The large-scale and mesoscale structure of the Great Salt Lake–effect snowstorm of 7 December 1998 is examined using radar analyses, high-density surface observations, conventional meteorological data, and a simulation by the Pennsylvania State University–National Center for Atmospheric Research fifth generation Mesoscale Model (MM5). Environmental conditions during the event were characterized by a lake–700-hPa temperature difference of up to 22.5°C, a lake–land temperature difference as large as 10°C, and conditionally unstable low-level lapse rates. The primary snowband of the event formed along a land-breeze front near the west shoreline of the Great Salt Lake. The snowband then migrated eastward and merged with a weaker snowband as the land-breeze front moved eastward, offshore flow developed from the eastern shoreline, and low-level convergence developed near the midlake axis. Snowfall accumulations reached 36 cm and were heaviest in a narrow, 10-km-wide band that extended downstream from the southern shore of the Great Salt Lake. Thus, although the Great Salt Lake is relatively small in scale compared to the Great Lakes, it is capable of inducing thermally driven circulations and banded precipitation structures similar to those observed in lake-effect regions of the eastern United States and Canada.
The prediction of lake-effect snowstorms that develop over and downwind of the Great Salt Lake (GSL) is one of the major forecast challenges facing meteorologists in northern Utah. Occurring several times each year, Great Salt Lake–effect (GSLE) snowstorms last from a few hours to more than a day, frequently produce snowfalls of 20–50 cm, and have contributed to the state record 129-cm lowland storm-total snowfall that was observed near Salt Lake City (SLC) from 24 to 28 February 1998 (Carpenter 1993; Slemmer 1998; Steenburgh et al. 2000). Despite significant improvement in observational technologies and numerical forecast systems, GSLE snowstorms remain difficult to predict with lead times of more than a few hours.
Previous studies have identified the climatological characteristics, large-scale conditions, and mesoscale precipitation structures associated with GSLE snowstorms. Based on lake-effect events identified by visual observations and spotter reports, Carpenter (1993) found that GSLE snowstorms were associated with post-cold-frontal northwesterly flow at 700 hPa, a lake–700-hPa temperature difference of at least 17°C (which approximately represents a dry adiabatic lapse rate), and an absence of stable layers or inversions near or below 700 hPa.1 Steenburgh et al. (2000) used observations from a recently installed National Weather Service Weather Surveillance Radar-1988 Doppler (WSR-88D) to identify GSLE events between September 1994 and May 1998. During this period, 16 well-defined GSLE events were observed, with the synoptic, mesoscale, and convective characteristics of these events examined using National Centers for Environmental Prediction (NCEP) Rapid Update Cycle version 2 analyses (RUC2;Benjamin et al. 1991, 1994), SLC radiosonde observations, and local WSR-88D radar observations. In addition to supporting the findings of Carpenter (1993), Steenburgh et al. (2000) also found that GSLE events tend to occur during periods of positive lake–land temperature differences, usually exceeding 6°C, and are most active during the overnight and early morning hours. It was hypothesized that the positive lake–land temperature difference results in the development of land breezes and low-level convergence that focus the development of convection over the GSL. The greater frequency of lake-effect precipitation during the overnight and early morning hours may be related to the diurnal modulation of the lake–land temperature difference and associated land-breeze convergence, similar to that suggested by Passarelli and Braham (1981) over Lake Michigan.
GSLE snowstorms share many similarities with lake-effect snowstorms over the Great Lakes region of the United States (Carpenter 1993; Steenburgh et al. 2000). Wiggin (1950) described the general characteristics of lake-effect snowstorms in the Great Lakes region, including their potential for large accumulations and significant variations in snowfall over short spatial scales. Additionally, Wiggin (1950) noted that such storms were favored in polar continental air masses during periods of large lake–air temperature differences, near-adiabatic lapse rates, and long overwater fetches. Peace and Sykes (1966) studied a lake-effect snowband using a mesoscale surface observing network over the eastern end of Lake Ontario. It was found that a narrow convergence line accompanied the snowband and it was hypothesized that surface sensible heating caused the formation of the snowband, with winds aloft controlling the location and movement of the band. Subsequent studies over the Great Lakes have identified a variety of lake-effect precipitation structures including (i) broad area coverage, which may include multiple wind-parallel bands or open cells (Kelly 1982, 1984); (ii) shoreline bands that form roughly parallel to the lee shore due to the convergence of a land breeze with the large-scale wind field (Ballentine 1982; Braham 1983; Hjelmfelt and Braham 1983; Hjelmfelt 1990); (iii) midlake bands that form when the large-scale flow is parallel to the long axis of a lake and a lake–land temperature contrast exists (Peace and Sykes 1966; Passarelli and Braham 1981; Braham 1983; Hjelmfelt 1990; Niziol et al. 1995); and (iv) mesoscale vortices that form in a polar air mass under conditions of a weak surface pressure gradient and large lake–air temperature differential (Forbes and Merritt 1984).
Precipitation during GSLE events is most frequently characterized by the irregular development of radar echoes over and downstream of the GSL (Steenburgh et al. 2000). The most commonly observed organized precipitation structures are solitary wind-parallel bands resembling midlake bands found over the Great Lakes, and broad-area coverage precipitation shields that form near the lee shoreline. In addition, GSLE precipitation sometimes occurs in concert with orographic precipitation, or within a broader-scale precipitation shield associated with synoptic-scale lifting. Significant enhancement of GSLE events can occur when lake-induced precipitation features, such as solitary wind-parallel bands, extend over the downstream orography.
