1. Introduction

Numerical weather prediction (NWP) and mesoscale modeling products are routinely used by national defense agencies around the world for applications far beyond weather forecasting. In addition to simulating spatial and temporal variability of the atmosphere at high resolution over regions of military interest, the modeled fields are often ingested by a variety of tactical decision aids yielding weapon and radar performance and communication and surveillance guidance.

Many naval applications utilizing mesoscale models involve at-sea and coastal scenarios in which discontinuities between land and ocean can create sharp gradients in air temperature and water vapor content. Vertical gradients, particularly of water vapor, can have a significant impact on the propagation of electromagnetic (EM) waves, bending them toward or away from the earth’s surface (Wagner 1960; Bean and Emmanuel 1969). The vertical gradient of modified refractivity, based upon Snell’s law, denotes how EM energy will travel from its point of origin through a given layer (Bean and Dutton 1968). The four refractivity regimes—subrefraction, normal refraction, superrefraction, and trapping—are each characterized by a range of modified refractivity slope values as shown schematically in Fig. 1 along with definitions of ducting-layer characteristics. Surface-based ducts are an important subset of ducting events, defined as those having a duct base height equal to 0 m, because of their potential to alter the propagation environment at low elevations and thus to affect surface radars and communications.

This study presents the coastal refractivity and ducting characteristics represented by four mesoscale forecasting systems during a 7-day intensive observation period (IOP) near Wallops Island, Virginia, offshore the Delmarva Peninsula in April and May of 2000. Hereinafter we refer to the experiment as Wallops-2000, or Wallops-2000 Microwave Propagation Measurement Experiment (MPME). The field study was designed by the Naval Surface Warfare Center Dahlgren Division (NSWCDD) to measure both environmental and propagation conditions contributing to the emergence of refractive features in a challenging coastal environment (Stapleton et al. 2001).

The intercomparison team was formed from four countries: the United States of America, United Kingdom, Canada, and New Zealand. These so-called ABCANZ countries have an international exchange agreement on scientific research in several specialized areas of technology. Each country employed the numerical weather prediction tool used to support their national defense and navy missions to simulate the entire 7-day Wallops-2000 IOP. The hindcast nature of this study, 8 years after the field campaign, created some limitations in model initialization and data assimilation that were unavoidable. We gained considerable insight from those limitations, however, and identified several sensitivity tests as a result. The team’s primary objective was to identify the strengths and weaknesses of each country’s complete forecasting system that contribute to accuracy in simulating refractive structures in a littoral environment. We utilized the intercomparison to establish a validation benchmark of mesoscale modeling capabilities of coastal refractivity and atmospheric ducting. This process is of vital importance because atmospheric heterogeneity is often depicted within EM propagation codes by use of high-resolution NWP fields. Further, such information provides valuable guidance on where to focus system development for substantive improvements to refractivity forecasts.

Large-scale influences on ducting have been studied extensively and were first documented by Rosenthal and Helvey (1979) and then Helvey et al. (1995), who developed a schematic synoptic–refractive relationship model. They found increased duct frequency over the eastern and equatorward sectors of midlatitude high pressure areas coinciding with the strongest subsidence inversions and near-standard refraction behind low pressure troughs. Those findings were reinforced by von Engeln and Teixeira (2004) from a 6-year global ducting climatological dataset from the European Centre for Medium-Range Weather Forecasts (ECMWF). Rosenthal and Helvey’s synoptic–refractive relationship model was noted as being more appropriate for the open ocean, however, because of mesoscale features commonly observed near coastlines.

In a littoral setting, vertical gradients generated by large-scale processes in the atmosphere and ocean are frequently affected by embedded mesoscale structures. These structures may be imparted by complex topography and coastline geometry creating topographic flows and diurnally driven sea and land breezes. During weak background flow, for example, land breeze circulations were reported to influence radar observations in the Wallops Island area (Meyer 1971). In another coastal region, severe radio signal fades were coincident with the formation and inland advection of a sea breeze, as documented by Reddy and Reddy (2007). Abrupt changes in surface stability, from a spatially complex and highly resolved sea surface temperature, can also have dramatic effects on the overlying marine atmospheric boundary layer (MABL). High-resolution sea surface temperature (SST) analyses, when compared to climatological SST forcing, were found to alter coastal processes and MABL structure in the vicinity of the Gulf Stream (Doyle and Warner 1993). Such complexities make the characteristics of atmospheric ducts a challenge to predict and validate.

