fig. 3 and correspond to the four major tributaries that impact the estuaries (as determined by a study of 2.1.5. Distribution to end-users historical storms): the Tar/Pamlico and Neuse Rivers and Con- Results from ASGS-STORM are shown in real-time in various tentnea and Fishing Creeks. As stated above, these four handoff forms on three different web locations. One of these is the NSSL CI- points were chosen, based on historical records, to be above zones FLOW web page, which shows hydrographs and maximum inunda- of tidal and surge influence. tion figures in real-time. Second, results in kmz format are provided The entire ADCIRC domain is shown in Fig. 4(a), with zooms of on an OPENDAP server from the Renaissance Computing Institute the North Carolina area shown in Fig. 4(b) and (c). ADCIRC utilizes 84 K.M. Dresback et al. / Continental Shelf Research 71 (2013) 78–94

14 and concluding with Advisory 35, the last advisory of NOAA/ NWS/NHC (Avila and Cangialosi, 2011). In advisories prior to

Best Track number 14, the path of the storm did not approach the North Adv. 30 0d 0h Carolina area, thus, ASGS-STORM did not provide guidance for Adv. 29 0d 0h these advisories. Adv. 28 0d 3h Adv. 27 0d 9h 3.1. Track of storm from advisories Adv. 26 0d 15h Adv. 25 0d 21h Adv. 24 1d 3h Fig. 5 shows the forecast tracks from 16 different advisories, Adv. 23 1d 9h along with the post-analysis “best track” developed by the NOAA/ Adv. 22 1d 15h NWS/NHC of Hurricane Irene. As can be seen in Fig. 5, the advisory Adv. 21 1d 21h tracks of Hurricane Irene were close to the actual track of the Adv. 20 2d 3h hurricane, even for early advisories. Advisory tracks 15 through 19 Adv. 19 2d 9h show the hurricane just skirting the tip of the Outer Banks of Adv. 18 2d 15h North Carolina (east of the actual track of the storm), while the Adv. 17 2d 21h – Adv. 16 3d 3h other advisories (20 30) show a path similar to the best track, Adv. 15 3d 9h with some of the advisories predicting that the storm's path would pass slightly to the west or east of the actual track. In their Irene report, Avila and Cangialosi (2011) indicated that the track forecast was better than average, which is evident in the advisory tracks Fig. 5. Storm tracks for Advisories 15 to 30 from the NOAA/NWS/NHC and the best passing close to the best track. track for the North Carolina area. The legend provides the information on the color of the tracks, the related advisory and time before landfall in North Carolina. Note that Advisory 29 and 30 are included here as the hurricane was in North Carolina 3.2. Intensity of storm from advisories during these advisories, but had already made landfall. (For interpretation of the fi references to color in this gure caption, the reader is referred to the web version of The intensity of Hurricane Irene peaked at a Category 3 and this article.) decreased to a Category 1 storm as it made landfall in the United States. However, the forecast intensity of the storm at landfall varied from a Category 3 to Category 1, with the forecast of the correct a large domain that covers the Western Atlantic Ocean, Caribbean intensity occurring as the storm made landfall. Fig. 6 shows three of 1 Sea and Gulf of Mexico waters that lay to the west of the 60 W the advisories from Irene. Each of these advisories predicted a longitude. The large spatial extent is utilized in order to minimize different storm intensity as Irene made landfall. Fig. 6(a) shows the errors associated with specifying open boundary conditions. It also track and intensity associated with Advisory 23, while Fig. 6(b) shows allows for tracking of the hurricane formation through the domain Advisory 25 and Fig. 6(c) shows Advisory 28. The intensity of Irene and the resulting basin-to-basin interaction. Fig. 4(a) shows the was over-predicted in the early advisories. As will be shown in unstructured, triangular grid utilized in this study that contains Sections 4 and 5, these intensity errors (and track errors to a lesser 295,328 computational nodes. Resolution within the domain, extent because they were more consistent for Irene) manifest shown in Fig. 4(b), varies from 50 km in the deeper portion of themselves as surge prediction errors. the domain to approximately 40 m within the riverine areas of North Carolina. Fig. 4(c) shows the bathymetry and topography of the coastal areas of North Carolina, which is referenced to mean 4. Comparison to real-time data sea level (MSL). The extent of the domain inland is up to the 8-m land contour for the Tar/Pamlico and Neuse Rivers, along with the As mentioned above, Fig. 6 shows three different advisory Contentnea and Fishing Creeks. A historical review of the inland forecasts, with the strength of Hurricane Irene shown by the dots extent of flooding led to the inclusion of the riverine areas up to along the path of the storm. These three advisories were chosen the 8-m land contour. Outside of the riverine areas, the domain due to their different strengths and tracks. The strength of the extends up to the 15-m contour, which allows for inland flooding storm forecasted by NOAA/NWS/NHC for these different advisories within these areas. changed significantly as the storm approached the coast of North Carolina. Advisory 23 (shown in Fig. 6(a)) predicted a Category 3 hurricane upon landfall in North Carolina, while Advisory 25 3. Hurricane Irene (shown in Fig. 6(b)) showed a decrease to Category 2, and Advisory 28 (shown in Fig. 6(c)) predicted a further decrease to a Category 1, Hurricane Irene became the ninth named storm of the 2011 which Irene was at landfall. The path of the storm forecasted by season and the first to make landfall in the United States since NOAA/NWS/NHC for the three different advisories varied to a Hurricane Ike in 2008. The storm tracked through the Caribbean certain extent as the storm approached the coast of North Islands of Puerto Rico, Hispaniola, Dominican Republic, Haiti, Carolina. The predicted track of Advisory 23 took the storm into Bahamas and the Turks and Caicos before heading to the East the inland areas of North Carolina, while Advisory 25 showed the Coast of the United States. Irene peaked as a Category 3 in the storm track over the estuarine area of the North Carolina coast, Bahamas, and it made its initial landfall in the United States along and Advisory 28 predicted the storm track over the outer banks of the Outer Banks of North Carolina near Cape Lookout as a Category the North Carolina coast. The best track of Irene, developed in the 1 hurricane on August 27 at 1200 UTC, bringing with it 39 m/s post-analysis of the storm, showed it falling between the tracks of wind speeds. It then made a second United States landfall on Advisory 25 and 28. August 28 at 0935 UTC in New Jersey and was downgraded to a The results from ASGS-STORM shown in this manuscript will tropical storm near New York City. It caused damage in many areas compare the wind, wave and total water levels during the three of the United States and the Caribbean, with estimates of mone- advisories shown in Fig. 6, along with the wind, wave and total tary damage ranging from 10 to 15 billion dollars. ASGS-STORM water levels for the best track. These results will then be compared provided guidance during Hurricane Irene, starting with Advisory to observations taken from National Ocean Service (NOS) water K.M. Dresback et al. / Continental Shelf Research 71 (2013) 78–94 85

