Tropical Cyclone Prediction Using COAMPS-TC

Abstract : A new version of the Coupled Ocean/ Atmosphere Mesoscale Prediction System for Tropical Cyclones (COAMPS -TC) has been developed for prediction of tropical cyclone track, structure, and intensity. The COAMPS-TC has been tested in real time in both uncoupled and coupled modes over the past several tropical cyclone seasons in the Western Pacific and Atlantic basins at a horizontal resolution of 5 km. An evaluation of a large sample of forecasts in the Atlantic and Western Pacific basins reveals that the COAMPS-TC intensity predictions are competitive with, and in some regards more accurate than, the other leading dynamical models, particularly for lead times beyond 36 hours. Recent real-time forecasts of Hurricane Sandy (2012) highlight the capability of COAMPS-TC to capture both intensity and multiscale structure in agreement with observations. Results from the air-ocean coupled COAMPS-TC simulations of Typhoon Fanapi (2010) and Super Typhoon Jangmi (2008) in the Western Pacific indicate accurate predictions of the track and intensity, as well as the sea surface temperature cooling response to the storm, in agreement with satellite measurements. The air-ocean- wave coupled simulations of the Atlantic Hurricane Frances (2004) highlight the capability of the COAMPS-TC system to realistically capture not only sea surface temperature cooling following storms but also characteristics of ocean surface waves and their interactions with boundary layers above and below the ocean surface.

track prediction through the use of global prediction models (e.g., Goerss, 2007;Hamill et al., 2011).A three-day hurricane track forecast today is as skillful as a one-day forecast was 30 years ago.Evacuating coastal areas before a hurricane is estimated to cost $1 million for every mile of coastline evacuated (e.g., Whitehead 2003).These dramatically improved track forecasts have reduced the size of evacuation areas and mitigated costs.However, intensity prediction remains a significant challenge, and progress has been considerably slower (DeMaria et al., 2005(DeMaria et al., , 2014;;Rogers et al., 2006).The slow improvement in TC intensity and structure forecasts has been attributed to a variety of reasons, ranging from a lack of critical observations in the TC inner core and the surrounding environment to inaccurate representations of physical processes in numerical weather prediction (NWP) models.Marks and Shay (1998) note that track prediction depends more on largescale processes, while intensity depends on both inner-core dynamics and its relationship to the environment.This motivates the requirement for accurate representation of the key physical and dynamical processes within the storm itself and in the larger-scale environment.The need to explicitly resolve the inner part of the storm, including the eye, the eyewall, and spiral rainbands, has motivated modeling of the inner core at high horizontal resolution (e.g., Zhu et al., 2004Zhu et al., , 2006;;Braun et al., 2006;Chen et al., 2007;Davis et al., 2008).One distinct advantage of applying models at high resolution (grid increments of 5 km or less) is that convection can be explicitly represented in the model, which precludes the need for a convection parameterization and results in more accurately resolved convective characteristics (e.g., structure, morphology, propagation), at least for continental locations (Fowle and Roebber, 2003;Done et al., 2004).
The Coupled Boundary Layer Air-Sea Transfer (CBLAST) field program (Black et al., 2007), conducted from 2002 to 2004, provided important air-sea interaction observations in hurricanes and motivated new approaches to parameterization of these processes in tropical cyclone models.Coupled air-ocean and air-ocean-wave tropical cyclone modeling systems more realistically represent these key air-sea interaction processes (e.g., Bao et al., 2000;Bender et al., 2007;Chen et al., 2007Chen et al., , 2010)).The coupling to an ocean-circulation model can improve the storm intensity forecast through a more realistic representation of storm-induced cooling in the upper ocean and sea surface.Inclusion of ocean waves and their feedback to the atmosphere and ocean boundary layers of a hurricane can yield more realistic momentum fluxes across the air-sea interface (e.g., Bao et al., 2000;Doyle, 2002) and improve the model-predicted maximum wind speed-central pressure relationship (e.g., Chen et al., 2007)

COAMPS-TC DESCRIPTION
The COAMPS-TC system is composed of data quality control, analysis, initialization, and forecast model subcomponents (Doyle et al., 2011) (Daley and Barker, 2000).As part of the TC analysis procedure, the pre-existing circulation in the COAMPS-TC first guess fields is relocated to allow for accurate representation of TC position for the analysis background following Liou and Sashegyi (2012).

