Introduction:

The East Coast snowstorm of 4-6 March 2001 will be a memorable event for many. Ultimately, it produced a widespread swath of 1-2 feet of snow from Pennsylvania northeastward into New England. An impressive area of 2-3 feet of snow with some isolated higher amounts were observed across east-central New York and New England. From a forecast perspective, the storm was neither a resounding success nor complete failure. Forecasts were dictated by models, which, despite the strength of the storm, seemed to have difficulties forecasting the storms ultimate evolution.

This storm occurred near the peak time of a period of a negative North Atlantic Oscillation (NAO). Negative NAO’s have been correlated to cold and relatively dry periods. The Great Atlantic Low of January 1956 (Ludlum 1956) occurred during an unusual period of low NAO and was the strongest storm between 1948 and 2001 along the eastern United States and the adjacent western Atlantic Ocean. Similar to the 1956 storm, this event was associated with a deep upper-level cut-off low ever the western Atlantic.

From an anomaly point of view, this storm had an estimated MTOTAL of only 3.5, stronger then most the Kocin and Uccellini (1990) East Coast storms ranked. This event was a textbook Kocin and Uccellini (1990) storm including the strong anchoring anticyclone to the north. The 500 hPa height and 850 hPa wind anomalies associated with this event were on the order of 4 standard deviations from normal. However, the lack of significant moisture departures kept the MTOTAL for this storm in the unusual rather then the historic regime.

Model forecast of this storm, starting around 28 February to 1 March suggested a strong surface cyclone would track across the northern Gulf of Mexico eastward to the Carolina coast and then northeastward out to sea. Initially, it looked more like a Great Atlantic Low evolution then a major East Coast storm. However, subsequent model forecasts showed a continued trend to a more northeastward track. By 2 March, it was clear that the storm would affect the Mid-Atlantic region and southern New England. From 2-3 March, the focus seemed to be for heavy precipitation across southern New England. Unlike older studies of the Eta and AVN (Grumm and Siebers 1993), this storm track trended north and west and verified north and west of earlier forecasts. The traditional and anticipated south and east model error was not present in this event.

This page will examine the forecasts of this event and some tools to improve forecasting of similar events in the future. The majority of imagery used were available in real-time and were taken from the NWS-PSU eyewall website. Satellite imagery was provided by the Colorado State University RAMM team.

For this event, due to the great ensemble and LAF potential, all available GRIB files from the ETA, AVN, MRF, UKMET, ECMWF, and RUC were archived beginning with the forecasts from 28 February 2001. Select ETA and AVN GRIB files were used in the figures shown.

Additional data archived include the NCEP-MRF ensembles for the month of February and March 2001.WSR-88D data from the KCCX radar is available for the 4-5 March time period. The maximum forecast low-level wind anomalies (around 5 standard deviations above normal) occurred in the KOKX and KBOX forecast areas. These may contain evidence of orographic snows, which north and west of the surface cyclone.

Lagged Average Forecasts (LAFs) were computed from the AVN and ETA GRIB files using GrADS. For each verifying time, four panel charts of the MSLP and QPFs were made to show the dProg/dt. In some examples, 500 hPA charts were also made. All forecasts, from the same verifying time, where then averaged to compute the dispersion of these forecasts.

Initial GOES imagery was provided by CIRA-RAM. A local archive of GEMPAK viewable data was made.

1. Long range forecasts

The forecast from the 01 March 2001 MRF Ensembles valid at 1200 UTC 5 March is shown in Figure 1. This image shows the storm well out over the western Atlantic Ocean. The forecast from 02 March is shown in Figure 2. And the forecast from 3 March is shown in Figure 3. The trend in all these forecasts was for a cyclone to develop closer to the coast of the eastern United States. The multi-model ensembles offered similar forecasts and trends and only the forecasts from 3 March are shown in Figure 4. Early multi-model ensembles were showing a more eastward track.