Fig. 6. Forecast information of the track and storm intensity (shown by the color of the dots along the track) for three different advisories: a) Advisory 23, b) Advisory 25, and c) Advisory 28. The legend shows the storm strength and the track information, along with the predicted maximum water surface elevation. The cone of uncertainty is also shown in each figure by the gray areas on either side of the official forecast track. Storm position is shown by the black hurricane symbol. (For interpretation of the references to color in this figure caption, the reader is referred to the web version of this article.) level stations, National Data Buoy Center (NDBC) buoys, and USGS 4.1. Wind and waves gauge locations. Table 1 gives the longitude and latitude location of the stations analyzed herein. Model results are presented in the Fig. 7 shows the location of the NDBC buoy stations that are following subsections with an initial discussion of the results and a utilized to evaluate the wind and wave performance of ASGS- more thorough discussion occurring in Section 5. STORM, along with the forecast tracks for three different 86 K.M. Dresback et al. / Continental Shelf Research 71 (2013) 78–94 advisories (Advisories 23, 25 and 28) and the best track. The speeds to a rapid increase in winds) over the area for many of stations used to compare the wind performance are shown in the stations analyzed. In both cases, the results tend to capture the the left panel of Fig. 7, while the wave stations are shown in the trend of the wind speeds, but there is an over-estimation of right panel. the actual peaks of the wind speeds, with Advisory 25 showing a greater over-prediction due to the over-estimation in the strength 4.1.1. NOAA National Data Buoy Center—Wind of the storm. In all the advisories and the best track results, the Fig. 8 presents the wind speed results from ASGS-STORM for wind speeds are above the actual observed wind speeds after the stations along coastal North Carolina. Wind direction results were storm passes. This leads to the water levels remaining high after also evaluated, and they are presented in Fig. 9. The wind speed the storm passes through the area. This over-estimation of the plots for the three different advisories are shown in Fig. 8, with the wind speeds is being investigated as part of improving future line types matching the track lines as shown in Fig. 7, along with operations. the best track result and the observations from the NDBC buoy The wind direction results indicate that for the different stations. Wind results show that the earlier advisories had wind advisories and the best track many of the changes in direction speeds that were larger than actual speeds recorded during the are captured well. At some of stations (DUKN7 and CLKN7), the storm due to the over-prediction of the strength of the hurricane. direction changes faster than those of the observed due to the The best track, as well as Advisory 25, capture the passing of the easterly path of the storm and the location of the eye of eye of the hurricane (denoted by the sudden decrease of wind the storm.

Table 1 4.1.2. NOAA National Buoy Data Center—Waves Latitude and longitude information for the NDBC, NOS, and USGS stations. Fig. 10 presents the results for significant wave heights from Station Latitude (1N) Longitude (1E) ASGS-STORM for stations along coastal North Carolina. Similar to the wind results, the wave height plots show results from the fi Signi cant wave heights stations (NDBC) three different advisories shown in Fig. 6, with the line types 41001 34.561 72.631 41013 33.436 77.743 matching the track lines shown in Fig. 7, along with the best track 41110 34.141 77.709 and the observations from the NDBC buoy stations. Results are 44014 36.611 74.842 only shown here for the significant wave heights; however, the 41036 34.206 76.949 wave periods and direction present similar patterns. The wave 44100 36.258 75.591 heights were over-predicted in most of the advisories due to the Wind speed and direction stations (NDBC) over-estimation of the strength of the hurricane. This over- DUKN7 36.183 75.747 estimation leads to increases in the wind speeds, which translates ORIN7 35.795 75.548 HCGN7 35.208 75.703 to the increase in the heights of the waves. The best track and BFTN7 34.720 76.670 Advisory 28 results show only a slight over-prediction of the wave CLKN7 34.622 76.525 heights. Improvements in the prediction of the wind speeds JMPN7 34.210 77.795 should translate to better prediction in the significant wave Water level stations (NOS) heights along the coastal areas. 8651370 36.183 75.747 8652587 35.795 75.548 8656483 34.720 76.670 4.2. Water levels 8658163 34.213 77.786 8638863 36.967 76.113 4.2.1. USGS gauge stations—HL-RDHM 8510560 41.048 71.960 Fig. 11 shows four snapshots of ensemble predictions from HL- Water level stations (USGS) RDHM produced during the hurricane event for the Tar/Pamlico 02092162 35.109 77.033 02084472 35.543 77.062 River at Tarboro, North Carolina. This site is in close proximity to one of the ADCIRC boundary condition locations; as such, it is

Fig. 7. Location of the NDBC buoy stations for comparison of the wind (left panel) and wave (right panel) results, along with the tracks for three different advisories and the best track: long dashed line – Advisory 23, dotted line – Advisory 25, dot-dashed line – Advisory 28, and black line – Best track. Actual longitude and latitude locations can be found in Table 1. K.M. Dresback et al. / Continental Shelf Research 71 (2013) 78–94 87