"
with increasing distance from the RMW to the outermost ring at 600 km from the center.
Sea surface temperature is analyzed directly on the model computational grid using the Navy Coastal Ocean Data Assimilation (NCODA) system, which makes use of all available satellite, ship, float, and buoy observations (Cummings, 2005).In coupled applications, both the NCODA and NAVDAS systems are applied using a data assimilation cycle in which the first guess from the analysis is derived from the previous short-term forecast.The COAMPS-TC system has the capability to operate in a fully coupled air-sea interaction mode (Chen et al., 2010(Chen et al., , 2011;;Doyle, et al., 2011).The atmospheric module within COAMPS-TC is coupled to the Navy Coastal Ocean Model (NCOM; Martin, 2000;Martin et al., 2006) to represent air-ocean interaction processes.The COAMPS-TC system has an option to predict ocean surface waves and the interactions between the atmosphere, ocean circulation, and waves using either the Simulating WAves Nearshore (SWAN) or WAVEWATCH III models.
Wave and current interaction is parameterized through inclusion of Stokes drift currents and Langmuir turbulence (Kantha and Clayson, 2004;see Allard et al., 2014, in this issue).The wind-wave interaction is represented through the prediction of a sea-state-dependent Charnock parameter (Moon et al., 2004) that is a function of wave age and wind speed.A sea spray parameterization (Fairall et al., 1994(Fairall et al., , 2009;;Bao et al., 2011)  interface.An advanced version of a sea spray parameterization developed by Fairall et al. (1994Fairall et al. ( , 2009) ) and a drag parameterization where C d is limited for wind speed above 30 m s -1 (hereafter referred to as the "limited C d " experiment) following Powell et al. (2003) and Donelan et al. (2004)  pattern in Figure 1d, which further suggests that these processes impact not only the TC's intensity but also its structure.In general, the current state-of-the-science sea spray parameterizations are still limited by numerous uncertainties, in part due to the lack of reliable and accurate measurements of sea spray (e.g., Andreas et al., 2008;Bao et al., 2011).Bougeault and André, 1986;Grinier and Bretherton, 2002).This approach has significantly improved COAMPS-TC intensity forecasts due to its suitability for turbulent mixing in deep convection.
The radius of the 34 kt winds, however, is too large, a result of the excessive mixing in the boundary layer.In order to improve the prediction of the TC wind field within the boundary layer, the Bougeault and André mixing length is replaced with a Mellor-Yamada mixing length representation (Mellor and Yamada, 1982) in the lowest 3 km, where wind shear dominates in the turbulence production, while maintaining the Bougeault and André mixing length above the boundary layer.This new method leads to a reduction in the intensity bias (Figure 2a) in a large sample of Atlantic basin TC forecasts verified against best-track values.In addition, as Figure 2b shows, the radii of the 34 kt wind speed mean absolute error and mean error decrease for all lead times.
In particular, at 96 h, the mean absolute error is reduced by 20% and the mean error almost by half.Both the coupled and uncoupled simulations accurately predicted Jangmi's track for the four days prior to landfall (Figure 3a).However, the intensity is considerably different between the two A M E S D .D O Y L E , R I C H A R D M .H O D U R , S U E C H E N , Y I J I N , J O N AT H A N R .M O S K A I T I S , S H O U P I N G WA N G , E R I C A .H E N D R I C K S , H A O J I N , A N D T R AV I S A . S M I T H ABSTR ACT.A new version of the Coupled Ocean/ Atmosphere Mesoscale Prediction System for Tropical Cyclones (COAMPS®-TC) has been developed for prediction of tropical cyclone track, structure, and intensity.The COAMPS-TC has been tested in real time in both uncoupled and coupled modes over the past several tropical cyclone seasons in the Western Pacific and Atlantic basins at a horizontal resolution of 5 km.An evaluation of a large sample of forecasts in the Atlantic and Western Pacific basins reveals that the COAMPS-TC intensity predictions are competitive with, and in some regards more accurate than, the other leading dynamical models, particularly for lead times beyond 36 hours.Recent real-time forecasts of Hurricane Sandy (2012) highlight the capability of COAMPS-TC to capture both intensity and multiscale structure in agreement with observations.Results from the airocean coupled COAMPS-TC simulations of Typhoon Fanapi (2010) and Super Typhoon Jangmi (2008) in the Western Pacific indicate accurate predictions of the track and intensity, as well as the sea surface temperature cooling response to the storm, in agreement with satellite measurements.The airocean-wave coupled simulations of the Atlantic Hurricane Frances (2004) highlight the capability of the COAMPS-TC system to realistically capture not only sea surface temperature cooling following storms but also characteristics of ocean surface waves and their interactions with boundary layers above and below the ocean surface.
, which is an important aspect of the hurricane intensity and structure relationship.Inclusion of wave-current interaction through the use of an observation-based momentum drag formulation in the wave model, the wind source generation, and the total volumetric dissipation is shown by Smith et al. (2013) to reduce coupled model forecast errors in significant wave height and wave period for Hurricane Ivan (2004).Advances in high-resolution TC modeling and data assimilation are thought to be necessary in order to significantly improve the intensity and structure prediction.To this INTRODUCTION The demand for more accurate forecasts of tropical cyclone track and intensity with longer lead times is greater than ever due to the enormous economic and societal impact.A dramatic example occurred during October 2012 as Hurricane Sandy threatened many communities along the US Eastern Seaboard.Basic questions such as where Sandy would track and how strong it would become had profound implications for the millions of people in its path and billions of dollars of vulnerable assets.With an estimated total damage amount of $50 billion USD or more, Hurricane Sandy is the second costliest hurricane since 1900, and the deadliest hurricane to hit the northeastern United States in four decades.The potential impact of tropical cyclones on military operations can also be enormous.Typhoon Cobra, also known as Halsey's Typhoon after Admiral William Halsey, struck the US Navy's Pacific Fleet in December 1944 during World War II.Three destroyers were lost, and a total of 790 sailors perished.More recently, during Sandy, the decision to sortie Navy assets from Norfolk, VA, and other ports along the Eastern Seaboard days in advance of the storm was critically dependent on forecasts of Sandy's track, intensity (maximum sustained wind speed at the surface), and storm structure (i.e., the size of the storm or radius of key wind speed thresholds).In the western North Pacific basin, an area of strategic importance for the US Navy, the Navy Pacific Fleet in the Philippine Sea has been affected by numerous storms, such as