The first highly idealized case studies of refractive effects were presented by Silveira et al. (1995) over the São Paulo river basin and water reservoir in Brazil, showing the effects of lake–land breeze circulations on ducting and line-of-sight microwave propagation links. Burk and Thompson (1997) obtained more realistic coastal refractivity by initializing model runs with real data for a 5-day period over the Southern California Bight. They found that the trapping-layer depth and strength evolved with the diurnal cycle and that the entrance of a synoptic low pressure trough eliminated the ducting layer entirely. Additional high-resolution real-data model runs along the California coast during summertime were performed by Haack and Burk (2001) and showed modulation of refractive layers by the MABL interaction with topography.

Refractivity studies were also conducted in the Persian Gulf using the Ship Antisubmarine Warfare Readiness/Effectiveness Measuring exercise (SHAREM-115) data. Atkinson et al. (2001) revealed that simulating correct ducting structures depended upon inhomogeneous model initialization using an appropriate profile for an underlying land or water surface. A more comprehensive set of modeling experiments by Atkinson and Zhu (2006) identified four factors influencing propagation in the Persian Gulf—the sea breeze, coastal configuration, orography, and ambient wind, and an observational study by Brooks et al. (1999) showed sensitivity of duct depths to SST.

Forecasting refractivity in the coastal zone relies upon representation of evolving synoptic-scale systems along with a detailed, high-resolution description of the lower boundary (topography, coastline, land types, and sea surface temperature) and sophisticated surface and boundary layer parameterization schemes to capture the inhomogeneity and nonlinearity in the lower atmosphere. Previous modeling studies were limited somewhat by low spatial resolutions or fairly idealized initialization methods. The present work not only takes advantage of the advancements in model parameterization schemes and real-data initialization procedures over the last decade, but also utilizes high-resolution model grids to improve the mesoscale representation of coastal ducting layers. The proceeding analysis investigates the role of initialization, data assimilation, and SST forcing within four modeling systems. The impact of horizontal resolution on ducting is studied using nested grids from one model, each increasing in resolution from its parent grid by a factor of 3. The paper is organized as follows: Section 2 describes the Wallops-2000 MPME field campaign. In section 3, each NWP modeling setup is described. An overview of the large-scale environment is provided in section 4, and the model validation is given in section 5. Section 6 presents the mesoscale structure in the simulated refractivity fields and sensitivity tests used to discern the relative importance of initialization and boundary conditions. The conclusions are given in section 7.

2. Wallops-2000 MPME field campaign

The Wallops-2000 field program was conducted in April and May of 2000 to collect meteorological measurements and radar-frequency one-way propagation data along onshore–offshore radials extending up to approximately 60 km from the Virginia shoreline near Wallops Island (Fig. 2). During the 7-day IOP, between 28 April and 4 May 2000, a wide range of refractive conditions was observed.

The Naval Postgraduate School (NPS) deployed a fixed buoy approximately 13 km offshore from a shore-based meteorological observing tower. The platforms reported temperature, relative humidity, pressure, and winds at heights of approximately 4 and 10 m, respectively. Sea surface temperature was also collected at the NPS buoy. An instrumented helicopter (HELO), outfitted by the Johns Hopkins University Applied Physics Laboratory (JHU/APL), collected measurements of temperature, relative humidity, and pressure along the radials, primarily within the lowest 150 m of the atmosphere. The HELO measurements have been used in refractivity-related research by Babin (1995, 1996) who investigated surface ducting and subrefractive environments in and around the Wallops Island area. Details about the instrumentation, including measurement response times, accuracy, and resolutions, are given in Babin and Rowland (1992).

The helicopter radial positions were matched by an instrumented boat, the Sealion, which recorded temperature, relative humidity, pressure, winds, and SST. The Microwave Propagation Measurement System (II) developed by NSWCDD was used to collect one-way radio-frequency propagation loss between a transmitter mounted on the Sealion and a shore-based receiver. Analysis of the propagation data and EM ducting depends upon the details of the transmitter and receiver and will be undertaken in a subsequent paper. The focus of the study presented here is on the meteorological measurements, MABL refractivity, and atmospheric ducting conditions.