Several studies have used numerical models to examine lake-effect snowstorm dynamics (e.g., Lavoie 1972; Ballentine 1982; Hjelmfelt and Braham 1983; Hjelmfelt 1990). Using a three-layer primitive equation model, Lavoie (1972) found that frictional convergence due to land–water roughness contrasts, and surface sensible heating due to lake–air temperature differences, produce upward vertical motion and elevated inversion heights near the lee shoreline of Lake Erie. The lake–air temperature difference was found to be dominant. Hjelmfelt (1990, 1992) examined the importance of low-level instability, lake–land temperature difference, sensible and latent heat fluxes, topography, capping inversions, and upstream moisture in producing lake-effect snowstorms over Lake Michigan. He found that both shoreline-parallel and midlake snowbands were favored by strong lake–land temperature differences, weak stability, and the absence of capping inversions at low elevations. Moderate cross-lake flow enhanced land-breeze-induced convergence, thus strengthening shoreline-parallel bands. Midlake snowbands, however, were favored by strong wind flow parallel to the long axis of the lake. Weaker wind flows combined with strong lake–land temperature differences tended to produce mesoscale vortices instead of midlake bands. Upstream moisture was also found to be important in enhancing lake-effect precipitation and land-breeze strength due to latent heat release from condensation. Ballentine et al. (1998) described a successful simulation of a Lake Ontario snowband using the Pennsylvania State University–National Center for Atmospheric Research fifth generation Mesoscale Model (MM5). The simulation reproduced the observed precipitation distribution, although changes in snowband location in response to the evolving synoptic-scale flow had timing errors of a few hours.
The purpose of this paper, and the companion article by Onton and Steenburgh (2001), is to describe the evolution and physical processes responsible for a GSLE snowstorm that occurred on 7 December 1998. Snowfall accumulations of up to 36 cm were produced by the event, which featured a wind-parallel snowband that developed near the western shoreline of the GSL and became aligned along the midlake axis as it moved eastward and merged with a weaker snowband. Specific questions that will be addressed in the two papers include the following.
What are the underlying mesoscale dynamics responsible for the development of GSLE snowbands? Are solitary wind-parallel bands over the GSL produced primarily by thermally driven land-breeze convergence?
How important are topographic effects such as orographic uplift and low-level flow blocking and channeling? To what degree are GSLE events triggered or enhanced by such local orographic effects?
How do sensible and latent heat fluxes influence the development and intensity of lake-effect precipitation? Does the hypersaline composition of the GSL significantly affect latent heat flux (compared to freshwater) and snowband evolution or intensity?
Does frictional convergence due to land–water roughness contrasts influence the development of GSLE snowbands?
Can present-day mesoscale models accurately simulate the mesoscale circulations and precipitation patterns observed during GSLE snowstorms? Does the“fixed” surface forcing of the lake and surrounding topography extend predictability, or do small errors in surface characteristics and the upstream flow characteristics limit forecast skill?
The mesoscale structure and evolution of the 7 December 1998 GSLE snowstorm is examined in the present paper using conventional meteorological data, high-density surface observations provided by MesoWest, a collection of cooperating mesonets in the western United States, and a numerical simulation by the nonhydrostatic MM5. Section 2 describes the regional orography and unique characteristics of GSL hydrology, composition, and air–lake interactions. Section 3 presents a detailed observational analysis of the 7 December 1998 event using RUC2 analyses, radar observations, and MesoWest surface observations. Then, section 4 uses a mesoscale model simulation to further examine the mesoscale structure and evolution of the event. A summary and discussion of major results follow in section 5. Further diagnosis of the dynamics and predictability of the 7 December 1998 event is presented in Onton and Steenburgh (2001).
2. The Great Salt Lake and surrounding topography
There are several unique aspects of the land surface properties and orography of northern Utah that influence the development of lake-effect precipitation (Fig. 1). These include the region’s intense and complex vertical relief, and the varying hydrologic structure, thermal characteristics, and hypersaline composition of the GSL. The GSL is the largest body of water in the United States west of the Great Lakes. It currently occupies an area of ∼4400 km2, is about 120 km long and 45 km wide, and has an average (maximum) depth of only 4.8 (10) m. Due to the lack of a drainage outlet, the lake’s size fluctuates due to interseasonal and interannual variations in precipitation and evaporation, and has ranged from 2500 to 6200 km2 in area and from 1278 to 1284 m in surface elevation since the mid-1850s (Arnow 1980; Wold et al. 1996).
Due to the GSL’s shallow depth, climatological lake-surface temperatures exhibit little lag relative to climatological mean air temperatures at Salt Lake City (Steenburgh et al. 2000; see their Fig. 2). The average lake-surface temperature exhibits a maximum (minimum) near 1 August (1 February), similar to the timing of the maximum (minimum) mean air temperature at SLC on 24 July (5 January). From late winter through summer, the mean lake temperature is similar to the mean air temperature at SLC, but during the fall through early winter, the mean lake-surface temperature exceeds the mean air temperature by 2°–3°C.
Carpenter (1993) suggested that lake-surface temperatures may correlate with the preceding week’s mean air temperature. In the past, estimates of lake-surface temperature using this method were necessary for operational forecasting due to the lack of real-time observations. However, starting in late summer 1998, lake-surface temperatures have been observed at a MesoWest site installed at Hat Island (HAT; see Fig. 1 for location). A comparison between the mean daily lake-surface temperature and mean daily air temperatures at SLC and HAT for the initial five-month observation period is presented in Fig. 2. This figure clearly illustrates the seasonal decline in both lake-surface and mean air temperature. Note, however, that from September to December, lake-surface temperatures were generally 2°–3°C greater than the mean air temperature, while in January, lake-surface temperatures were similar to the mean air temperature. This is in rough agreement with the twice-monthly observations presented by Steenburgh et al. (2000), although they showed higher lake-surface temperatures persisting into mid-January. Also evident in Fig. 2 are more rapid lake-surface and mean air temperature changes associated with transient synoptic weather systems. Specifically, lake-surface temperature changes of as much as 3.3°C (5°C) in 24 h (48 h) were observed following the intrusion of cold air masses into the region in September and October.