Typhoon Nanmadol (2011) that exhibited erratic movement and was poorly forecasted.There has been remarkable improvement in tropical cyclone (TC) end, the Naval Research Laboratory (NRL) in Monterey, CA, developed the Coupled Ocean/Atmosphere Mesoscale Prediction System for Tropical Cyclones (COAMPS-TC), a new version of COAMPS designed specifically for high-resolution tropical cyclone prediction (Doyle et al., 2011).The model builds on the existing COAMPS infrastructure and provides the framework to add new capabilities and advancements in data assimilation, vortex initialization, physical parameterization, and air-sea coupling appropriate for high-resolution tropical cyclone prediction.This paper provides an overview of the main capabilities of COAMPS-TC through examination of several different representative tropical cyclone cases and statistical analysis of real-time and retrospective forecasts that were conducted as part of a pre-operational evaluation of the system.We next provide a brief description of the COAMPS-TC system, after which we address aspects of the air-sea coupling, including boundary layer and surface flux parameterizations and fully coupled air-ocean and airocean-wave options.The final section provides an overview of the statistical performance of the system, and we conclude with a summary of the results.
Synthetic observations or profiles are used to incorporate TC structure and intensity into the initial conditions based on National Hurricane Center (NHC) and the Joint Typhoon Warning Center (JTWC; Liou and Sashegyi, 2012) specifications.This is accomplished by using these synthetic observations in NAVDAS to represent the hurricane's characteristics, and then blending the synthetic observations with other available real-time observations that describe the larger-scale environment outside the TC circulation.The TC synthetics are necessary due insufficient real-time in situ observations of tropical cyclones.Recently, a new method of generating TC synthetics was introduced into COAMPS-TC that is based on TC position, maximum winds, radius of maximum winds, mean radius of the 34 knot winds, and recent storm motion.The TC synthetics are generated at one point in the center of each TC, and at eight points around each of nine concentric circles centered over the TC, for a total of 73 individual points.At each of these points, profiles of the u-and v-components are prescribed at 1000, 925, 850, 700, 600, 500, and 400 hPa, along with geopotential height at 1,000 hPa.The winds' horizontal structure is generated from a modified Rankine wind vortex that best fits the observed value of the maximum winds, the radius of the maximum winds, and the radius of the 34-knot winds.A Rankine vortex is a simplified model of the radial structure of the tangential wind field.The prescribed vertical profile of the winds follows the method described by Liou and Sashegyi (2012).The mean storm motion is added to the Rankine-vortex winds.The 1,000 hPa geopotential heights are created by solving the equation for the Rankine vortex for the geopotential field.The innermost ring of synthetics can be set to either the reported radius of maximum wind (RMW) or to the location of the RMW in the COAMPS-TC first-guess fields.The latter is typically used to minimize the size of the analysis increments.The remainder of the rings is evenly spaced " AN EVALUATION OF A LARGE SAMPLE OF FORECASTS IN THE ATLANTIC AND WESTERN PACIFIC BASINS REVEALS THAT THE COAMPS-TC INTENSITY PREDICTIONS ARE COMPETITIVE WITH, AND IN SOME REGARDS MORE ACCURATE THAN, THE OTHER LEADING DYNAMICAL MODELS, PARTICULARLY FOR LEAD TIMES BEYOND 36 HOURS.
The COAMPS-TC atmospheric model uses the nonhydrostatic and compressible form of the dynamics and has prognostic variables for the three components of the wind (two horizontal wind components and the vertical wind), the perturbation pressure, potential temperature, water vapor, cloud droplets, raindrops, ice crystals, snowflakes, graupel (soft hail), and turbulent kinetic energy(Hodur, 1997).Physical parameterizations include representations of cloud microphysical processes, convection, radiation, boundary layer processes, and surface layer fluxes.The COAMPS-TC model contains a representation of dissipative heating near the ocean surface, which has been found to be important for tropical cyclone intensity forecasts (Jin et al., 2007).The COAMPS-TC system uses a flexible nesting design, which has proven useful when more than one storm is present in a basin at a given time, as well as special options for moving nested grid families that independently follow individual tropical cyclone centers of interest.In the applications shown in this paper (unless otherwise noted), the atmospheric portion of COAMPS-TC uses three nested grids of 45 km, 15 km, and 5 km horizontal resolution and 40 vertical levels that extend from 10 m to approximately 30 km.The inner two grid meshes follow the storm.
can be used to represent the injection of droplets into the atmospheric boundary layer due to ocean surface wave breaking and shearing of the crest of breaking waves.The spray droplets impact the momentum and enthalpy fluxes through increased mass loading, air flow stratification, and evaporation and/or condensation.BOUNDARY L AYER AND AIR-SEA INTER ACTION SENSITIVIT Y Momentum exchange at the sea surface is dependent on the sea-state-dependent drag coefficient, C d .Prior to the past decade, the characteristics of C d had not been observed in a tropical cyclone and were primarily based on extrapolations from field campaign measurements conducted in much weaker wind conditions.In a seminal study, Powell et al. (2003) analyzed data from GPS dropwindsondes deployed from aircraft into hurricanes; they found that the mean wind speed varied logarithmically with height in the lowest 200 m and was a maximum at 500 m.They estimated the surface stress, roughness length, and neutral stability C d , and found a markedly reduced C d at wind speeds above 30 m s -1 .Their analysis showed a leveling off of the surface momentum flux as the winds increase above the hurricane threshold and even a slight decrease of the C d with further increases in wind speed.Donelan et al. (2004) extended the Powell et al. (2003) study through a series of windwave tank experiments and found that C d saturation occurs when wind speed exceeds 33 m s -1 .Beyond this wind speed threshold, the surface roughness no longer increases.Donelan et al. (2004) found a C d saturation level of 0.0025, similar to the saturation value of 0.0026 found by Powell et al. (2003).Both surface drag and sea spray processes play major roles in regulating energy exchange at the air-sea James D. Doyle (james.doyle@nrlmry.navy.mil) is a meteorologist and head of the Mesoscale Modeling Section in the Marine Meteorology Division, Naval Research Laboratory (NRL), Monterey, CA, USA.Richard M. Hodur is a scientist at Science Applications International Corporation, Monterey, CA, USA.Sue Chen, Yi Jin, Jonathan R. Moskaitis, Shouping Wang, Eric A. Hendricks, and Hao Jin are all meteorologists in the Marine Meteorology Division, NRL, Monterey, CA, USA.Travis A. Smith is an oceanographer in the Oceanography Division, NRL, Stennis, MS, USA.

