Abstract

This study initiates the application of the maturing Weather Research and Forecasting (WRF) model to the polar regions in the context of the real-time Antarctic Mesoscale Prediction System (AMPS). The behavior of the Advanced Research WRF (ARW) in a high-latitude setting and its ability to capture a significant Antarctic weather event are investigated. Also, in a suite of sensitivity tests, the impacts of the assimilation of Moderate Resolution Imaging Spectroradiometer (MODIS) atmospheric motion vectors on ARW Antarctic forecasts are explored. The simulation results are analyzed and the statistical significance of error differences is assessed. It is found that with the proper consideration of MODIS data the ARW can accurately simulate a major Antarctic event, the May 2004 McMurdo windstorm. The ARW simulations illuminate an episode of high-momentum flow responding to the complex orography of the vital Ross Island region. While the model captures the synoptic setting and basic trajectory of the cyclone driving the event, there are differences on the mesoscale in the evolution of the low pressure system that significantly affect the forecast results. In general, both the ARW and AMPS’s fifth-generation Pennsylvania State University–National Center for Atmospheric Research Mesoscale Model (MM5) tend to underforecast the wind magnitudes, reflecting their stalling and filling of the system near Ross Island. It is seen, however, that both targeted data assimilation and grid resolution enhancement can yield improvement in the forecast of the key parameter of wind speed. It is found that the assimilation of MODIS observations can significantly improve the forecast for a high-impact Antarctic weather event. However, the application to the retrievals of a filter accounting for instrument channel, observation height, and surface type is necessary. The results indicate benefits to initial conditions and high-resolution, polar, mesoscale forecasts from the careful assimilation of nontraditional satellite observations over Antarctica and the Southern Ocean.

1. Introduction

The Weather Research and Forecasting (WRF) model (Skamarock et al. 2005) has been developed as a next-generation mesoscale modeling system for both operational prediction and atmospheric research. The nonhydrostatic WRF (information available online at http://www.wrf-model.org/index.php) has arguably the largest user base of any current mesoscale model and is replacing the fifth-generation Pennsylvania State University–National Center for Atmospheric Research (PSU–NCAR) Mesoscale Model (MM5; Grell et al. 1995) in that user community.1 It is being used by official forecasting centers [e.g., the National Centers for Environmental Prediction (NCEP)], while seeing research applications ranging from large eddy simulations, to severe weather at convection-permitting resolutions, to tropical cyclogenesis, to regional climate modeling (see, e.g., Moeng et al. 2007; Davis et al. 2006; Done et al. 2006).

The developmental applications of WRF have primarily been in the midlatitudes, and to date the high latitudes have been largely ignored. In light of this, and given the growing use of WRF worldwide and the maturation of the model’s capabilities, this study initiates the application of WRF to Antarctica. Although addressing the interest of the WRF user community in the model’s behavior in the high latitudes, the more specific concern is investigating its ability to capture a significant polar weather event. The target is the May 2004 windstorm at McMurdo Station, Antarctica (Powers et al. 2005; Steinhoff and Bromwich 2005). While this was an extraordinary episode that inflicted winds of up to 71 m s−1 (139 kt) on the main American research base in Antarctica (McMurdo Station, Fig. 1), the accurate forecasting of surface flow in Antarctica is important for a number of reasons. The lifeline of flight operations depends on reliable wind forecasts (see, e.g., Holmes et al. 2000), as flows beyond specified thresholds can exceed crosswind tolerances for takeoffs and landings and blowing snow can obscure runways. Both situations are of critical importance in the case of aircraft that absolutely must land at McMurdo after passing their points of safe return. Second, activities in the harsh polar field may face life-threatening conditions dependent on the strength and duration of surface winds. Third, operations at the focal facility of McMurdo may simply be shut down under excessive velocities (see footnote 4). Beyond these concerns is the need to be able to simulate and analyze, for both forecasting and research purposes, the ubiquitous katabatic and other outflows endemic to Antarctica. A prime example is the Ross Ice Shelf Air Stream (RAS; Bromwich and Parish 2002).2

Fig. 1.

AMPS domains and Antarctic locations. (a) 90- (60) and 30- (20) km grids. (b) 30- (20) km grid with 10- (6.7) and 3.3- (2.2) km grids inset. (c) 10- (6.7) km grid with 3.3- (2.2) km grid inset. Dots mark observation/AWS sites discussed in the text. Grid spacings in parentheses refer to the 60-/20-/6.7-/2.2-km (MOD1_60) setup.

Fig. 1.

AMPS domains and Antarctic locations. (a) 90- (60) and 30- (20) km grids. (b) 30- (20) km grid with 10- (6.7) and 3.3- (2.2) km grids inset. (c) 10- (6.7) km grid with 3.3- (2.2) km grid inset. Dots mark observation/AWS sites discussed in the text. Grid spacings in parentheses refer to the 60-/20-/6.7-/2.2-km (MOD1_60) setup.

The May 2004 case serves as a vehicle for the first-time Antarctic application of the Advanced Research WRF (ARW; Skamarock et al. 2005). Sharing the WRF software framework, this features the ARW dynamics solver (originally referred to as the “Eulerian mass core”) and its mass-based vertical coordinate, a grid-nesting capability, numerous physical process schemes, and the WRF-Var (variational) data assimilation system (Barker et al. 2004). In addition to beginning to understand and advance the capability of the ARW for polar prediction, this work is concerned with the improvement of a real-time Antarctic numerical weather prediction (NWP) facility known as the Antarctic Mesoscale Prediction System (AMPS; Powers et al. 2003). Funded by the National Science Foundation, AMPS is an experimental mesoscale modeling system that provides forecast guidance in support of the flight, scientific, and logistical activities of the U.S. Antarctic Program (USAP) and international Antarctic efforts. While originally established to improve the numerical guidance available to the USAP forecasters3 at McMurdo Station, over the years AMPS has expanded to serve a broad range of international groups and activities (including emergency rescues) across Antarctica (see, e.g., Monaghan et al. 2003). Historically AMPS has relied on the MM5 (Powers et al. 2003; Bromwich et al. 2003), but it has begun also to employ the ARW. (AMPS forecasts may be accessed online at http://www.mmm.ucar.edu/rt/amps/wrf_pages.)

A challenge to Antarctic NWP is the lack of conventional (e.g., surface observational and radiosonde) data over the Southern Ocean and the continent. Antarctica has approximately 10 radiosonde sites, and, except for the Amundsen–Scott Station at the South Pole, these are situated around the continent’s edge. Satellite measurements can populate the data void, however, and a promising source is the Moderate Resolution Imaging Spectroradiometer (MODIS) instrument aboard the National Aeronautics and Space Administration (NASA) Terra and Aqua polar-orbiting platforms. Over the polar regions MODIS imagery can yield atmospheric motion vectors (AMVs; Velden et al. 2005; Key et al. 2003) offering vector winds at varying heights. Available in near–real time, MODIS polar winds have shown promise in improving performance metrics in global NWP models (Key et al. 2003; Bormann and Thépaut 2004). Given the potential for these measurements to enhance polar forecasting, this study also investigates the impact of the assimilation of MODIS winds on the skill of a mesoscale model, (viz. the ARW) in the Antarctic.

This investigation thus provides the growing WRF user community with an initial test and analysis of this new modeling capability in the polar regions. The May 2004 McMurdo windstorm is the vehicle for the ARW’s application to a challenging forecast for this critical area, for the ARW’s comparison with the MM5 in AMPS, and for sensitivity experiments to explore model performance with different MODIS data assimilations and grid resolutions. The foci are how well the ARW can simulate a high-impact polar weather event and whether mesoscale ARW forecasts in AMPS may be improved by MODIS data assimilation. Section 2 of this paper describes the observed event, while section 3 covers the ARW configuration and experiments. Section 4 examines the ARW event simulations and analyzes the MODIS sensitivity experiments. Section 5 pursues a statistical evaluation of the results, and section 6 presents a summary and conclusions.

2. The May 2004 McMurdo windstorm

On 15 May 2004 extreme winds battered the McMurdo Station, Antarctica region (see Fig. 1 for maps of Antarctica and the McMurdo area) and locked down base activities as “condition 1” status4 was declared. The event winds, out of the south, were sustained at over 44 m s−1 (86 kt) and exceeded 52 m s−1 (102 kt) in gusts in town. One building on a hill above the base center claimed gusts to 71 m s−1 (139 kt). The winds pounded structures and equipment, blew in doors, tore up roofs, and peeled siding off dormitories. The velocities hit abruptly, and the peak hours were 1800 UTC 15 May–0000 UTC 16 May 2004. Figure 2 presents time series of wind speeds at sites in the area. A number of the observational records from the vicinity are truncated, reflecting the winds’ damaging of the meteorological instruments [e.g., Crater Hill and Helo Pad (helicopter pad, in town)].