Station DUKN7 Station ORIN7 40 50

40 30 s s

m m 30 20 20

WindSpeed 10 WindSpeed 10

0 0 8 25 8 26 8 27 8 28 8 29 8 30 8 25 8 26 8 27 8 28 8 29 8 30 Date in 2011 Date in 2011

Station HCGN7 Station BFTN7 50 50

40 40 s s

m 30 m 30

20 20 WindSpeed 10 WindSpeed 10

0 0 8 25 8 26 8 27 8 28 8 29 8 30 8 25 8 26 8 27 8 28 8 29 8 30 Date in 2011 Date in 2011

Station JMPN7 Station CLKN7 40 50

40 30 s s

m m 30 20 20

WindSpeed 10 WindSpeed 10

0 0 8 25 8 26 8 27 8 28 8 29 8 30 8 25 8 26 8 27 8 28 8 29 8 30 Date in 2011 Date in 2011

Fig. 8. Results for the wind speed from ASGS-STORM for Hurricane Irene. The panels show the wind speed results for NDBC buoy stations for the six stations in North Carolina shown in Fig. 7 (left panel) for three advisories, along with the best track: long dashed line – Advisory 23, dotted line – Advisory 25, dot-dashed line – Advisory 28, black line – best track, and gray dots - NBDC observations.

critical to have accurate flows to pass to the ADCIRC domain. The period. At this point in the simulation, the rainfall hyetograph mean of the forecast streamflows is derived from the 128-member indicates approximately half of the rainfall forcing comes from ensemble (discussed in Section 2.1.1), and the ensemble mean estimated radar plus gauge rainfall from QPE, while the other half streamflow is provided to ADCIRC at the four hand-off locations is from QPF. The simulations at later time in Fig. 11(c) and (d) are shown in Fig. 3. At gauged locations, the best member, which all forced with QPE rainfall estimates. Once the response due to performed best compared to observed streamflow, as measured by the event is observed at the gauging station, the assimilation the Nash–Sutcliffe Coefficient of Efficiency (Nash and Sutcliffe, technique and selection of the best member is able to provide 1970) and bias in the last 24 h, is highlighted in the red dashed more accurate forecasts (cf. red dashed and blue dotted lines in line. In short, this best member assimilates observed streamflow Fig. 11c). In Fig. 11d, the accuracy of the best member finally into its selection. It should be cautioned that this member only has exceeds that of the ensemble mean. relevance when the gauge observes event-based streamflow (e.g., A post-event analysis was conducted, which utilized the full the rising limb of hydrograph); during base flow conditions, the observed hydrograph for the Tar/Pamlico River at Tarboro to quantify ensemble approach has difficulty identifying a best member. the relative skill of the best member vs. the ensemble mean. Fig. 12 Fig. 11(a) shows that the forecast peak flow from the ensemble illustrates the results of this skill assessment. It shows the peak water mean is in close agreement with the observed streamflow. The level error (observed-predicted) for the ensemble mean and the simulation at this stage in the forecast cycle was driven purely by member with streamflow assimilation are plotted as a function of forecast rainfall, providing a lead time of 3 days prior to the lead time. This summary plot demonstrates that the ensemble mean maximum observed streamflow. It can also be seen that the best performed quite well, with errors of approximately 20% when the member (in the red dashed line) overestimates the observed peak precipitation forcing was from forecasts alone. As the forcing became flow because there is very little time-varying data to assimilate in dominated by radar plus gauge rainfall, errors were generally less streamflow observations within the past 24 h. Fig. 11(b) shows than 10%. Overall, the performance of the HL-RDHM ensemble can be similar performance from both the ensemble mean and the best judged as accurate for this high-impact event, thus giving confidence member due to base flow conditions dominating the 24-hour toourprocedureofusingtheensemblemeanatADCIRChandoff 88 K.M. Dresback et al. / Continental Shelf Research 71 (2013) 78–94

Station DUKN7 Station ORIN7 250 350 300 200 true true 250 150 200

100 150 100 Wind Direction Wind Direction 50 50

8 26 8 27 8 28 8 29 8 30 8 26 8 27 8 28 8 29 8 30 Date in 2011 Date in 2011

Station HCGN7 Station BFTN7 350 300 300 250 true true 250 200 200 150 150 100 100 Wind Direction Wind Direction 50 50

8 26 8 27 8 28 8 29 8 30 8 26 8 27 8 28 8 29 8 30 Date in 2011 Date in 2011

Station JMPN7 Station CLKN7 350 350 300 300 true true 250 250 200 200 150 150 100 100 Wind Direction Wind Direction 50 50

8 26 8 27 8 28 8 29 8 30 8 26 8 27 8 28 8 29 8 30 Date in 2011 Date in 2011

Fig. 9. Results for the wind direction from ASGS-STORM for Hurricane Irene. The panels show the wind direction results for NDBC buoy stations for the six stations in North Carolina shown in Fig. 7 (left panel) for three advisories, along with the best track: long dashed line – Advisory 23, dotted line – Advisory 25, dot-dashed line – Advisory 28, black line – best track, and gray dots – NBDC observations.