The GSL is a terminal lake (i.e., it has no outlet) and can be up to eight times as saline as ocean water. Currently, the lake is divided by an earthen railroad causeway that limits mixing between the northern and southern sections (Sturm 1980; Butts 1980; Newby 1980), named Gunnison Bay and Gilbert Bay, respectively. Gunnison Bay has only limited freshwater inflow and generally features salinity near saturation (27%). Salinity in Gilbert Bay, which has several freshwater inlets, has ranged from 6% to 15% and during December 1998 was near 9%. Due to the high salinity, the lake never freezes over except near freshwater inlets. Because the lake never freezes over and can warm rapidly, lake-effect snow is possible from early fall through late spring (Steenburgh et al. 2000). The salinity also acts to reduce saturation vapor pressure and latent heat fluxes compared to those found under similar conditions over freshwater (Steenburgh et al. 2000; see their Fig. 3). Given the current salinity of Gunnison and Gilbert Bays, the ratio of saturation vapor pressure over saline water to saturation vapor pressure over freshwater is approximately 0.70 and 0.94, respectively. Due to this reduction in saturation vapor pressure, upward moisture fluxes calculated using a bulk aerodynamic formula [Krishnamurti and Bounoua 1996, their Eq. (8.2)] would be eliminated or negative over Gunnison (Gilbert) Bay if the difference between the lake-surface temperature and near-surface dewpoint temperature was 5°C (0.9°C) or smaller.2 The implications of salinity on moisture fluxes and precipitation will be examined further in the companion article by Onton and Steenburgh (2001).
Several steeply sloped mountain ranges extending to over 3000 m are located south and east of the GSL (Fig. 1). To the east and southeast are the Wasatch Mountains, which are oriented roughly meridionally and rise abruptly to elevations of 2500–3500 m. South of the GSL, the Oquirrh Mountains rise directly from the south shore to heights of 2500–3250 m, while to the southwest, the Stansbury Mountains reach similar altitudes. Lowland regions between these mountain ranges, including the Salt Lake and Tooele Valleys, are approximately 25 km wide and feature broadly sloped relief that may also produce orographic precipitation enhancement. For example, the city of Tooele (TOO), 17 km from the GSL shoreline, is located 215 m above lake level, while broadly sloped benches on the western, southern, and eastern sides of the Salt Lake Valley are 150–400 m above lake level. This lowland relief is comparable to that found east of Lake Ontario and northern Lake Michigan where significant orographic enhancement of lake-effect precipitation occurs (Muller 1966; Hjelmfelt 1992; Niziol et al. 1995), while the adjacent mountain ranges described above are substantially higher. Other important orographic features include the Great Salt Lake Desert, a lowland area west of the lake, and the Raft River Mountains northwest of the lake. Thus, flow from the northwest, which is associated with lake-effect storms (Carpenter 1993; Steenburgh et al. 2000), must traverse substantial topography before moving over the GSL.
3. Observational analysis of the 7 December 1998 snowband
a. Large-scale analysis
To examine the large-scale evolution of the 7 December 1998 snowband event, regional-scale analyses from the RUC2 and upper-air soundings from SLC are presented in Figs. 3–6. At 1200 UTC 6 December 1998, roughly 12 h prior to the onset of lake-effect precipitation, a large-scale upper-level trough was located over the western United States, with the 500-hPa trough axis extending equatorward from eastern Washington into southern California (Fig. 3c). The 700-hPa trough axis was just west of the Utah–Nevada border, with a region of significant moisture [i.e., relative humidity (RH) > 70%] collocated with and upstream of this feature (Fig. 3b). There was a weak contrast in temperature across the trough with 700-hPa temperatures over southern Utah near −12°C, compared to −16°C over western Washington and Oregon. A sea level pressure low center was located northwest of Las Vegas beneath a region of 500-hPa cyclonic absolute vorticity advection (Figs. 3a,c). Weak sea level pressure troughing extended northeastward from the low center into northern Utah. The observed sounding at SLC showed veering winds with height from the surface to 700 hPa implying low-level warm advection ahead of the trough (Fig. 3d). Conditions were not favorable for lake-effect precipitation with southerly to southwesterly flow, a series of stable layers, and 5°–20°C dewpoint depressions evident at low levels.
Twelve hours later at 0000 UTC 7 December, shortly after the onset of lake-effect precipitation, the 500-hPa (700-hPa) trough axis had moved over (downstream of) SLC (Figs. 4b,c). Although the signature of this trough was relatively weak at the surface, low-level winds gradually became northwesterly to northerly (Fig. 4a), and low-level cold advection developed over northern Utah as inferred from backing winds in the SLC sounding near and below 650 hPa (Fig. 4d). In fact, the lowest 700-hPa temperatures were now located just upstream of northern Utah (Fig. 4b). Visible satellite imagery showed the passage of a band of clouds across the GSL with the 700-hPa trough between 1400 and 1900 UTC, but no precipitation was reported over northern Utah (not shown). The large-scale pattern described above is similar to that found at the onset time of lake-effect events by Steenburgh et al. (2000).
Other characteristics of the environment were also favorable for the development of lake-effect precipitation. With lake-surface (HAT) and 700-hPa temperatures (SLC) of 5°C and −15.9°C, respectively, the lake–700-hPa temperature difference of 20.9°C (12.4 K km−1) exceeded the 16°C threshold required for GSLE precipitation identified by Steenburgh et al. (2000). Although an upper-level sounding upstream of the GSL was not available, the SLC sounding that was taken downstream of the GSL showed small dewpoint depressions throughout most of the troposphere (Fig. 4d). Low-level lapse rates were near moist adiabatic and, although the observed surface parcel at SLC had no convective available potential energy, a surface parcel defined using air temperature and dewpoint observations from HAT exhibited a limited amount of positive buoyancy (not shown). Finally, the lake–land temperature difference, calculated using the SLC air temperature and HAT lake-surface temperature, was 8°C, near the mean value for GSLE events (Steenburgh et al. 2000). Such conditions favor localized surface heating, boundary layer destabilization, and the development of land-breeze circulations and low-level convergence over the GSL.
By 1200 UTC 7 December 1998, the 500-hPa trough was located well downstream of Utah and an upper-level ridge was building over the western United States (Fig. 5c). At this time, lake-effect precipitation was occurring in a solitary wind-parallel band extending from the GSL into the Tooele Valley. Over northern Utah, moist (RH > 80°%) north to northwesterly flow was evident at 700 hPa with the lowest temperatures at this level located just south of the GSL (Fig. 5b). Sea level high pressure was found over eastern Nevada with light surface winds over northern Utah (Fig. 5a). The sounding was moist (dewpoint depressions <5°C) and conditionally unstable below 650 hPa, with a strong inversion near 500 hPa (Fig. 5d). In addition, the lake–700-hPa temperature difference was 22.5°C (13.0 K km−1) and the lake–land temperature difference was 10°C.
