CRH SSD
December 1996

 

CENTRAL REGION TECHNICAL ATTACHMENT 96-09

 

A Method for Forecasting Lake Effect Snow Using Synoptic-Scale Model Forecasts of 850 mb Temperature, 850/700 mb Vertical Velocity, and 850/700 mb Relative Humidity

Michael S. Evans
National Weather Service Forecast Office
Detroit/Pontiac, Michigan

 

 

 

1.   INTRODUCTION

Lake effect snow across southwest lower Michigan occurs when cold air flows over the relatively warm waters of Lake Michigan. The most common kind of lake effect snow pattern occurs when a northwest through southwest lower-tropospheric flow occurs roughly parallel to the short axis of the lake, resulting in multiple snow bands of light to moderate intensity across all of western lower Michigan.

Dockus (1985) has stated that a minimum over-water fetch of at least 100 miles is necessary for significant lake effect snow in the absence of synoptic-scale forcing. Since Lake Michigan is only about 50-70 miles wide in the east-west direction, over-water fetch lengths tend to be marginal (about 80-120 miles) for accumulating lake effect snow when the wind is southwest through northwest. Because of these marginal fetch lengths, it can be hypothesized that lake effect snow across southwest Michigan is strongly dependant on how favorable the local thermodynamic environment is for convection. It follows then, that synoptic-scale vertical velocity likely plays a crucial role in determining lake effect snow amounts, by modifying the height of the subsidence inversion that is typically present, and thereby favorably (or unfavorably) conditioning the environment for moist convection.

Evans (1996) examined data from four winter seasons (1991-1992 through 1994-1995) across southwest lower Michigan, and demonstrated that synoptic-scale model forecasts of vertical velocity and relative humidity are closely related to snowfall amounts when the lower tropospheric flow is roughly parallel to the short axis of lake Michigan. It was concluded that examining model forecasts of lower-tropospheric vertical velocity and relative humidity may be a quick way to roughly assess how favorably conditioned the lower troposphere will be for moist convection, and therefore how favorable the environment will be for lake effect snow.

Shortly after the above study was completed, the new National Weather Service Forecast Office (WFO) at Grand Rapids, Michigan (GRR) expanded the existing lake effect snow spotter network across southwest lower Michigan for the 1995-1996 winter season. As a result, it was decided to expand on some work done by Evans, using the new data. The purpose of the study presented in this paper is twofold: 1) to verify the work done by Evans (1996) using the expanded data-set of observations from the 1995-1996 winter season, and 2) to develop scatter diagrams of snowfall as a function of model forecasts of 850 mb/lake temperature difference vs. vertical velocity and relative humidity. In Section 2, the method for integrating data from the 1995-1996 season into the preexisting data base developed by Evans (1996) will be described. Results will be presented in Section 3, examples of a forecasting technique derived from the results will be given in Section 4, and a summary and discussion will be given in Section 5.

2.   METHOD

Evans (1996) created a database of snowfall amounts and model forecast gridded output from 34 24-hour periods during which several parameters were favorable for multiple band lake effect snow from 1991-1995. Periods were selected during which Arctic high pressure was building southeast from Canada toward the Great Lakes, with no major synoptic-scale cyclones or surface troughs located near western Michigan. Each 24-hour period included an interval during which an 850-mb temperature of #-12°C was forecast in combination with northwesterly through southwesterly lower-tropospheric winds, and minimal directional wind shear.

For each period, 24-hour snowfall data was taken from nine cooperative observation stations that report 24-hour snowfalls within three hours of 12 UTC. In addition, data from GRR and the Weather Service Office (WSO) at Muskegon (MKG) were incorporated into the study. Snowfall for each period was categorized as either trace, light, or heavy based on the following definitions:

Trace: No more than one station reported a 24-hour snowfall of an inch or more.
Light: Two or more stations reported a 24-hour snowfall of an inch or more, with fewer than two stations reporting 3 inches or more.
Heavy: Two or more stations reported a 24-hour snowfall of 3 inches or more.

