Abstract

The Pennsylvania State University–NCAR Mesoscale Model version 5 (MM5), running on a triply nested grid, was used to simulate the intense lake-effect snowstorm of 4–5 January 1995. On the finest grid (5-km resolution) centered over Lake Ontario, MM5 produced a snowband in the correct location having a size and orientation similar to the band observed by the WSR-88D radar at Binghamton, New York. The simulated precipitation distribution agreed well with the observed snowfall during the first 18 h during the time when the snowband was in its midlake position extending into the Tug Hill plateau. During the last 12 h of the simulation, when both the observed and simulated snowbands lay along the south shore of Lake Ontario, the simulated snowfall at inland locations of Oswego County was less than observed. During this period, the simulated precipitation over Lake Ontario appeared to be excessive, although no radar data or ground truth was available to confirm this.

Two short-wave troughs interacted with the Lake Ontario snowband. The temporary weakening of the snowband after passage of the first trough was simulated well in the triply nested MM5 simulation. A comparison was made between the operational Eta Model run and an MM5 simulation on a grid of comparable resolution (80 km) in handling the passage of the second more vigorous short wave. Both the Eta and the 80-km MM5 were a few hours too early with the passage of this trough. The nested-grid version of MM5 was correct in simulating the rapid southward movement of the band to Oswego County just after the second trough moved east of the lake. However, because of the timing error with the trough, MM5 was premature by a few hours in the southward shift of the snowband.

Results on the 15-km grid indicated that moisture plumes from Lake Huron and Georgian Bay fed into the Lake Ontario band. In the lowest few hundred meters, these plumes were deflected around the Shelburne Plateau, which lies between Lake Huron and Lake Ontario. Future research will focus on interactions between circulations downwind of Lake Huron and snowbands that form over Lake Ontario.

The results of the 4–5 January 1995 simulation are sufficiently encouraging to suggest that MM5 may be used to make real-time forecasts of lake-effect snowstorms. The lead author is participating in a COMET cooperative project to provide lake-effect snow forecasts, in GEMPAK format, to the National Weather Service Forecast Offices at Buffalo and Binghamton using a 20-km nested grid over Lakes Huron, Erie, and Ontario. Despite relatively coarse resolution, MM5 has produced useful predictions of snowband location and movement during the 1996/97 and 1997/98 lake-effect snow seasons.

1. Introduction

The accurate prediction of the onset, location, movement, duration, and intensity of lake-effect snowstorms is one of the most challenging forecast problems facing meteorologists in the Great Lakes region. Common in late fall and winter, lake-effect snows occur when cold air crosses the relatively warm Great Lakes (Wiggin 1950) or other large bodies of water such as the Great Salt Lake (Carpenter 1993). The cold air is destabilized as it is heated and moistened by the water, allowing convective cells and “steam devils” to organize into stratocumulus cloud streets (Pease et al. 1988). One or more cloud streets develop into well-organized convective bands that sometimes produce heavy snow. Radar and satellite imagery indicate that the bands are narrow (typically 5 to 20 km wide) and elongated (usually 50 to 300 km long). In a persistent, heavy lake-effect event, one location may receive over 100 cm of snow, while locations 20 km away may get barely a trace. These snowstorms are especially troublesome to millions of residents in the Great Lakes region because the deep accumulation of low-density snowflakes and the occasional occurrence of near-zero visibility produce extremely hazardous driving conditions. An intense, well-placed heavy snowband located over the Buffalo and Rochester metropolitan areas of western New York has the potential to bring the activities of over 2 million people to a standstill.

Early observational studies gave initial insight into the lake-effect forecasting problem. Wiggin (1950) observed that the width of snowbands is roughly proportional to the depth of the cold air. A later field study by Peace and Sykes (1966) documented narrow confluence zones and pressure troughs near the surface associated with lake-effect snowbands. The importance of instability was shown by Holroyd (1971), who found that nearly all significant lake-effect snowstorms occur when the air at 850 hPa is at least 13°C colder than the lake surface temperature. The role of steering winds was documented in a study by Jiusto and Kaplan (1972), which showed that the location of the axis of heaviest snowfall is strongly controlled by the gradient-level wind direction. Passarelli and Braham (1981) used a detailed study of radar, satellite, and aircraft data to conclude that a winter land breeze can organize low-level convergence and cloud formation parallel to the Lake Michigan shoreline. A mobile sounding study by Byrd et al. (1991) showed the importance of the depth of instability to the intensity of lake-effect snowfall events. Recently the Lake Ontario Winter Storms project (Reinking et al. 1993) demonstrated the utility of Doppler radar, wind profilers, the Radio Acoustic Sounding System, and portable sounding systems to monitor and nowcast lake-effect snowband development over and downwind of Lake Ontario.

Drawing upon these observational efforts, a series of numerical modeling studies focused on the lake-effect process in even greater detail. An initial study by Lavoie (1972) used a mixed-layer model to show that thermal convergence was the dominant factor in organizing snowbands along the south shore of Lake Erie. Later Ellenton and Danard (1979) used a primitive equation model to show that overlake heat and moisture fluxes were the principal contributors to lake-effect precipitation downwind of Lake Huron. Ballentine (1982) and Hjelmfelt and Braham (1983) successfully simulated the formation of land-breeze-induced snowbands along the Lake Michigan shoreline. The movement of small disturbances that formed over Lake Michigan near curved coastlines was simulated by Hsu (1987), who attributed the development of convergence zones to the interaction of the mean wind and local circulations forced by thermal contrasts near curved shorelines. Pease et al. (1988) noted in satellite imagery the presence of a mesoscale vortex embedded within a lake-effect snowband over Lake Michigan, and model simulations showed that the formation of the snowband and vortex was the result of intense surface heating. Hjelmfelt (1990) simulated four types of lake-effect storms and evaluated the relative importance of environmental factors such as air–lake temperature difference, prevailing wind speed, and upwind humidity. Ballentine et al. (1992) performed sensitivity studies to determine the significance of water temperature, land and air temperature, humidity, wind speed, and wind direction on the intensity of a lake-effect snowband.

Building upon both observational studies and numerical model results, several forecasting procedures have been developed to improve the short-range prediction of lake-effect snow. Based on a 10-yr study of lake-effect events, Dewey (1979) used stepwise multiple discriminant analysis to develop a forecast scheme for Lake Ontario using several dominant snow predictors, the most notable being the lake–850 hPa temperature difference. Niziol (1987), Dockus (1988), and Murphy (1989) used the decision-tree approach to prepare forecast guidance for weather offices in the United States and Canada. Niziol concluded that the alignment of winds in the lower troposphere and the height of the capping inversion are important factors to consider in forecasting the location and intensity of lake-effect snowbands. Lake-effect snow forecast guidance using decision trees and multiple discriminant analysis was developed by Burrows (1991), and he included output from operational numerical models among the predictors.

Forecasters at the Buffalo (BUF) National Weather Service (NWS) office have taken this idea a step further by devising “locator charts” (Niziol et al. 1995) to predict the location of snowbands. A graphics package called BUFKIT was developed by forecasters at NWS Buffalo (Mahoney and Niziol 1997) to help predict the location, movement, and duration of lake-effect snowbands. BUFKIT allows forecasters to produce loops of hourly model soundings, time–height cross sections of wind shear, relative humidity, and inversion height, and hourly displays of most probable snowband location based on predictions from several of the operational models from the National Centers for Environmental Prediction (NCEP).

With the recent advances in computing capability, researchers and forecasters have turned to mesoscale models to shed further light on the lake-effect forecast problem. The Pennsylvania State University–National Center for Atmospheric Research (PSU–NCAR) Mesoscale Model version 5 (MM5) has figured prominently in these studies. A 30-km version of MM4 was used by Warner and Seaman (1990) to simulate a variety of mesoscale circulations, including lake-effect snowstorms in the northeastern United States. Later Sousounis and Fritsch (1994) used MM4 to show the significant role of the lake aggregate in modifying the synoptic-scale environment downwind. The present study aims to determine the degree to which a high-resolution version of MM5 can simulate the behavior of the intense snowband of 4–5 January 1995, which formed in a strong-wind environment, in the lee of Lake Ontario. To our knowledge this is the first multigrid application of MM5 to simulate such a lake-effect snowband. Specifically, this study will assess the degree to which the model successfully describes the development, structure, and movement this intense snowband. A primary objective is to determine the ability of MM5 to represent the effects of short-wave troughs propagating through the Great Lakes region on the development and behavior of this snowband. As part of this objective, we will compare MM5 and Eta Model simulations on a comparable grid (80-km resolution) of the larger-scale factors that affect snowband evolution. A second objective is to compare characteristics of the model snowband (e.g., location, orientation, movement, precipitation rate) with data from the Binghamton, New York (BGM), WSR-88D; IR satellite imagery; and the network of snowfall observers. These results will also provide insight as to the potential utility of MM5 for real-time lake-effect snow prediction, which is of primary interest to operational forecasters in the Great Lakes region.