By 0000 UTC 8 December 1998, lake-effect precipitation had ended. At this time, the 500-hPa ridge axis was moving over northern Utah and the sea level high pressure system was centered over eastern Utah (Figs. 6a,c). At 700 hPa, temperatures had climbed to −12°C (Fig. 6b), presumably from large-scale subsidence beneath the building upper-level ridge and, as can be inferred from veering winds with height at SLC (Fig. 6d), warm advection in the lower and middle troposphere. The SLC sounding also shows that the inversion base that was previously located near 500 hPa had lowered to 700 hPa (cf. Figs. 5d and 6d). In addition, the lake–700-hPa temperature difference was 18.5°C (10.5 K km−1), and the lake–land temperature difference was under 5°C. These values were near or below the minima observed during lake-effect events by Steenburgh et al. (2000). Correspondingly, only shallow, nonprecipitating cumulus were observed over the region.
b. Mesoscale structure
Observations from the Salt Lake City WSR-88D (KMTX) radar and MesoWest surface networks, presented in Figs. 7–9, illustrate the mesoscale structure of the GSLE event. Between 2200 UTC 6 December and 0400 UTC 7 December, after the passage of the surface and upper-level troughs, disorganized convective cells forming primarily over the lake and moving downstream to the southeast were observed in radar analyses (not shown). By 0400 UTC 7 December, the last of these cells were drifting into the Tooele Valley and the first long-lived snowband (snowband A) began to form near the western shoreline of the GSL (Fig. 7a). This snowband was roughly parallel to the wind flow on Promontory Point (PRP), a mountaintop observing site approximately 800 m above lake level that roughly represents a steering-layer wind for lake-effect convection.3 Weak low-level confluence into the northern end of the snowband, as observed during similar banded events over Lakes Michigan and Ontario (e.g., Peace and Sykes 1966; Passarelli and Braham 1981; Braham 1983), was suggested by a shift in surface winds from northerly to northwesterly as the snowband passed over Gunnison Island (GNI; Fig 8a). Elsewhere, surface winds were generally light and northwesterly to northeasterly. Surface temperatures ranged from −2° to −10°C, with the highest temperatures found over and near the GSL.
Over the next 75 min, snowband A intensified and at 0515 UTC was located near the western shoreline (Fig. 7b). Meanwhile, a second snowband (snowband B) developed over the southernmost arm of the GSL and northeast portion of the Tooele Valley. The wind flow in the northern Tooele Valley was confluent toward the northern half of this band. Weak confluence into the northern portion of snowband A is also suggested by the northwest surface wind at GNI and north-northeast surface wind at HAT.
By 0630 UTC snowbands A and B were beginning to merge into a solitary snowband (Fig. 7c). Snowband A had just passed over HAT where winds shifted from northerly to northwesterly, suggesting low-level confluence along the northern portion of the snowband axis (Fig. 8b). Significant changes in temperature or dewpoint were, however, not observed (cf. Figs. 7b,c; dewpoint not plotted). Fifteen minutes later snowband A moved westward back across HAT resulting in a wind shift back to northerly (Fig. 8b; radar analysis not shown). Farther downstream, surface winds beneath snowband A appeared to be divergent over the western Tooele Valley, perhaps due to convective outflow (Fig. 7c). In the eastern Tooele Valley, surface winds remained confluent toward the axis of snowband B.
The radar reflectivity analysis for 0815 UTC shows the solitary snowband that developed from the merger of snowbands A and B at one of its most organized stages (Fig. 7d). At this time the snowband extended from just west of HAT southeastward over the Tooele Valley and was nearly parallel to the flow at PRP. Reflectivity values of 20–30 dBZ composed much of the snowband and likely represent moderate to heavy snow. Isolated reflectivity values approaching 40 dBZ were observed within the band over the GSL, Tooele Valley, and western slopes of the Oquirrh Mountains. At this time, confluent flow that was previously observed over the Tooele Valley beneath snowband B was weakening as winds were becoming northerly or northwesterly.
Over the next 135 min the snowband became more meridionally oriented and by 1030 UTC extended from near the center of the GSL southward into the Tooele Valley (Fig. 7e). On the mesoscale, surface wind observations continued to suggest that the northern portion of the snowband was associated with low-level confluence. Surface winds at HAT veered from northerly to westerly with snowband passage, as occurred between 0600 and 0700 UTC, although the wind shift appeared to follow the passage of the reflectivity band by 15–30 min (Fig. 8b). In addition, overlake convergence was suggested by the westerly wind at HAT and north-northeasterly wind at Antelope Island (Fig. 7e; see Fig. 1 for locations). This mesoscale wind pattern may have been related to the development of land-breeze circulations due to localized heating over the lake surface. Temperatures over and near the GSL were generally higher than those at surrounding locations, but a lack of wind observations prevented diagnosis of wind flows along the western and eastern shorelines. Farther downstream, over the northern Tooele Valley, winds were strongly diffluent (Fig. 7e), and at the observing site near the GSL shoreline where the Oquirrh Mountains rise abruptly, surface winds had shifted from northeasterly to westerly (cf. Figs. 7d,e). Although the cause of the diffluent wind pattern over the northern Tooele Valley at this time was not clear, it is possible that it was produced by convective outflow associated with precipitation and diabatic cooling beneath the downstream portion of the snowband. Compared to the relatively steady confluent flow beneath the upstream portion of the snowband over the GSL, surface winds throughout the event were more variable and occasionally diffluent near the downstream portion of the snowband over the northern Tooele Valley.
By 1315 UTC the snowband extended southeastward from the GSL over the western Salt Lake Valley (Fig. 7f). With clearing skies, temperatures dropped rapidly to −10°C or lower in the central and western Tooele Valley, substantially lower than temperatures over the GSL. As a result, thermally driven downvalley and offshore winds developed in this area. Overall, the regional wind pattern suggests the presence of low-level convergence over the GSL and near the axis of the snowband.
