Ensemble Forecasts of the Blizzard of 22-23 January 2005 By Richard H. Grumm National Weather Service State College Pennsylvania 1. Introduction A fast moving upper-level disturbance produced a wide area of snow, and heavy snow from the Great Lakes to the northeastern United States from 21 to 23 January 2005. This storm has become known as the blizzard of 2005 snowfall produced by this storm over the 3 day period is shown in Figure 1. Though hard to infer from the image, Areas of intense snowfall, in excess of 10 inches (25 cm) were observed west of Lake Michigan in Illinois, in southern Michigan and south of Lake Ontario in New York. Very heavy snowfall, with over 20 inches (50 cm) and in some instances 30 inches (75 cm) was observed in southern New England. This heavy snowfall set the city of Boston up for its snowiest January on record. Though there were several medium-range forecasts of this system 4-7 days in advance, the exact details of the snowfall areas remained elusive. Most forecast systems showed a clipper-like low which had the potential to produce heavy snow from the central Appalachians to the East Coast. Within 4 days of the event, some forecasts of extremely heavy snow were indicated. Clipper-type lows are not normally associated with extremely heavy snowfalls in the eastern United States. This storm was forecast to be an exception to this general rule. Though the medium and short-range models did forecast a significant East Coast snow storm, they missed the details. This storm showed many errors typical of recent large snow storms along the East Coast. As in the Presidents Day Weekend storm of 16-17 February 2003 (Grumm 2003), the storm was initially not forecast to affect New England with significant snow. This storm shared other characteristics with previous storms in failing to produce significant snow far enough northward. Whether these apparent problems reflect systematic errors or represent anecdotal evidence is uncertain. It is clear that several large East Coast storms in recent years were initially forecast not to affect the regions farther which were ultimately affected by the storm. This storm was a relative success for both the deterministic and ensemble forecast systems (EPS) in that they correctly forecast a significant winter storm; they correctly forecast the meso-α features associated with this storm, and the correctly forecast the general timing of the event. This event also was a failure in that the areas of heavy snow in New York and New England were poorly forecast with any significant lead times. It is beyond the scope of this paper to determine if this was an initial conditions issue or is related to model physics issues. This paper will serve to document the Blizzard of 2005. It will also demonstrate the strengths and the weaknesses of the current National Centers for Environmental Predictions (NCEP) Medium- Range Ensemble Forecast (MREF) system and the Short-Range Ensemble Forecast (SREF) system. The emphasis is on how these two systems correctly forecast a significant snowstorm, but failed to capture and correctly depict the details of the heavy snow areas. 2. Method All SREF and SREF data were archived in real-time. All images were recreated after the event using GrADS (Doty and Kinter 1992). Standard displays of spaghetti and spread are shown. The spread is normally shown with gray shading. The spaghetti plots show distinct contours for each parameter. Lower panels for some fields show the ensemble mean with color shading showing the ensemble mean in standard deviations from normal, referred to as standardized anomalies (Grumm and Hart 2001;Hart and Grumm 2001).
Precipitation displays show the critical thresholds over a time period, normally 24 hours and 0.50 inches of quantitative precipitation forecast (QPF). Upper panels normally show the probabilities and the consensus of the specified thresholds. Lower panels show the position of the specified threshold for each EPS member and shading shows the ensemble mean QPF. The spaghetti plots follow the standard ensemble concepts outlined by Sivillio et al (1997). This paper is a good reference point Figure 1 Large view of the snow fall observed during the blizzard of January 2005. Snowfall shows totals in inches from 20-23 January 2005. Lower panel has zoomed in view of southern New England and southern New York.