points where gauge information is lacking. Future work will analyze of the water levels peaks are captured, but water levels are higher performance at additional stations and provide comparative results than that recorded. For example, results from Advisory 23 for Duck to operationally produced forecasts from the SERFC. Pier show the over-prediction of the storm surge due to the estimated category 2 strength of the storm; however, it also displays that the timing of the peaks coincides with the actual 4.2.2. NOAA NOS Stations—ADCIRC peaks. In Advisory 28 and the best track results, ASGS-STORM Fig. 13 (left panel) shows the location of the NOS water level captures the surge and peaks of the storm surge; however, the best stations that are utilized to evaluate the performance of ASGS- track does better at obtaining the duration of flooding and STORM, along with the forecast tracks for three different advi- reproducing the receding water levels after the storm versus sories (Advisories 23, 25 and 28) and the best track. As can be seen Advisory 28. An example of this can be seen in the Oregon Inlet in Fig. 13, the best track of Irene (solid black line in the figure) buoy (cf., solid black line and gray dots in Fig. 14). Future changed directions several times during its forward progression improvements in the prediction of the wind and wave information toward landfall. during the storm may lead to further improvements in capturing Fig. 14 shows results from six different NOS stations along the the flooding that occurs in the coastal areas. In particular, North Carolina, Virginia and New York coasts. The figures include improvements in these two areas may lead to a more accurate the total water level predicted from the three different advisories representation of the secondary peak that occurs in the Oregon presented previously, with the line types matching the track lines Inlet area that is currently missed by ASGS-STORM. shown in Fig. 13, along with the observations from the NOS stations. Total water level results from four stations along the North Carolina coast are shown in the first four panels, with the last two panels providing results from the real-time system in 4.2.3. USGS gauge stations—ADCIRC Virginia and New York, which are outside the focus area of the The panel on the right side of Fig. 13 shows the location of project. Results are shown in relation to MSL (the ADCIRC datum). the two USGS gauge stations that are utilized to evaluate the In both Advisories 23 and 25, ASGS-STORM over-predicts performance of ASGS-STORM, along with the forecast tracks for the storm surge due to the over-estimation of the strength of three different advisories (Advisories 23, 25 and 28) and the the hurricane in the advisories. In most of the cases, the timing best track. K.M. Dresback et al. / Continental Shelf Research 71 (2013) 78–94 89

Station 41001 Station 41013 14 10 12 8

10 m m 8 6

6 4 4 Wave Heights Wave Heights 2 2 0 0 8 25 8 26 8 27 8 28 8 29 8 25 8 26 8 27 8 28 8 29 Date in 2011 Date in 2011

Station 41110 Station 44014 7 14 6 12 m m 5 10 4 8 3 6 2 4 Wave Heights Wave Heights 1 2 0 0 8 25 8 26 8 27 8 28 8 29 8 25 8 26 8 27 8 28 8 29 Date in 2011 Date in 2011

Station 41036 Station 44100 12 10

10 8 m 8 m 6 6 4 4 Wave Heights Wave Heights 2 2

0 0 8 25 8 26 8 27 8 28 8 29 8 25 8 26 8 27 8 28 8 29 Date in 2011 Date in 2011

Fig. 10. Results for the signiﬁcant wave heights from ASGS-STORM for Hurricane Irene. The different panels show results for the NDBC buoy stations for the stations shown in Fig. 7 for three advisories and the best track: long dashed line – Advisory 23, dotted line – Advisory 25, dot-dashed line – Advisory 28, black line – best track, and gray dots – NBDC observations.

Fig. 15 provides the total water results for the two USGS gauge accurately capture the receding water levels after the storm. stations shown in Fig. 13 (right panel). The figures include the total Currently, model resolution in riverine areas may be limiting its water level predicted from the three different advisories presented ability to capture the timing and magnitude of the peaks previously, with the line types matching the track lines shown in accurately. This is an area that is being investigated to improve Fig. 13, along with the observations from the USGS gauge stations. future operations. The New Bern station resides on the Neuse River, while the Washington station is on the Tar/Pamlico River, thus providing an initial test of the ASGS-STORM as a whole. However, Irene did 4.2.4. High water marks—total water levels not have a significant freshwater discharge component associated Fig. 16 shows the difference between high-water marks with it; therefore, it is not currently possible for us to make (HWMs), collected after Hurricane Irene by the USGS and other conclusions about the role of freshwater discharges on total water partners in North Carolina, and the total water level results from levels in estuarine zones. On-going work will examine historical ASGS-STORM using the best track for Irene. There are 74 HWM storms, such as Floyd, where the freshwater discharge played a locations utilized in this comparison. Results show that, in most significant role. areas of coastal North Carolina, the differences between the two Advisories 23 and 25 both over-predicted the water levels at range from 70.5 to 71.0 m, with only a few areas showing a these two USGS stations, and for Advisory 23, the predicted peak greater than 71.5 m difference. In particular, 81% of the HWMs is delayed slightly (cf. long-dashed line in Fig. 15). The over- are within 70.5 m difference, while 16% and 3% of the HWMs are prediction of the water levels is related to the over-estimation of within 71.0 m and 71.5 m, respectively. Most of the large the strength of the storm, while the delayed peak for Advisory differences between the HWMs and ADCIRC results occur in areas 23 is related to the more westerly track of the storm. In contrast, where more grid resolution may be needed. Previous studies in for Advisory 28, the predicted track of the storm is easterly, Louisiana (Bunya et al., 2010) have shown that increased grid which leads to the peak occurring slightly earlier than the actual resolution can provide improvement in the accuracy of the model. timing of the peak at both USGS gauge stations. The best track Additionally, the best track information does not incorporate the results show that in both stations the peak is slightly higher data-assimilated wind products for the hurricane. Thus, improved than the actual recorded peak, but the best track results are skill may be seen when evaluating the data-assimilated wind 90 K.M. Dresback et al. / Continental Shelf Research 71 (2013) 78–94

Fig. 11. Hydrographs of predicted flow for the Tar/Pamlico River at Tarboro (USGS Station 02083500) throughout the landfall of Hurricane Irene. The four panels (a–d) capture different snapshots in time during the forecast time period, progressing from August 26 to August 30, 2011. The gray curves represent the maximum and minimum from the 128-member ensemble, while the gray shading depicts the range. The blue dotted curve corresponds to the mean of the 128-member ensemble. The red dashed curve corresponds to the member that realized the best performance, compared to observed streamflow, over the last 24 hours. The black curve shows the observed streamflow for the entire time analyzed to determine how well the model is capturing the changes. The straight vertical line divides the observed versus forecast periods, with the forecast period on the right side and the observed on the left side. Lastly, the forecast and observed rainfall hyetographs are shown on the secondary x- and y-axes. (For interpretation of the references to color in this figure caption, the reader is referred to the web version of this article.)