2. Description of the 4–5 January 1995 snowstorm

The most significant lake-effect snow event of the 1994–95 season in New York State occurred downwind of the Great Lakes just after New Year’s weekend. The following case study description and subsequent model simulation will focus on a 36-h period commencing at 0000 UTC 4 January during which very heavy snow fell within this protracted lake-effect event. The total observed snowfall for the event (2–5 January 1995) for the counties east of Lake Ontario is shown in Fig. 1b. The heaviest reported snowfall was 157 cm (62 in.) at Montague on Tug Hill. Nearby Hooker (see locator map, Fig. 1a) reported 115 cm (45 in.) of snow of which 89 cm (35 in.) fell during the period of the model simulation. A secondary maximum appears in the analysis in southern Oswego County. Fulton (formerly NOO, now KFZY) reported 66 cm (26 in.) during the period 2100 UTC 4 January to 1200 UTC 5 January.

Fig. 1.

(a) Map showing locations referred to in the text; (b) observed snowfall 3–5 Jan 1995.

Fig. 1.

(a) Map showing locations referred to in the text; (b) observed snowfall 3–5 Jan 1995.

At 0000 UTC 4 January, a cold front extended from a surface cyclone over southeastern Ontario through eastern Lake Ontario and then southward through Pennsylvania (Fig. 2a). Two weak troughs at 850 hPa, one over central Pennsylvania extending into western New York (trough A in Fig. 3a), and a second over Wisconsin (trough B in Fig. 3a), were analyzed using GEMPAK weather analysis software (desJardins et al. 1991). The role of these troughs in modulating the intensity of the Lake Ontario snowband will be investigated in section 4 using output from the PSU–NCAR MM5. Moderately strong cold advection at 850 hPa was evident behind the front, extending from western Lake Ontario across the western Great Lakes region. Southwesterly flow was present throughout the lower troposphere ahead of a short-wave trough at 500 hPa located over Lake Superior (Fig. 4a). While clouds were present over and downwind of all the lakes, the rapid development of convective activity east of Lake Superior (apparent in the satellite imagery, not shown) shortly after 0000 UTC on 4 January, and then over northern Lake Huron beginning around 0800 UTC, is of particular interest in the present study.

Fig. 2.

Surface analysis for 4–5 Jan 1995. Sea level pressure (hPa), solid; temperature (°C), dashed; wind barb, 5 m s−1; half barb, 2.5 m s−1: (a) 0000 UTC 4 Jan, (b) 1200 UTC 4 Jan, (c) 0000 UTC 5 Jan, (d) 1200 UTC 5 Jan.

Fig. 2.

Surface analysis for 4–5 Jan 1995. Sea level pressure (hPa), solid; temperature (°C), dashed; wind barb, 5 m s−1; half barb, 2.5 m s−1: (a) 0000 UTC 4 Jan, (b) 1200 UTC 4 Jan, (c) 0000 UTC 5 Jan, (d) 1200 UTC 5 Jan.

Fig. 3.

850-hPa analysis for 4–5 Jan 1995. Height (dam), solid; temperature (°C), dashed; wind flag, 25 m s−1; full barb, 5 m s−1; half barb, 2.5 m s−1: (a) 0000 UTC 4 Jan, (b) 1200 UTC 4 Jan, (c) 0000 UTC 5 Jan, (d) 1200 UTC 5 Jan. Locations of trough A and trough B (discussed in the text) at 0000 UTC are shown in (a).

Fig. 3.

850-hPa analysis for 4–5 Jan 1995. Height (dam), solid; temperature (°C), dashed; wind flag, 25 m s−1; full barb, 5 m s−1; half barb, 2.5 m s−1: (a) 0000 UTC 4 Jan, (b) 1200 UTC 4 Jan, (c) 0000 UTC 5 Jan, (d) 1200 UTC 5 Jan. Locations of trough A and trough B (discussed in the text) at 0000 UTC are shown in (a).

Fig. 4.

500-hPa analysis for 4–5 Jan 1995. Height (dam), solid; absolute vorticity (10−5 s−1), dashed; winds same as in Fig. 3: (a) 0000 UTC 4 Jan, (b) 1200 UTC 4 Jan, (c) 0000 UTC 5 Jan, (d) 1200 UTC 5 Jan.

Fig. 4.

500-hPa analysis for 4–5 Jan 1995. Height (dam), solid; absolute vorticity (10−5 s−1), dashed; winds same as in Fig. 3: (a) 0000 UTC 4 Jan, (b) 1200 UTC 4 Jan, (c) 0000 UTC 5 Jan, (d) 1200 UTC 5 Jan.

The Buffalo sounding (Fig. 5a) indicated a shallow mixed layer up to a weak inversion at 925 hPa, with a nearly saturated moist-adiabatic layer extending above that level to 730 hPa. The average water temperature of Lake Ontario was estimated to be about 4°C based on satellite data from the Great Lakes Environmental Research Laboratory. An 850-hPa temperature of −12°C combined with a lake temperature of 4°C exceeded the Holroyd (1971) criterion (TlakeT850 > 13°C) for an unstable lapse rate. The Weather Surveillance Radar-1988 Doppler (WSR-88D) at BGM and IR satellite imagery confirmed that the Lake Ontario snowband formed over southern Jefferson County at about 0300 UTC. The counties referred to in the paper are identified in Fig. 1b.

Fig. 5.

Soundings for Buffalo, NY, for 4–5 Jan 1995. Winds same as in Fig. 3: (a) 0000 UTC 4 Jan, (b) 1200 UTC 4 Jan, (c) 0000 UTC 5 Jan, (d) 1200 UTC 5 Jan. Temperature, solid, and dewpoint, dashed.

Fig. 5.

Soundings for Buffalo, NY, for 4–5 Jan 1995. Winds same as in Fig. 3: (a) 0000 UTC 4 Jan, (b) 1200 UTC 4 Jan, (c) 0000 UTC 5 Jan, (d) 1200 UTC 5 Jan. Temperature, solid, and dewpoint, dashed.

At 0600 UTC 4 January, the cold front passed to the east of Lake Ontario and the surface winds over western New York veered to westerly. The 6-h forecast by the Eta Model of vertical velocity (i.e., omega) at 750 hPa (Fig. 6) included an area of ascending motion over central New York with a center of upward motion over the eastern end of Lake Ontario just ahead of trough A. The Eta predicted a second slightly stronger core of upward motion over the eastern portion of Lake Huron ahead of trough B. The upward motion and cloud development with trough B increased rapidly after 0600 UTC. The IR imagery indicated cloud streets over and to the lee of the lakes upwind of Lake Ontario. Strong cold advection over the warmer waters of Lake Ontario had destabilized the environment sufficiently to initiate a strong east–west snowband parallel to the lower-tropospheric flow. Peak reflectivity detected by the BGM radar was 28 dBZ in a small area just offshore (Fig. 7b). This initial episode of heavy snowfall was associated with temperatures of −13° to −16°C at 850 hPa, which is close to the level of nondivergence in the convective mixed layer. This is a temperature range conducive to maximum dendritic crystal growth and subsequent snowflake production according to Auer and White (1982) and Wesley et al. (1990). While the IR imagery suggests a link between the Lake Superior snowband and a new band that formed over northern Lake Huron around 0500 UTC, there appears to be no connection between the Lake Ontario snowband and bands to the lee of the upwind lakes.

Fig. 6.

The 6-h prediction by the Eta Model of vertical velocity at 750 hPa and wind at 950 hPa verifying 0600 UTC 4 Jan 1995. Contour interval for vertical velocity is 1 dPa s−1 with dashed representing upward motion and solid representing downward motion. Winds same as in Fig. 2.

Fig. 6.

The 6-h prediction by the Eta Model of vertical velocity at 750 hPa and wind at 950 hPa verifying 0600 UTC 4 Jan 1995. Contour interval for vertical velocity is 1 dPa s−1 with dashed representing upward motion and solid representing downward motion. Winds same as in Fig. 2.

Fig. 7.

Conditions for 0600 UTC 4 Jan. (a) Infrared GOES-8 satellite image; (b) Binghamton WSR-88D base reflectivity; (c) MM5-predicted precipitation (water equivalent, mm h−1); (d) MM5-predicted vertical velocity at 950 hPa (dPa s−1) on the 5-km grid. Winds same as in Fig. 2.

Fig. 7.

Conditions for 0600 UTC 4 Jan. (a) Infrared GOES-8 satellite image; (b) Binghamton WSR-88D base reflectivity; (c) MM5-predicted precipitation (water equivalent, mm h−1); (d) MM5-predicted vertical velocity at 950 hPa (dPa s−1) on the 5-km grid. Winds same as in Fig. 2.

During the period 0800–1000 UTC, the Lake Ontario snowband moved north and weakened considerably as evidenced by a narrower area of precipitation and smaller reflectivity values shown by the BGM radar (Fig. 8b). This temporary weakening of the snowband coincided with a marked decrease in the area covered by clouds over the western half of the lake. Also during this period, there was a dramatic increase in cloud growth over the northern half of Lake Huron (as indicated by a rapid decrease in cloud-top temperatures) as trough B approached from the west.

Fig. 8.

Same as Fig. 7 except for 1000 UTC 4 Jan.

Fig. 8.

Same as Fig. 7 except for 1000 UTC 4 Jan.