For each 24-hour period, model forecasts of vertical velocity, relative humidity, wind direction, wind speed and 850-mb temperature were recorded. For periods that occurred before 11/93, values were taken from alphanumeric displays of NGM gridded forecasts, available at 6 hourly increments and interpolated between grid points to a point near Muskegon. For periods from 11/93 through 4/95, values were inferred by viewing ETA model forecast time-cross sections generated by PCGRIDDS at 43°N, 86°W (again, near Muskegon). For each event, values were taken from the model run that began 12 hours before the beginning of the designated 24-hour period. From each of these model runs, the values were taken from the 12 through 36 hour forecasts.

As was discussed in section 1, the new WFO in Grand Rapids expanded the snowfall reporting network during the 1995-1996 winter season by soliciting data from a large number of volunteer spotters. In this study, data from the expanded network was recorded for 13 potential lake effect snow events from the 1995-1996 season. The events were selected using the same criteria as in Evans (1996). Only data from the westernmost (closest to the lake) two tiers of counties within the WFO GRR area of responsibility were used. Also, for the sake of consistency, only data from stations that provided 24-hour snowfall observations within three hours of 12 UTC were used. Despite these restrictions, the expanded spotter network still resulted in many more reporting stations than in the 1991-1995 data-set. Therefore, the snowfall categories for the 1995-1996 data-set were redefined as follows:

Trace: No more than two stations reported a 24-hour snowfall of an inch or more.
Light: Three or more stations had 24-hour snowfalls $1 inch, and fewer than five stations had snowfalls $3 inches.
Heavy: Five or more stations had 24-hour snowfalls $3 inches.

For each of the 13 periods from the 1995-1996 season, model forecasts of vertical velocity, 850-mb temperature and relative humidity were recorded. All of the values were inferred by viewing ETA model forecast time-cross sections generated by PCGRIDDS at 43°N, 86°W. For each event, values were taken from the model run that began 12 hours before the beginning of the lake-snow event. From each of these model runs, the values were taken from the 12 through 36 hour forecasts. The resulting data base was then combined with the data base from Evans (1996).

3.   RESULTS

Combining the data from Evans (1996) and data from the 1995-1996 winter season resulted in an updated data base containing 47 events. Figures 1a-d show scatter diagrams of model forecast 850-mb temperature/lake temperature difference vs. (a) model forecast 850-mb relative humidity, (b) model forecast 700-mb relative humidity, (c) model forecast 850-mb vertical velocity and (d) model forecast 700-mb vertical velocity, for the 47 events in the study. The plotted letters indicate the snowfall category for each event. Events where "trace" amounts occurred are denoted by (T), "light" events are denoted by (L), and "heavy" events are denoted by (H). Data from the 1995-1996 season are indicated by bold lettering.

The following conventions were used to plot data for each 24-hour period (as in Evans 1996): The highest 24-hour forecast values of upward vertical velocity are indicated for each event, given that the vertical velocity must have coincided with a forecast 850-mb temperature forecast of # -12°C. The highest 24-hour forecast values of relative humidity are also indicated, given that the relative humidity must have coincided with a forecast 850-mb temperature forecast # -12°C. (Note: in Evans (1996), a different convention was used to plot "trace" events). Finally, the largest forecast 850-mb/lake temperature difference for each period is plotted, where the values for each event were determined by subtracting the monthly climatological average lake temperature for southern Lake Michigan from the model forecast 850-mb temperature.

Figures 1a and b show that the heaviest snows in the study tended to occur with a combination of large forecast air/lake temperature differences and high forecast values of low-level relative humidity. Likewise, the trace events mostly occurred with a combination of smaller forecast air/lake temperature differences and lower values of forecast relative humidity. Both diagrams are divided subjectively into "trace", "light" and "heavy" regions, with the resultant distribution of snowfall categorizations in each region listed below the diagram. Of the 47 cases in the data set, 38 were plotted within their expected regions on the graph in Figure 1a, and 34 were plotted within their expected regions in Figure 1b. Of the 13 cases from the 1995-1996 data set, 9 were plotted within their expected regions on the graph in Figure 1a, and 8 were plotted within their expected regions in Figure 1b.