Fig. 17 provides scatterplots of the measured to predicted total water levels for the NOS and USGS stations in the coastal North Carolina area, along with the HWMs for the three different advisories and the best track for Hurricane Irene. As can be seen in this figure, the results show an improvement with each advisory as the track and intensity of the storm start to approach the actual storm track and intensity. For each of the advisories and the best track results, a root mean square error (RMSE) was calculated using the following formula: sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ðy y Þ2 RMSE ¼ i t ð1Þ n

where yi are the modeled values, yt are the observed values and n represents the number of predictions. The RMSE improved (decreased towards zero) through all three of the advisories, with Fig. 12. Peak water level error for the ensemble mean and best member with the best track providing the lowest RMSE. It can be seen in Fig. 17 streamﬂow assimilation, as compared to post-event observations, with streamﬂow that the spread of the points decreases with each advisory, and the assimilation, shown as a function of time (lead time) that the simulation was early advisories show over-prediction of peak water levels, with conducted for Hurricane Irene. The observed rainfall hyetograph is shown at the many of the points residing above the one-to-one line. The best top to indicate the times over which the rainfall occurred. track results show the least amount of scatter and provide results that are slightly under-predicted compared to the observed peak water levels. It should be noted that there are two possible sources of errors products and best track together. A full study of the wind model, as in the water levels: errors within the hydrologic/hydrodynamic compared to the data-assimilated wind products, is the subject of models themselves or errors in the weather forecast that serve as an upcoming paper. forcing to the models. The latter is not within our control and K.M. Dresback et al. / Continental Shelf Research 71 (2013) 78–94 91

Fig. 13. Location of the NOS water level stations (left panel) and USGS gauge stations (right panel) for comparison of the total water level results, along with the tracks for three different advisories and the best track: long dashed line – Advisory 23, dotted line – Advisory 25, dot-dashed line – Advisory 28, and black line – best track. Note that for the NOS stations, Station 8510560 is not shown on the ﬁgure in the left panel, as it is located further north in latitude than is available in the ﬁgure. Actual longitude and latitude locations can be found in Table 1.

Duck Pier, NC Station 8651370 Oregon Inlet, NC Station 8652587 m m 2 3

2 1 1 0 0 1 1 Height above datum in MSL Height above datum in MSL 2 2 8 26 8 27 8 28 8 29 8 30 8 26 8 27 8 28 8 29 8 30 Date in 2011 Date in 2011

Beaufort, NC Station 8656483 Wrightsville Beach, NC Station 8658163 m m 2 2

1 1

0 0

1 1 Height above datum in MSL Height above datum in MSL 2 2 8 26 8 27 8 28 8 29 8 30 8 26 8 27 8 28 8 29 8 30 Date in 2011 Date in 2011

Chesapeake, VA Station 8638863 Montauk, NY Station 8510560 m m 2 2

1 1

0 0

1 1 Height above datum in MSL Height above datum in MSL 2 2 8 26 8 27 8 28 8 29 8 30 8 26 8 27 8 28 8 29 8 30 Date in 2011 Date in 2011

Fig. 14. Results for the total water levels from ASGS-STORM for Hurricane Irene. The different panels show result from the modeling system for the NOS water level stations, which are shown in Fig. 13 for three advisories and the best track: long dashed line – Advisory 23, dotted line – Advisory 25, dot-dashed line – Advisory 28, black line – best track, and gray dots - NOS observations. 92 K.M. Dresback et al. / Continental Shelf Research 71 (2013) 78–94

Fig. 15. Results for the total water levels from ASGS-STORM for Hurricane Irene. The different panels show results from the modeling system at USGS gauge station locations, which are shown in Fig. 13 for three advisories and the best track: long dashed line – Advisory 23, dotted line – Advisory 25, dot-dashed line – Advisory 28, black line – best track, and gray dots – USGS observations.

Fig. 16. Plot of the differences between High Water Mark (HWMs) collected after Irene by the USGS and other partners in North Carolina and the total water level results for the best track of Irene from ASGS-STORM. Scale is shown in the ﬁgure.

4 4

m 3 m 3

2 2

1 1 Modeled Peak Value Modeled Peak Value y 1.276 x y 1.135 x RMSE 0.847 RMSE 0.477 0 0 0 1 2 3 4 0 1 2 3 4 Measured Peak Value m Measured Peak Value m

4 4

m 3 m 3

2 2

1 1

Modeled Peak Value y 0.831 x Modeled Peak Value y 0.897 x RMSE 0.461 RMSE 0.406 0 0 0 1 2 3 4 0 1 2 3 4 Measured Peak Value m Measured Peak Value m

Fig. 17. Scatterplots of the measured and predicted total water levels for NOS stations along the Outer Banks of North Carolina (shown in the left panel of Fig. 13), the two USGS stations (shown in the right panel of Fig. 13), and the HWMs shown in Fig. 16 for three advisories and the best track of Hurricane Irene: (a) Advisory 23, (b) Advisory 25, (c) Advisory 28, and (d) Best track. K.M. Dresback et al. / Continental Shelf Research 71 (2013) 78–94 93 tends to produce larger errors with the earlier advisories and As mentioned in Section 4.2.4, two possible sources of errors lower errors with the later advisories. The errors within the exist in the water levels: errors within the models themselves, models themselves are something that can be controlled and is which are under our control and consistent over time, or errors usually constant over time. Thus, if the majority of the error is in the in the weather forecast, which are not under our control and can forecast, then the guidance should approach the measured values produce signiﬁcant errors. Results from ASGS-STORM have over the course of each advisory as the weather forecast improves. shown that as the forecast of the advisories improves, so does Results herein have shown that as the forecast of the advisories the accuracy of the models used in the study; therefore, accurate improves, so does the accuracy of the models used in the study. input from the weather forecast is a necessary, but not sufﬁ- cient, condition to ensure the accuracy of the guidance provided by the system.