Figure 2 MREF forecasts initialized at 0000 UTC 18 January 2005 showing a) MSLP (hpa) forecasts valid at 1200 UTC Sunday 23 January 2005 and b) 24-hour precipitation for the threshold contour of 0.50 inches of QPF. Pressure shows spaghetti plots and spread in the upper panel. The lower panel shows the ensemble mean and the departure of this field in standardized anomalies. The precipitation panel shows probability of exceeding 0.50 inches of QPF and the mean position of the 0.50 inch contour. The lower panels shows the spaghetti plot of each EPS members forecast position of the 0.50 inch contour and the EPS mean QPF for the 24 hours ending at 1200 UTC 23 January 2005. to those learning to use ensembles. 3. Results a) MREF Forecasts MREF forecasts as early as 16 January 2005 showed signals for a snow event along the East Coast. However, the overall structure showed a fast moving cyclone affecting the Mid-West and the Mid-Atlantic region. It was a slow evolution toward a significant snow event and most forecasts appeared to emphasized the Mid-Altantic region as the focus for the heavy snowfall. A few MREF images will be presented to show the evolution of these forecasts. a.1) 18 January Forecasts Forecast from 0000 UTC 18 January 2005 are shown in Figure 2. These forecasts show off the Carolina coast at 1200 UTC 23 January 2005 with most of the precipitation (Figure 2 right) over the southeastern United States. This run was too fare south with the cyclone track from the upper-mid West to the coast. Though not shown, the CMC forecasts at this time showed a higher threat of snow farther north. Subsequent 0600 and 1200 UTC MREF runs began to converge on a more northern storm track and a northward shift in the precipitation shield (not shown). a..2) 19 January 2005 forecasts Figure 3 shows the MREF forecasts initialized at 0000 UTC 19 January 2005. The panels on the left show the MSLP
Figure 3 MREF forecasts initialized at 0000 UTC 19 January 2005 showing a) MSLP (hpa) forecasts valid at 1200 UTC Sunday 23 January 2005 and b) 24-hour precipitation for the threshold contour of 0.50 inches of QPF. Pressure shows spaghetti plots and spread in the upper panel. The lower panel shows the ensemble mean and the departure of this field in standardized anomalies. The precipitation panel shows probability of exceeding 0.50 inches of QPF and the mean position of the 0.50 inch contour. The lower panels shows the spaghetti plot of each EPS members forecast position of the 0.50 inch contour and the EPS mean QPF for the 24 hours ending at 1200 UTC 23 January 2005. forecasts (hpa) and the panels on the right the QPF forecasts. Of interest is the cyclone off the Mid-Atlantic coast. In the mean, this cyclone was forecast to be -1.5SDs below normal. This implies good agreement between ensemble members. The spread, in the upper panel is on the order of 2-4 hpa with some dispersion in the position of the surface cyclone. A strong anticyclone is seen to the west. Though not shown, this cyclone was forecast to track out of Canada, across the Mid-West and to the Mid-Atlantic coastal area. The resulting QPF is shown in Figure 2(right). These data showed a high probability of precipitation; forecasts suggested cold air in place for snow, based on 2m and 850 hpa temperatures; for the 24 hours ending at 1200 UTC 23 January 2005. Note 1 member forecast the threat into central New England, based on Figure 1, suggesting that at least one member caught the potential of snowfall farther north. The focus at this time was for precipitation along the coast. Early POP forecasts showed the track of high threat running from Minnesota to Maryland but are not shown. These forecasts followed a trend which began in the 0600 UTC 18 January 2005 MREF forecast cycles showing a farther northward track and a significant threat for snowfall in the Mid-Atlantic region. Figure 4 shows the QPF for times the 24-hour periods ending at 0000 and 1800 UTC 23 January 2005. These two figures shows the threat for precipitation as the wave moved across the Mid West and the threat for snow into southern New England for the forecasts