3. Model setup

The four models used in this study were set up to be as similar as possible (Table 1). This included using similar latitude and longitude boundaries for the inner 4-km nest on which most of the subsequent analysis and data comparisons was conducted (Fig. 2). Roughly half of this nest resides over water and includes a portion of the Gulf Stream in its southeastern corner. The intricate coastline in this region contains the Potomac River basin and the Delmarva Peninsula on which Wallops Island is located.

For the United States, mesoscale model prognostic fields were produced by the Naval Research Laboratory Coupled Ocean–Atmosphere Mesoscale Prediction System (COAMPS).¹ Of the four models investigated, only COAMPS could be run in a manner consistent with its typical operational setup. The four COAMPS domains (36-, 12-, 4-, and 1.33-km grid spacing) were initialized 3 days prior to the start of Wallops-2000 from Navy Operational Global Analysis and Prediction System (NOGAPS) 1.0° × 1.0° analyses (27 pressure levels), allowing higher-resolution detail and vertical structure to develop. Lateral boundary tendencies for the outer 36-km grid were computed from 6-hourly NOGAPS fields. The 12-h forecast fields, and an analyzed SST, were corrected every 12 h by a multivariate optimum interpolation (MVOI) analysis of available satellite, station, aircraft, ship, and buoy data residing within each grid. This procedure produced twice-daily (0000 and 1200 UTC) 1–12-h forecasts forced at the lower boundary by 12-hourly updated SST analyses at each grid’s resolution from the Navy Coupled Ocean Data Assimilation (NCODA) system.

The United Kingdom utilized the Met Office’s Unified Model (MetUM). For their initial set of model runs, MetUM produced seven 24-h-long simulations using three grids (12-, 4-, and 1-km grid spacing), reinitializing daily at 1200 UTC by dynamical downscaling from ECMWF global analyses at T319L60 (∼0.5°; 60 levels). The boundary conditions for their outer grid (12 km) were generated from ECMWF data at 6-h intervals. Observational data had to be incorporated indirectly into the MetUM through the 4D variational (4DVAR) data assimilation on the global fields because the Met Office mesoscale data assimilation scheme could not easily be run in hindcast mode. The SST values on all grids were obtained from the coarser-resolution ECMWF SST field and were updated every 24 h at 1200 UTC.

The New Zealand Defense Technology Agency used the fifth-generation Pennsylvania State University–National Center for Atmospheric Research Mesoscale Model (MM5). The MM5 grids (36-, 12-, and 4-km grid spacing) were initialized at 0000 UTC 25 April from the National Centers for Environmental Prediction (NCEP) 1.0° × 1.0° global reanalysis (27 pressure levels). No subsequent reinitialization took place for MM5 so that observational data were only incorporated by nudging the boundaries of the MM5 outer (36 km) grid to the NCEP reanalysis every 6 h. A coarse NCEP SST analysis at 1.0° grid spacing was interpolated to all three MM5 grids and was updated daily at 0000 UTC.

Environment Canada ran their Global Environmental Multiscale model (GEM). Seven analyses were produced daily at 0000 UTC for the GEM simulation on a global variable-resolution grid (∼15-km grid spacing) initialized from the Canadian Meteorological Centre’s (CMC) global analysis fields (∼24-km grid spacing; 28 pressure levels). This 0000 UTC initialization allowed for a 6-h spinup of the global simulation to provide fields for a 12-km nest, which after another 6 h provided the fields for a 4-km nest. For the 4-km grid, the model was run for 12 h starting at 1200 UTC daily. Thus, GEM fields were not continuous over a full 24-h cycle. Boundary conditions were updated every hour on their 4- and 12-km nests. As in the MetUM and MM5 models, the Wallops hindcast prevented their 3DVAR data assimilation from being run on the GEM mesoscale grids. Hence, all observational information was provided by the 0000 UTC CMC analyses. The SST fields for GEM were produced daily at 0000 UTC from a coarse 100-km CMC SST analysis.

Terminology adopted throughout the paper includes the use of “model” to signify the complete forecast system run by each country for this study, “forecast” to refer to each model’s simulation, and “forecast length” to indicate the length of time from model initialization or update, which as indicated above was not the same for all models. Given those differences, we do not attempt to evaluate more subtle albeit important aspects of the mesoscale models, such as their numerics, dynamics, and physical parameterization schemes. Rather, we used the set of model predictions to draw common conclusions and point to areas for further study, a strategy that was effective in motivating the sensitivity tests detailed in section 6b.