5. General observations 6. Some concluding remarks and future work Several observations can be made from the results presented in Section 4: Hurricane Irene provided a rigorous initial test of ASGS- STORM for real-time operations. Total water level products were Results from Advisory 23, the long-dashed lines in Figs. 8, 9, 10, produced every 6 hours during the storm, with the results 14, 15, show over-prediction of the wind speeds, significant capturing the water levels with good skill. Wind results show wave heights, and surge at all the stations due to the over- slight over-prediction of the wind speeds ahead of the hurricane estimation of the strength of the storm upon landfall and after the hurricane passes through the area. Wind direction (cf., predicted 54 m/s vs. observed 38 m/s Avila and Cangialosi, results indicate that ASGS-STORM captures the changes accu- 2011), along with a more westerly track of the storm than the rately. Lastly, the significant wave heights are consistently over- actual track of Irene. The more westerly track of the storm, predicted from ASGS-STORM; however, this over-prediction is combined with the error in the intensity, alters the timing and most likely due to the wind fields. Additionally, as mentioned response in the coastal areas. earlier Irene provided a good real-time test of the ASGS-STORM Results from Advisory 25, the dotted line in Figs. 8, 9, 10, 14, 15, system; however, Irene did not produce significant freshwater show over-prediction of the wind speeds, significant wave discharges. Thus, definitive conclusions about the role of fresh- heights, and surge at some of the stations also due to the water discharges on total water levels in the estuarine zones over-estimation of the strength of the storm (cf., predicted cannot be determined from Irene. Future work will examine 46 m/s vs. observed 38 m/s Avila and Cangialosi, 2011). How- storms,suchasFloyd,forwhichthefreshwaterdischarge ever, these results did capture the duration of the peak storm played a meaningful role with the ASGS-STORM system to surge, because the forecast track was closer to the actual track determine the role of freshwater discharges on total water of Irene. levels. Results from Advisory 28, the dot-dashed line in Figs. 8, 9, 10, Since Irene, ASGS-STORM has continued to run 24 hours-a-day, 14, 15, show that ASGS-STORM accurately captures the peak 7 days-a-week and produces updates to the websites as results surge at most of the stations and only slightly over-predicts the become available. Improvements to the modeling system are on- wind speeds and significant wave heights, primarily due to the going, with changes being related to every aspect of the system. differences in the strength of the storm (cf., predicted 40 m/s For example, there will be upgrades to the hydrologic and hydro- vs. observed 38 m/s Avila and Cangialosi, 2011). However, the dynamic models. We are improving the data exchange between more easterly track of the storm decreased the predicted the two models. As a specificADCIRCexample,modifications duration of the peak storm surge. are currently addressing the precipitation input over the Results from the best track of Hurricane Irene, the black line in ADCIRC portion of the coastal plains. Further analysis of the Figs. 8, 9, 10, 14, 15, show that, given the actual track of the coupled modeling system for Hurricane Irene will entail ana- hurricane, ASGS-STORM shows good skill in capturing the peak lyzing the performance of HL-RDHM at additional stations, as surge and wind speeds measured along the coast, with only slight well as comparing these results to the operationally produced over-prediction in the significant wave heights. forecasts from the SERFC of NOAA NWS. Additional analysis of theADCIRCandSWANresultswilllookatthetotalwaterlevel Overall, ASGS-STORM did well at faithfully reproducing the and wave height outputs from simulations with data- measured surge along the coast and in the major estuarine/river assimilated wind products. systems in North Carolina, provided the NOAA/NWS/NHC advisories accurately captured the actual hurricane strength and path. It should also be noted that, while ASGS-STORM only provides Acknowledgments high-resolution and hydrologic modeling in the North Carolina study area, it also showed skill at other Mid-Atlantic stations. For example, Funding for this project was provided, in part, by NOAA (IOOS the bottom two plots in Fig. 14 showtheresultsfromthesamethree program, through contract #NA07NOS4730212), US Department of forecast advisories for NOS stations near Montauk, New York and Homeland Security Science and Technology Center of Excellence- Chesapeake, Virginia. As can be seen, the later advisories, along with Coastal Hazards Center (through award #2008-ST-061-ND0001), the best track, provide an accurate representation of the surge in NOAA/Office of Oceanic and Atmospheric Research under NOAA- these areas. ASGS-STORM also doesadecentjobofcapturingthe University of Oklahoma Cooperative Agreement #NA11OAR4320072, wind speeds and wind direction. However, the wind model does United States Department of Commerce, and the University of seem to have trouble with dissipating winds after the storm passes Oklahoma. The Renaissance Computing Institute at the University