Figure 1 .
Figure 1a,b), however, attains a minimum surface pressure that is 29 hPa higher and an intensity 20 m s -1 weaker than the best track analysis.In contrast, storm intensity is increased in the limited C d experiment (Figure 1, magenta line), likely a result of reduced surface friction.Including sea spray processes substantially increases the surface latent heat fluxes, and the additional energy input at the air-sea interface further enhances convection (not shown).The simulation with both the limited C d and the sea spray parameterization (red line) attains a minimum pressure of 931 hPa and a 68 m s -1 maximum wind speed at 120 h, a significant improvement over the standard Charnock parameterization.An increase in the energy input, attributable to the sea spray along with the higher wind speeds due to the limited C d at high wind speeds, is clearly apparent in Figure 1c,d, which compares the enthalpy flux from the standard Charnock C d with that from the experiment that used the limited C d and sea spray parameterization.The magnitude of the enthalpy flux is nearly doubled due to the stronger wind speeds and evaporation of sea spray drops.The maximum enthalpy flux reaches 2,000 W m -2 on the right side of the storm track where the wind speed is stronger due, in part, to the contribution by the translation speed of the storm.It should be noted that the differences in the enthalpy fluxes in Figure 1c,d correspond to a time when the two simulations are quite different (there is a ~ 20 m s -1 difference in the maximum wind speed at 110 h).Thus, the differences in the enthalpy fluxes are mostly due to integrated differences in the development pathways of each simulation.The impact of the limited C d and sea spray is also manifested by the strong asymmetry shown by the enthalpy flux It remains an outstanding challenge to develop a physically based and observationally verifiable sea spray parameterization that is fully interactive with ocean waves and the atmospheric boundary layer.The sensitivity of the predicted track, intensity, and structure of tropical cyclones to the representation of the atmospheric planetary boundary layer (PBL) is illustrated through a comparison of two different versions of the turbulence parameterization in COAMPS-TC.One version of the PBL parameterization makes use of a buoyancy-based nonlocal mixing length to represent turbulent mixing (e.g., As a demonstration of the COAMPS-TC air-ocean coupled capability, Figure 3 shows results from an air-ocean coupled simulation of Super Typhoon Jangmi, a 2008 category-5 storm in the western North Pacific basin that occurred during the THe Observing system Research and Predictability EXperiment (THORPEX) Pacific Asian Regional Campaign (T-PARC) and the Office of Naval Research's (ONR's) Tropical Cyclone Structure-08 (TCS-08) experiments.During its lifetime from September 24 to October 2, Jangmi formed in a region of deep ocean mixed layer northeast of the island of Yap, then moved northwestward, crossing over several warm eddies followed by a series of cold eddies, and then made landfall over northern Taiwan on September 28.Jangmi's intensity change is highlighted here to show the impact of the evolving ocean during the forecast through comparisons of high-resolution coupled and uncoupled simulations.The NCOM ocean model is employed, with a single domain at 15 km resolution and 36 vertical levels from the ocean surface to ~ 4 km depth.First guess and boundary conditions are from the US Navy Operational Global Atmospheric Prediction System (NOGAPS) for the atmosphere and global NCOM for the ocean for a series of five-day forecasts; here, we highlight the forecast initialized at 0000 UTC September 25, 2008.The same initial sea surface temperature (SST) is used for the coupled and uncoupled forecasts, but SST is unchanged during the uncoupled run, and the ocean model predicts SST during the coupled run.