During the next 90 min the snowband gradually deteriorated into a broad area of precipitation with embedded convective cores that was drifting northeastward by 1445 UTC (Fig. 7g). At this time, surface winds appeared convergent over the GSL, but the snowband structure and intensity were beginning to decay for two reasons. First, the near-steering-layer wind at PRP was weakening and beginning to veer to westerly (Fig. 8c), a direction with a much shorter overwater fetch. Second, warm advection and subsidence were producing rapid stabilization at midlevels, limiting the depth of surface-based convection (Figs. 5 and 6). By 1900 UTC the near-steering-layer winds at PRP were west-southwesterly, the lake-effect precipitation area had drifted eastward, and new cells were no longer forming (Fig. 7h).
c. Radar composite and snowfall distribution
To summarize the distribution and intensity of snowfall during this event, a composite radar image was generated from the 155 lowest-elevation (0.5°) radar scans taken from 0000 to 1455 UTC, which encompasses the period when lake-effect precipitation was falling over the Tooele Valley (Fig. 9). This involved computing the percentage of time that reflectivity values exceeded 10 dBZ at each point within each radar scan (hereafter the 10-dBZ frequency of occurrence or 10-dBZ FO). This method was originally developed by Slemmer (1998) and was used by Steenburgh et al. (2000) to describe the GSLE precipitation distribution as a function of various wind and thermodynamic variables. The composite reflectivity analysis shows that during the event a band of frequent returns stretched from near HAT to the western slopes of the Oquirrh Mountains (see Fig. 1 for locations), with a secondary 10-dBZ FO maximum in the western Salt Lake Valley where the snowband was resident for a shorter period of time. The highest 10-dBZ FO region (60%–80%) extended in a narrow band from near the southernmost tip of the GSL to TOO. Snowfall totals of 25, 30, and 36 cm (18.8-mm liquid equivalent for the latter) were observed at reporting sites in this region. Outside this band of heavy snowfall, accumulations were much lower, as indicated by snowfall accumulations of 5 and 8 cm to the south and west, and reports of trace amounts in the eastern Salt Lake Valley.
4. Model simulation
a. Mesoscale model description
Simulations by the MM5 were used to further examine the evolution of the 7 December 1998 snowband event. The MM5 is a nonhydrostatic finite-difference atmospheric model employing a terrain-following sigma vertical coordinate (Grell et al. 1995). Simulations featured four one-way nested domains with grid spacings of 54, 18, 6, and 2 km, respectively (Fig. 10). Thirty-six variably spaced full-sigma levels were used in the vertical with resolution varying from approximately 10 hPa in the boundary layer to 30 hPa in the upper troposphere.4 Precipitation processes were parameterized in all four domains using a mixed-phase microphysical parameterization that included predictive equations for cloud ice, cloud water, rain, and snow and allowed for supercooled water below 0°C and unmelted snow above 0°C (Grell et al. 1995). The Kain–Fritsch cumulus parameterization (Kain and Fritsch 1993) was used in the 54-, 18-, and 6-km domains to represent subgrid-scale convective precipitation. Boundary layer processes were parameterized using the so-called Blackadar scheme that accounts for the vertical mixing of horizontal wind, temperature, mixing ratio, cloud water, and cloud ice in the boundary layer (Blackadar 1976, 1979; Zhang and Anthes 1982). One modification was made to the boundary layer parameterization to account for the impact of lake salinity on saturation vapor pressure and surface moisture fluxes. North of the railroad causeway (Fig. 1), the saturation vapor pressure of lake water was set to 70% of that observed for freshwater. This reduction was based on recent salinity observations in the northern arm of the GSL (27%) and the saturation vapor pressure measurements obtained for lake water by Dickson et al. (1965) and presented in Steenburgh et al. (2000; see their Fig. 3). South of the railroad causeway, the saturation vapor pressure was set to 94% of that observed for freshwater based on the observed salinity (9%) and estimates of vapor pressure reduction obtained by Steenburgh et al. (2000) using Raoult’s law. Other model parameterizations included a long- and shortwave atmospheric radiation scheme that accounts for interactions with the atmosphere, clouds, precipitation, and surface (Dudhia 1989), and the Klemp and Durran (1983) radiative upper boundary condition.
Observed terrain data, bilinearly interpolated onto the MM5 grid and filtered with a two-pass smoother/desmoother, provided the model terrain. For the 6- and 2-km domains, a 30-s resolution dataset was used, while 10- and 5-min resolution data was used for the 54- and 18-km domains, respectively. All land use information was derived from a 10-min resolution dataset, though the land use and elevation near the GSL was corrected to match the lake shoreline. The topography for the 2-km domain represents most of the major terrain features of northern Utah, although mountain crest levels and slopes are somewhat lower and less steep than observed (cf. Figs. 1 and 11).
Analyses for initialization, data assimilation, and lateral boundary conditions were generated at 12-h intervals from 1200 UTC 6 December to 0000 UTC 8 December 1998 in the following manner. First, operational surface and upper-level analyses from the NCEP Eta model (Black 1994; Rogers et al. 1995, 1996), which were available at 80-km horizontal and 50-hPa vertical resolutions, were interpolated onto each domain’s horizontal grid. This provided a first guess for a modified Cressman-style analysis (Benjamin and Seaman 1985) that incorporated rawinsonde and surface data. After the removal of superadiabatic lapse rates below 500 hPa, the analysis was interpolated to sigma coordinates and the integrated mean divergence was removed to avoid the production of spurious gravity waves. Sea surface temperatures were generated from operational NCEP analyses that were available on a 1° lat × 1° long grid. The GSL temperature was set to 278 K, the mean lake-surface temperature at HAT during the event period.
Four-dimensional data assimilation (FDDA) was used to constrain large-scale error growth in the 54- and 18-km domains. Following Stauffer and Seaman (1990), this involved using Newtonian nudging to relax the model simulation to the gridded analyses that were generated using the methods described above. Linear interpolation in time was used between the analyses, which were at 12-h intervals. For the 54-km domain, FDDA was used during the entire 36-h simulation, while FDDA was used for the 18-km domain for only the first 12 h.