At 500 hPa, the models were forecasting a deep and anomalous 500 hPa low to develop under a blocking ridge. An omega block was present over the North America in all the forecasts as depicted in Figure 5. Note the anomalous 500 hPa ridge over Hudson Bay, a classic blocking pattern favoring a strong surface anticyclone over much of eastern North America. This feature was well forecast by the MRF ensembles and in the multi-model ensembles, as shown in Figure 6. The spaghetti plots suggested a high confidence forecast of a deep 500 hPa low and ridge system. The forecasts showed an intense East Coast cyclone by 1200 UTC 6 March (Fig. 7) associated with an intense and anomalous upper-level low (Fig 8).

From a long-range perspective, this storm was poorly forecast and had a distinct trend to the north and west of previous forecasts. NCEPs model page on predictability showed that there was a measurable degree of uncertainty in these forecasts. But, by 3 March, the models had clearly converged on a strong storm to impact the eastern United States. After 1200 UTC, the forecasts really were in the realm of NCEPs short-term models and then the devil was in the details.

2. AVN

The NCEP aviation run of the Global Spectral Model (AVN) and the medium range forecast run (MRF) had several glaring successes during the winter of 2000-2001. The 30 December 2000 storm marked a clear example of the AVN out performing its NCEP cousin, the stepped terrain, Eta model. No doubt, many forecasters had come to rely on the reliable AVN in this event too.

The forecasts from the 1200 UTC run on 2 March valid 1200 UTC 5 March of 700 hPa height and MSLP are shown in Figure 9. The anomalies are also shown in each panel. The corresponding 850 hPa wind anomalies are shown in Figure 10. The model clearly was forecasting an intense coastal storm with an anomalous easterly jet. These forecasts probably played heavily on the statements and watches issued along the East Coast on Friday afternoon.

The following model run at 0000 UTC continued the trend. The accumulated QPF is shown in panel 2 of Figure 11. The corresponding surface forecast is shown in Figure 12. The storm was offshore, producing a large area of precipitation along the Mid-Atlantic and southeastern US. As the storm moved northward, the precipitation was forecast to move into the major cities of the northeast, including NYC and Philadelphia (Figure 13). As for type, the model 850 hPa temperatures are shown in Figure 14, the AVN forecasts suggested a strong possibility for snow.

By 1200 UTC 4 March, the models were clearly forecasting a large storm with the potential for near record snow as shown in the QPF forecasts valid at 0000 UTC 6 March (Figure 15). Even the consensus forecast, and AVN and Eta blend, was forecasting record snow water equivalent over the eastern two thirds of Pennsylvania, New Jersey and into New England. Seemingly, a very high confidence forecast of heavy QPF, of mainly snow over the major population corridor of the eastern United States. An ensemble or consensus of bad forecasts appears to be, in itself, a bad forecast, at least in this case.

In many major storms, timing is everything. The perception on Monday morning 5 March was that the cities of the northeastern United States would awake to a significant amount of snow. This led to some negative press via the media during the day of 5 March. However, the model QPFs valid at 0000 UTC and 1200 UTC 5 March clearly showed that the significant QPF and related snow, was not forecast to reach the NYC are much before 1200 UTC 5 March. In fact, the heavy snow observed along the southern tier of New York State by 1200 UTC 5 March was totally unforecast and much of the heavy QPF in southeastern Pennsylvania and New Jersey was unobserved (a rigorous precipitation examination has not been conducted). The consensus (lower left in both panels) and the AVN (upper right) had some serious QPF errors.

 

The QPFs from the 0000 UTC and 1200 UTC 4 March AVN and Eta are shown in Figures 19 and 20 respectively. Over Pennsylvania, there was clearly a trend toward a more westward shift in the precipitation axis between the 1200 UTC 3 March and 0000 UTC 4 March forecasts. Then, the later forecasts from 12 March shifted the QPF farther to the west. The threat for eastern Pennsylvania and New Jersey was diminishing with time and the snowfall amounts, if all the QPF all fell as snow were no longer much above the winter storm warning criteria. The model 850 hPa temperatures (Figure 21) clearly showed an intrusion of warm air well inland. The models still seemed unable to handle the potential snowfall across the southern tier and Finger Lakes district of New York State.