Figure 2 .
Figure 2. Comparison of two planetary boundary layer (PBL) parameterizations for: (a) forecast intensity, and (b) forecast radius of the 34 kt wind.The mean absolute errors (MAE, solid) and mean errors (ME, dashed) are shown for a large sample of western Atlantic Ocean storms.PBL-a uses aBougeault and André (1986) mixing length only, while PBL-b features a combination ofBougeault and André (1986) andMellor and Yamada (1982) mixing lengths.The number of samples for each forecast lead time is shown along the top of the plot, with a greater number of samples in (b) because the evaluation is performed for each quadrant.

Figure 3 .
Figure4shows for a forecast initialized at 0000 UTC September 15, 2010.There is general agreement in SST distribution as well, derived from satellite observations (Figure4a) and the forecast (Figure4b).Both predicted and satellite-observed cold wakes form on the right side of the track, with a similar amount of cooling (~ 3°C) relative to SST prior to the storm.The most intense cooling occurs on the eastern flank of the wake, in both the observations and the forecast, due to the initial slow movement of the storm in the September 15-17, 2010, time period.It is noteworthy that cold water is advected to the north at both the western and the eastern ends of the wake, surrounding the warm water near 126°E and 24.5°N.Note that cooling in the wake evolves during the forecast and is not present in the ocean initial conditions.The capabilities of the air-oceanwave coupled COAMPS-TC system are highlighted for Atlantic Hurricane