4. Synoptic forcing

The SST field for the region in springtime depicts cold coastal waters over the continental shelf that form a sharp gradient with the Gulf Stream as it separates from the coast near Cape Hatteras, North Carolina. Monthly mean fields of Advanced Very High Resolution Radiometer (AVHRR) SST indicate that temperatures increase throughout the spring, more rapidly over the continental shelf, maintaining about a 10-K SST front with the warm waters of the Gulf Stream in April and May (Mesias et al. 2007). Satellite data also reveal a propensity for warm- or cold-core eddies between the continental shelf and the Gulf Stream (Lee and Cornillon 1996). The Gulf Stream is located about 200–300 km offshore Wallops Island, which can be seen in the 3-day composite SST ending 2 May 2000 (Fig. 2) within the southeastern corner of the 4-km grids.

The sharp SST front and two cyclonic eddies can impart abrupt changes in surface fluxes and stability. One of the earliest papers on air–sea interaction (Sweet et al. 1981) described the modifications to the lower atmosphere, and to the atmosphere’s refractive index, by the stability change across the Gulf Stream. Figure 3 presents the SST field used by each model near the middle of the Wallops IOP on 1 May. The NCODA SST analysis used by COAMPS (Fig. 3a) closely depicted the observed distribution with a meandering SST front, cold coastal SST, and a warm Gulf Stream. The SST gradients were not well resolved in MetUM (Fig. 3b) or MM5 (Fig. 3d), being roughly 70% weaker than were observed, and MM5’s SST gradient was oriented in a more north–south direction. The GEM SST gradient was only 30% weaker than observed but lacked the sharp front and eddies composing the Gulf Stream. These SST differences permitted an examination of atmospheric ducting sensitivities to the ocean boundary condition from which subsequent SST sensitivity tests were devised.

Rapid transitions in synoptic weather regimes are well documented over the Wallops Island region during the spring season (Babin 1996; Goldhirsh et al. 1994; Goldhirsh and Musiani 1999). The present study begins on 28 April with a deep trough over the eastern United States that slowly moved out of the area early on 30 April. During these first two days, shortwave passage, upper-level disturbances, and convective activity complicated weather patterns with rapid adjustments in near-surface winds and in the free atmosphere above the MABL. A frontal passage on 30 April marked the transition to a more tropical air mass as weak ridging combined with a surface anticyclone off the coast of North Carolina, advecting warm, dry air over the study area. This began a period of sustained, increasingly strong ducting from 1200 UTC 30 April until 2 May when an approaching low pressure cell elevated the MABL inversion and a cold front mixed away vertical gradients. The last two days of the IOP (3–4 May) were controlled by weak synoptic forcing associated with a surface high pressure cell offshore the Canadian coast that brought in cold, subsiding air in a long, overwater fetch.

Time series of hourly surface meteorological parameters at the NPS buoy (diamond in Fig. 2) reveals many of the synoptic features discussed above (Fig. 4). The observed SST (shown by the black line) displays the warming trend over the 7-day IOP as well as a maximum diurnal perturbation of about 2 K. All models except COAMPS had nearly constant SST and contained large 2–3-K biases in temperature. The NCODA analysis produced the observed warming trend, but its weak 0.5-K diurnal signature was out of phase with the observations. The remaining variables generally followed the observed trends, revealing two episodes of offshore flow with warmer temperatures and lower relative humidity (30 April and 2 May). Both days were followed by high pressure (1 and 3 May), although each corresponded to very different air masses over Wallops Island, as suggested by the wind shift to southerlies on 1 May and to northeasterlies on 3 May. The MetUM correctly predicted the magnitude of low relative humidity during the two offshore flow periods and overall had more accurate large-scale evolution, while synoptic transitions appear to be delayed in GEM and COAMPS by 3–6 h.

The time series of model-computed duct strength at the buoy location (Fig. 5) is used to assess temporal variations in ducting and associate them with the changing synoptic environment. These data suggest that ducting is a common feature for this time period and location, occurring more than 40% of the time in all of the models. Periods of observed high pressure were conducive to strong ducting as was present in the models on 1 May, although the high pressure event on 3–4 May was problematic for some of the models. Weaker ducts developed during periods of offshore flow, occurring after about 1800 UTC 29 and 30 April and 2 May, by a heated planetary boundary layer advecting over much colder coastal waters. These stable internal boundary layer (SIBL) ducts began at the coastline as shallow, thin surface-based ducts that expanded downwind, gradually subsiding after ∼0300 UTC. Periods with no ducting coincided with a negative pressure tendency and an approaching low pressure center or front, which was the case early on each of 28 April, 29 April, and 2 May. During these times the MABL inversion was lifted and gradients were weakened by dynamic processes and vertical mixing. In general, the evolution of model-simulated ducts was in accord with the expected relationships developed by Rosenthal and Helvey’s (1979) for their synoptic–refractivity schematic model.