By 1200 UTC, a strong westerly flow of Arctic air was evident over the eastern Great Lakes region at both the surface (Fig. 2b) and at 850 hPa (Fig. 3b). The rapid modification of the arctic air mass resulted in an elongated thermally induced trough that extended from southeastern Ontario to Upper Michigan. This feature was a reflection of the large temperature contrast that resulted from passage of one air stream over land and an adjacent air stream over the Great Lakes. For example, Muskoka, Ontario, Canada (YQA), just to the east of Georgian Bay (see Fig. 1a), and close to the axis the thermal trough, reported −12°C, while Earlton, Ontario (YXR), located about 300 km north of YQA, reported −27°C. The contrast in dewpoint temperatures was even larger.

The 500-hPa vorticity maximum had moved over Lake Huron by 1200 UTC (Fig. 4b). Behind this short wave, there was dry air advection, synoptic-scale subsidence, and winds that backed with height. Drying and stabilizing of the air mass coincided with the weakening of the Lake Superior band. The backing of winds with height was most likely the result of a combination of cold-air advection and reduced vertical mixing in the more stable air mass over Lake Superior. Immediately ahead of this short wave, a rapid intensification of the Lake Huron–Georgian Bay snowband was occurring, especially during the 0800–1000 UTC period. This is the period during which the Lake Ontario band narrowed and weakened following the passage of trough A. As trough B progressed toward the east end of Lake Huron, the Lake Ontario band reintensified dramatically shortly after 1100 UTC with peak reflectivities of 24–28 dBZ (not shown) measured just inland from the shore. The band orientation continued to parallel the direction of the height contours at 850 hPa.

The 1200 UTC Buffalo sounding showed a mixed layer extending from the surface up to about 900 hPa (Fig. 5b), topped by a weak 50-hPa stable layer, then a nearly saturated moist-adiabatic layer from 850 to 715 hPa. Between 0000 and 1200 UTC 4 January, the 850-hPa temperature decreased to −18°C, creating moderate instability. During the period 1000–1500 UTC, the BGM radar and the IR satellite imagery (not shown) revealed that the east–west-oriented Lake Ontario snowband meandered north and south over the southern half of Jefferson County. The area of cloudiness seen on the IR satellite sequence propagating eastward from Lake Huron to northeastern New York during the period 1500–2200 UTC 4 January (Figs. 10a, 11a, and 12a later) was located in a region of ascending motion just ahead of trough B. Based on the 500-hPa wind and vorticity analyses (Fig. 4c), trough B had moved into northeastern New York by 0000 UTC on 5 January. The Lake Ontario snowband did not begin moving southward until just after this cloud mass, coincident with the trough, moved east of the lake. Time series of numerical model output will be examined in section 4 to help clarify the relation between the Lake Ontario snowband and the short-wave troughs crossing the Great Lakes.

Fig. 10.

Same as Fig. 7 except for 1500 UTC 4 Jan.

Fig. 10.

Same as Fig. 7 except for 1500 UTC 4 Jan.

Fig. 11.

Same as Fig. 7 except for 1900 UTC 4 Jan.

Fig. 11.

Same as Fig. 7 except for 1900 UTC 4 Jan.

Fig. 12.

Same as Fig. 7 except for 2200 UTC 4 Jan.

Fig. 12.

Same as Fig. 7 except for 2200 UTC 4 Jan.

During the period 1200–1800 UTC, the single east–west Lake Huron–Georgian Bay band broke into two separate weaker bands with a northwest-to-southeast orientation, one over the main portion of Lake Huron and the other over Georgian Bay. At the same time, the band over Lake Ontario remained very strong. By 1500 UTC, the Lake Ontario band had increased its length, extending from the western end of the lake to just west of Lake Champlain. Geostationary Operational Environmental Satellite-8 (GOES-8) 3.9-μm imagery at 1500 UTC 4 January (Fig. 9) revealed brighter (higher albedo) liquid water clouds over the center of Lake Ontario, with darker (lower albedo) glaciated clouds extending from eastern Lake Ontario inland. The utility of the 3.9-μm imagery is apparent in its clear delineation of glaciated regions where the Bergeron–Findeisen process is active, which correlates well with the region of heavy snowfall in this case. At 1500 UTC, the coldest clouds over and to the east of Lake Huron formed an elbow-shaped pattern (Fig. 10a). This is consistent with the presence of a northwest-to-southeast lake-effect snow band over Lake Huron and enhanced synoptic-scale ascent to the east of the lake.

Fig. 9.

GOES-8 3.8-μm (water vapor) image for 1500 UTC 4 Jan.

Fig. 9.

GOES-8 3.8-μm (water vapor) image for 1500 UTC 4 Jan.

By 1900 UTC, the very intense Lake Ontario snowband had moved northward into central Jefferson County, assuming a WSW–ENE orientation (Fig. 11a). The peak reflectivity measured by the BGM radar was between 20 and 24 dBZ (Fig. 11b). This is rather impressive considering that the snowband was nearly 200 km north of the radar. At this distance, the center of the 0.5° elevation beam slice was approximately 4300 m above ground level and was probably overshooting the highest reflectivity region. During the next few hours, the band moved rapidly southward in response to a wind shift with the passage of the short-wave trough (trough B).

By 2200 UTC (Figs. 12a,b), the band had moved to southern Oswego County, stalling near Fulton. The elbow-shaped appearance of the band at this time suggests the influence of the trough with southwest winds ahead and west-northwest winds behind it. During its rapid movement southward, the band produced a brief but intense snowburst, dumping 2.5–7.5 cm (1–3 in.) as it passed over the city of Oswego. The band made landfall on the southeast shore of Lake Ontario, passing inland from northeast Wayne County along a line to southern Oneida County. It also appears to have broadened and become more diffuse, but still exhibited moderate-to-strong intensity with peak reflectivities observed just offshore. As a result of the rapid movement of the band, the city of Oswego received less than 7.5 cm (3 in.) of snow, while Fulton, only 15 km away, received 66 cm (26 in.) in the 15 h ending 1200 UTC 5 January.

At 0000 UTC 5 January, the surface (Fig. 2c) and the 850-hPa (Fig. 3c) flow had veered to west-northwesterly, coincident with the migration of the short-wave trough through the region. The 500-hPa trough axis extended from southern Ontario southwestward through western New York to southwestern Ohio (Fig. 4c). The BUF sounding (Fig. 5c) showed a deep mixed layer extending up to 730 hPa. Cold advection had lowered the 850-hPa temperature to −22°C, with a lake-effect instability condition bordering on “extreme” according to Niziol (1987).

During the period 2300 UTC 4 January (not shown) to 0500 UTC 5 January there was little change in the orientation of the band, but there was a significant migration of the snowband axis. At 0000 UTC the snow band reached its southernmost position in northern Onondaga County, but then it moved rapidly back northward, a distance of about 20 km, by 0200 UTC. During the next several hours, it drifted slowly southward again (Figs. 13a,b) to near the Oswego–Onondaga County border where it remained for the rest of the night.

Fig. 13.

Same as Fig. 7 except for 0500 UTC 5 Jan.

Fig. 13.

Same as Fig. 7 except for 0500 UTC 5 Jan.

By 1200 UTC 5 January, BGM radar and IR imagery (not shown) suggest that the band had pivoted to a NNW–SSE orientation and was still located over central Oswego County near Fulton. The band was weakening in response to stabilization of the boundary layer due to lower-tropospheric warm advection (Fig. 3d). Examination of the 1200 UTC Buffalo sounding (Fig. 5d) showed a stable boundary layer with veering winds consistent with the warm advection in the 925–500-hPa layer. The substantial warming aloft had weakened the lake-effect instability to the “conditional” state (Niziol 1987). This, in combination with a substantial 50°–60° directional shearing of the lower-tropospheric winds, contributed to the observed weakening of the band. After 1200 UTC, the band was observed to continue a northward migration assuming a more east–west orientation prior to passing out of the BGM radar range.

3. Description of the model

The model used in this investigation was the nonhydrostatic version of the PSU–NCAR MM5 (Grell et al. 1994). Options chosen included the following: the high-resolution planetary boundary layer scheme (Zhang and Anthes 1982); the explicit moisture scheme with simplified treatment of ice and snow (Dudhia 1989); the upper radiative boundary condition (Dudhia 1993); and the Grell convective scheme (Grell 1993), moist vertical diffusion (Durran and Klemp 1982).

The model was run on a triply nested domain (Fig. 14) using 23 vertical layers. The largest grid (135-km resolution) was chosen to include all features that could propagate into the Lake Ontario region during the 36-h integration period. The 45-km grid provided increased horizontal resolution over the Great Lakes basin. The 15-km grid covered Lake Huron, Lake Erie, and Lake Ontario with sufficient resolution to identify interactions between the three eastern lakes. The 5-km grid provided more detail on the structure of the Lake Ontario snowband. Time-dependent lateral boundary conditions were applied to the 135-km grid, whereas two-way interactive nesting was used in the interior of the domain.

Fig. 14.

Map showing model domain and three nested grids. Largest domain is 135-km grid spacing, next largest is 45-km grid spacing, next smaller is 15-km grid spacing, and smallest is 5-km grid spacing.

Fig. 14.

Map showing model domain and three nested grids. Largest domain is 135-km grid spacing, next largest is 45-km grid spacing, next smaller is 15-km grid spacing, and smallest is 5-km grid spacing.