Frances ( 2004
), which was observed during the ONR CBLAST field campaign near the northern shore of the Bahamas Archipelago.The three-way coupled COAMPS model configuration includes the three grid meshes (45/15/5 km), a single 3 km horizontal resolution ocean mesh, and a single 10 km horizontal resolution wave model.In this application, the atmospheric model has 60 vertical sigma levels and the ocean model has 49 vertical levels with 35 sigma layers.The wave model, SWAN, has 36 discrete directional and 33 frequency bands.For the forecast highlighted here, the coupled COAMPS-TC is initialized on 1200 UTC August 31, 2004.The initial and boundary conditions for the atmospheric and ocean models are provided by NOGAPS and global NCOM, respectively.

Figure 5 Figure 4 .
Figure 5 shows the air-ocean-wave coupled COAMPS-TC forecast SST, significant wave height, 10 m wind, total surface currents, and Stokes currents at the 48 h time.Comparisons with the best track show that the 48 h forecast track error is 100 nm and the intensity

Figure 5 .
Figure 5. Air-ocean-wave coupled COAMPS-TC 48 h forecast of (a) sea surface temperature (°C) and (b) significant wave height (m; shading), wind (black arrows), surface current (blue arrows), and surface Stokes current (magenta arrows) vectors for Hurricane Frances (2004).COAMPS-TC is initialized on 1200 UTC August 31, 2004.The unit vectors for the wind, surface current, and surface Stokes current vectors are shown in (b).A subset of the domain is shown in (b) as well for clarity purposes.The model and best track are by the black and red lines and squares, respectively (every 6 h).

FigureFigure 6 .Figure 7 .
Figure 6 summarizes the statistics comparing the performance of COAMPS-TC forecasts and real-time forecasts from other operational regional dynamical tropical cyclone models.The operational model forecasts were sourced from the Automated Tropical Cyclone Forecast system (Sampson and Schrader, 2000) archive, along with the "best-track" analyses of TC position and intensity used as verification.Following conventional TC forecast validation procedures, a forecast case is only included in the validation sample if the best-track indicates the storm is a TC at the forecast initial time and valid time, and also that all models in the comparison made a forecast (i.e., the sample is homogeneous).For the Atlantic sample, COAMPS-TC forecasts are compared against those from the Geophysical Fluid Dynamics Laboratory model (GFDL) and the Hurricane Weather Research and Forecasting model (HWRF), both run by the National Centers for Environmental Prediction. Figure 6a displays track accuracy results and Figure 6b shows intensity accuracy (solid lines) and bias (dashed lines) results.The COAMPS-TC mean absolute error for track is slightly higher than the corresponding values for the operational models.However, the COAMPS-TC intensity forecasts are more skillful than HWRF and GFDL for all lead times beyond 24 h.The mean intensity error for COAMPS-TC is generally closer to zero than for the operational models, indicating that COAMPS-TC also performs better in terms of intensity bias as well as intensity accuracy.Figure 6c shows intensity error and bias results for the western