In Table 2, the mean ducting conditions at the nearshore buoy location are contrasted with those 100 km offshore as simulated by each model. For profiles having multiple ducting layers, the strongest duct was retained, which may not have been the lowest. In comparing mean values between the buoy and offshore location, the models generally had less than 10% difference with respect to duct frequency (DFQ) and strength (DST). The COAMPS model was the exception, having 25% stronger ducting at the offshore location relative to near shore. In all models, duct thicknesses (DTK) tended to increase with distance from the coast by 30%–40%. Mean duct base heights (DBH) conveyed the most variability between models and locations with all but GEM having higher DBH offshore. Roughly half to two-thirds of these ducts were surface-based ducts (SBD) in the models, with slightly fewer SBD occurring at the offshore location.

The difference between nearshore and offshore DBH was considerably less in COAMPS (20%) than in the other models (65%–50%), which was likely due to the uniformly low SST of COAMPS over this 100-km distance. The cold expanse of water and the sharp SST front in COAMPS were also responsible for its prediction of stronger ducting at the offshore location, which resulted from the MABL rapid adjustment to the change in surface stability, an abrupt lowering of the boundary layer, and intensification of the inversion on the cold side of the front. This process is more gradual in the other models, which feature weaker SST gradients. It is important to note that GEM statistics are weighted more toward the SIBL ducts that form in the afternoon and evenings with offshore flow because their simulations were only from 1200 to 0000 UTC. Hence, their statistics did not include the strongest ducting event between 0000 and 1200 UTC 1 May, giving the impression that GEM had fewer and weaker ducts than the other models.

5. Model validation

Validation of the simulations is based upon standard statistical methods including means, biases (defined as observation minus model), and root-mean-square (rms) errors along with duct event contingency tables used to establish a benchmark set of statistics by which to judge the impact of sensitivity tests to follow. Means were computed for the near-surface variables at the buoy location using the hourly time series of buoy measurements and for the vertical profiles at the HELO locations using the subset of times corresponding to each HELO descent leg.

The mean statistics for buoy-measured quantities provide an estimate of overall predictive skill for near-surface variables (Table 3). Despite biases in air temperature, which were mostly a reflection of those in SST, MetUM had more accurate synoptic evolution yielding lower rms errors—in particular, for relative humidity. Both COAMPS and GEM had difficulties with some of the synoptic transitions (Fig. 4), resulting in slightly larger rms errors. Although MM5 displayed the general large-scale trends over the 7-day IOP, significant errors in near-surface fields were the result of the long simulation length, with large-scale adjustments made only at the outer grid’s lateral boundaries.

We evaluated profiles of potential temperature, specific humidity, and modified refractivity simulated by each model against those computed from HELO measurements. The model profiles were extracted from fields at the nearest hour and gridpoint location corresponding to the average latitude and longitude position at the start and end of each HELO descent. Mean statistics for the vertical profiles were obtained by subsampling the HELO data to the model heights. If the HELO profile did not extend below the lowest model level, data were duplicated below the lowest observational height so as not to artificially introduce any ducting by extrapolation or interpolation to the surface. Because GEM and MM5 were run at different vertical resolutions, their models’ profiles were first interpolated to the vertical grid identified for analysis (70-level distribution).

Although the 190 HELO profiles had high vertical resolution of typically less than 10 m, only about 30 profiles reached 500 m. Fewer than 10 profiles extended above 650 m, thus limiting the ability to adequately validate the model fields above the lowest 100 m. Moreover, if a model produced a duct but it was above the height of the observed profile, it was not considered in the ducting statistics to follow. The data were also limited temporally to daytime hours between 1200 and 2300 UTC. As enumerated in Table 4, some days were heavily observed and others were not. As a consequence, mean statistics for the vertical profiles might suggest poor model performance if it did not simulate the timing or distribution of ducting particularly well on either 29 April or 4 May, when nearly 50% of all HELO data were collected. Profiles were not discarded if they failed to represent an independent observation or in instances whereby consecutive HELO descent legs corresponded to the same model profile. This strategy was adopted to retain consistency in postprocessing methods used by all countries.