Fig. 2.

Observed wind speeds (m s−1) at sites in the McMurdo area for 0000 UTC 15 May–0000 UTC 17 May 2004. Cosray, Arrival Heights, Helo Pad, and Crater Hill are all in the immediate McMurdo vicinity; Black Island AWS is about 34 km to the south.

Fig. 2.

Observed wind speeds (m s−1) at sites in the McMurdo area for 0000 UTC 15 May–0000 UTC 17 May 2004. Cosray, Arrival Heights, Helo Pad, and Crater Hill are all in the immediate McMurdo vicinity; Black Island AWS is about 34 km to the south.

The event was motivated by the passage of a deep synoptic low pressure system to the east of Ross Island. Figure 3 shows the track of this system, based on the analysis of satellite imagery.5 Over 14–15 May 2004, after having moved in from the Amundsen Sea, the center traveled from Marie Byrd Land, across the Siple Coast, and onto the Ross Ice Shelf (Fig. 1b). The presence of the low in the Siple Coast region is consistent with that area exhibiting a maximum in cyclone density for the winter months (Simmonds et al. 2003). After crossing 180° at around 1500 UTC 15 May 2004, the low turned northward. Seasoned operational forecasters at McMurdo know that that area is subject to its strongest winds when low pressure systems track from the south over the ice shelf and remain east of Ross Island (R. Hennig, Space and Naval Warfare Systems Center, 2005, personal communication; see also Holmes et al. 2000).

Fig. 3.

Track of the observed low. Times (UTC) of central position of low (marked “L”) indicated. Italic L reflects positions during the observed wind event at McMurdo.

Fig. 3.

Track of the observed low. Times (UTC) of central position of low (marked “L”) indicated. Italic L reflects positions during the observed wind event at McMurdo.

Figure 4 presents IR imagery of the system. At 1455 UTC 15 May 2004 (Fig. 4a) the low sits in midshelf. The 2125 UTC image reveals the center during the event, south and east of McMurdo, positioned east of Minna Bluff and White Island (Fig. 4b). The system propagates northward east of Ross Island, and after 0000 UTC 16 May 2004 it weakens and moves across Franklin Island to the Terra Nova Bay area.

Fig. 4.

IR satellite imagery (MODIS, 5 km) for Ross Sea sector, 15 May 2004. The L indicates the center of the surface low: (a) 1455 and (b) 2125 UTC. MB and WI indicate Minna Bluff and White Island.

Fig. 4.

IR satellite imagery (MODIS, 5 km) for Ross Sea sector, 15 May 2004. The L indicates the center of the surface low: (a) 1455 and (b) 2125 UTC. MB and WI indicate Minna Bluff and White Island.

While the migrating cyclone is essential for the winds in the McMurdo area, the forcing pressure gradient, enhanced on the mesoscale, is a reflection of the synoptic conditions established by the low and a stationary high pressure system over East Antarctica. This broader picture appears in Figs. 5 and 6, which present surface and 500-hPa analyses for 15–16 May 2004. These analyses are from the European Centre for Medium-Range Weather Forecasts (ECMWF) global model. Note in Fig. 5a the strong (>1032 hPa) surface high over East Antarctica at 1200 UTC 15 May 2004. At this time the surface low sits downstream of the 500-hPa trough over the western the Ross Ice Shelf; that trough is rotating around the upper-level cutoff over the Ross Sea (Figs. 5a and 6). Sea level pressures over the Ross Sea sector are low relative to the East Antarctic high. Because of the potential for cumulative error in deriving below-ground SLP analyses over the elevated Antarctic continent, Fig. 5b presents the 600-hPa height analysis to confirm the synoptic gradient.

Fig. 5.

(a) SLP and (b) 600-hPa heights at 1200 UTC 15 May 2004 from ECMWF analyses. The letters L and H in (a) mark the surface low and the East Antarctica high and in (b) mark the upper-level Ross Sea and East Antarctica centers. Contour interval is 6 hPa in (a) and 60 gpm in (b).

Fig. 5.

(a) SLP and (b) 600-hPa heights at 1200 UTC 15 May 2004 from ECMWF analyses. The letters L and H in (a) mark the surface low and the East Antarctica high and in (b) mark the upper-level Ross Sea and East Antarctica centers. Contour interval is 6 hPa in (a) and 60 gpm in (b).

Fig. 6.

500-hPa heights and winds from ECMWF analyses. Heights (solid lines); contour interval is 60 gpm. Wind vectors (arrows); magnitudes are 25 m s−1 (vector length interval)−1. Dashed line marks trough axis. (a) 1200 UTC 15 May and (b) 0000 UTC 16 May 2004.

Fig. 6.

500-hPa heights and winds from ECMWF analyses. Heights (solid lines); contour interval is 60 gpm. Wind vectors (arrows); magnitudes are 25 m s−1 (vector length interval)−1. Dashed line marks trough axis. (a) 1200 UTC 15 May and (b) 0000 UTC 16 May 2004.

The synoptic-scale pressure gradient across the ice shelf, the Transantarctic Mountains (TAM), and East Antarctica is thus established by the high and the migrating low, and the wind event at McMurdo attends the imposition of an enhanced mesoscale gradient on the Ross Island region. The axis of the 500-hPa trough (dashed line in Fig. 6) remains upstream of the surface low until approximately 0000 UTC 16 May 2004. This offers upper-level support through divergence and positive vorticity advection aloft in accordance with quasigeostrophic theory (see, e.g., Holton 1992; Bluestein 1992; Carlson 1998).

3. Model configurations and experiments

The model experiments here employ the ARW, version 2 (Skamarock et al. 2005), while the real-time AMPS forecast for the event used the MM5. Figures 1a,b show the grids for the experiments. The primary four-domain configuration has horizontal grid spacings of 90 km (Southern Hemisphere/Southern Ocean), 30 km (Antarctic continent), 10 km (western Ross Sea), and 3.3 km (Ross Island; “90/30” setup). All nesting is two-way interactive. One higher-resolution ARW experiment is run with the same grid areas as in Fig. 1, but with 60-, 20-, 6.7-, and 2.2-km spacings (“60/20” setup). In the vertical the model is run with 31 half-levels with a model top at 50 hPa. The initialization time is 0000 UTC 15 May 2004. The 90/30 grid configuration employed here has been chosen to be consistent with that of the real-time AMPS running at the time of the event.

The experiments enlist the WRF single-moment, five-species microphysics scheme (WSM5) (Hong et al. 2004), the Mellor–Yamada–Janjic (aka Eta) planetary boundary layer (PBL) scheme (Mellor and Yamada 1982; Janjic 2002), and the Kain–Fritsch cumulus parameterization (Kain and Fritsch 1993; see also Skamarock et al. 2005). The 3.3-km (2.2 km) domain is run fully explicit. This scheme configuration is chosen because the counterpart package was that which was run successfully in AMPS (see Bromwich et al. 2005). The Noah land surface model (Chen and Dudhia 2001) is also active. Capabilities to ingest sea ice analyses and to represent sea ice concentration fractionally in grid cells do not exist at present in the ARW. Thus, for these experiments full sea ice coverage is assumed at oceanic points where the skin temperature is less than 271.4 K.

Initial and boundary conditions are derived from the NCEP Global Forecast System (GFS) output. In the data assimilation experiments, the GFS first-guess field is reanalyzed with observations using the WRF-Var data assimilation system (Barker et al. 2004; Skamarock et al. 2005). WRF-Var offers a three-dimensional variational data assimilation (3DVAR) capability within the WRF software framework. It is run on the outer (90/30 and 60/20 km) grids, as in the real-time AMPS. The background error covariances were generated via the NMC method. For the assimilation, the observations referred to as “standard” consists of the conventional data acquired from the Global Telecommunication System (GTS) circuit and used in a regular AMPS run: reports from manned surface stations (e.g., SYNOP, METAR), surface automatic weather stations (AWSs), upper-air stations, ships, buoys, pilot and Aircraft Meteorological Data Relay (AMDAR) reports, and (geostationary) satellite cloud-track winds.