Figures 1c and d show that the heaviest snows in the study tended to occur with a combination of large forecast air/lake temperature differences and negative forecast values of vertical velocity (upward motion). Likewise, the trace events mostly occurred with a combination of smaller forecast air/lake temperature differences and positive forecast values of vertical velocity (subsidence). As in Figures 1a and b, both diagrams are divided subjectively into "trace", "light" and "heavy" regions, with the resultant distribution of snowfall categorizations in each region listed below the diagram. Of the 47 cases in the data set, 35 were plotted within their expected regions on the graph in Figure 1c, and 34 were plotted within their expected regions in Figure 1d. Of the 13 cases from the 1995-1996 data set, nine were plotted within their expected regions on the graph in Figure 1c, and eight were plotted within their expected regions in Figure 1d.

Figure 1a. A scatter diagram showing the relationship between snowfall (by category), model forecast 850 mb/lake temperature difference (vertical axis) in °C, and model forecast 850-mb relative humidity (horizontal axis) in percent, for 47 events from 1991-1996.

Figure 1b. A scatter diagram showing the relationship between snowfall (by category), model forecast 850 mb/lake temperature difference (vertical axis) in °C, and model forecast 700-mb relative humidity (horizontal axis) in percent, for 47 events from 1991-1996.

Figure 1c. A scatter diagram showing the relationship between snowfall (by category), model forecast 850 mb/lake temperature difference (vertical axis) in °C, and model forecast 700-mb vertical velocity (horizontal axis) in µb/s, for 47 events from 1991-1996.

Figure 1d. A scatter diagram showing the relationship between snowfall (by category), model forecast 850 mb/lake temperature difference (vertical axis) in °C, and model forecast 700-mb vertical velocity (horizontal axis) in µb/s, for 47 events from 1991-1996.

4.   EXAMPLES

Figure 2 shows a PCGRIDDS-generated time-cross section forecast generated at 43°N, 86°W (near Muskegon) for the period from 12 UTC 12/10/94 through 12 UTC 12/12/94. The period from 12 UTC 12/11/94 through 00 UTC 12/12/94 (24-36 hours) was selected as a potential lake effect snow event, since the 850-mb temperature fell to as low as -12°C, with a northwest flow and minimal low-level directional shear. Given that the monthly climatological average lake temperature for December is 5°C over southern Lake Michigan, the estimated maximum forecast air/lake temperature difference during the period was -17°C. The forecast data on the time-cross section showed a maximum 850-mb relative humidity during the period of interest of 60% (at 24 hours, at 12 UTC 12/11/94). Figure 1a indicates that the combination of an air/lake temperature difference of -17°C and an 850-mb relative humidity of 60% yields a forecast snowfall amount in the "trace" category. Figure 3 shows 24-hour snowfalls that occurred during the period of interest.

Figure 2. A PCGRIDDS-generated time-cross section of ETA forecast temperature (dashed) in°C and relative humidity in percent from 12 UTC December 10, 1994 through 12 UTC December 12, 1994, at 43°N, 86°W.

Figure 3. 24-hour snowfall amounts in inches for the period ending on the morning of December 12, 1994 (snowfall values marked with asterisks indicate 24-hour reports received later in the day on December 12).

Figure 4 shows a PCGRIDDS-generated time-cross section forecast generated at 43°N, 86°W from 00 UTC 12/25/95 to 00 UTC 12/27/95. The period from 12 UTC 12/25/95 to 12 UTC 12/26/95 (12-36 hours) was selected as a potential lake effect snow event, since the 850-mb temperature was below -12°C through the period, in combination with a northwest wind and minimal shear. The forecast minimum 850-mb temperature was -14°C. Assuming a lake temperature of 5°C, the air/lake temperature difference was -19°C. Meanwhile, the forecast data on the time-cross section shows a maximum 850-mb relative humidity of 82%. Figure 1a indicates that the combination of an air/lake temperature difference of -19°C and an 850-mb relative humidity of 82% yields a forecast snowfall amount barely into the "heavy" range. Figure 5 shows 24-hour snowfalls that occurred during the period of interest.