The results of 190 HELO observations (Table 5) indicate that the lower atmosphere was on average 1.5 g kg⁻¹ drier and 2.5 K warmer at 112 m than near the surface. The minimum in mean modified refractivity occurred at a height of ∼75 m, suggesting a propensity for surface-based ducting in the observed profiles. In general, all of the models had increasingly cold and wet biases with height, leading to larger rms errors at 112 m than near the surface. The vertical variation of the moisture field largely defined the slope of the modified refractivity in each layer, thus characterizing the layer’s refractivity regime. In the statistics, the models had large differences in the vertical variation of mean temperatures, but all models tended to overestimate the moisture above the surface layer. The largest decrease in mean moisture was given by COAMPS (1.0 g kg⁻¹), followed by the MetUM (0.7 g kg⁻¹), while the remaining models had less than 0.5 g kg⁻¹.

Because mean statistics only give a broad picture of the models’ performance, we further investigated three specific times for individual profile comparisons, so chosen because each contained an observed surface-based duct developing under very different background flow characteristics. Figure 6a shows the SIBL structure that developed in the offshore flow at 1700 UTC 30 April following a frontal passage. All four models predicted the duct despite significant discrepancies in temperature (Fig. 6a). The ducting layer was well represented in the models because it was predominately a function of the relatively dry, warm mixed layer over land that was advected over a near-saturated layer adjacent to the much colder ocean surface.

For 1500 UTC 1 May, the observed profiles depict a strong, shallow inversion and sharp moisture decrease generated by upper-level subsidence (Fig. 6b). Both GEM and COAMPS had larger changes in temperature and humidity with height than did the models forced by a warmer ocean. As evidenced by the time series at the NPS buoy location (Fig. 5), all of the models captured strong ducting associated with high pressure ridging during 1 May but had large fluctuations in duct strength. Temporal fluctuations were commonplace in the models’ time series and also in duct-strength animation loops. COAMPS developed pulses in the strength of the duct after ∼1200 UTC 1 May as veering winds became more southerly aloft. The wind shift caused continental flow to slope down from the topography and to interact with an abruptly lowered MABL across the sharp SST front to the south. This finding suggests that model errors can also arise during synoptic transitions not only because of timing errors, but also because of errors in the interaction of large-scale processes with local mesoscale forcing.

The 1800 UTC 4 May event is a case of shoreward-directed flow, also crossing the warm Gulf Stream to colder nearshore ocean waters. While this period was dominated by high pressure, a cold, dry air mass aloft originated over Wallops from the east. The observed profiles display a complicated vertical structure with a mixed layer and embedded internal boundary layer (Fig. 6c). At this time, the upwind SST distribution and the local sea-breeze return flow contributed to the layering in the lower atmosphere. Each model successfully simulated the cold, dry air aloft; however, COAMPS also predicted a warmer upwind ocean relative to the other models, leading to a well-defined mixed layer and sharp drop in specific humidity, albeit above the level of the HELO profile. None of the models captured the weak surface-based duct present in the observations at this time. We return in the next section to the above three cases to discuss ubiquitous mesoscale features present in simulated ducting.

The mean ducting characteristics from each model at the HELO locations (Table 6) indicate that ducts occurred about one-half of the time they were observed for COAMPS and one-third of the time for the MetUM and GEM. The relatively little ducting depicted by MM5 was due to insufficient observational information in that model run as discussed in section 3. The DFQ percentages are roughly half those reported at the buoy location (Table 2), and the majority of these are surface based. The HELO statistics are more representative of daytime ducting phenomena within the lower MABL being restricted to times and heights of the HELO measurements. In terms of mean DST, the COAMPS overprediction stemmed from an inability to capture the timing of transitions on 1 May, whereas the other models tended to underpredict mean DST because of weaker upper-level subsidence. The cold shelf waters in COAMPS also yielded more stable surface fluxes within 100 km of the coast, thus augmenting the nearshore subsidence and leading to greater DFQ and SBD percentages and stronger ducts with lower DBH in COAMPS.