The real-time AMPS MM5 grid configuration featured the 90-, 30-, 10-, and 3.3-km domains shown in Fig. 1, as well as 10-km grids over the South Pole and the Antarctic Peninsula (not shown). The MM5 was run with 31 half-σ levels from the surface to 50 hPa, and its initial and boundary conditions were derived from the GFS. WRF-Var assimilated the standard GTS observations. Sea ice analyses from the National Snow and Ice Data Center (NSIDC) initialized the sea ice coverage.

Note that AMPS employs the “Polar MM5” (Bromwich et al. 2001; Cassano et al. 2001). This is a modified version of the model that was developed by the Polar Meteorology Group of the Byrd Polar Research Center (at The Ohio State University) and contains adjustments to improve performance in the polar regions and to better capture features unique to extensive ice sheets. The Polar MM5 modifications include accounting for a separate sea ice category with specified thermal properties, use of forecast cloud species in the radiation scheme, representing fractional sea ice coverage in grid cells, using the latent heat of sublimation for calculations of latent heat fluxes over ice surfaces, and assuming ice saturation when calculating surface saturation mixing ratios over ice.

This study also investigates the assimilation of MODIS wind data on ARW forecasts of a significant Antarctic weather episode. This is motivated by the potential of MODIS wind measurements for high-latitude forecast improvement, given their focus on the relatively data-sparse polar regions, and the absence of an examination of MODIS data’s impacts on a high-resolution, mesoscale model simulation of a critical event (cf. global models or period studies; e.g., Key et al. 2003; Velden et al. 2005). The MODIS wind data are produced from sequential imagery tracking of height-assigned cloud and water vapor features detected by the MODIS instrument’s infrared (IR) and water vapor (WV) channels (Velden et al. 2005; Key et al. 2003). For this study, the Cooperative Institute for Meteorological Satellite Studies (CIMSS; at the University of Wisconsin) MODIS data are used. Examining the assimilation of a 30-day MODIS dataset on the ECMWF global model and the NASA Data Assimilation Office model, Key et al. (2003) found significant improvements in skill scores for geopotential height forecasts. In a subsequent study, Bormann and Thépaut (2004) found that the incorporation of MODIS data using four-dimensional variational data assimilation (4DVAR) had a positive effect on medium-range forecasts by the ECMWF global model. Neither of these studies, however, dissected case impacts in a high-resolution mesoscale model forecast.

We performed ARW experiments involving the assimilation of the conventional (or standard) AMPS observations and the CIMSS MODIS data. The following tests are conducted:

  • CTRL—No data assimilation;

  • STD—Standard AMPS data only;

  • ALL—Standard AMPS data plus all MODIS data;

  • MOD1—Standard AMPS data plus filtered MODIS data; and

  • MOD1_60—As in MOD1, but for a 60-, 20-, 6.7-, and 2.2-km domain setup.

CTRL is a run involving no data assimilation. In STD the standard AMPS data only are assimilated. ALL entails the incorporation of all standard data plus all of the MODIS observations in an assimilation window around 0000 UTC 15 May 2004. There is no exclusion of any MODIS data apart from the possible rejection by WRF-Var’s quality control (QC) criteria.

In MOD1 a subset of MODIS observations, one remaining after the application of a filter, is used with the standard data. Reflecting lower confidence in the estimates in certain regimes, the filtering follows the suggestion of Key et al. (2003), later applied by Bormann and Thépaut (2004), in which retrieval height, surface type, and source MODIS channel (IR or WV) are considered in accepting a measurement. Specifically, the probability of poorer-quality retrievals below certain levels is the reason for the restrictions (Key et al. 2003). For MOD1, the filtering criteria are as follows: over land, both IR and WV data above 400 hPa are retained; over ocean, IR data above 700 hPa and WV data above 550 hPa are retained; other measurements are rejected. A number of operational forecasting centers (e.g., ECMWF, the Met Office, Japan Meteorological Agency, Deutscher Wetterdienst) have adopted modifications of these based on their own models and experience (see, e.g., Forsythe and Berger 2004). For the experiments here, the filtering reduces the number of measurements considered by approximately 33%.

The impact of the filtering may be seen by comparing the MODIS data used in ALL and MOD1 and the difference in their initializations. Figures 7a,b reveal how the filtering reduces the number of the MODIS winds, at all heights, considered for ingest in MOD1 (e.g., over East Antarctica). Aloft (Fig. 7c), the differences are manifested in ALL’s greater 500-hPa heights over Queen Maud Land and East Antarctica and lower heights over West Antarctica. The positive height difference over Queen Maud Land is collocated with 500-hPa low centers in both analyses (not shown) and reflects a slightly weaker upper-level low in ALL (ALL heights 22 gpm higher than a 4994-gpm low center in MOD1). The maximum wind differences at this level reach about 10 m s−1.

Fig. 7.

MODIS observations for ALL and MOD1 and 500-hPa height/wind analysis differences for ALL and MOD1. (a) MODIS data available for ALL. Circles mark wind data points at all heights. (b) Same as in (a), but for MOD1. (c) Difference in 500-hPa analyses (ALL — MOD1) for 0000 UTC 15 May 2004. Positive (negative) geopotential height differences are indicated by solid (negative) contours; interval is 5 gpm. Wind difference vectors (arrows); magnitudes are 8.5 m s−1 (vector length interval)−1. Maximum vector magnitude is ≈10 m s−1.

Fig. 7.

MODIS observations for ALL and MOD1 and 500-hPa height/wind analysis differences for ALL and MOD1. (a) MODIS data available for ALL. Circles mark wind data points at all heights. (b) Same as in (a), but for MOD1. (c) Difference in 500-hPa analyses (ALL — MOD1) for 0000 UTC 15 May 2004. Positive (negative) geopotential height differences are indicated by solid (negative) contours; interval is 5 gpm. Wind difference vectors (arrows); magnitudes are 8.5 m s−1 (vector length interval)−1. Maximum vector magnitude is ≈10 m s−1.

A final experiment, MOD1_60, mimics MOD1, but has horizontal grid spacings enhanced 33%: its grids are 60, 20, 6.7, and 2.2 km. This experiment investigates the impact of such enhanced resolution on the forecast of a major weather event in the Ross Island region. The effectiveness of such resolution increase has been a question of interest to the USAP forecasters and international users of AMPS.

4. Model results

Satellite and surface observations from the McMurdo area for the 15 May 2004 event indicate the passage of a deep cyclone through the region, with local AWS data showing minimum pressures (<960 hPa) occurring from 1200 to 1600 UTC 15 May 2004. After 1500 UTC, with the center passing east of Ross Island (Fig. 3), the trajectory results in the mesoscale enhancement of the pressure gradient and the barrier influence of the TAM, producing intense southerly flow through the McMurdo region.

The ARW experiments and the AMPS MM5 forecast simulate the transit of a strong low pressure system from Marie Byrd Land across the ice shelf (Fig. 8). While the model trajectories on the synoptic scale are similar to that observed, meso-β scale (Orlanski 1975) differences from the observed evolution in the Ross Island region have a crucial bearing on the wind event. Consider first the AMPS MM5 forecast (Fig. 8f). The MM5 low is driven too quickly westward, to south of Minna Bluff by 1500 UTC 15 May 2004, when in contrast the observed system was still in midshelf. The center stalls east of Minna Bluff at 1800 UTC, and filling follows. This results in a cyclone that is rapidly weakening as it approaches Ross Island. While the AMPS system persists north of Ross Island, the low is but a remnant.

Fig. 8.

Tracks of ARW and AMPS MM5 lows. Times (UTC) of the central position of the low (marked L) are indicated. Italic L reflects the time during the observed wind event at McMurdo. (a) CTRL, (b) STD, (c) ALL, (d) MOD1, (e) MOD1_60, and (f) AMPS MM5.

Fig. 8.

Tracks of ARW and AMPS MM5 lows. Times (UTC) of the central position of the low (marked L) are indicated. Italic L reflects the time during the observed wind event at McMurdo. (a) CTRL, (b) STD, (c) ALL, (d) MOD1, (e) MOD1_60, and (f) AMPS MM5.

In the WRF results in CTRL (no data assimilation; Fig. 8a) one sees a system driven too far westward, into the coast south of Minna Bluff. In STD (Fig. 8b), the counterpart to the AMPS MM5 forecast in terms of data assimilated, the low remains east of Minna Bluff. This shift compared to CTRL will be reflected in an improved wind event depiction. ALL (Fig. 8c) produces the poorest results, sending the low into Minna Bluff where it fills in situ. The filtered MODIS experiments MOD1 and MOD1_60 (Figs. 8d,e) yield the low track and evolution closest to observation. The system turns northward in the best agreement with the observed motion, and the simulations correctly center the low off of Ross Island’s east end at 2300–0000 UTC 15–16 May 2004.