The initial data and boundary conditions for the model were computed by running the standard NCAR initialization programs. The program TERRAIN was used to specify terrain elevation and land-use categories, and DATAGRID provided a first guess on constant pressure surfaces from the NCEP global analysis on a 2.5° lat × 2.5° long grid for 0000 UTC 4 January 1995. The program RAWINS was used to refine the first guess by use of upper-air soundings and surface observations, and INTERP was used to interpolate the initial data and boundary data from pressure coordinates to the sigma coordinates chosen in MM5.

Runs were carried out using both the mixed phase ice scheme and the simple ice physics scheme of Dudhia (1989). While the simulated snowfall was about 20% less with the mixed phase scheme, the location and orientation of the precipitation contour pattern was virtually identical. Since the results of the two simulations were so similar, we chose the more efficient Dudhia scheme to approach the primary objectives of this study, that is, to determine the effects of the short-wave troughs on snowband behavior and to compare the characteristics of the simulated snowband with observations. Both of the runs failed to match the maximum snowfall reported in the Tug Hill plateau even assuming a 30:1 snow-to-liquid ratio. This is not surprising since even with 5-km resolution it is not possible to account for all of the precipitation physics.

Since the MM5 model produced negligible convective precipitation over the 5-km grid using the Grell scheme, we note that the model precipitation was almost entirely due to the explicit moisture scheme. According to Molinari and Dudek (1992), it remains uncertain whether a suitable cumulus parameterization exists for grid spacings between 3 and 20 km. The resolution of the grids for the simulations over the Great Lakes region (5 and 15 km) lie in Molinari and Dudek’s questionable range. Molinari and Dudek state that when grid-scale forcing is large and when instability is small or moderate, the explicit approach may be sufficient. Grid-scale forcing is very large with lake-effect snowbands due to strong low-level convergence. Lapse rates with soundings in the Great Lakes region were close to the moist-adiabatic lapse rate (e.g., see Figs. 5a–c), which represents at most a small degree of conditional instability.

The Blackadar planetary boundary layer (PBL) scheme (Zhang and Anthes 1982) was chosen in order to provide a reasonably accurate estimate of surface fluxes and turbulent diffusion within the boundary layer. With our choice of vertical levels in MM5, there were typically 10 model layers between the surface and 700 hPa. Therefore, use of the high-resolution Blackadar PBL scheme was justified despite the substantial increase in CPU time. It is worth noting that another choice (MRF option), based on the PBL scheme used in NCEP’s Medium-Range Forecast model (MRF), recently became available with MM5V2. The MRF scheme is similar to the Blackadar scheme and produces comparable results using about one-quarter of the CPU time.

In order to gain insight into how large-scale factors influenced the evolution of the Lake Ontario snowband, an MM5 simulation on a single 80-km grid (comparable to the resolution of that version of the “early Eta” Model, which was operational in January 1995) was also run using the same physics options described above. Since results from the 80-km simulation show less mesoscale detail than results from the nested-grid simulation, it is easier to establish the timing of the propagation of the short-wave troughs through the Great Lakes region. It will be shown that the timing of these features matches closely the periods of weakening and subsequent strengthening of the Lake Ontario snowband.

4. Description of results

Results from both the Eta Model and a single-grid 80-km MM5 simulation will be discussed for a 36-h run initialized at 0000 UTC 4 January 1995. Then we will present results of the MM5 nested-grid simulation, and discuss the results on the 15- and 5-km grids.

a. Initial analyses for the Eta and MM5

Data used to construct the MM5 initial analysis were derived from the NCEP global analysis on a 2.5° grid. As a result, short-wavelength features such as the weak 500-hPa trough over eastern lower Michigan (apparent in the Eta Model analysis, Fig. 15b) were absent in the MM5 analysis (Fig. 15a). The NCEP data was interpolated to an 80-km single grid covering approximately the same area as the Eta. Only minor differences between the Eta and MM5 can be seen in the initial 500-hPa vorticity field (Figs. 15a,b), with maxima over northwestern Wisconsin, northeastern Lake Superior, and just north of Lake Ontario. However, 500-hPa heights were generally lower over the Great Lakes region in the MM5. The 850-hPa height analysis was a little smoother with MM5 than with the Eta (Figs. 15c,d). Troughs at 850 hPa over Wisconsin and eastern Michigan in the Eta merged into a single trough extending from western Lake Superior to northwestern Pennsylvania in the MM5 analysis (Fig. 15c). Relative humidity was a little higher over the Great Lakes with the Eta despite slightly warmer air as implied by higher 500-hPa heights and higher 950–700-hPa thicknesses (Figs. 15e,f). The surface frontal trough over Pennsylvania was sharper and analyzed a little farther east with the Eta.

Fig. 15.

MM5 initial analysis on the 80-km grid and Eta Model initial analysis for 0000 UTC 4 Jan 4: (a) 500-hPa heights (dam), solid;absolute vorticity (10−5 s−1), dashed for MM5 analysis; winds as in Fig. 3. (b) Same as (a) except for Eta analysis. (c) 850-hPa temperature (°C), solid; and relative humidity, dashed for MM5; winds as in Fig. 3. (d) Same as (c) except for Eta analysis. (e) Sea level pressure (hPa), solid; 950–700-hPa thickness (dam), dashed for MM5. (f) Same as (e) except for Eta analysis.

Fig. 15.

MM5 initial analysis on the 80-km grid and Eta Model initial analysis for 0000 UTC 4 Jan 4: (a) 500-hPa heights (dam), solid;absolute vorticity (10−5 s−1), dashed for MM5 analysis; winds as in Fig. 3. (b) Same as (a) except for Eta analysis. (c) 850-hPa temperature (°C), solid; and relative humidity, dashed for MM5; winds as in Fig. 3. (d) Same as (c) except for Eta analysis. (e) Sea level pressure (hPa), solid; 950–700-hPa thickness (dam), dashed for MM5. (f) Same as (e) except for Eta analysis.

b. MM5 and Eta Model simulations on 80-km grids

It is of interest to compare the MM5 and the Eta Model simulation of synoptic-scale features. Of note were the lower-tropospheric troughs, which moved through the Great Lakes region during the period of the simulation. By 1200 UTC 4 January, an area of cloud enhancement over eastern Lake Huron (not shown) coincided with an area of moderate positive vorticity advection at 500 hPa just ahead of the trough axis (Fig. 4b). The sequence of IR satellite images between 0800 and 1200 UTC 4 January (only 1000 UTC image shown, Fig. 8a) clearly depicted the propagation of trough B (Fig. 3b) across the northern portion of Lake Huron. During this period, convective activity increased explosively in the Lake Huron region (cf. Figs. 7a and 8a), and the area covered by clouds decreased over western and southern Lake Ontario. As noted in section 2, the passage of trough B across Lake Huron had deformed the coldest clouds into an elbow-shaped pattern by 1500 UTC (Fig. 10a). Both models were accurate in placing the vorticity maximum over southern Lake Huron (Figs. 16a,b).

Fig. 16.

Same as Fig. 15 except for MM5 and Eta Model simulations verifying 1200 UTC 4 Jan.

Fig. 16.

Same as Fig. 15 except for MM5 and Eta Model simulations verifying 1200 UTC 4 Jan.

The surface winds predicted at 1200 UTC and at later times were somewhat stronger with MM5 than with the Eta. This difference can be attributed to the definition of “surface” wind, which is the 10-m wind in the Eta, and the lowest sigma level wind in MM5 (σ = 0.995 equivalent to about 40 m). Simulated mean sea level pressure in the St. Lawrence Valley region was slightly lower with the Eta. The simulated surface trough extending westward across the northern Great Lakes was similar in both models. However, the ridge along the Ohio River Valley and extending to the east was sharper, better defined, and located farther to the north in the Eta forecast.

To obtain higher temporal resolution, plots of vertical velocity at 750 hPa (Fig. 17) and divergence at 900 hPa (not shown) were made every 3 h. The 80-km MM5 run produced the strongest ascent of −5.5 dPa s−1 (−5.5 μb s−1) at 0900 UTC just east of Lake Huron during the time (and coinciding with the location) of the onset of most rapid cloud growth as shown in the IR satellite sequence. The MM5 model simulated the eastward propagation of a cell of upward vertical velocity at 750 hPa from the eastern shore of Lake Huron at 0900 UTC 4 January to the northern Adirondacks at 2100 UTC. Comparing the location of the cloud mass on the IR satellite sequence with the core of upward vertical velocity shown in the 80-km simulations, it is apparent that both the Eta and MM5 were a few hours too fast in moving this disturbance eastward. One of the results of this timing error shows up in the time series (Fig. 29c) of wind direction at Nine Mile Point (NMP). The wind at Nine Mile Point shifted to the northwest at around 2200 UTC, whereas the wind shift simulated by MM5 at NMP occurred about 1800 UTC. This timing error also explains the premature southward shift of the snowband simulated on the MM5 nested grids as compared to the movement of the snowband detected by the BGM radar (Fig. 18).

Fig. 17.

Same as Fig. 6 except for 80-km MM5 simulation verifying (a) 0600, (b) 0900, (c) 1200, (d) 1500, (e) 1800, and (f) 2100 UTC 4 Jan.

Fig. 17.

Same as Fig. 6 except for 80-km MM5 simulation verifying (a) 0600, (b) 0900, (c) 1200, (d) 1500, (e) 1800, and (f) 2100 UTC 4 Jan.