Statistics derived from a contingency table comprising duct/no duct events from each NWP model reveal that the models all have a greater percentage of missed ducts than hits and, with the exception of COAMPS, have a greater percentage of errors than percent correct (Table 7). The hit rate minus the false-alarm rate yields discrimination scores of 25 or less. The highest discrimination score of 100 is attained when the hit rate is 100% and there are no false alarms. Discrimination scores were somewhat sensitive to the method used for computing ducting statistics. Scores could be improved by removing duplicate profiles, increasing discrimination scores by about 14 points in COAMPS, and by extending model profiles by one level (to include modeled ducts that were slightly deeper than observed). The latter had more of an impact on the coarse grid (discussed in section 6a below), increasing the COAMPS 36-km-grid discrimination score by 9 points. The above changes tended to increase DFQ percentages and hit rates but otherwise had little effect on the mean ducting statistics.

6. Mesoscale structure

To show the gridwide horizontal variability in simulated ducts, we revisit the three profile dates and times presented in Fig. 6. Much of the structure in the 30 April SIBL ducts arose from complexity in the coastline and bays (Fig. 7); however, the nearshore ducting in GEM was less extensive than in the other models. This feature was likely related to GEM’s deeper surface layer, perhaps because of their coarse vertical resolution. Ducting also occurred over the SST front at this time. In the wake of the frontal passage, upper-level subsidence moved southeastward across the region, forming ducts where the moist, mixed layer over the warm ocean surface met with the advancing subsidence. Because the subsidence was more pronounced in the simulations with colder shelf waters, and more moisture could be fluxed into the MABL by a warm Gulf Stream, GEM and especially COAMPS had stronger ducting across the SST front.

In the alongshore case of 1 May, warm, dry air in the free atmosphere combined with surface southerlies around the backside of a high pressure cell located to the southeast of the study area. The duct-strength maps in Fig. 8 reveal substantial differences between models and within a given model over very short distances that can be explained by differences in synoptic transitions and in SST. Veering winds promoted ducting across much of the ocean domain aided by the large, downward sensible heat flux on this day. Ducting was inhibited by upward sensible heat flux over warm-water regions. The absence of any appreciable ducting in MM5 at this time clearly demonstrated the importance of correcting mesoscale model fields with observations or reinitializing to obtain a more accurate description of the large-scale environment. When combined with their coarsely resolved SST, little vertical structure developed in that model as depicted in their profiles (Fig. 6b).

In the other models, the abrupt transition to ducting occurred downwind of the surface stability change, driven by the collapse of the MABL and reduction in turbulent mixing over colder ocean. This same mechanism was responsible for strong coastal ducting south of Wallops Island in GEM and COAMPS. Away from the stable surface flux transition, inversions lifted and weakened, diminishing or eliminating ducts along the coast north of Wallops Island. Shallow inversions were reestablished in COAMPS over the cold waters offshore of Delaware Bay and to the northeast, increasing duct strengths there. The MetUM synoptic transition was more rapid than the other models, having already advected a deep layer of warm, moist air into the southern portion of the domain, virtually eliminating their ducting south of Wallops Island. Because of the rapid evolution of ducting events that contain substantial mesoscale structure, even subtle timing or position errors can negatively characterize model performance when compared with observations at specific locations and times.

For the onshore case of 4 May, the upwind SST created a deep MABL capped by large-scale subsidence. Differences in ducting distributions were substantial on this day (Fig. 9) and were largely related to the upwind SST east of the 4-km grid’s eastern boundary. The Gulf Stream SST front in that region created a deep, moist mixed layer that resulted in strong, elevated ducting over most of the ocean domain in COAMPS. Ample mesoscale structure resulted from changes in MABL depth and inversion strength as the layer approached the coast and interacted with the local sea-breeze return flow. While the elevated duct was predicted by MetUM over the warm-water regions, it lacked subsidence closer to shore to sustain ducting there. In contrast, weak near-surface ducting developed in GEM and MM5 only over their cold-water regions, but neither model’s simulation contained enough of a mixed layer to support the elevated ducting offshore.

7. Conclusions

This modeling study examines the 7-day Wallops-2000 IOP data collected off the eastern shores of the Delmarva Peninsula in April and May of 2000. The primary observational dataset used to evaluate four mesoscale model simulations (COAMPS, MetUM, MM5, and GEM) included instrumented fixed-buoy time series and helicopter vertical profiles. This dataset provided a unique description of the temporal and spatial changes in atmospheric ducting associated with rapidly evolving synoptic and mesoscale forcing. Comparisons of simulated fields from each model’s 4-km grid were made with observed meteorological conditions and with the mean diagnosed modified refractivity and ducting characteristics for the full 7-day period. The study included traditional verification methods such as mean, bias, and rms error statistics. Combined with ducting-event contingency table statistics, these metrics provide the basis by which we assessed each model’s simulation and the effects of different grid resolutions and sensitivity tests.