As seen in the low positions after 1200 UTC (hour 12), all of the ARW runs move the system too swiftly across the shelf. This westward bias in the experiments is the largest in CTRL. ALL’s track performance suffers compared to both MOD1 and STD, suggesting that perhaps MODIS observations should be filtered prior to assimilation (MOD1), and that if not, it may be preferable to exclude them (STD).

Sea level pressure (SLP) analyses from MOD1 give a fuller picture of the path and evolution of the system. The run initializes with a 941-hPa center in Marie Byrd Land (Fig. 9a) By hour 12 (1200 UTC) the cyclone (955 hPa) has traveled to midshelf, sitting over the 180° meridian (Fig. 9b). The packing of contours poleward and southwest of the center is associated with the strongest surface winds, which are 20–30 m s−1 south of the center on the shelf boundary. The localized maximum along the TAM is consistent with an enhanced, barrier flow. Adams (2005) examined this case with the University of Wisconsin Nonhydrostatic Modeling System and concluded the enhanced flow and an attendant surface cold wind surge along the TAM to have characteristics of a trapped wave. An analysis of the wind event at McMurdo, however, was not the focus of that broader investigation, which targeted the RAS.

Fig. 9.

SLP from MOD1. (a), (b) Window of 30-km domain; (c), (d), (e) 10-km domain. Contour interval in (a), (b) is 2 hPa to 1000 hPa, 4 hPa above; contour interval in (c), (d), (e) is 1 hPa to 1000 hPa, 2 hPa above. South Pole is denoted by SP. (a) Hour 0 (0000 UTC 15 May 2004), (b) hour 12 (1200 UTC), (c) hour 18 (1800 UTC), (d) hour 21 (2100 UTC), and (e) hour 23 (2300 UTC).

Fig. 9.

SLP from MOD1. (a), (b) Window of 30-km domain; (c), (d), (e) 10-km domain. Contour interval in (a), (b) is 2 hPa to 1000 hPa, 4 hPa above; contour interval in (c), (d), (e) is 1 hPa to 1000 hPa, 2 hPa above. South Pole is denoted by SP. (a) Hour 0 (0000 UTC 15 May 2004), (b) hour 12 (1200 UTC), (c) hour 18 (1800 UTC), (d) hour 21 (2100 UTC), and (e) hour 23 (2300 UTC).

By hour 18 (1800 UTC) the low, at 967 hPa, has reached the Ross Island region and sits east of Minna Bluff (Fig. 9c). An enhanced pressure gradient lies to the south and west of the system (i.e., toward the coast/TAM). The strongest surface winds are in this zone, with speeds of 20–30 m s−1 over the shelf and over 40 m s−1 in the mountains. Marilyn AWS (79.95°S, 156.13°E; Fig. 1c) is in this area and recorded winds of approximately 24 m s−1 from 180° at this time, comparing well to MOD1’s simulation of 26 m s−1 from 169°.

The model’s slowing and filling of the cyclone in the Ross Island area are evident in MOD1 at hour 21 (Fig. 9d). The center has moved but little to the north from its hour-18 position shelfward (east southeast) of Minna Bluff, and the central pressure has risen 10 hPa to 977 hPa. The strong SLP gradient to the south and southwest is still present, however. Note that the gradient along the south face of Ross Island is a fingerprint of stagnating flow and higher surface pressure on the windward side of this obstacle, the southern embayment of Windless Bight. As documented in O’Connor and Bromwich (1988), static stability in the PBL air encountering the steep topography results in a stagnation zone with relatively high pressure. At hour 23 (Fig. 9e) the low is losing its identity, having further filled to 984 hPa. While all of the experiments display this rapid cyclolysis, satellite imagery shows that the observed system maintained a circulation through 1105 UTC 16 May 2004.

The ARW’s ability to recreate the synoptic low is thus apparent. The further question for forecasting, however, is how well the event winds are simulated. Figure 2 presents time series of wind speeds in the McMurdo region (locations in Fig. 1c). The sites around McMurdo proper are Helo Pad, Arrival Heights, Crater Hill, and Cosray, while Pegasus North and Black Island lie to the south. Arrival Heights and Pegasus North are chosen here for highlighted analysis.6 As for verification of wind directions (not presented), the model event winds are, as observed, consistently southerly.

Figure 10 presents the results for Arrival Heights. Note first that all of the simulations delay and underpredict the winds representing the event at this location. CTRL (Fig. 10a), for example, misses the onset and magnitude of the southerly blast. CTRL’s strongest wind is 21.6 m s−1, compared to a measured 48 m s−1. In STD (Fig. 10b) the assimilation of standard AMPS data improves the flow intensity, which reaches 28 m s−1. ALL (Fig. 10c) simulates conditions less accurately, both in terms of the maximum velocities (26 m s−1) and their duration and in the poorer correspondence of the early wind trace (0000–1800 UTC 15 May 2004).

Fig. 10.

Observed (solid) vs ARW (dashed) wind speed (m s−1) at Arrival Heights. Abscissa shows the time in hours from 0000 UTC 15 May 2004. Missing observed values are not plotted. (a) Observed vs CTRL, (b) observed vs STD, (c) observed vs ALL, (d) observed vs MOD1, and (e) observed vs MOD1_60.

Fig. 10.

Observed (solid) vs ARW (dashed) wind speed (m s−1) at Arrival Heights. Abscissa shows the time in hours from 0000 UTC 15 May 2004. Missing observed values are not plotted. (a) Observed vs CTRL, (b) observed vs STD, (c) observed vs ALL, (d) observed vs MOD1, and (e) observed vs MOD1_60.

Among the 90/30 runs MOD1 (Fig. 10d) best approximates the intensity, timing, and duration of the high winds. The strong flow arrives at 2000 UTC and hits 33.6 m s−1. The higher-resolution version of MOD1, MOD1_60 (Fig. 10e), yields a somewhat improved profile, with a truer pre-event trace (1200–1800 UTC), a stronger, slightly longer event, and a maximum of 35.5 m s−1. In CTRL and STD the event is about 5–6 h delayed; ALL reduces this to 4 h; MOD1 and MOD1_60 exhibit the least delay, about 2 h.

Figures 11a–e present the traces for Pegasus North. CTRL (Fig. 11a) produces an event at Pegasus, but underforecasts and delays it. The maximum simulated wind speed is 24.6 m s−1, compared to 39.6 m s−1 observed. In contrast to CTRL, STD produces a significantly stronger event (Fig. 11b). After a first pulse [O (15 m s−1)] at 1900 UTC, the winds decrease, then the strong southerlies hit at up to 31.5 m s−1. Again, there is a timing lag. ALL (Fig. 11c) forecasts a weaker, more delayed episode.

Fig. 11.

Same as in Fig. 10, but at Pegasus North.

Fig. 11.

Same as in Fig. 10, but at Pegasus North.

MOD1 (Fig. 11d) produces a strong wind episode of timing and magnitude that compares reasonably well with the observations. The event begins just after 1900 UTC and peak at 36.6 m s−1 (cf. 39.6 m s−1 observed). MOD1 also simulates the increase in flow seen in the pre-event period of ∼1000–1800 UTC. MOD1_60 (Fig. 11e) is similar to MOD1, but its peak velocity is slightly higher (37.2 m s−1), and the profile through the event is, overall, closer to that observed. In short, from the results of these representative McMurdo area sites (and others, not shown), MOD1 best reproduces the event of the 3.3-km grid experiments. In addition, in MOD1_60 the resolution increase to 2.2-km spacing appears to be a further improvement.

While the focus has been on the ARW experiments, consider now the performance of the operational AMPS MM5. For Arrival Heights and Pegasus North (Fig. 12), one first sees that the MM5 significantly underpredicts the wind maxima. At Arrival Heights (Fig. 12a) there is a wind speed increase at observed onset, but this is not sustained. At Pegasus North (Fig. 12b) there are sustained relatively strong winds after 1600 UTC, although they peak at only 21.6 m s−1 (2100 UTC). Comparing the counterpart experiment STD in Figs. 10b and 11b with the MM5, the ARW produces a more distinct and less underestimated, albeit more delayed, event at these sites.