Fig. 29.

Same as Fig. 26 except for NMP.

Fig. 29.

Same as Fig. 26 except for NMP.

Fig. 18.

Time series of the latitude of the maximum radar reflectivity (solid) at longitude 76.1°N (east end of Lake Ontario), latitude of maximum upward vertical motion on 15-km grid (dotted), and on 5-km grid (dashed). The times for the radar extend 2 h beyond the period of the simulation.

Fig. 18.

Time series of the latitude of the maximum radar reflectivity (solid) at longitude 76.1°N (east end of Lake Ontario), latitude of maximum upward vertical motion on 15-km grid (dotted), and on 5-km grid (dashed). The times for the radar extend 2 h beyond the period of the simulation.

A time series of vertical motion (Fig. 19a) from the 80-km MM5 simulation at points S, C, and N in Fig. 1a indicates that there was a sharp decrease of upward motion between 0800 and 1000 UTC 4 January. There was also a sharp decrease in 900-hPa convergence during the period 0800–1100 UTC (Fig. 19b). This implied decrease in synoptic-scale forcing over Lake Ontario coincided with passage of trough A to the east of the lake and with the explosive cloud growth over northeastern Lake Huron related to the approach of trough B from the west. The timing of this apparent minimum in synoptic-scale forcing over Lake Ontario also coincided with the time (1000 UTC) of lowest reflectivity and smallest snowband width indicated by the BGM radar (Fig. 8b). The behavior of the snowband during this time period as simulated with the nested-grid version of MM5 will be discussed in section 4e.

Fig. 19.

Time series of output from 80-km MM5 simulation of (a) upward vertical velocity at 750 hPa (dPa s−1) and (b) divergence at 900 hPa (10−5 s−1) at the points S, C, and N identified in Fig. 1a.

Fig. 19.

Time series of output from 80-km MM5 simulation of (a) upward vertical velocity at 750 hPa (dPa s−1) and (b) divergence at 900 hPa (10−5 s−1) at the points S, C, and N identified in Fig. 1a.

By 0000 UTC 5 January, the 500-hPa trough axis in both simulations was located along the St. Lawrence River Valley extending southwestward to southern Indiana (Figs. 20a,b). At 850 hPa, both models predicted a pocket of cold air (T < −24°C) over southern lower Michigan and northwestern Ohio. The temperature simulated by MM5 was a little lower over southeastern New York. Both models simulated high relative humidity in the Great Lakes region with large gradients of moisture southeast of Lake Superior, across Lake Michigan, northern Indiana, central Ohio, and southeastern Pennsylvania. The Eta predicted a very large area of relative humidity greater than 80% covering northern New England, most of New York, Pennsylvania, and Michigan, while MM5 produced a narrow and discontinuous 80% zone, which extended from near Portland, Maine, westward to Lake Huron. Observed soundings in the Great Lakes region (e.g., Fig. 5c) suggest that the Eta prediction was too moist. The 850-hPa winds predicted by the Eta were west-northwesterly across New York and Pennsylvania while the MM5 winds were more westerly. The observed 850-hPa wind at 0000 UTC at Buffalo (Fig. 3c) was west-northwest, which was closer to the Eta solution. However, the MM5 winds over Lake Ontario at 850 hPa lined up better with the orientation of the snowband at that time. With regard to the sea level pressure forecasts at 0000 UTC 5 January, the Eta predicted more ridging over and to the northeast of Lake Huron (Fig. 20f), and slightly more ridging in northeastern Pennsylvania than MM5. Compared to the sea level pressure analysis (Fig. 2c), the Eta had too much ridging north of the eastern Great Lakes with pressures about 4 hPa too high near the north shore of Lake Huron. MM5 also produced excessive ridging north of Lake Huron with pressures about 2 hPa too high. Both models produced a weak trough from northern New England to northern New York similar to the trough analyzed in Fig. 2d. Since the excessive ridging was mainly located downwind of Lake Superior, a possible explanation is that the simulations with the 80-km resolution underestimated the bulk heating effects of the upwind lakes and, therefore, the magnitude of the thermally induced pressure deficit over and downwind of Lake Superior and Lake Huron (Sousounis and Fritsch 1994).

Fig. 20.

Same as Fig. 15 except for MM5 and Eta Model simulations verifying 0000 UTC 5 Jan.

Fig. 20.

Same as Fig. 15 except for MM5 and Eta Model simulations verifying 0000 UTC 5 Jan.

At 1200 UTC 5 January, both models produced northwesterly flow at 500 hPa over the Great Lakes region behind the trough axis along the East Coast (Figs. 21a,b). The models were in agreement in predicting negative vorticity advection (NVA) over eastern Lake Ontario. Despite the NVA, the Lake Ontario snowband continued to produce moderate snowfall in Oswego County. Comparing the 850-hPa temperature output (Figs. 21c and 21d, solid) with the 850-hPa temperature analysis (Fig. 3d, dashed), it appears that both the Eta and MM5 correctly placed the thermal trough (axis of cold air) along a line from Maine to West Virginia. Both models also correctly simulated the warm advection over all of the Great Lakes region under westerly winds. At the surface, both models produced a westerly gradient over the upper Midwest between a strong high over Kentucky and a low north of Lake Superior (Figs. 21e and 21f). Comparison with the surface analysis (Fig. 2d) indicates that both models also correctly produced a ridge over eastern New York. However, the pressures with the Eta were about 4 hPa too high from Ohio to central New York. The MM5 pressure pattern (Fig. 21e) was much closer to the observed pattern over northern Ohio, Pennsylvania, and New York State, although MM5 also produced too strong a high over Kentucky. The 225-dam contour of 950–700-hPa thickness was farther south and west with the Eta forecast than with MM5 consistent with the Eta’s slightly colder 850-hPa temperature and higher sea level pressure over northeast Pennsylvania and New York State.

Fig. 21.

Same as Fig. 15 except for MM5 and Eta Model simulations verifying 1200 UTC 5 Jan.

Fig. 21.

Same as Fig. 15 except for MM5 and Eta Model simulations verifying 1200 UTC 5 Jan.

c. Model soundings

The simulated Buffalo temperature profiles at 1200 UTC 4 January computed from the MM5 and the Eta model output (Figs. 22a,c) are similar to the observed sounding (Fig. 25e) indicating a nearly moist neutral stratification up to 710 hPa. Model soundings (especially MM5) were too dry in the 900–700-hPa layer. The MM5 simulations on the 80-km grid (Fig. 16c) and the 15-km grid (not shown) indicate a large north–south gradient of relative humidity ranging from 50% to 80% at 1200 UTC near Buffalo. The low dewpoints above 900 hPa in the MM5 solution appear to be the result of an incorrect placement of dry air advected eastward from lower Michigan into western New York. The wind profiles predicted by both models agreed well with observations above 850 hPa, but both models predicted too much southerly component below 900 hPa. It is likely that the west-northwesterly winds shown in the observed Buffalo sounding below 900 hPa (Fig. 22e) represented the inflow into a well-developed Lake Erie snowband (see Fig. 10a), whereas the west-southwest winds in the 880–720-hPa layer represented mesoscale outflow. The higher dewpoints observed in the 900–700-hPa layer may be the result of air advected with the outflow from the observed band just south of the airport. This suggests that the Lake Erie snowband (simulated by MM5 on the 15-km grid, Fig. 23a) was too weak and too far south of Buffalo at 1200 UTC 4 January.

Fig. 22.

MM5-predicted soundings, Eta-predicted soundings, and observed soundings at Buffalo, NY. Winds as in Fig. 3: (a) MM5 sounding at 1200 UTC 4 Jan, (b) MM5 sounding at 0000 UTC 5 Jan, (c) Eta sounding at 1200 UTC 4 Jan, (d) Eta sounding at 0000 UTC 5 Jan, (e) observed Buffalo sounding 1200 UTC 4 Jan, (f) observed Buffalo sounding 0000 UTC 5 Jan.

Fig. 22.

MM5-predicted soundings, Eta-predicted soundings, and observed soundings at Buffalo, NY. Winds as in Fig. 3: (a) MM5 sounding at 1200 UTC 4 Jan, (b) MM5 sounding at 0000 UTC 5 Jan, (c) Eta sounding at 1200 UTC 4 Jan, (d) Eta sounding at 0000 UTC 5 Jan, (e) observed Buffalo sounding 1200 UTC 4 Jan, (f) observed Buffalo sounding 0000 UTC 5 Jan.

Fig. 25.

Time series of maximum precipitation rate (solid) mm h−1 and maximum upward vertical velocity (dotted) dPa s−1 over the eastern third of the 5-km domain.

Fig. 25.

Time series of maximum precipitation rate (solid) mm h−1 and maximum upward vertical velocity (dotted) dPa s−1 over the eastern third of the 5-km domain.

Fig. 23.

MM5 predictions on the 15-km grid: (a) 6-h precipitation (water equivalent, mm) for the period ending 1200 UTC 4 Jan; (b) 950-hPa specific humidity (g kg−1) at 1200 UTC 4 Jan, winds as in Fig. 3; (c) same as (a) except for 1800 UTC 4 Jan; (d) same as (b) except for 1800 UTC 4 Jan; (e) same as (a) except for 0000 UTC 5 Jan; (f) same as (b) except for 0000 UTC 5 Jan.