throughtheNorthCarolinaarea(notethetailofthewindspeeds of North Carolina at Chapel Hill provided computing resources and shown in Fig. 8). The over-prediction of the waves may be due to operational support for the ASGS. We also thank our CI-FLOW errors in the wind fields, or they may indicate model errors. Possible collaboratorsatNOAA'sNationalSevereStormsLaboratory(NSSL) sources of the latter are that the parameters used in ADCIRC may for helping in bringing this project to fruition. The views and need further calibration. conclusions contained in this document are those of the authors 94 K.M. Dresback et al. / Continental Shelf Research 71 (2013) 78–94 andshouldnotbeinterpretedasnecessarilytheofficial policies, Jelesnianski, C., Chen, J., Shaffer, W.A., April 1992. SLOSH: Sea, Lake, and Overland either expressed or implied, of the US Department of Homeland Surges from Hurricanes. Technical Report, National Weather Service. Kinnmark, I.P.E., 1986. The shallow water wave equations: formulations, analysis Security or other funding agencies. and application. Lecture Notes in Engineering, vol. 15. Springer-Verlag. Kolar, R.L., Gray, W.G., Westerink, J.J., Luettich, R.A., 1994. Shallow water modeling in spherical coordinates: equation formulation, numerical implementation and References application. J. Hydraul. Res. 32 (1), 3–24. Koren, V., Reed, S., Smith, M., Zhang, Z., Seo, D., 2004. Hydrology Laboratory Research Modeling System (HL-RMS) of the US National Weather Service. Avila, L.A., Cangialosi, J., 2011. Tropical Cyclone Report: Hurricane Irene (AL092011) J. Hydrol. 291, 297–318. 21-28 August 2011. Technical Report, National Hurricane Center, National Link, L.E., Jaeger, J.J., Stevenson, J., Stroupe, W.A., Mosher, R.L., Martin, D., Garster, Weather Service. J.K., Zilkoshi, D.B., Ebersole, B.A., Westerink, J.J., Resio, D.T., Dean, R.G., Sharp, M. Blain, C.A., Cambazoglu, M.K., Kourafalou, V.H., 2010. Modeling the Dardanelles K., Steedman, R.S., Duncan, J.M., Moentenich, B.L., Howard, B., Harris, J., strait outflow plume using a coupled model system. In: OCEANS 2009, MTS/ Fitzgerald, S., Moser, D., Canning, P., Foster, J., Muller, B., 2006. Performance IEEE, pp. 1–8. Valuation of the New Orleans and Southeast Louisiana Hurricane Protection Blain, C.A., Cambazoglu, M.K., Linzell, R.S., Dresback, K.M., Kolar, R.L., 2012. The System. U.S. Army Corps of Engineers. predictability of near-coastal currents using unstructured grid models. Ocean Luettich, R.A., Hench, J.L., Fulcher, C.W., Werner, F.E., Blanton, B.O., Churchill, J.H., – Dyn. 62 (3), 411 437. 1999. Barotropic tidal and wind-driven larval transport in the vicinity of a Blain, C.A., Preller, R.H., Rivera, A.P., 2002. Tidal prediction using the advanced Barrier Island Inlet. J. Fish. Oceanogr. 8, 190–209. circulation model (ADCIRC) and a relocatable PC-based system. Oceanography Luettich, R.A., Westerink, J.J., Scheffner, N.W., 1992. ADCIRC: An Advanced Three- – 15 (1), 77 87. Dimensional Circulation Model for Shelves, Coasts and Estuaries; Report 1: fl Blain, C.A., Westerink, J.J., Luettich, R.A., 1994. The in uence of domain size on the Theory and Methodology of ADCIRC-2DDI and ADCIRC-3DL. Technical Report, response characteristics of a hurricane storm surge model. J. Geophys. Res. 99 U.S. Army Corp of Engineers, U.S. Department of the Army. – (18), 18467 18479. Mattocks, C.A., Forbes, C., Ran, L., 2006. Design and Implementation of a Real-Time Blain, C.A., Westerink, J.J., Luettich, R.A., 1998. Grid convergence studies for the Storm Surge and Flood Forecasting Capability for the State of North Carolina. – prediction of hurricane storm surge. Int J Numer Methods Fluids 26, 369 401. Technical Report, University of North Carolina. Booij, N., Ris, R.C., Holthuijsen, L.H., 1999. A third-generation wave model for coastal Nash, J.E., Sutcliffe, J.V., 1970. River flow forecasting through conceptual models— regions 1. Model description and validation. J. Geophys. Res. 104 (C4), part IA discussion of principles. J. Hydrol. 10 (3), 282–290. – 7649 7666. Oliveira, A., Fortunato, A.B., Pinto, L., 2006. Modeling the hydrodynamics and the Bunya, S., Dietrich, J.C., Westerink, J.J., Ebersole, B.A., Smith, J.M., Atkinson, J.H., fate of passive and active tracers in the Guadiana Estuary. Estuar. Coast. Shelf Jensen, R., Resio, D.T., Luettich, R.A., Dawson, C.N., et al., 2010. A high-resolution Sci. 70, 76–84. fl coupled riverine ow, tide, wind, wind wave, and Storm Surge Model for Powell, M.D., Houston, S.H., Amat, L.R., Morisseau-Leroy, N., 1998. The HRD real- Southern Louisiana and Mississippi. Part I: model development and validation. time hurricane wind analysis system. J. Wind Eng. Ind. Aerodyn. 77 and 78, – Month. Weather Rev. 138 (2), 345 377. 53–64. Burnash, R., 1995. The NWS River Forecast System-catchment modeling. In: Singh, Ramakrishnan, L., Blanton, B.O., Lander, H., Luettich, R.A., Reed, D.A., Thorpe, S.R., V. (Ed.), Computer Models of Watershed Hydrology. Water Resources Publica- 2006. Real-time storm surge ensember modeling in a grid environment. In: – tions, pp. 311 366. Second International Workshop on Grid Computing Environments (GCE). Chu, P., Blain, C.A., Linzell, R.S., 2009. Development of a relocatable coastal forecast Reed, S., Maidment, D., 1999. Coordinate transformations for using NEXRAD data in — system for the US navy. In: OCEANS 2009, MTS/IEEE Biloxi Marine Technology GIS-based hydrologic modeling. J. Hydrol. Eng. 4, 174–182. – for Our Future: Global and Local Challenges, pp. 1 8. Reyns, N.B., Eggleston, D.B., Luettich, R.A., 2006. Secondary dispersal of early Dietrich, J.C., Bunya, S., Westerink, J.J., Ebersole, B.A., Smith, J.M., Atkinson, J.H., juvenile blue crabs within a wind-driven estuary. Limnol. Oceanogr. 