Fig. 12.

Observed (solid) vs AMPS MM5 (dashed) wind speed (m s−1) at (a) Arrival Heights, (b) Pegasus North, and (c) Marilyn. Abscissa shows time in hours from 0000 UTC 15 May 2004. Missing observed values are not plotted.

Fig. 12.

Observed (solid) vs AMPS MM5 (dashed) wind speed (m s−1) at (a) Arrival Heights, (b) Pegasus North, and (c) Marilyn. Abscissa shows time in hours from 0000 UTC 15 May 2004. Missing observed values are not plotted.

One may also examine the evolution of the winds associated with the synoptic system (i) away from the immediate Ross Island area and its complex topography and (ii) indicative of conditions on the ice shelf. For this, consider Marilyn AWS. The winds peaked at Marilyn from 1600 to 2200 UTC, reaching 26.7 m s−1 (Fig. 13). CTRL (Fig. 13a) reproduces the onset of the event winds and the observed magnitude (26.2 m s−1), although it does not sustain the high winds for as long as observed (e.g., 2000–0000 UTC). STD is better at this site (Fig. 13b). The timing of the increase is captured, as well as the peak winds, with even a bit of an overforecast (28.8 m s−1). An arguable improvement over this, in terms of wind strength from before the event (0000–1800 UTC) through the strongest wind phase, is MOD1 (Fig. 13d). MOD1 avoids an overforecast, and the wind decrease after the peak hours agrees with the observed trace. MOD1_60 (Fig. 13e) is similar to MOD1. ALL displays the poorest results (Fig. 13c). It suffers from a lag and a significant underforecast. Last, although in the AMPS MM5 (Fig. 12c) the amplitude of the maxima is not captured as well as in STD and MOD1, the mean profile does approximate the episode to the level of those ARW runs.

Fig. 13.

Same as in Fig. 10, but at Marilyn.

Fig. 13.

Same as in Fig. 10, but at Marilyn.

To illuminate the wind event and flow pattern on the mesoscale, as opposed to the point views from the time series, Fig. 14 presents the ARW surface (lowest model level) winds from the Ross Island grids at 2300 UTC (hour 23) 15 May 2004. CTRL exhibits strong southerlies striking Minna Bluff and the TAM (Fig. 14a), but this momentum is not reaching McMurdo. In contrast, STD (Fig. 14b) generates a well-defined flow around Ross Island, with a broader and stronger (>40 m s−1) wind field. One revelation from the high-resolution output is the strong shadowing effect that Minna Bluff and the associated TAM topography can exert on Hut Point Peninsula (the extension of Ross Island on which McMurdo sits, marked in Fig. 14a) and the McMurdo area. Due to this, in STD McMurdo is not yet experiencing the southerlies. And, ALL (Fig. 14c) shows only a narrow stream of higher-velocity flow (>12.5 m s−1) getting into McMurdo.

Fig. 14.

Surface winds and SLP from ARW experiments for hour 23 (2300 UTC 15 May 2004). Wind speed shaded (scale at right), medium gray shading is ∼15 m s−1. Arrows indicate wind directions and magnitudes are 22 m s−1 (vector length interval)−1. SLP contour interval is 1 to 988 hPa, 2 hPa for SLP > 988 hPa; maximum SLP contour is 1008 hPa. The italic L indicates the surface low and V indicates the von Karman vortex. (a) CTRL with Minna Bluff (MB) and the TAM indicated, the asterisk shows the location of McMurdo, which lies on Hut Point Peninsula; (b) STD; (c) ALL; (d) MOD1, the open circle shows the location of Cape Crozier; and (e) MOD1_60.

Fig. 14.

Surface winds and SLP from ARW experiments for hour 23 (2300 UTC 15 May 2004). Wind speed shaded (scale at right), medium gray shading is ∼15 m s−1. Arrows indicate wind directions and magnitudes are 22 m s−1 (vector length interval)−1. SLP contour interval is 1 to 988 hPa, 2 hPa for SLP > 988 hPa; maximum SLP contour is 1008 hPa. The italic L indicates the surface low and V indicates the von Karman vortex. (a) CTRL with Minna Bluff (MB) and the TAM indicated, the asterisk shows the location of McMurdo, which lies on Hut Point Peninsula; (b) STD; (c) ALL; (d) MOD1, the open circle shows the location of Cape Crozier; and (e) MOD1_60.

MOD1 (Fig. 14d) simulates a 984-hPa low center east of Cape Crozier and an associated pressure gradient that has engaged the McMurdo area. MOD1’s wind speeds are over 32 m s−1 at Arrival Heights (32 m s−1 observed) and over 34 m s−1 at Pegasus North (33 m s−1 observed) (marked in Fig. 1c). The southerly momentum field has engulfed Ross Island. This is inducing notable secondary phenomena—flow splitting in the surface layer and a von Kármán vortex (e.g., Heinemann 1986) to the north (marked “V” in Fig. 14d). Model-generated von Kármán vortices resulting from strong southerly flow around Ross Island have been previously described by Powers et al. (2003).

MOD1_60 (Fig. 14e) displays conditions at McMurdo similar to MOD1, but with even greater coverage and intensity of the high-momentum flow enveloping Ross Island, Minna Bluff, and the TAM. Characteristics of these two most successful experiments, MOD1 and MOD1_60, are the low’s maintaining its integrity and the associated pressure gradient being relatively strong as it skirts Ross Island. Note, as in MOD1, the well-defined, leeside von Kármán vortex (Fig. 14c). Although the satellite imagery does not reveal such vortices at this time, midlevel and higher cloud are shielding the lower levels north of Ross Island (see, e.g., Fig. 4b).

The AMPS MM5 hour-23 forecast is seen in Fig. 15. While over the Ross Island area in general there is strong southerly flow, McMurdo itself is not experiencing a wind event, as Hut Point Peninsula and most of the south side of the island are sheltered. The MM5 result is most similar to that of STD (Fig. 14b). From a forecast perspective, the MM5 wind field depiction is not as good as that of MOD1, which does capture strong winds impacting McMurdo at this time (Fig. 14d).

Fig. 15.

Surface winds and SLP from the AMPS MM5 forecast for hour 23 (2300 UTC 15 May 2004). Shading, vectors, and contours are the same as in Fig. 14.

Fig. 15.

Surface winds and SLP from the AMPS MM5 forecast for hour 23 (2300 UTC 15 May 2004). Shading, vectors, and contours are the same as in Fig. 14.

To briefly investigate the variations in the low tracks, and thus the simulated wind events in the experiments, the upper-level synoptic differences in the runs are now considered. As 500 hPa is taken as a representative level for the steering flow (SPAWAR forecasters 2005, personal communication), Figs. 16a,b show the ECMWF analysis and the CTRL 12-h forecast at this level for 1200 UTC 15 May 2004. First apparent is that the trough over the Ross Ice Shelf is a sharper, higher-amplitude feature in the analysis than in CTRL. Correspondingly, downstream of the trough axis the analyzed flow is more southerly and less easterly over the eastern ice shelf than the forecast. The model, furthermore, displays less ridging over the eastern edge of the shelf and over the TAM.

Fig. 16.

500-hPa heights and winds for 1200 UTC 15 May 2004. Heights (solid lines); contour interval is 50 gpm. Wind vectors (arrows); magnitudes are 18 m s−1 (vector length interval)−1. (a) ECMWF analysis and (b) CTRL 12-h forecast.

Fig. 16.

500-hPa heights and winds for 1200 UTC 15 May 2004. Heights (solid lines); contour interval is 50 gpm. Wind vectors (arrows); magnitudes are 18 m s−1 (vector length interval)−1. (a) ECMWF analysis and (b) CTRL 12-h forecast.

Figure 17a shows the 500-hPa vector wind difference between CTRL and the analysis at 1200 UTC. In the analysis the midshelf flow is more southerly, while the eastern shelf flow is more westerly (from the TAM side). The actual track would thus tend to be more to the north and away from the western shelf edge, while conversely, in the model, the trajectory would be less northerly and more westward (i.e., toward the Minna Bluff coast and the TAM). This is indeed seen compared to observations (cf. Figs. 3 and 8a).

Fig. 17.