Fig. 23.

MM5 predictions on the 15-km grid: (a) 6-h precipitation (water equivalent, mm) for the period ending 1200 UTC 4 Jan; (b) 950-hPa specific humidity (g kg−1) at 1200 UTC 4 Jan, winds as in Fig. 3; (c) same as (a) except for 1800 UTC 4 Jan; (d) same as (b) except for 1800 UTC 4 Jan; (e) same as (a) except for 0000 UTC 5 Jan; (f) same as (b) except for 0000 UTC 5 Jan.

At 0000 UTC 5 January, the simulated temperature profiles (Figs. 25b,d) were similar to the observed profile (Fig. 22f) indicating a well-mixed layer from the surface to 730 hPa. Both models were a little too cold at 800 hPa, and MM5 was a little too warm at 900 hPa. The Eta was a little too moist below 800 hPa. The models both predicted a veering wind with time of about 25° between 1200 UTC 4 January and 0000 UTC 5 January in the layer 850–500 hPa.

d. Simulations of precipitation and specifichumidity on the 15-km grid

It is of interest to inspect the MM5 results on the 15-km grid to study how the air was modified upwind of Lake Ontario. The maximum precipitation simulated on the 15-km grid for the 6 h ending 1200 UTC 4 January was greater than 8 mm at a location just east of Lake Ontario (Fig. 23a). The MM5 model produced a broad secondary maximum exceeding 4 mm along the southeast coast of Georgian Bay where west-northwesterly winds were forced up steeply sloping terrain south of the eastern end of Georgian Bay. Comparing the winds shown in Fig. 23b with the surface analysis at 1200 UTC 4 January (Fig. 2b), the model handled the movement of the surface trough reasonably well during the first 12 h of the run with west winds to the south of Georgian Bay and north winds to the east of Georgian Bay. The highest specific humidity at 950 hPa coincided with the snowband just east of Lake Ontario. The plumes of maximum specific humidity downwind of Lake Huron and Georgian Bay were separated by a relative minimum just to the east of the Shelburne Plateau, which is located between Toronto and the Bruce Peninsula (see Fig. 1a). The simulation on the 15-km grid indicates that the moisture from Lake Huron was channeled around this barrier.

The distribution of precipitation for the 6 h ending at 1800 UTC was similar to the distribution for the previous 6 h, although the area exceeding 8 mm east of Lake Ontario was slightly larger. The location of the 6-h precipitation was in good agreement with the observed location of maximum snowfall. The location of the model snowband was close to the axis of maximum simulated specific humidity over the southern portion of Lake Ontario. Tongues of drier air had advected eastward (in the model) during the 6-h period ending 1800 UTC with a core of driest air over the north shore of Lake Ontario and another plume of dry air over the Finger Lakes. These tongues contained air that never had a significant fetch over the open waters of the upwind lakes. The west-northwest winds that were simulated over the northern half of Lake Ontario funneled dry air into the region east of the lake a few hours too early as evidenced by the decrease in simulated dewpoint after 2200 UTC at Nine Mile Point (Fig. 29b, section 4). Air that crossed over Lake Superior and Lake Huron had a much higher moisture content. The snowband over Lake Ontario was augmented by this moisture-laden air from the upwind lakes. Although the predicted hourly (and 6 h) precipitation rates downwind were greater with the Lake Ontario band, the storm total precipitation was greatest at a grid point just south of Georgian Bay because there was little movement of the Lake Huron band during the period of the model simulation, and because of the lift induced along the northwest side of the Shelburne Peninsula.

e. Simulations of precipitation and verticalvelocity on the 5-km domain

The accumulated precipitation (water equivalent) simulated by MM5 (Fig. 24) featured a double maximum with 27.5 mm simulated near the Jefferson County–Lewis County line and a secondary maximum over Lake Ontario just north of Rochester. This was roughly similar to the distribution of observed snowfall (Fig. 1b). However, assuming a 25:1 snow-to-water ratio, the simulated precipitation amount was less than half of that observed in the Tug Hill region.

Fig. 24.

Storm total water equivalent precipitation (mm) predicted by MM5 on the 5-km grid.

Fig. 24.

Storm total water equivalent precipitation (mm) predicted by MM5 on the 5-km grid.

The model snowband first appeared with a double-band structure over the northern half of Lake Ontario at around 0300 UTC 4 January (not shown). Since this development occurred shortly after model initialization, we suspect that this bifurcation reflected the adjustment process rather than a representation of a physical reality.

By 0600 UTC (Fig. 7c), MM5 had developed a single midlake band extending eastward through southern Jefferson County. Greatest simulated precipitation rate was more than 3.5 mm h−1 at a point just offshore. Using an ice-to-water ratio of 25:1, which is a conservative estimate for the very low density snow that falls in lake-effect storms (Hill 1971), this translates to a snowfall rate of about 9 cm h−1 (3.5 in. h−1). Maximum upward motion at 950 hPa (Fig. 7d) exceeded −60 dPa s−1 (−60 μb s−1) in a small area just inland from the coast. The vertical motion pattern was similar at 850 hPa (not shown), but the maximum upward vertical motion exceeded −150 dPa s−1. By comparison, the 0600 UTC BGM (Fig 7b) radar data indicated an east–west band over northern Oswego County in good agreement with the simulated vertical velocity pattern. The MM5 model produced a distinct convergence zone at 950 hPa in northern Oswego County (Fig. 7d). Over Lake Ontario, the convergence line separated west-northwesterly winds over the northern half of the lake from west-southwesterlies over the southern part. The wind shift line, apparent over the central portion of the 5-km domain at 0600 UTC, may have been a reflection of a weak trough (trough A, Fig. 3a) passing through this region. The Eta Model forecast, verifying at 0600 UTC 4 January, indicated a weak cell of upward motion centered over the eastern end of Lake Ontario (Fig. 6). The 80-km MM5 simulation produced a zone of convergence at 950 hPa (not shown) extending from the St. Lawrence River southwestward over eastern Lake Ontario. This weak disturbance appeared to enhance low-level convergence and precipitation over the eastern portion of the lake and in counties east of the lake. However, the stronger short-wave trough (trough B) was still over the western Great Lakes at this time.

At 1000 UTC the BGM radar (Fig. 8b) indicated that the Lake Ontario snowband drifted northward and weakened considerably. The model precipitation rate (Fig. 8c) diminished greatly over the water and slightly over land. A time series of greatest hourly precipitation rate and maximum upward motion over the eastern third of Lake Ontario and the counties directly to the east of the lake (Fig. 25) indicated a minimum in snowband intensity between 0800 and 1000 UTC. This coincided with the minimum in upward motion and low-level convergence simulated on the single-domain 80-km grid (Figs. 19a,b). The strong implication is that the minimum in synoptic-scale forcing, apparent in the 80-km MM5 run, was responsible for a temporary weakening in snowband intensity just after trough A moved east of Lake Ontario and trough B was beginning to intensify over Lake Huron during the period 0800–1100 UTC. Simulated vertical velocity (Fig. 8d) diminished considerably near the eastern shore in agreement with the radar imagery.

During the period 1000–1200 UTC 4 January, there was a large increase in reflectivity, although some of the increase was likely due to decreased distance from the radar. At 1500 UTC, the BGM radar (Fig. 10b) showed a slight southward drift of the snowband. The model precipitation (Fig. 10c) agreed very well with the BGM radar in southern Jefferson County, but MM5 developed a core of heavier snowfall over northern Lewis County close to the area of greatest observed snowfall (see Fig. 1b). The BGM radar displayed relatively low reflectivity in the region where observed precipitation was greatest (i.e., Lewis County). This could be due to greater distance from the antenna and to the drift of snowflakes with the strong west-to-southwest winds. At 950 hPa, MM5 produced a narrow core of upward motion exceeding −20 dPa s−1 extending across the lake into Jefferson County in an area of strong convergence (Fig. 10d).

The simulated snowband moved rapidly southward between 1700 and 1900 UTC. By 1900 UTC (Fig. 11c), the model snowband was located along the south shore of Lake Ontario with an elbow extending northeastward into the Tug Hill region. In contrast, the radar and IR imagery (Figs. 11a,b) indicated that the band was still in southern Jefferson County. The model winds shifted to west-northwest over the northern portion of the lake indicating that trough B, as simulated in the model, had passed through most of Lake Ontario.

At 2200 UTC (Fig. 12d), the model snowband was aligned along the south shore of Lake Ontario with greatest simulated upward motion over the lake. The simulated precipitation rate over land was considerably smaller than it was earlier when the model band was in its midlake position. It appears from the radar and spotter reports that the model underestimated snowfall in southern Oswego County after 2200 UTC although the model snowband was in approximately the correct location.

By 0500 UTC 5 January (Fig. 13), the MM5 snowband extended from the southern shore of Lake Ontario east-southeastward into Onondaga County. This was close to the observed position, but the simulated precipitation was much less than observed. During the period 0500–1200 UTC 5 January, the snowband weakened as it drifted northward while maintaining a northwest–southeast orientation.

f. Time series comparisons

The time series of observed and simulated surface data for Watertown (ART), Syracuse (SYR), Rochester (ROC), and NMP are shown in Figs. 26–29. The simulated surface values are the mean value in the lowest model layer, representing conditions at about 40 m above the ground. The MM5 temperature, dewpoint, and wind direction for ART agreed well with observations during the first 29 h of the simulation. After 0600 UTC 5 January (forecast hour 30), the observed wind speed decreased to less than 3 m s−1 and the wind direction shifted abruptly from northwest to northeast apparently as the result of a land breeze. In addition, the observed temperature at ART dropped by 10°C between 0500 and 1000 UTC on 5 January. The model failed to capture either the wind shift or the temperature drop at ART, which may have been augmented by cold-air drainage from higher terrain to the east.