Figure 4. MREF forecasts initialized at 0000 UTC 19 January 2005 showing the probability of exceeding 0.50 inches of QPF for the 24 hour periods ending at 0000 and 1800 UTC 23 January 2005. ending at 18000 UTC 23 January 2005. Again, note that at least one member showed the threat for at least 0.50 inches of QPF into central New England for the period valid ending at 1800 UTC 23 January 2005. Clearly, there was sufficient information to alert the public of the threat for the potential heavy snow as far north southern New England to include the city of Boston. a.3) 20 January 2005 forecasts The 20 January MREF forecasts continued to show the threat for a precipitation event, mainly snow from the Mid-West to the East Coast (Fig. 5). The POP and QPF is shown for the 24 hour periods ending at 0000 and 1200 UTC 23 January. The earlier times show the high threat for the Chicago area and portions of Ohio. The latter forecasts show the high threats for the Mid-Atlantic region. It is interesting that 2 ensemble members continued to show a threat farther north into New England suggesting some members were capturing the more northern cyclone track and the more northern precipitation shield with this system. Figure 6 shows the MSLP and 850 hpa wind forecasts valid at 1200 UTC 23 January 2005. The much below normal easterly winds were a significant sign of a potentially significant event. Grumm and Hart (2001) showed the impact of much below normal easterly winds with winter storms. Though the QPF values were low, the signal with the jet suggested the potential for a significantly above normal event. Though not shown, the single GFS runs showed the 850 hpa jet to be as much as -5SDs below normal over the region. Other fields showed similarly high anomalies. b) SREF forecasts The QPF forecasts for the 24 hour periods ending at 0000 and 1800 UTC 23 January 2005 from forecasts initialized at 0900 UTC
Figure 5 As in Figure 4 except for MREF forecasts initialized at 0000 UTC 20 January 2005 for time 24 hour intervals ending at a) 0000 and b) 1200 UTC 23 January 2005. 21 January 2005 are shown in Figure 7. These data show a high probability of 0.50 or greater QPF over the Chicago and Washington DC areas on the 22 nd. Later the forecasts suggest a high probability of 0.50 or greater QPF for southern New York and extreme southern New England. There is a sharp cut-off of precipitation to the north. Local maximum in the consensus forecasts show some areas of over 1.0 inches of QPF and several SREF member forecast over 1.0 inches of QPF (not shown). Based on temperatures profiles, forecasts indicated 10:1 snow ratios along the southern edge of the storm and 20:1 ratios in the colder air. This the 0.50 area of QPF implied 10 inches of snow over most areas. Snow ratios are beyond the scope of this paper. Figure 8 shows SREF 850 hpa wind forecasts valid at 0600 UTC 23 January 2005 from forecasts initialized on the 21 st at 0900 UTC. These data show a significantly above normal low-level jet feature. Over New Jersey the forecast showed easterly wind anomalies in excess of -5SDs below normal. The anomalous jet extended well westward with the are of -3 to -4SD anomalies covering most of Pennsylvania and the area of -2 to -3SD below normal easterlies extending into Illinois. This suggests two things first, a very anomalous and strong easterly jet was forecast and second, there was general agreement between forecast members to produce such large anomalies in a 42-hour forecast. The latter point suggests the SREF may have lacked spread in the forecast members. The SREF MSLP forecasts from the 0900 UTC 21 January cycle are shown in Figure 9. These data show generally good agreement with the primary features including the anomalously strong surface anticyclone. The only area of spread is associated with secondary cyclone along the coast. This is due to the effects central pressure and placement differences between SREF members.