500-hPa wind and wind component differences for analysis − CTRL, 1200 UTC 15 May 2004. (a) Wind vector differences, magnitudes are 9 m s−1 (vector length interval)−1. (b) The u-component differences; contour interval is 2 m s−1; solid contours reflect reduced easterly/stronger westerly component, dashed contours reflect stronger easterly/reduced westerly component, and large arrows indicate net momentum relative to CTRL. (c) The υ-component differences; contour interval is 2 m s−1; solid contours reflect stronger southerly component, dashed contours reflect weaker southerly/stronger northerly component, and large arrows indicate net momentum relative to CTRL.

Fig. 17.

500-hPa wind and wind component differences for analysis − CTRL, 1200 UTC 15 May 2004. (a) Wind vector differences, magnitudes are 9 m s−1 (vector length interval)−1. (b) The u-component differences; contour interval is 2 m s−1; solid contours reflect reduced easterly/stronger westerly component, dashed contours reflect stronger easterly/reduced westerly component, and large arrows indicate net momentum relative to CTRL. (c) The υ-component differences; contour interval is 2 m s−1; solid contours reflect stronger southerly component, dashed contours reflect weaker southerly/stronger northerly component, and large arrows indicate net momentum relative to CTRL.

Figures 17b,c isolate the westerly and southerly wind component differences between the analysis and CTRL. The broad arrows indicate the momentum in the analysis that is not represented in the model. The ARW lacks a relative westerly component, one that would retard the propagation of the system to the Ross Island side of the shelf (Fig. 17b), as well as the stronger southerly flow (Fig. 17c). In short, the steering was relatively northward and eastward for the observed system compared to the ARW systems, and the CTRL track most distinctly exemplifies the behavior of the latter. The improvement in low trajectory in MOD1 and MOD1_60 indicates that the assimilation of filtered MODIS AMVs aloft can better the forecast on the mesoscale from a modified upper-level synoptic initialization.

5. Statistical evaluations

The ARW’s performance in simulating the local flow amplitude is now analyzed statistically. To this end wind speed errors have been quantified through verification at six locations across the Ross Island region: Arrival Heights, Pegasus North, Black Island, Minna Bluff, Marilyn, and Schwerdtfeger (located in Fig. 1c). Wind speed bias [or mean error (ME)], mean absolute error (MAE), and root-mean-square error (RMSE) have been calculated for two periods: 0000 UTC 15 May–UTC 17 May 2004 and 1200 UTC 15 May–0600 16 May 2004. The former is the whole period of simulation (hours 0–48), while the latter represents the subperiod centered on the event, from 6 h prior through 6 h afterward (hours 12–30). The purpose in considering two intervals is to see whether the picture of model performance varies for the event proper versus the entire forecast.

Tables 1 and 2 present the results for the full simulation and event periods. For both periods the lowest biases, MAEs, and RMSEs are seen for experiments STD, MOD1, and MOD1_60. The AMPS MM5 compares well with these ARW runs. Note that the wind speed biases are consistently negative: the ARW and the MM5 underpredict the wind speeds, both for the event and the whole forecast period.

Table 1.

Model wind speed errors (m s−1) for hours 0–48 (0000 UTC 15 May–0000 UTC 17 May 2004) for McMurdo region: Arrival Heights, Minna Bluff, Pegasus North, Black Island, Marilyn, and Schwerdtfeger.

Model wind speed errors (m s−1) for hours 0–48 (0000 UTC 15 May–0000 UTC 17 May 2004) for McMurdo region: Arrival Heights, Minna Bluff, Pegasus North, Black Island, Marilyn, and Schwerdtfeger.
Model wind speed errors (m s−1) for hours 0–48 (0000 UTC 15 May–0000 UTC 17 May 2004) for McMurdo region: Arrival Heights, Minna Bluff, Pegasus North, Black Island, Marilyn, and Schwerdtfeger.
Table 2.

Same as in Table 1, but for hours 12–30 (1200 UTC 15 May–0600 UTC 16 May 2004).

Same as in Table 1, but for hours 12–30 (1200 UTC 15 May–0600 UTC 16 May 2004).
Same as in Table 1, but for hours 12–30 (1200 UTC 15 May–0600 UTC 16 May 2004).

Table 3 presents the averages of the errors. For the ARW experiments, MOD1 and MOD1_60 have the lowest error means. Following these is STD, while the ALL results are relatively poor. The AMPS MM5 scores are among the best and are comparable to those for MOD1.

Table 3.

Average McMurdo region wind speed errors in model runs. Errors averaged over values from locations in Tables 1 and 2. Full simulation (hours 0–48) and event period (hours 12–30) results shown.

Average McMurdo region wind speed errors in model runs. Errors averaged over values from locations in Tables 1 and 2. Full simulation (hours 0–48) and event period (hours 12–30) results shown.
Average McMurdo region wind speed errors in model runs. Errors averaged over values from locations in Tables 1 and 2. Full simulation (hours 0–48) and event period (hours 12–30) results shown.

To objectively determine whether the errors and error differences are significant, statistical significance testing has been performed. The first analysis is that of whether each average error can be concluded to be statistically distinct from zero. A two-tailed Student’s t test (see, e.g., Walpole and Myers 1985) reflects the null hypothesis (H0) that the error populations of the experiments have means of zero and the alternative hypothesis that the means do not equal zero. The results find that for both the 48- and the 18-h periods, all the average MAEs for the experiments are significantly different from zero at the 95% confidence level.

To assess the significance of experiment differences, testing on the differences of the mean errors has been performed. The null hypothesis is that the difference in error means is zero, while the alternate hypothesis is that the error mean of one experiment is less than that of the other. The t value is computed as

 
formula

where d is the difference of the means, d0 is the null hypotheses difference in means (here 0), ν is the degrees of freedom, and sd is the standard deviation of the sample of mean differences (see, e.g., Walpole and Myers 1985; Panofsky and Brier 1968; Wilks 1995). The computed t value is compared to the critical value of tα/2, where the primary confidence level considered here is 95% (α = 0.05).

Table 4 presents the results. Here the error means of the experiments in the first column are compared with those in the second. The experiment with the concluded lower mean error at the 95% level is shown under the error type (bias, MAE) column. If the null hypothesis cannot be rejected the table entry is “I.” If the alternate hypothesis may be accepted for experiment 1 or 2 at the 90% level (but not the 95% level), then the entry given is E190 or E290 (for experiment 1 or experiment 2 having the lower mean error).

Table 4.

Statistical comparisons of experiments. Under “Bias” and “MAE,” the listed experiment’s mean error for the period indicated (forecast hours 12–30 or 0–48) is concluded to be lower than that of the compared experiment at the 95% confidence level. EXPT90 indicates that mean error is lower at 90% confidence level. The I indicates that the test is inconclusive at either the 95% or 90% confidence level. M1_60 indicates MOD1_60.

Statistical comparisons of experiments. Under “Bias” and “MAE,” the listed experiment’s mean error for the period indicated (forecast hours 12–30 or 0–48) is concluded to be lower than that of the compared experiment at the 95% confidence level. EXPT90 indicates that mean error is lower at 90% confidence level. The I indicates that the test is inconclusive at either the 95% or 90% confidence level. M1_60 indicates MOD1_60.
Statistical comparisons of experiments. Under “Bias” and “MAE,” the listed experiment’s mean error for the period indicated (forecast hours 12–30 or 0–48) is concluded to be lower than that of the compared experiment at the 95% confidence level. EXPT90 indicates that mean error is lower at 90% confidence level. The I indicates that the test is inconclusive at either the 95% or 90% confidence level. M1_60 indicates MOD1_60.

For the first pair of STD and CTRL, the mean biases are significantly different at the 95% level over both the full 48-h forecast and the 18-h event subperiod, with the mean error for STD less than that of CTRL. As seen in results such as in Figs. 11, 12 and 14, STD did verify better than CTRL, and such a difference is significant. In terms of wind speed MAEs, however, the mean errors in STD and CTRL are statistically indistinguishable. For MOD1 and MOD1_60 compared to CTRL, these filtered MODIS experiments are both concluded to have significantly lower mean biases and MAEs for both periods considered. In contrast, in comparing the approach of assimilating all of the available MODIS data without filtering (ALL) with either no data (CTRL) or conventional observation (STD) assimilation, the results argue against the former approach (ALL). STD exhibits significantly lower biases and MAEs than ALL for both periods, while CTRL exhibits such lower errors for the wind event subperiod.