Initial analyses for the 6- and 2-km domains were generated by interpolation of analyses from their parent grids since the density of available observations was not sufficient to adequately resolve features on scales consistent with their grid resolutions. Four-dimensional data assimilation was not used on these domains, although degradation of forecast skill from large-scale error growth should be reduced because of the superior lateral boundary conditions provided by the use of FDDA on the outer domains (Vukicevic and Paegle 1989). The 6-km domain was initialized at the same time as the 54- and 18-km domains (1200 UTC 6 December), while the 2-km domain was initialized 12 h later at 0000 UTC 7 December. Because of computational resource limitations, the 2-km domain was run after the integration of the coarser-resolution domains was complete, with boundary conditions provided by hourly output files from the 6-km domain.
b. Simulated large-scale evolution
Analyses from the 18-km domain are presented in Figs. 12–14 to examine the large-scale evolution of the model simulation. At 0000 UTC 7 December, the simulated 500-hPa trough axis extended from Arizona to northern Idaho (Fig. 12c) and the lowest 700-hPa temperatures were located upstream of the GSL (Fig. 12b). Over northern Utah, northwesterly flow was found at 700 hPa and the surface, with the relative humidity at the former level exceeding 70% (Figs. 12a,b).5 The most notable differences between the simulation and the RUC2 analyses were the lack of a well-defined 700-hPa trough extending northward through eastern Utah and the placement of the 500-hPa trough axis approximately 50–100 km too far west in the vicinity of the GSL (cf. Figs. 4b,c and 12b,c). The model-derived sounding at SLC showed northwesterly winds extending from the surface to 500 hPa, where the winds abruptly backed to southwesterly (Fig. 12d). A conditionally unstable lapse rate was found below ∼750 hPa. The simulated sounding agreed well with the observed, although some minor differences were evident (cf. Figs. 4d and 12d). In particular, the observed layer of backing winds near 700 hPa was not found in the model sounding and the simulated surface temperature appeared to be too low (−7.0°C) compared to the observed (−3.3°C). The latter was mainly a reflection of the elevation of the model terrain, which for the 18-km domain was 422 m (45 hPa) higher than the actual elevation. At a given pressure level, the simulated temperature closely resembled the observed temperature. The sounding derived from the higher-resolution 2-km domain, in which the terrain was only 34 m (4 hPa) higher than the actual terrain, had a near-surface temperature of −4.3°C (not shown).
At 1200 UTC 7 December, the simulated 500-hPa trough axis was located downstream of Utah and an upper-level ridge was building over the western United States (Fig. 13c). At 700 hPa, the lowest temperatures were located near northern Utah where northwesterly flow was found (Fig. 13b). In this region, the simulated relative humidity was slightly lower than analyzed by the RUC2 (Fig. 5b). At the surface, sea level pressure in the higher-resolution MM5 showed more mesoscale structure than the RUC2 (cf. Figs. 5a and 13a), but there were no substantial differences in the placement of synoptic-scale features, including the position of the sea level pressure high that was centered over the Great Basin. Comparison of the simulated and observed soundings (cf. Figs. 5d and 13d) revealed a model warm bias between 500 and 700 hPa and cold bias near the surface, resulting in a more stable low-level lapse rate than observed. It should be noted, however, that modification of the low-level temperature and dewpoint due to heat and moisture fluxes from the GSL was likely underrepresented in the 18-km domain since at this grid spacing only 12 grid points represent the GSL.
By 0000 UTC 8 December, the simulated 500-hPa ridge axis extended from southern California northeastward to central Montana and was just upstream of northern Utah (Fig. 14c). This position was well forecast, although the simulated ridge was slightly less amplified than analyzed by the RUC2 (cf. Figs. 6c and 14c). At 700 hPa, the simulated flow remained northwesterly over northern Utah with temperatures rising to near −14°C over SLC in response to low-level warm advection and middle-tropospheric subsidence (Fig. 14b). The MM5 correctly centered the sea level pressure high over Utah and also produced more mesoscale structure than analyzed by the coarser-resolution RUC2 (cf. Figs. 6a and 14a). The simulated SLC sounding showed veering winds with height at low levels, implying warm advection, and an isothermal layer between 700 and 600 hPa (Fig. 14d). A weaker stable layer was located between a shallow surface-based mixed layer and the base of the isothermal layer. These features captured the general character of the SLC sounding, although the static stability of the simulated isothermal layer was much weaker than the observed 5°C inversion (cf. Figs. 6d and 14d). Low-level temperatures were also ∼3°C lower than observed.
c. Simulated mesoscale structure
To illustrate the simulated mesoscale structure of the 7 December 1998 event, Fig. 15 presents analyses from the 2-km domain, including the model-diagnosed 10-m wind, lowest half-sigma-level (∼40 m AGL) temperature, and vertically integrated precipitation (VIP). For purposes of model validation, station plots of wind and temperature from several MesoWest observing sites are overlaid on the model analysis. The VIP is the total mass of parameterized rain and snow in a model column and is used to illustrate the instantaneous position of the snowband at each analysis time. The modeled VIP can be qualitatively compared to radar reflectivity analyses presented in Fig. 7, although it should be noted that the former is a column-integrated quantity while the latter represents an observation primarily from the radar sample volume and is affected by particle size and shape, as well as other factors such as beam attenuation, refraction, and spreading.
At 0400 UTC 7 December, a region of low-level confluence was oriented along the western shoreline of the GSL (Fig. 15a). Model diagnostics at this and subsequent times showed this region of confluence was convergent and will hereafter be referred to as a convergence zone. Simulated VIP was located near the southern portion of this convergence zone and extended downstream along the eastern slopes of the Stansbury Mountains. Comparison with the corresponding WSR-88D radar reflectivity and mesonet analysis shows that this feature represented snowband A, which in the simulation appeared to be forming correctly near the western GSL shore, but with the VIP region located south of the radar reflectivity band (cf. Figs. 7a and 15a). This discrepancy could be due to model error, although, as noted above, VIP and radar reflectivity are not entirely consistent. Precipitation was also indicated in the radar analysis over the Tooele Valley. Three weak VIP features were found in this region. Simulated low-level temperatures over the GSL were above −4°C, approximately 2°C greater than over surrounding regions of a similar elevation (Fig. 15a). This model low-level temperature analysis agreed well with observed temperatures at most sites, with differences generally less than 2°C (cf. Figs. 7a and 15a). The most notable difference was over the Great Salt Lake Desert where the observed (simulated) temperature was −3°C (−6° to −8°C). Wind directions and magnitudes near the convergence zone and over other regions were also in good agreement.