Figure 6 MREF forecasts initialized at 0000 UTC 20 January 2005 valid at 1200 UTC 23 January 2005 showing a) MSLP forecasts and b) 850 hpa winds. The 850 hpa winds show the ensemble mean and the U(upper) and V (lower) wind anomalies in standard deviations from normal. MSLP shows spaghetti and spread in the upper panel and the ensemble mean and standardized anomalies in the lower pane.. Overall, these forecasts showed the threat and captured the regions susceptible to the threat. However, when compared to Figure 1, these forecasts failed to capture the heavy snowfall observed farther north in New York State and central New England. iii) SREF initialized at 2100 UTC 21 January Forecasts initialized at 2100 UTC 21 January 2005 showed several significant differences than the previous forecasts. This is best illustrated by the MSLP forecast which showed the cyclone at 23/0000 UTC off the Delmarva and New Jersey coastal areas (Figure 10a). The cyclone was forecast to be deeper and farther north. The forecasts valid 6 hours later showed a strong cyclone (992 hpa) in the ensemble mean south of Long Island, New York (Fig. 10b). This resulting northward shift pushed the anomalous low-level jet northward (Fig. 11) now focusing the inflow and the cold conveyor belt over southern New England and into southeastern New York State. These forecasts represented a dramatic shift northward in the region threatened by heavy snow as seen in Figure 12. Note the increased probabilities of 0.50 inches or greater QPF with the clipper as it zipped from Illinois to the coast and the marked northward shift in the snow shield over New York and New England. The impact of the cyclone track shift was dramatic. 6. Conclusions The Blizzard of 2005 struck the Mid West and the East Coast from 20 to 23 January 2005. This storm brought heavy snow from the upper Mid West to the northeastern United States. Heavy snow was observed in
Figure 7 SREF forecasts initialized at 0900 UTC 21 January 2005 showing 24-hour accumulated precipitation for the 24-hour periods ending at 0000 and 1800 UTC 23 January 2005. Upper panels show the percentage of exceeding 0.50 inches of QPF and the consensus position of the 0.50 inch contour. The lower panels show each members location of the 0.50 inch contour and the consensus QPF. Chicago, Detroit, Philadelphia, New York, and Boston. Snow fall totals in excess of 12 inches were common in the Mid West and along the East Coast. A rare event when a Clipper-like cyclone can produce such high snowfall totals. Overall, the NCEP and CMC EPS s performed well with this event, providing long-lead times for a potential snow storm with a high probability of heavy snow in close proximity to where the snow was observed. Two distinct problems were observed in this event. First, all forecasts, including the MREF and SREF, initially had the storm track and heavy precipitation shield too far south compared to what was observed. However, the signals were well forecast. The QPF forecasts along the northern edges of the storm were particularly poor for interior New York State and New England. In the SREF, the more northward track shift at 2100 UTC 21 January produced pushed the precipitation shield farther north. The cause of these more northward positions of the cyclone, wind anomalies, and resulting precipitation shield may have been due to the impacts of initial conditions and possibly the impacts of convection in the models. Zhang et al (2003) showed how the upscale impacts of model convection can impact the larger scale forecasts. How convection affects the development and track of the surface cyclone and the evolution of the precipitation shield is an potential avenue of future research. The MREFs by 0000 UTC 19 January began to show, with as many as 2 member, the potential for heavy snow into central New York and New England. This implies to some measure that the initial conditions may have picked up on the pattern that made this
Figure 8 SREF forecasts initialized at 0900 UTC 21 January 2005 showing EPS mean 850 hpa winds and a) U wind anomalies and b) V wind anomalies in standard deviations from normal. event as the forecasts appeared to capture the outcome which was observed. Why the SREFs appeared to lack similar spread and missed the more northward precipitation shield and cyclone track is an intriguing question. It may be due to initial conditions and/or model physics. But the impact on the precipitation shield was dramatic. Despite the under forecasting of the precipitation amounts and the northward extend of the precipitation shield, the models forecast the event quite well. They produced signals which should have, and in many instances did, alert forecasters to the potential of a significant winter storm. The strong low-level 850 hpa jet in both the longer range MREF (Fig 6) and shorter