Comparing MOD1 and STD it is seen that the addition of filtered MODIS data yields an improvement over the entire simulation and results in significantly improved biases and MAEs for the 48-h period. Compared to ALL, the MOD1 filtering yields better model performance for both periods. In terms of relative performance among the 90/30 ARW experiments, overall MOD1 shows the lowest mean errors for both periods. It is followed by STD, then CTRL. ALL is statistically the poorest. It is thus again seen that MODIS data can improve the ARW simulations for this extreme polar event, but under the conditions of filtering.

With respect to the increase in resolution from a 90/30/10/3.3- to a 60/20/6.7/2.2-km configuration, the finer grids do improve scores somewhat. MOD1_60 exhibits a significantly lower wind speed bias than MOD1 for the episode at the 95% level, while at a lower 90% confidence level it is better in terms of MAE for both periods.

With respect to the performance of the AMPS MM5, the MM5 has lower mean biases and MAEs, in general, than ALL (Table 4, bottom). Compared to CTRL, the MM5 is superior in terms of bias, but no conclusion can be made for MAE. Compared to those of its ARW counterpart run, STD, the MM5’s biases for the 48-h period are significantly better (95% level), with the confidence in this conclusion reduced to 90% considering just the 18-h event subperiod. The MM5 MAE, however, is not significantly better than the STD MAE. Compared to the best ARW 90/30 experiment, MOD1, it cannot be concluded that the MM5’s errors are significantly different. The deduction is mostly the same with respect to MOD1_60, although at the 90% confidence level MOD1_60 is superior in terms of MAE.

6. Summary and conclusions

In the setting of the Antarctic Mesoscale Prediction System (AMPS), the Weather Research and Forecasting (WRF) model has been applied for the first time to Antarctica. In this study the abilities of the ARW to forecast a major Antarctic weather event and of MODIS retrievals to improve such forecasts are explored in simulations of the windstorm that struck McMurdo Station on 15 May 2004. The suite of tests primarily uses a nested 90/30/10/3.3-km domain setup, with an additional run using higher-resolution grids of 60/20/6.7 down to 2.2 km over the critical Ross Island area. In addition, a comparison of the AMPS MM5 forecast with the ARW simulations provides a first look at the relative performance in a polar case study of these two models, currently both in real-time use.

Considering first its ability to capture the synoptic setting and evolution of the event, the ARW simulates the evolution of the motivating low pressure system, and the accuracy of the track forecasts on the larger scale is confirmed. The best experiments with respect to these forecast elements are MOD1 and MOD1_60, which both assimilate filtered MODIS data. A mesoscale examination, however, reveals that the ARW (as well as the AMPS MM5) tends to move the low across the Ross Ice Shelf too strongly and rapidly to the west, not precisely capturing the timing of the backing to the north. Analyses show the discrepancy in the synoptic trajectory to be associated with long-wave pattern errors and weaker westerly and southerly upper-level flow components in the model. The model also fills the system relatively abruptly near Ross Island. Both of these developments significantly affect the wind simulations in the McMurdo region.

The high-resolution (3.3 and 2.2 km) ARW grids reveal the regional character of the 15 May 2004 windstorm and show how strong southerly momentum associated with a low pressure system moving to the east of Ross Island impacts McMurdo. The surface wind depictions from the most successful simulations (MOD1, MOD1_60) present a pattern of high-momentum flow encountering and responding to the local topography (i.e., Minna Bluff, the Transantarctic Mountains, and Ross Island). As the surge arrives in the McMurdo area, the wind velocities at the base increase substantially. The realistic model results further show that, farther afield, circulations such as von Kármán vortices may be spawned from the interaction of such flow and Ross Island. While it is possible that the latter are ubiquitous in strong, stable, southerly flow regimes, their prevalence cannot as yet be confirmed.

The ARW can successfully forecast the strong southerly flow defining the event [e.g., provide guidance alerting forecasters to condition-1 (≥55 kt) and condition-2 level (48–55 kt) winds]. The 90/30 run best reproducing the wind timing and intensity, MOD1, reflects the assimilation of the filtered MODIS AMVs. From the 60/20 MOD1_60 experiment, however, it is found that increasing the ARW’s grid resolution by 33% does yield statistically significant improvement in the wind event simulation. Despite the model’s ability, in general the surface wind speed amplitudes for the event tend to be underforecast. As noted, this reflects track errors on the mesoscale and a stalling and filling of the cyclone responsible for the event near Ross Island.7 The assimilation of filtered MODIS data, however, mitigates the errors in low trajectory and evolution and in the resultant McMurdo wind forecast. This shows the potential for benefits to the initial conditions and forecasts of Antarctic high-resolution NWP systems from the assimilation of nontraditional satellite observations over the continent and the Southern Ocean.

The performance of the ARW and the AMPS MM5 has also been examined statistically, through significance testing of event wind speed errors. It is found that the assimilation of conventional observations and select MODIS AMV data can improve the mesoscale forecast in statistically significant terms. The application of a filter to the MODIS retrievals, however, is necessary for such benefit, as the assimilation of unfiltered measurements is found to actually degrade model performance. The filtering criteria used here follow from previously published suggestions and account for instrument channel, surface type, and observation height.

In summary, this study has provided an initial investigation of the behavior of the emerging Advanced Research WRF model in a polar region. It has also illuminated the impact of the high-potential MODIS AMV data on polar mesoscale NWP and on the reproduction of a major Antarctic weather event. Most importantly, it is found that the ARW can realistically simulate such an event and that assimilation of MODIS polar winds can significantly improve the forecast on the mesoscale. That the application of a data filter is necessary would confirm—in the context of a specific, high-impact forecast and for a mesoscale model—previous work involving global model applications (Key et al. 2003; Bormann and Thépaut 2004). Overall, it is seen that WRF shows promise for both research and operational applications over Antarctica.

Acknowledgments

This study and AMPS have been supported by the National Science Foundation, Office of Polar Programs. The author thanks Michael G. Duda of NCAR for his assistance in data processing and graphics support. The author thanks Udo Voight and Ralf Brauner of Deutscher Wetterdienst for their provision of ECMWF analyses. The author also thanks SPAWAR for computing hardware and the SPAWAR forecasters for helpful discussions.