51 (5), Jensen, R., Resio, D.T., Luettich, R.A., Dawson, C.N., et al., 2010. A high-resolution 1982–1995. fl coupled riverine ow, tide, wind, wind wave, and Storm Surge Model for Skamarock, W., Klemp, J., Dudhia, J., Gill, D., Barker, D., Duda, M., Huang, X., Wang, Southern Louisiana and Mississippi. Part II: synoptic description and analysis of W., Powers, J., 2008. A Description of the Advanced Research WRF Version 3. Hurricanes Katrina and Rita. Month. Weather Rev. 138 (2), 378–404. Technical Report NCAR/TN-475þSTR, National Center for Atmospheric Dietrich, J.C., Trahan, C.J., Howard, M.T., Fleming, J.G., Weaver, R.J., Tanaka, S., Yu, L., Research. Luettich, R.A., Dawson, C.N., Westerink, J.J., Wells, G., Lu, A., Vega, K., Kubach, A., Slinn, D., 2008. Wave Setup Methodology for the Fema Mississippi Flood Study. Dresback, K.M., Kolar, R.L., Kaiser, C., Twilley, R., 2012. Surface trajectories of oil Technical Report, Federal Emergency Management Agency. transport along the northern coastline of the Gulf of Mexico. Cont. Shelf Res. 41, Tanaka, S., Bunya, S., Westerink, J.J., Dawson, C.N., Luettich, R.A., 2010. Scalability of 17–47. an unstructured grid continuous galerkin based Hurricane Storm Surge Model. Dietrich, J.C., Westerink, J.J., Kennedy, A.B., Smith, J.M., Jensen, R.E., Zijlema, M., J. Sci. Comput., 1–30. Holthuijsen, L.H., Dawson, C.N., Luettich, R.A., Powell, M.D., Cardone, V.J., Cox, A. Tromble, E.M., Kolar, R.L., Dresback, K.M., Hong, Y., Vieux, B.E., Luettich, R.A., T., Stone, G.W., Pourtaheri, H., Hope, M.E., Tanaka, S., Westerink, L.G., Westerink, Gourley, J.J., Kelleher, K.E., Van Cooten, S., 2011. Aspects of coupled hydrologic– H.J., Cobell, Z., 2011a. Hurricane Gustav (2008) waves and storm surge: hydrodynamic modeling for coastal flood inundation. In: Spaulding, M.L. (Ed.), hindcast, synoptic analysis, and validation in Southern Louisiana. Month. Estuarine and Coastal Modeling, Proceedings of the Eleventh International Weather Rev. 139, 2488–2522. Conference, vol. 1. ASCE, pp. 724–743. Dietrich, J.C., Zijlema, M., Westerink, J.J., Holthuijsen, L.H., Dawson, C.N., Luettich, R. Tromble, E.M., Kolar, R.L., Dresback, K.M., Luettich, R.A., 2012. River flux boundary A., Jensen, R.E., Smith, J.M., Stelling, G.S., Stone, G.W., 2011b. Modeling considerations in a coupled hydrologic-hydrodynamic modeling system. In: hurricane waves and storm surge using integrally-coupled scalable computa- Spaulding, M.L. (Ed.), Estuarine and Coastal Modeling, Proceedings of the tions. J. Coast. Eng. 58, 45–65. Twelfth International Conference. ASCE, pp. 510–527. Environmental Modeling Center, 2003. The GFS Atmospheric Model. Technical Van Cooten, S., Kelleher, K.E., Howard, K., Zhang, J., Gourley, J.J., Kain, J.S., Report NCEP Office Note 442, Global Climate and Weather Modeling Branch. Nemunaitis-Monroe, K., Flamig, Z., Moser, H., Arthur, A., et al., 2011. The Fleming, J.G., Fulcher, C.W., Luettich, R.A., Estrade, B.D., Allen, G., Winer, H., 2008. A CI-FLOW project: a system for total water level prediction from the summit to real time storm surge forecasting system using ADCIRC. In: Spaulding, M.L. the sea. Bull. Am. Meteorol. Soc. 92 (11), 1427–1442. (Ed.), Estuarine and Coastal Modeling, Proceedings of the Tenth International Westerink, J.J., Feyen, J.C., Atkinson, J.H., Roberts, H.J., Kubatko, E.J., Luettich, R.A., Conference. ASCE, pp. 373–392. Dawson, C.N., Powell, M.D., Dunion, J.P., Pourtaheri, H., 2008. A basin-to Forbes, C., Luettich, R.A., Mattocks, C.A., Westerink, J.J., 2010. A retrospective channel-scale unstructured grid hurricane storm surge model applied to evaluation of the storm surge produced by Hurricane Gustav (2008): Forecast Southern Louisiana. Month. Weather Rev. 136 (3), 833–864. and hindcast results. Weather Forecast. 25, 1577–1602. Westerink, J.J., Luettich, R.A., Muccino, J.C., 1994. Modelling tides in the Western Holland, G.J., 1980. An analytic model of the wind and pressure profiles in North Atlantic using unstructured graded grids. Tellus A 46 (2), 178–199. hurricanes. Month. Weather Rev. 108 (8), 1212–1218. Zhang, J., Howard, K., Langston, C., Vasiloff, S., Kaney, B., Arthur, A., Van Cooten, S., Houston, S.H., Shaffer, W.A., Powell, M.D., Chen, J., 1999. Comparisons of HRD and Kelleher, K., Kitzmiller, D., Ding, F., et al., 2011. National Mosaic and Multi- SLOSH surface wind fields in hurricanes: implications for Storm Surge Model- Sensor QPE (NMQ) System: description, results, and future plans. Bull. Am. ing. Weather Forecast. 14 (5), 671–686. Meteorol. Soc. 92, 1321–1338. Janjic, Z., Gerrity, J., Nickovic, S., 2001. An alternative approach to nonhydrostatic Zijlema, M., 2010. Computation of wind wave spectra in coastal waters with SWAN modeling. Month. Weather Rev. 129, 1164–1178. on unstructured grids. Coast. Eng. 57, 267–277.