Fig. 26.

Observed and simulated time series of surface conditions for ART: (a) temperature (°C), (b) dewpoint (°C), (c) wind direction (°), (d) wind speed (kt).

Fig. 26.

Observed and simulated time series of surface conditions for ART: (a) temperature (°C), (b) dewpoint (°C), (c) wind direction (°), (d) wind speed (kt).

The MM5 simulations of the temperature were reasonably accurate at ROC and SYR except during the daylight hours (roughly 1400 to 2200 UTC) when MM5 was a few degrees too cold. One possible explanation for the lower MM5 temperatures is that Rochester (and to a lesser extent Syracuse) was located under the southern edge of a simulated cloud deck that appeared farther south in the model than in the observed satellite imagery. The simulated wind direction was generally good at SYR and ROC, but the model developed excessive wind speeds especially at ROC during the second half of the simulation. It should be noted that the simulated and the observed wind direction at ROC always maintained a southerly component even after passage of the short-wave trough. A possible explanation is that this southerly component was part of an isallobaric wind directed toward the area of mesoscale convergence over the lake.

The time series of simulated temperatures at Nine Mile Point was in generally good agreement with the observations throughout the period, but the MM5 dewpoint was too low after 2200 UTC. Some possible explanations are 1) MM5 may have underestimated moisture flux from Lake Ontario; 2) MM5 may have underestimated the relative humidity of the air upwind of Lake Ontario; 3) MM5 may have overestimated the precipitation over the southern portion of Lake Ontario, which would have resulted in some decrease in water vapor content over that portion of the lake just upwind of Nine Mile Point; and 4) the dewpoint at Nine Mile Point may not have been representative of conditions along the eastern shore. The fact that MM5 brought the trough through too early in the triply nested simulation is apparent in the wind direction plot showing a wind shift from west-southwest to west-northwest just after 1800 UTC in the model compared to just after 2200 UTC in the observed winds.

5. Discussion

The results presented in section 4 show that MM5 is capable of simulating the formation of the intense lake-effect snowband that formed on 4 January 1995 under strong west-to-northwest winds over Lake Ontario. On the 5-km grid, the model snowband had a width, length, and orientation similar to the actual snowband as depicted by the Binghamton WSR-88D radar and GOES-8 satellite imagery. The snowbands simulated by MM5 on the 15-km grid (over Lake Huron and Lake Erie) are less intense than the bands on the 5-km grid, but the size and orientation appear to be realistic based on satellite imagery. In another study, we found that MM5 captured the mesoscale convergence zone of the snowband quite well even with 20-km resolution, although the predicted precipitation was only about one-third the amount predicted on the 5-km grid and only about one-sixth of the observed snowfall (assuming a 25:1 snow-to-liquid ratio).

MM5 accurately simulated the location and movement of the Lake Ontario snowband during the first half of the run, moved the band southward about 3–4 h too early, then kept the band in about the correct location during the last 14 h. The southward shift of the model snowband appears to have resulted from the premature development of northwesterly winds over the northern half of Lake Ontario beginning around 1700 UTC 4 January. Such a wind regime was observed after the short-wave trough (trough B in Fig. 3a) crossed Lake Ontario, which suggests that MM5 brought the trough through too early. Comparing Figs. 20e and 2c, it is clear that MM5 produces a trough east of Lake Huron that is slightly too intense with pressures about 1–2 hPa too low. Since the timing error was also present in the 80-km simulation, it is possible that the premature movement of the trough and wind-shift line over Lake Ontario was the result of insufficient resolution in the NCEP 2.5° data used to initialize MM5. Although the timing was in error, the high-resolution MM5 simulation was able to move the snowband southward in response to passage of the synoptic-scale trough.

The MM5 winds at NMP were stronger than the observed winds during the period 1000–1800 UTC 4 January (Fig. 29d). The observed and the simulated snowbands were both north of NMP during this period. The MM5 winds were also too strong at ROC after 2100 UTC 4 January (Fig. 28d). This suggests that the predicted pressure gradient was excessive on the south side of the model snowband. One possible explanation for this excessive pressure gradient was that the simulated temperature was too cold over land just south of Lake Ontario. For example, at both SYR (Fig. 27a) and ROC (Fig. 28a), MM5 was more than 3°C too cold at 1800 UTC. Therefore, the model simulated too large a thermal contrast (and therefore too strong a pressure gradient) between land and water during the daytime. The lower-than-observed temperatures at SYR and ROC may have resulted from inaccurate placement of cloud cover in the MM5 simulation, premature passage of the simulated trough, and/or inaccurate representation of the diurnal heating cycle by MM5.

Fig. 28.

Same as Fig. 26 except for ROC.

Fig. 28.

Same as Fig. 26 except for ROC.

Fig. 27.

Same as Fig. 26 except for SYR.

Fig. 27.

Same as Fig. 26 except for SYR.

When the model snowband was located east of Lake Ontario, the maximum precipitation rate was comparable to typical maximum snowfall rates of about 10 cm h−1 (4 in. h−1) observed with intense snowbands. However, when the model snowband reached the south shore of Lake Ontario, the simulated precipitation appeared to be excessive over the lake and was much less than observed at inland locations such as Fulton, New York. One possibility is that snowflakes fell out too rapidly over the lake. However, it is not clear why this would have been more of a problem when the band moved close to the south shore. Another possibility is that the stronger-than-observed winds along the south shore of Lake Ontario may have contributed to the unrealistically large vertical velocities predicted in that portion of the model snowband over the lake just to the northwest of ROC. Small cells of upward motion exceeding 150 dPa s−1 were simulated at 850 hPa (not shown) during the period 0000–0600 UTC 5 January. We suspect that a feedback mechanism, similar to the conditional instability of the second kind that operates in tropical cyclones, caused MM5 to produce excessive precipitation over the lake after 0000 UTC 5 January. As a result, there would have been less water vapor and cloud water available in the model to produce the heavy snowfall that was observed over land at locations such as Fulton after 0000 UTC.

Based on MM5 predictions on the 15-km grid, there were no direct connections between the snowbands that formed downwind of Lake Huron–Georgian Bay and the band that formed over Lake Ontario. However, the simulated moisture plumes downwind of Lake Huron and Georgian Bay appear to have fed directly into the Lake Ontario snowband. This probably enhanced the precipitation from the simulated Lake Ontario band by enriching the environmental moisture field. These moisture plumes were deflected around both sides of the Shelburne Plateau. Future research will focus on this lake-to-lake interaction.

The results of the 4–5 January 1995 lake-effect snowstorm simulation are sufficiently encouraging to indicate that MM5 may be used to make real-time forecasts of lake-effect snowstorms. The lead author is participating in a COMET cooperative project between State University of New York (SUNY) Oswego, SUNY Brockport, Cornell University, and the National Weather Service Forecast Offices at Binghamton and Buffalo. One of the goals of the project is to provide hourly MM5 output to NWS forecasters. On a 715/75MHz Hewlett-Packard workstation, it takes MM5 7.5 h to complete a 24-h simulation over a 60-km large grid and a 20-km nested grid over Lakes Huron, Erie, and Ontario. Model predictions are converted to GEMPAK format (Schleede et al. 1995). Initial data and lateral boundary conditions are computed using the early Eta Model grids available from Unidata. Despite relatively coarse resolution, MM5 has made useful forecasts of snowband location and movement during the winter of 1996/97 (Waldstreicher et al. 1998). Relevant material related to the 4–5 January 1995 case, including some MM5 output, is available to the operational and academic communities online from the Joint Office of Science Suppport (http://www.comet.ucar.edu/resourses/cases/c5_05jan95/).

6. Summary and conclusions

The Pennsylvania State University–NCAR Mesoscale Model can simulate the development of the 4–5 January 1995 lake-effect snowband on grids ranging in horizontal resolution from 5 to 20 km. On the 5-km grid, the model snowband has a size and orientation similar to that depicted by the WSR-88D radar. The simulated precipitation distribution matches the observed precipitation very well during the first half of the run when the snowband was over central Lake Ontario and extended inland to the Tug Hill plateau. The model underpredicted snowfall at Fulton, New York, and appears to have overpredicted precipitation over the lake during the time when the snowband was located along the south shore of Lake Ontario. This suggests that the physics relating to shoreline contrasts needs some further study.

Hourly plots of vertical motion at 750 hPa (approximate cloud level) and horizontal divergence at 900 hPa from the 80-km MM5 simulation were used to help establish the timing of the two short-wave troughs that crossed the Great Lakes region during the period of the simulation. As the first trough (trough A) moved east of Lake Ontario a little after 0700 UTC 4 January, MM5 correctly simulated the temporary weakening of the snowband as depicted by the BGM radar. Both the upward vertical motion and low-level convergence simulated on the nested grid decreased sharply after passage of trough A, then increased gradually during the next few hours as trough B approached from the west. The model was correct in moving the snowband into southern Oswego County late on 4 January just after the simulated trough B moved east of Lake Ontario. However, the simulated trough passage and the southward shift of the snowband were a few hours too early.