REFERENCES

REFERENCES
Adams
,
A. S.
,
2005
:
The relationship between topography and the Ross Ice Shelf air stream. Ph.D. thesis, University of Wisconsin—Madison, 125 pp
.
Barker
,
D. M.
,
W.
Huang
,
Y. R.
Guo
, and
Q. N.
Xiao
,
2004
:
A three-dimensional (3DVAR) data assimilation system for use with MM5: Implementation and initial results.
Mon. Wea. Rev.
,
132
,
897
914
.
Bluestein
,
H. B.
,
1992
:
Synoptic-Dynamic Meteorology in Midlatitudes.
Vol. 1. Oxford University Press, 431 pp
.
Bormann
,
N.
, and
J-P.
Thépaut
,
2004
:
Impact of MODIS polar winds in ECMWF’s 4DVAR data assimilation system.
Mon. Wea. Rev.
,
132
,
929
940
.
Bromwich
,
D. H.
, and
T. R.
Parish
,
2002
:
Ross Island Meteorology Experiment (RIME) detailed science plan. BPRC Miscellaneous Series M-424, Byrd Polar Research Center, The Ohio State University, 39 pp. [Available from Byrd Polar Research Center, The Ohio State University, 1090 Carmack Rd., Columbus, OH, 43210.]
.
Bromwich
,
D. H.
,
J. J.
Cassano
,
T.
Klein
,
G.
Heinemann
,
K. M.
Hines
,
K.
Steffen
, and
J. E.
Box
,
2001
:
Mesoscale modeling of katabatic winds over Greenland with the Polar MM5.
Mon. Wea. Rev.
,
129
,
2290
2309
.
Bromwich
,
D. H.
,
A. J.
Monaghan
,
J. G.
Powers
,
J. J.
Cassano
,
H-L.
Wei
,
Y-H.
Kuo
, and
A.
Pellegrini
,
2003
:
Antarctic Mesoscale Prediction System (AMPS): A case study from the 2000/2001 field season.
Mon. Wea. Rev.
,
131
,
412
434
.
Bromwich
,
D. H.
,
A. J.
Monaghan
,
K. W.
Manning
, and
J. G.
Powers
,
2005
:
An evaluation of the Antarctic Mesoscale Prediction System (AMPS).
Mon. Wea. Rev.
,
133
,
579
603
.
Carlson
,
T. N.
,
1998
:
Mid-Latitude Weather Systems.
Amer. Meteor. Soc., 507 pp
.
Cassano
,
J. J.
,
J. E.
Box
,
D. H.
Bromwich
,
L.
Li
, and
K.
Steffen
,
2001
:
Evaluation of Polar MM5 simulations of Greenland’s atmospheric circulation.
J. Geophys. Res.
,
106
,
33867
33890
.
Chen
,
F.
, and
J.
Dudhia
,
2001
:
Coupling an advanced land-surface/hydrology model with the Penn State/NCAR MM5 modeling system.
Mon. Wea. Rev.
,
129
,
569
585
.
Davis
,
C. A.
, and
Coauthors
,
2006
:
Advanced Research WRF developments for hurricane prediction. Extended Abstracts, Seventh WRF Users’ Workshop, Boulder, CO, National Center for Atmospheric Research
.
Done
,
J. M.
,
L. R.
Leung
, and
B.
Kuo
,
2006
:
Understanding error in the long-term simulation of warm season rainfall using the WRF model. Extended Abstracts, Seventh WRF Users’ Workshop, Boulder, CO, National Center for Atmospheric Research
.
Forsythe
,
M.
, and
H.
Berger
,
2004
:
MODIS winds: Latest results at the Met Office. Proc. Seventh Int. Winds Workshop, Helsinki, Finland, EUMETSAT
.
Grell
,
G. A.
,
J.
Dudhia
, and
D. R.
Stauffer
,
1995
:
A description of the fifth-generation Penn State/NCAR Mesoscale Model (MM5). NCAR Tech. Note TN-398+STR, 122 pp. [Available from UCAR Communications, P.O. Box 3000, Boulder, CO 80307.]
.
Heinemann
,
G.
,
1986
:
Lee-vortices in the Antarctic.
Beitr. Phys. Atmos.
,
59
,
599
602
.
Holmes
,
R. E.
,
C. R.
Stearns
,
G. A.
Weidner
, and
L. M.
Keller
,
2000
:
Utilization of automatic weather station data for forecasting high wind speeds at Pegasus Runway, Antarctica.
Wea. Forecasting
,
15
,
137
151
.
Holton
,
J. R.
,
1992
:
An Introduction to Dynamic Meteorology.
Academic Press, 511 pp
.
Hong
,
S-Y.
,
J.
Dudhia
, and
S-H.
Chen
,
2004
:
A revised approach to ice microphysical processes for the bulk parameterization of clouds and precipitation.
Mon. Wea. Rev.
,
132
,
103
120
.
Janjic
,
Z. I.
,
2002
:
Nonsingular implementation of the Mellor–Yamada level 2.5 scheme in the NCEP Meso Model. NCEP Office Note 437, 61 pp
.
Kain
,
J. S.
, and
J. M.
Fritsch
,
1993
:
Convective parameterization for mesoscale models: The Kain–Fritsch scheme.
The Representation of Cumulus Convection in Numerical Models, K. A. Emanuel and D. J. Raymond, Eds., Amer. Meteor. Soc., 165–170
.
Key
,
J. R.
,
D.
Santek
,
C. S.
Velden
,
N.
Bormann
,
J-N.
Thépaut
,
L. P.
Riishojgaard
,
Y.
Zhu
, and
W. P.
Menzel
,
2003
:
Cloud-drift and water vapor winds in the polar regions from MODIS.
IEEE Trans. Geosci. Remote Sens.
,
41
,
482
492
.
Mellor
,
G. L.
, and
T.
Yamada
,
1982
:
Development of a turbulence closure model for geophysical fluid problems.
Rev. Geophys. Space Phys.
,
20
,
851
875
.
Moeng
,
C-H.
,
J.
Dudhia
,
J. B.
Klemp
, and
P. P.
Sullivan
,
2007
:
Examining two-way grid nesting for large eddy simulation of the PBL using the WRF model.
Mon. Wea. Rev.
,
135
,
2295
2311
.
Monaghan
,
A. J.
,
D. H.
Bromwich
,
H-L.
Wei
,
A. M.
Cayette
,
J. G.
Powers
,
Y-H.
Kuo
, and
M.
Lazzara
,
2003
:
Performance of weather forecast models in the rescue of Dr. Ronald Shemenski from the South Pole in April 2002.
Wea. Forecasting
,
18
,
142
160
.
O’Connor
,
W. P.
, and
D. H.
Bromwich
,
1988
:
Surface airflow around Windless Bight, Ross Island, Antarctica.
Quart. J. Roy. Meteor. Soc.
,
114
,
917
938
.
Orlanski
,
I.
,
1975
:
A rational subdivision of scales for atmospheric processes.
Bull. Amer. Meteor. Soc.
,
56
,
527
530
.
Panofsky
,
H. A.
, and
G. W.
Brier
,
1968
:
Some Applications of Statistics to Meteorology.
The Pennsylvania State University Press, 224 pp
.
Powers
,
J. G.
,
A. J.
Monaghan
,
A. M.
Cayette
,
D. H.
Bromwich
,
Y-H.
Kuo
, and
K. W.
Manning
,
2003
:
Real-time mesoscale modeling over Antarctica: The Antarctic Mesoscale Prediction System (AMPS).
Bull. Amer. Meteor. Soc.
,
84
,
1533
1546
.
Powers
,
J. G.
,
K. W.
Manning
, and
M. M.
Lamberston
,
2005
:
Application of the Weather Research and Forecasting (WRF) model in Antarctica. Preprints, Eighth Conf. on Polar Meteorology and Oceanography, San Diego, CA, Amer. Meteor. Soc., CD-ROM, 3.5
.
Simmonds
,
I.
,
K.
Keay
, and
E-P.
Lim
,
2003
:
Synoptic activity in the seas around Antarctica.
Mon. Wea. Rev.
,
131
,
272
288
.
Skamarock
,
W. C.
,
J. B.
Klemp
,
J.
Dudhia
,
D. O.
Gill
,
D. M.
Barker
,
W.
Wang
, and
J. G.
Powers
,
2005
:
A description of the Advanced Research WRF, version 2. NCAR Tech. Note NCAR/TN-468+STR, 88 pp. [Available from UCAR Communications, P.O. Box 3000, Boulder, CO 80307.]
.
Steinhoff
,
D. F.
, and
D. H.
Bromwich
,
2005
:
Extreme winds and rapid degeneration of the May 2004 McMurdo Antarctica storm: Analysis using the AMPS forecast model, AWS data, and MODIS data. AMPS Users’ Workshop 2005, Columbus, OH, Byrd Polar Research Center, 43–51
.
Velden
,
C.
, and
Coauthors
,
2005
:
Recent innovations in deriving tropospheric winds from meteorological satellites.
Bull. Amer. Meteor. Soc.
,
86
,
205
223
.
Walpole
,
R. E.
, and
R. H.
Myers
,
1985
:
Probability and Statistics for Engineers and Scientists.
Macmillan, 639 pp
.
Wilks
,
D. S.
,
1995
:
Statistical Methods in the Atmospheric Sciences.
Academic Press, 467 pp
.

Footnotes

Corresponding author address: Dr. Jordan G. Powers, Mesoscale and Microscale Meteorology Division, Earth and Sun Systems Laboratory, National Center for Atmospheric Research, P.O. Box 3000, Boulder, CO 80307. Email: powers@ucar.edu

1

As of this writing, WRF has over 5000 registered users in over 90 countries.

2

The RAS is the broad, northward-moving airstream adjacent to the Transantarctic Mountains. It extends across the Ross Ice Shelf from its southern edge and through the western Ross Sea. It is a primary transport channel between Antarctica and the Southern Hemisphere (Bromwich and Parish 2002).

3

The forecasters are employed/contracted by the Space and Naval Warfare Systems Center (SPAWAR), Charleston, SC.

4

Condition 1 is defined by visibility less than 100 ft (31 m), or winds exceeding 55 kt (28 m s−1), or wind chill temperatures colder than −100°F/−73°C. A declaration of condition 1 means that personnel must remain indoors where they are at the time.

5

The orientation of Fig. 3 and subsequent figures—with the South Pole at the top—corresponds to the available satellite imagery perspective (Fig. 4). Figure 3 (and Fig. 17) provides directional information (N, S, E, W indicated relative to Ross Island) to assist in the references to compass directions.

6

These two sites, and Marilyn site (discussed below), have anemometers at 3 m. Accuracies are O(0.25 m s−1) for Pegasus North and Marilyn and 5% for Arrival Heights.

7

While the PBL and land surface representations may also be factors, sensitivity investigations of the associated schemes (kept constant in the experiment suite here) are left for future work.