Over the next 90 min, simulated snowband A remained quasi-stationary and at 0530 UTC the VIP band extended from about the midpoint of the western GSL shoreline to the southeastern slopes of the Stansbury Mountains (Fig. 15b). Meanwhile, the second snowband (snowband B) began to organize over the eastern Tooele Valley and western slopes of the Oquirrh Mountains. The simulated position of both snowbands was excellent, although they did not extend as far poleward as the corresponding radar reflectivity band (cf. Figs. 7b and 15b 6). Wind and temperature observations at this time indicate that the model was in general agreement with observations, although simulated temperatures were still too low near the west boundary.
At 0630 UTC (Fig. 15c), snowband A was becoming less organized and diminishing in precipitation intensity, although the convergence zone along the western shoreline was in nearly the same position and possessed a similar magnitude as at 0530 UTC. The VIP analysis did not show a continuous band of precipitation; however, the cloud mass associated with snowband B extended poleward toward Antelope Island in a well-organized band (not shown). The initial eastward movement of the simulated shoreline convergence zone and merger of snowbands A and B appeared to be slower than observed (cf. Figs. 7c and 15c). Temperatures in most locations, including the Tooele and Salt Lake Valleys, were in good agreement, although the simulated temperatures in the Great Salt Lake Desert had dropped to well below observed. The modeled wind field verified well against most land-based stations. Wind directions at GNI and HAT winds were off by roughly 60° due to the model placing the convergence zone too close to the western shoreline.
The simulated precipitation field at 0830 UTC (Fig. 15d) was significantly different from observed (Fig. 7d;0815 UTC). At this time, observed snowbands A and B had merged into a solitary snowband that extended from HAT to the Oquirrh Mountains. The simulated snowbands, however, were in one of their least organized stages and were just beginning to merge (Fig. 15d). Nevertheless, the simulated convergence zone was still evident, had moved offshore, and appeared to be well positioned based on the observation from HAT.
Reintensification and merger of simulated snowbands A and B occurred over the next few hours in a manner that was similar to observed but delayed. This is illustrated by the evolution of the VIP between 1000 and 1300 UTC (Fig. 16), which can be compared with the radar analyses presented in Fig. 7. This sequence illustrates some of the difficulties of mesoscale quantitative precipitation forecasting with existing modeling systems. Although surface winds and temperatures were generally well simulated, and the model simulation was reasonably accurate earlier in the period, errors related to the timing of the merger and propagation of the bands were still apparent.
At 1500 UTC the simulated convergence zone and snowband were aligned along the major axis of the GSL (Fig. 15e). The overall flow pattern resembled that associated with midlake bands over Lake Michigan (e.g., Peace and Sykes 1966; Braham and Kelly 1982; Hjelmfelt 1990), with land breezes from the opposing lake shorelines converging near the lake axis. The largest simulated lake–land temperature differences were found at this time with a narrow tongue of warm air (>−4°C) oriented along the convergence zone axis. Over land, a shallow nocturnal inversion had formed and was strongest over the Great Salt Lake Desert where near-surface temperatures were −10°C or lower (Figs. 15e and 17). Temperatures were similar to observed except at the Great Salt Lake Desert observing point (S17; see Fig. 1 for location) where the simulated temperature was several degrees too low. The persistent model cold bias at this location may be related to errors in the specification of land surface properties. The Great Salt Lake Desert land surface is composed primarily of salt flats, which at this time of year can be wet enough to form a salt slurry. Such a salt slurry would likely have a thermal inertia closer to water (0.06 cal cm−2 K−1 s−1/2) than the desert land surface that was specified in the model simulation (0.02 cal cm−2 K−1 s−1/2).7 Since the areal coverage of the salt slurry is poorly known, it could not be accurately specified in the simulation. During the remainder of the simulation, the area of precipitation drifted northeastward and weakened as low-level winds became southerly to southwesterly and conditions stabilized (Fig. 15f).
The total precipitation (liquid water equivalent) produced by the model simulation from 0000 to 1500 UTC 7 December 1998 is presented in Fig. 18. In comparison with Fig. 9, the model precipitation band stretching from just east of Stansbury Island into the eastern Tooele Valley was very close to the observed position. Maximum simulated precipitation in this band was 19.3 mm, comparable to, but slightly higher than, the observed maximum of 18.8 mm at TOO. The simulation also captured the distribution of the precipitation just east of the Oquirrh Mountains. Based on radar analyses, the model appears to have overpredicted precipitation over the western Tooele Valley and Stansbury Mountains, although no surface snowfall measurements were available for direct validation. Precipitation in this region occurred earlier, was shifted farther west, and extended farther downstream in the model simulation than was observed (cf. Figs. 7b and 15b).
5. Discussion and conclusions
The observational and model-derived analysis described above illustrates the importance of thermally driven circulations in producing the 7 December 1998 GSLE snowstorm. Specifically, the primary snowband of the event (snowband A) first formed along a land-breeze front near the western shoreline and eventually aligned along the midlake axis as the land-breeze front pushed eastward, flow along the eastern shoreline became increasingly offshore, and convergence developed along the midlake axis. Thus, despite the relatively small size of the GSL and presence of intense vertical relief, the underlying mesoscale dynamics responsible for this event appear to be analogous to shoreline and midlake snowband events over the Great Lakes (e.g., Peace and Sykes 1966; Passarelli and Braham 1981; Ballentine 1982; Braham 1983; Hjelmfelt and Braham 1983; Hjelmfelt 1990; Niziol et al. 1995).
The large-scale environment for the event was similar to that identified as favorable for the development of GSLE precipitation in previous climatological studies (e.g., Carpenter 1993; Steenburgh et al. 2000). Prior to the onset of lake-effect snow, an upper-level trough axis passed from west to east across the GSL, causing winds below 500 hPa to veer from southwesterly to northwesterly, low-level lapse rates to destabilize, and higher relative humidity air to move into northern Utah. Environmental conditions during the event were characterized by a lake–700-hPa temperature difference of up to 22.5°C, a lake–land temperature difference as large as 10°C, and conditionally unstable low-level lapse rates.