The 15-km results indicate that plumes of moisture from Lake Huron were channeled around the Shelburne Peninsula before they fed into the western end of Lake Ontario, and a plume of moisture from Georgian Bay was advected to the northwest shore of Lake Ontario. Since it is likely that these plumes had an impact on the location and intensity of the Lake Ontario bands, we plan to carry out sensitivity experiments in a future study wherein Georgian Bay and/or the main body of Lake Huron are “frozen” and/or removed completely.

Acknowledgments

We acknowledge the Mesoscale and Microscale Meteorology Division of NCAR for providing training in the use of the workstation version of MM5. We are grateful to the COMET program for providing surface and upper-air data and satellite imagery for the 4–5 January 1995 storm. This research was supported by NSF Grant ATM-9224384. This paper is funded in part from COMET Outreach Program Subaward (S96-75666) under a cooperative agreement between the National Oceanic and Atmospheric Administration (NOAA) and the University Corporation for Atmospheric Research (UCAR). The views expressed herein are those of the authors and do not necessarily reflect the views of NOAA, its subagencies, or UCAR.

REFERENCES

REFERENCES
Auer, A. H., Jr., and J. M. White, 1982: The combined role of kinematics, thermodynamics and cloud physics associated with heavy snowfall episodes. J. Meteor. Soc. Japan, 60, 500–507.
Ballentine, R. J., 1982: Numerical simulation of land-breeze-induced snowbands along the western shore of Lake Michigan. Mon. Wea. Rev., 110, 1544–1553.
——, E. C. Chermack, A. Stamm, D. Frank, M. Thomas, and G. Beck, 1992: Preliminary numerical simulations of the 31 January 1991 lake-effect snowstorm. Proc. 49th Annual Eastern Snow Conf., Oswego, NY, CRREL, 115–123. [Available from U.S. Army, CRREL, 72 Lyme Rd., Hanover, NH 03755.]
.
Burrows, W. R., 1991: Objective guidance for 0–24-hour and 24–48-hour mesoscale forecasts of lake-effect snow using CART. Wea. Forecasting, 6, 357–378.
Byrd, G. P., R. A. Anstett, J. E. Heim, and D. M. Usinski, 1991: Mobile sounding observations of lake-effect snowbands in western and central New York. Mon. Wea. Rev., 119, 2323–2332.
Carpenter, D., 1993: The lake-effect of the Great Salt Lake: Overview and forecast problems. Wea. Forecasting, 8, 181–193.
desJardins, M. L., K. F. Brill, and S. S. Schotz, 1991: Use of GEMPAK on UNIX workstations. Proc. Seventh Int. Conf. on Interactive Information and Processing Systems for Meteorology, Oceanography, and Hydrology, New Orleans, LA, Amer. Meteor. Soc., 449–453.
Dewey, K. F., 1979: An objective forecast method developed for Lake Ontario–induced snowfall systems. J. Appl. Meteor., 18, 787–793.
Dockus, D. A., 1988: DDT II: Computerized lake-effect snow forecasts. Natl. Wea. Dig., 13, 18–26.
Dudhia, J., 1989: Numerical study of convection observed during the Winter Monsoon Experiment using a mesoscale two-dimensional model. J. Atmos. Sci., 46, 3077–3107.
——, 1993: A nonhydrostatic version of the Penn State–NCAR mesoscale model: Validation tests and simulation of an Atlantic cyclone and cold front. Mon. Wea. Rev., 121, 1493–1513.
Durran, D. R., and J. B. Klemp, 1982: On the effects of moisture on the Brunt-Väisälä frequency. J. Atmos. Sci., 39, 2152–2158.
Ellenton, G. E., and M. B. Danard, 1979: Inclusion of sensible heating in convective parameterization applied to lake-effect snow. Mon. Wea. Rev., 107, 551–565.
Grell, G. A., 1993: Prognostic evaluation of assumptions used by cumulus parameterizations. Mon. Wea. Rev., 121, 764–787.
——, J. Dudhia, and D. R. Stauffer, 1994: A description of the fifth-generation Penn State/NCAR mesoscale model (MM5). NCAR Tech. Note TN-398+STR, 122 pp.
Hill, J. D., 1971: Snow squalls in the lee of Lake Erie and Lake Ontario: A review of the literature. NOAA Tech. Memo. NWS ER-43, 20 pp. [NTIS COM-72-00959.]
.
Hjelmfelt, M. R., 1990: Numerical study of the influence of environmental conditions on lake effect snowstorms over Lake Michigan. Mon. Wea. Rev., 118, 138–150.
——, and R. R. Braham Jr., 1983: Numerical simulation of the airflow over Lake Michigan for a major lake-effect snow event. Mon. Wea. Rev., 111, 205–219.
Holroyd, E. W., III, 1971: Lake-effect cloud bands as seen from satellites. J. Atmos. Sci., 28, 1165–1170.
Hsu, H., 1987: Mesoscale lake-effect snowstorms in the vicinity of Lake Michigan: Linear theory and numerical simulations. J. Atmos. Sci., 44, 1019–1040.
Jiusto, J., and M. Kaplan, 1972: Snowfall from lake-effect storms. Mon. Wea. Rev., 100, 62–66.
Lavoie, R. L., 1972: A mesoscale numerical model of lake-effect storms. J. Atmos. Sci., 29, 1025–1040.
Mahoney, E. A., and T. A. Niziol, 1997: BUFKIT: A software application tool kit for predicting lake-effect snow. Preprints, 13th Int. Conf. on Interactive Information and Processing Systems for Meteorology, Oceanography, and Hydrology, Long Beach, CA, Amer. Meteor. Soc., 388–391.
Molinari, J., and M. Dudek, 1992: Parameterization of convective precipitation in mesoscale numerical models: A critical review. Mon. Wea. Rev., 120, 326–344.
Murphy, B. P., 1989: Forecasting lake effect snow in Ontario. Ontario Regional Tech. Note 89-7. Atmospheric Environment Service, Toronto, ON, Canada, 21 pp. [Available from Environment Canada Printing Services, 4905 Dufferin Street, Toronto, ON M3H 5T4, Canada.]
.
Niziol, T. A., 1987: Operational forecasting of lake-effect snowfall in western and central New York. Wea. Forecasting, 2, 310–321.
——, W. R. Snyder, and J. S. Waldstreicher, 1995: Winter weather forecasting throughout the eastern United States. Part IV: Lake effect snow. Wea. Forecasting, 10, 61–77.
Passarelli, R. E., and R. R. Braham Jr., 1981: The role of the winter land breeze in the formation of Great Lakes snowstorms. Bull. Amer. Meteor. Soc., 62, 482–491.
Peace, R. L., and R. B. Sykes, 1966: Mesoscale study of a lake-effect snowstorm. Mon. Wea. Rev., 94, 495–507.
Pease, S. R., W. A. Lyons, C. S. Keen, and M. R. Hjelmfelt, 1988: Mesoscale spiral vortex embedded within a Lake Michigan snow squall band: High resolution satellite observations and numerical model simulations. Mon. Wea. Rev., 116, 1374–1380.
Reinking, R. F., and Coauthors, 1993: The Lake Ontario Winter Storms (LOWS) project. Bull. Amer. Meteor. Soc., 74, 1828–1849.
Schleede, D., J. Case, L. Keshishian, R. Ballentine, and G. Byrd, 1995: Processing of mm5 output files into an interactive graphical environment called GEMPAK. Fifth PSU/NCAR Mesoscale Model Users Workshop, Boulder, CO, NCAR, 11.
Sousounis, P. J., and J. M. Fritsch, 1994: Lake aggregate mesoscale disturbances. Part II: A case study of the effects on regional and synoptic-scale weather systems. Bull. Amer. Meteor. Soc., 75, 1793–1811.
Waldstreicher, J. S., E. A. Mahoney, R. Ballentine, D. Schleede, J. Maliekal, and S. Colucci, 1998: Operational use of a mesoscale model for predicting lake effect snow in upstate New York. Preprints, 17th Conf. on Weather Analysis and Forecasting, Phoenix, AZ, Amer. Meteor. Soc., 393–396.
Warner, T. T., and N. L. Seaman, 1990: A real-time, mesoscale numerical weather prediction system used for research, teaching, and public service at the Pennsylvania State University. Bull. Amer. Meteor. Soc., 71, 792–805.
Wesley, D. A., J. F. Weaver, and R. A. Pielke, 1990: Heavy snowfall during an Arctic outbreak along the Colorado Front Range. Natl.Wea. Dig., 15, 2–19.
Wiggin, B. L., 1950: Great snows of the Great Lakes. Weatherwise, 3, 123–126.
Zhang, D. L., and R. A. Anthes, 1982: A high resolution model of the planetary boundary layer—Sensitivity tests and comparisons with SESAME-79 data. J. Appl. Meteor., 21, 1594–1609.

Footnotes

Corresponding author address: Dr. Robert J. Ballentine, Department of Earth Sciences, SUNY Oswego, Oswego, NY 13126.