Figure 4. A PCGRIDDS-generated time-cross section of ETA forecast temperature (dashed) in °C and relative humidity in percent from 00 UTC December 25, 1995 through 00 UTC December 27, 1995, at 43°N, 86°W.

Figure 5. 24-hour snowfall amounts in inches for the period ending on the morning of December 26, 1995 (snowfall values marked with asterisks indicate 24-hour reports received later in the day on December 26).

The snowfall amounts indicated in Figure 5 show that snowfalls across most of southwest Michigan where in the "light" (1 to 3 inches) category. However, a few bands of "heavy" (>3 inches) snow did occur, south of Muskegon and Grand Rapids. Because of the shape of Lake Michigan, winds from the north of west usually result in the heaviest lake effect snows falling south of Muskegon and Grand Rapids (southwest winds often result in the heaviest snow falling in the Grand Rapids area). Apparently in this case, the atmosphere was only marginally favorable for "heavy" snow to fall; therefore, "heavy" snow fell only across areas where the wind trajectories were most favorable.

5.   SUMMARY AND DISCUSSION

Relationships between snowfall amounts and gridded model forecast output have been studied for 47 potential multiple band lake effect snow events across southwest lower Michigan. As a result, scatter diagrams of snowfall as a function of model forecasts of 850-mb/lake temperature differences vs. model forecasts of relative humidity and vertical velocities have been developed. The results indicate that the scatter diagrams can be used to forecast snowfall amounts across southwest lower Michigan. Specifically, the results indicate that applying model forecasts to any one of the four graphs presented in this study would result in a correct snowfall forecast (by category) roughly 70 to 80 percent of the time.

Forecasts of 850-mb/lake temperature difference can be considered an approximation of the lower-tropospheric lapse rate, while forecasts of lower-tropospheric relative humidity and vertical velocity can be related to the height of the subsidence inversion. Therefore, the results of this study are consistent with the idea that snowfall amounts in a multiple band environment east of Lake Michigan are strongly dependent on lower tropospheric thermodynamic stability.

As forecasters gain experience with the graphs presented in this study, it is likely that forecasts could be improved by using multiple graphs for any given event. In addition, the forecasts can be fine-tuned by area, based on local knowledge of the effects of lower-tropospheric wind direction and speed on the placement of the heaviest snow bands.

One factor that should be considered when using these graphs is that the forecast duration of the maximum relative humidity or upward vertical motion will naturally have an effect on snow amounts. For example, consider two 24-hour events, both with forecast 850-mb/lake temperature differences continuously near -25°C and both with a forecast maximum 850-mb relative humidity of 90 percent. In one event, the relative humidity is forecast to remain from 80 to 90 percent through most of the 24-hour period, while in the second event the forecast relative humidity remains at 90 percent for only a couple of hours, then falls rapidly to around 40 percent, where it remains for the rest of the forecast period. Clearly, the first case would be more favorable for heavy snow, yet applying the graphs in this study would yield the same forecast for both events. This example shows that using the technique outlined in this study still leaves plenty of room for careful analysis of the model data, and for forecaster judgement.

The example given in the last paragraph suggests that scatter diagrams similar to the ones shown in this paper could be constructed from averaged forecast values of relative humidity and vertical velocity, instead of maximum or minimum forecast values. An effort has been made to produce such graphs, with the results (not shown) indicating that they would be useful, although probably no more so than the graphs shown in this paper.

6.   REFERENCES

Dockus, D.A., 1985: Lake Effect Snow Forecasting in the Computer Age, Nat. Wea. Dig., 10, 4, 5-19.

Evans, M.S., 1996: A Study on the Relationship Between Synoptic-Scale Vertical Motion and Snow Amounts Across Southwest Lower Michigan in Lake Effect Snow Environments, CR-ARP volume 17 (pending publication).

 


USA.gov is the U.S. government's official web portal to all federal, state and local government web resources and services.