From analysis of observed data in conjunction with the models’ predictions, we found that the broad-scale ducting patterns followed the general synoptic–refractivity model of Rosenthal and Helvey (1979), in which favorable ducting conditions occurred during periods of high pressure due to upper-level subsidence and unfavorable conditions occurred during periods of negative sea level pressure tendency indicating approaching low pressure systems. We used the models to explore commonality in the simulated ducting patterns to gain an understanding of important sensitivities of atmospheric refractivity to the large-scale and mesoscale forcing present in each modeling system. The study was designed at the outset to have each country use their complete forecasting system, including model initialization procedures (global fields, data assimilation, and pre-forecast-period spinup) and boundary conditions (global fields and SST forcing). Sensitivity tests were then conducted to show the impact of a longer preforecast spinup and a high-resolution, evolving SST. Additional comparative analyses of ducting structures were made for each of four horizontal grid resolutions, ranging in grid spacing from 36 to 1.333 km.

The four models were found to generally overpredict the mean moisture above the surface layer, resulting in a weaker vertical gradient in specific humidity and thus producing fewer and weaker MABL ducts than were observed. However, some of the errors resulted from either timing or position inaccuracies in the simulated ducting layers, neither of which reflects well upon model performance. Improved ducting was achieved by earlier initialization of mesoscale simulations in MetUM and by using the high-resolution, twice-daily updated NCODA SST as a lower boundary condition in MetUM and GEM. Improvements to COAMPS were made by eliminating a problematic moisture correction on 4 May from the single Wallops Island sounding in the MVOI data assimilation scheme.

Considerable mesoscale variability was present in simulated ducting events during Wallops-2000 IOP. Using case study examples of offshore, alongshore, and onshore flow, we illustrated the complexities of the meteorological conditions in this region and their effect on atmospheric refractivity and ducting. Events characterized by weak background winds had ample mesoscale structure in ducted layers. The horizontal variability was associated with changes in the strength of gradients defining the layer and in their depth. More abrupt transitions occurred across changes in surface stability. Mesoscale forcing evolved from the spatially complex SST by the development of surface layer and MABL structure over and downwind of the meandering Gulf Stream and cold shelf waters. In the coastal zone, interactions with sea-breeze return flows induced additional mesoscale variability in ducted layers within 50–100 km of shore (4 May). On other days, topographically induced local pressure gradients developed over the diurnal cycle (29 April), and, during a period of rapid synoptic transition, oscillations in the free atmosphere were excited (1 May), both affecting the timing and location of coastal ducts. Analysis of COAMPS’s four grids showed that the mesoscale features and intricate vertical structure were only possible for grid spacing of 4 km or higher and also required a concurrent high-resolution SST analysis.

The findings from the initial simulations and subsequent sensitivity tests revealed some of the most critical aspects of mesoscale model systems necessary for simulating atmospheric refractivity and ducting. Listed in order of importance as established by this modeling study, these include but are not limited to

• accurate large-scale forcing in initial fields and at lateral boundaries,

• grid spacing of at least 5 km in the horizontal plane and an average of 60 m or better in the lowest 1 km of the atmosphere in the vertical direction,

• mesoscale structure retained in analysis or allowed to spin up on finer grids,

• accurate and evolving SST fields of equivalent resolution to the model grid, and

• 3D- and 4DVAR data assimilation techniques for proper moisture analysis.

Some aspects of the modeling intercomparison remain for future work. A study of model physics, numerics, and dynamics could be made by initializing all models with the same global fields, and an evaluation of the propagation measurements during Wallops-2000 will help to determine the degree to which predicted refractive layers yield accuracy in microwave radar signals. Of more importance, new observations are currently being analyzed from a recent field campaign in the Bay of Plenty, New Zealand. Over a span of 14 days, the four ABCANZ countries archived the 4-km-grid forecasts generated in near–real time utilizing their typical operational configuration, providing an opportunity for further evaluation of national defense forecasting capabilities of mesoscale coastal refractivity.