range SREFs (Fig. 8) implied an unusually strong low-level jet. A feature well know to be associated with significant winter storms and heavy rainfall in the eastern United States (Grumm and Hart 2001). Though not shown, the 250 hpa wind anomalies were on the order of -2.5 SD at below normal in the SREF. Stronger anomalies were found Figure 9 SREF forecasts initialized at 0900 UTC 21 January 2005 showing EPS mean MSLP (hpa) forecasts valid at 0000 UTC 23 January 2005. Upper panel shows select contours and spread and lower panel shows the ensemble mean and the departure of this field in standard deviations from normal. in single deterministic runs. The strength of the upper-level jet was significantly above normal in this event. The 850 hpa winds were in the nearly historic levels as forecast by the SREF, comparable to the Blizzard of 19'78. However, the upper-level anomaly was weaker than the -3.85SD anomaly at 250 hpa which persisted for over 30 hours in that event (Stuart in review). It should be noted that the National Weather Service office (WFO) in Taunton, Massachusetts made the following observation during the event: The first blizzard since the April fools storm of 1997 has blanketed the area with a top 5 historic snowfall... inside 24 hours... along with high winds and bitterly cold temperatures. Clearly, this was one of the most significant storms in recent history along the East Coast. These data suggest that an extreme winter storm forecast index (EWSFI) could be
Figure 10 SREF forecasts initialized at 2100 UTC 21 January 2005 showing EPS mean MSLP (hpa) forecasts valid at a) 0000 UTC and b) 0600 UTC 23 January 2005. Upper panel shows select contours and spread and lower panel shows the ensemble mean and the departure of this field in standard deviations from normal. produced from both ensemble forecast and deterministic forecast data to gauge the threat for a significant snow storm. These results also suggest that the models had some difficulty getting the details write of the system and missed the more northerly cyclone track. However the more significant problem is focused on the QPF forecasts. How we employ EPS QPFs in the forecast process is still a significant problem and clearly, the ensemble mean and probabilities should be used with caution and knowledge about uncertainty. The fact that 20% of the MREF members showed the threat for heavy snow farther north was a signal. This system was forecast within the realm of the ensembles. How this information was used is significant. The author used these data to change a flight from Washington Dulles to San Paulo, Brazil from the 22 nd to the 21 st and avoided the storm and any potentially delays. The question arises (adopted from Ken Mylne, UKMO), would you risk missing a trip if there was a 20% chance that the weather would cause a delay or cancellation of the flight you were on? 7. ACKNOWLEDGEMENTS John LaCorte for snowfall data and maps. Mike Evans, Neil Stuart, Walt Drag, and Jeff McQueen for insights and observations. 8. References Doty, B. and J.L. Kinter III, 1992: The Grid Analysis and Display System (GrADS): A practical tool for earch science visualization. Eighth International Conference on Interactive Information and Procession Systems, Atlanta, Georgia, 5-10 January, 1992.
Grumm, R.H., and RE Hart. 2001: Standardized Anomalies Applied to Significant Cold Season Weather Events: Preliminary Findings. Weather and Forecasting,16, 736 754. Grumm, R. H., and Hart R., 2001: Anticipating heavy rainfall events: Forecast aspects. Preprints, Symp. on Precipitation Extremes: Prediction, Impacts, and Responses, Albuquerque, NM, Amer. Meteor. Soc., 66 70. Hart, R.E., and R H. Grumm. 2001: Using Normalized Climatological Anomalies to Rank Synoptic-Scale Events Objectively. Monthly Weather Review,129,2426 2442. Hart, R., and Grumm R. H., 2001a: Anticipating heavy rainfall events: Climatological aspects. Preprints, Symp. on Precipitation Extremes: Prediction, Impacts, and Responses, Albuquerque, NM, Amer. Meteor. Soc., 271 274. Sivillo J.K, JE. Ahlquist and Z Toth. 1997: An Ensemble Forecasting Primer. Weather and Forecasting,12, 809 818. Zhang, F, C.Snyder, and R. Rotunno, 2003: Effects of moist convection on mesoscale predictability. JAS,60,1173-1184.
Figure 11 SREF forecasts initialized at 2100 UTC 21 January 2005 showing EPS mean 850 hpa winds and a) U wind anomalies and b) V wind anomalies in standard deviations from normal.
Figure 12 SREF forecasts initialized at 2100 UTC 21 January 2005 showing 24-hour accumulated precipitation for the 24-hour periods ending at 0000 and 1200 UTC 23 January 2005. Upper panels show the percentage of exceeding 0.50 inches of QPF and the consensus position of the 0.50 inch contour. The lower panels show each members location of the 0.50 inch contour and the consensus QPF.