Lake-effect precipitation began ∼2200 UTC 6 December when unorganized convective cells formed over the lake and moved downstream to the south and east. At 0400 UTC 7 December, an organized snowband began to form near the western shoreline of the GSL. This band was aligned parallel to the steering-layer wind and was associated with an abrupt wind shift and line of confluence produced by a land-breeze front. This kinematic structure was analogous to that found during similar events over the Great Lakes (e.g., Peace and Sykes 1966; Passarelli and Braham 1981; Braham 1983). As the event progressed, a second region of precipitation formed over the southern GSL and eastern Tooele Valley, and by 0815 UTC merged with the original snowband to form a solitary midlake snowband. The snowband was aligned along the surface confluence zone, which was now located near the midlake axis, and was generally oriented parallel to the steering-layer flow.
By 1445 UTC, the snowband had deteriorated into an area of precipitation with embedded convective cores, drifting northeastward over the GSL. Although surface winds appeared to be convergent over the GSL, steering-layer winds were veering to westerly and temperatures were increasing aloft as an upper-level ridge developed over the region. Significant lowering of the equilibrium level for convection occurred during this period as the base of a strong inversion that was located near 500 hPa at 1200 UTC 7 December lowered to 700 hPa by 0000 UTC 8 December. As a result, environmental conditions were becoming less favorable for GSLE snowfall due to the shorter overwater fetch and reduced depth of convection (Carpenter 1993; Steenburgh et al. 2000). By 1900 UTC, precipitation cells were no longer forming over the GSL.
The heaviest storm-total snowfall was found in a 10-km wide band that extended from the south shore of the GSL to the city of Tooele. The maximum observed storm-total snowfall of 36 cm (18.8-mm liquid equivalent) occurred in the city of Tooele. Only trace amounts of snow were reported 30 km from the accumulation band.
The nonhydrostatic model simulation, which featured an inner nest with 2-km grid spacing and employed four-dimensional data assimilation on the 54-km domain for the entire simulation, closely matched the large-scale evolution of the event, with only small timing or placement errors of synoptic systems. The model run also produced snowbands that were similar in structure to radar reflectivity patterns observed by the KMTX WSR-88D, although errors in timing of up to 5 h were observed and the simulated snowbands appeared to be located farther downstream than the observed reflectivity bands. Surface winds and temperatures were also well simulated compared to surface observations, with the exception of stronger than observed nocturnal surface cooling over the Great Salt Lake Desert that may have resulted from errors in the specification of surface properties in that region. The simulated storm-total precipitation agreed well with radar reflectivity composites and snowfall observations. The maximum simulated precipitation was 19.3 mm, slightly greater than the observed 18.8 mm, and was found in approximately the same location as observed. Although the accuracy of the quantitative precipitation forecast at 2-km grid spacing may raise optimism regarding potential predictive skill of future high-resolution forecast models, the use of data assimilation to constrain large-scale error growth on the coarser-resolution model grids represents a significant advantage that would not be available in a real-time environment. Even with this advantage, timing errors were observed in the movement and merger of the snowbands that would affect forecast skill on mesoscale temporal scales.
The companion paper by Onton and Steenburgh (2001) further describes the processes responsible for this GSLE snowstorm using model diagnostics and sensitivity studies. The predictability of this event in a real-time environment is also examined with a series of simulations incorporating varying environmental conditions.
This research was supported by National Science Foundation Grant ATM-9634191 and NOAA Grants NA67WA0465 and NA77WA0572 to the NOAA Cooperative Institute for Regional Prediction at the University of Utah. Surface observations were provided by MesoWest, a collection of cooperating mesonets in the western United States. MesoWest data were collected and processed by John Horel, Mike Splitt, and Bryan White of the University of Utah, and Larry Dunn and David Zaff of the National Weather Service. Additional observational data were provided by the Data Support Section of the Scientific Computing Division of NCAR, which is supported by the National Science Foundation. Use of the MM5 was made possible by the Mesoscale and Microscale Meteorology Division of NCAR. Computer time for the model simulation was provided by the University of Utah Center for High Performance Computing. Special thanks to Justin Cox, Larry Dunn, John Horel, Steve Krueger, Jan Paegle, Tom Potter, Andy Siffert, and David Schultz for their contributions, advice, and scientific support. We gratefully acknowledge the efforts of two anonymous reviewers, whose constructive evaluations greatly improved the manuscript.
Corresponding author address: Dr. W. James Steenburgh, Department of Meteorology, University of Utah, 135 South 1460 East Room 819, Salt Lake City, UT 84112-0110.
Due to the elevation of the GSL (∼1280 m above mean sea level), surface and 700-hPa observations are used instead of surface and 850-hPa observations as is commonly done in studies of lake-effect snowstorms over the Great Lakes (e.g., Niziol et al. 1995).
This example was calculated using a lake temperature of 5°C, the most common temperature observed during GSLE events (Steenburgh et al. 2000). Over the range of lake temperatures observed during GSLE events, this result varies by 13% or less.
Following Steenburgh et al. (2000) and the experience of local meteorologists, the steering layer for lake-effect convection is generally considered to be 800–600 hPa. PRP is located at roughly 780 hPa.
Specifically, the full-sigma levels were located at σ = 1.0, 0.99, 0.98, 0.96, 0.93, 0.90, 0.87, 0.84, 0.81, 0.78, 0.75, 0.72, 0.69, 0.66, 0.63, 0.60, 0.57, 0.54, 0.51, 0.48, 0.45, 0.42, 0.39, 0.36, 0.33, 0.30, 0.27, 0.24, 0.21, 0.18, 0.15, 0.12, 0.09, 0.06, 0.03, 0.0, with the model top at 100 hPa.
To facilitate comparison with RUC2 surface-wind analyses, which are for 10 m above ground level (AGL), and MesoWest surface-wind observations, which are generally taken at 10 m AGL, MM5 surface winds presented in this paper are 10-m winds that were diagnosed from the lowest half-sigma-level wind (∼40 m AGL) by assuming a logarithmic wind profile.
Since model output was not available at 0515 UTC, there is a 15-min difference between these two figures.
The thermal inertia, χ, is defined as χ = (λCs)1/2, where λ is the thermal conductivity of the land surface layer and Cs is the heat capacity per unit volume (Grell et al. 1995).