 

The 0000 UTC 5 March run of the AVN (Figure 22) finally showed the QPF over New York State and the much diminished forecast over coastal plain. Too late to salvage the forecasts and warnings issued on 3 and 4 March. It was this model run that began to show the potential record event over southern New England, if all the forecast QPF (Figure 23) fell as snow.

3. Eta forecast

The Eta is operationally available to 60 hours and on the web to 72-hours. Traditionally, the Eta is viewed as a detailed mesoscale model, so the focus will be on the shorter term forecasts.

Forecasts from the 0000 UTC 4 March cycle showed a strong surface cyclone developing along the Delmarva peninsula around 1200 UTC 5 March and deepening as it slowly drifted eastward by 0000 UTC 6 March (Figures 24 and 25). The intense low-level (850 hPa) jet associated with this low is shown in Figure 26. This jet was forecast to be in excess of 5 standard deviations from normal, quite an impressive feature should it have verified. Twelve hours later, the Eta forecast a slightly weaker jet farther north (Figure 27). The Eta verifying low-level jet was not as intense nor did it extend as far to the west as forecast (Figure 28.). The implied difference in the convergence and deformation zone over central Pennsylvania and New York is dramatic. Model diagnostics of Q-vectors and frontogenesis from the forecasts on 4 March 2001 may have told an interesting story as to why it would snow and where. The reality was, the model atmosphere used for any diagnostics was of no value based on the verifying winds. The model had no hint of the strong westerly winds penetrating Pennsylvania and the more southeasterly jet moving into New York State.

The 00-h 1200 UTC Eta surface analysis also showed a weaker cyclone farther east the previous forecasts (Figure 29). Clearly, the early stages of this event represented serious forecast problems. It was during this forecast cycle that the Eta finally got a handle on the short wave that would produce the deep surface cyclone along the East Coast. The Eta forecast this feature to be of eastern Long Island by 1200 UTC 6 March (Figure 30). Finally, the forecast of a surface cyclone to bring heavy snow to New England and eastern New York. Unfortunately, by this time, the remote sensing data also showed the rapid development off the Carolina coast.

The Eta QPF forecasts were shown along with the AVN above. In generally, the Eta was wetter and faster to bring the QPF northward then the AVN. Clearly, the reliance on any one model could have biased a forecast however, in this case, and consensus was not much of a tool.

5. Lagged Average Forecasts.

Under developed! Suffice to say there were some clear trends pointing to numerous problems.

6. Satellite data

1. Sunday 4 March 7, 2001

 

This phase of the event was the most poorly forecast. The first round of snow over the Mid-Atlantic region developed ahead of a short-wave that moved rapidly northward out northeastern Mexico and Texas on the morning of 3 March 2001. This feature can be in the GOES-8 imager channel-3 data valid at 1345 UTC 3 March (Figure X). By 1945 UTC the dry enhancement with the wave could seen over eastern Texas (Figure Y) reaching Arkansas early on 4 March (Figure Z). By 0745 UTC 4 March, the feature was clearly visible over northwestern Kentucky and an enhanced area of overrunning enhanced clouds was visible over West Virginia ahead of the strong wave. Snow broke out a few ours later over central Pennsylvania.

 

By 0145 UTC 5 March (Figure ZZ), the short-wave and overrunning clouds had washed out. A deformation band was forming over western Pennsylvania, and the short-wave that would produce the rapidly developing cyclone, could be seen moving across the Gulf Coast States. A loop of these data shows the second area of drying behind the thin line of moisture across Georgia northward into New York. The satellite image at this time was quite complex showing short-waves of varying scales and intensity.

2. Monday 5 March 7, 2001

The GOES-8 imager channel 3 data is used to show the evolution. At 1345 UTC 5 March (Figure 31) and intense developing short-wave can be seen off the Carolina coast. This wave would rapidly expand its cold cloud shield as it moved toward Long Island and southern New England. Six hours later (figure 32), the short-wave could be seen just south of Long Island, with enhanced clouds over New England. Note the strong deformation band over central New York. This band had slowly moved eastward in the 6-hour period. Surface data (not shown) clearly indicated a rapid transition from mixed precipitation and rain in the New York metropolitan area as the strong short wave approached from the south during the afternoon hours of 5 March.

By 0145 UTC 6 March (figure 33), the enhanced clouds, or deep moisture were over central New England while the low-level circulation was over coastal Connecticut. A persistent larger scale deformation band was still present over central New York State, well west of the now well developed surface cyclone and short-wave. By 1345 UTC figure 34), the enhanced clouds had moved into northern New England and the large vortex, as forecast, dominated the eastern United States and western Atlantic. A third short wave could be seen off the Carolina coast. At 1945 UTC (figure 34),) the cold conveyor belt north of the vortex could be seen across central and northern New England.

This was a storm of contrasts of success. Overall, the models correctly forecast a massive upper-level low and surface cyclone to develop over the western Atlantic. This forecast was a success about 7 days in advance. The heavy rains and severe weather that impacted the Gulf States and Florida represented extremely successful forecasts. The potential for heavy rain and severe weather over Florida during the weekend of 3-4 March was much anticipated. However, the location and intensity of the storm over the western Atlantic was in doubt for quite a long time. On about 2 March, the models converged on a solution of a storm, which would impact the East Coast. The devil was in the details.

As the storm fell into the range of the shorter-term models, there appeared to be some convergence of solutions. However, a signal developed showing a clear trend toward a more northwestward storm track and the resulting shift in the precipitation shield associated with the storm. On 2 March, precipitation type did not seem to be a significant issue. But as the storm track shifted westward, it became one.

The models also made gross over estimates of the QPF over many areas. For example, the large area of 1-2 inches of QPF over southeastern Pennsylvania did not verify. Most areas of southeastern Pennsylvania received 0.5 to 0.8 inches of total QPF. Most of the precipitation fell prior to 1200 UTC 5 March. The combination of too much QPF and the precipitation type issue both limited the accumulated snowfall over southeastern Pennsylvania. To the west and north, the models under forecast the amount of QPF, playing a slow game of catch up with each successive forecast. The heavy snow in northwestern Pennsylvania and central New York arrived faster and in greater amounts then initially forecast.

In this event, Eta and AVN consensus forecasts offered little hope of improving upon the forecasts. The consensus was generally less then the Eta forecasts. But most forecasters are aware that the finer scale models often produce more QPF and know to caveat their forecast appropriately. Surely, more QPFs would be required to make a consensus with better skill. It would be interesting to see if the RSM and ETA ensembles produce better forecasts. I speculate they would not have offered much hope due to the seemingly close agreement between the Eta and AVN.

Another aspect of forecasting has been on using model soundings and model diagnostics. Based on the mass and thermal field errors of Pennsylvania and New York around 1200 UTC, these types of data allow forecasters to make precise inaccurate forecasts. Model diagnostics and model soundings represent the best or the worst of a single deterministic model. The Eta diagnostics and soundings from 03-04 March clearly represented the latter. 

Finally, the models did correctly forecast the development of a major cyclone of the East Coast. NCEPs ensemble showed some uncertainty with these forecasts. However, the real atmosphere did it in an explosion over the warm Atlantic Ocean during the morning of 5 March, whereas the models did as a more progressive and continuous evolution. By the evening of 5 March, the models and the atmosphere appeared to be in concert again, too late for too many. Truly, the timing and evolution were not very good and the devil was in the details.

Surface analyses are provided by the CTP surface analysis team members. In this case,

John LaCorte is still working on analyses.

Many thanks to Brian Motta (CIRRA), Walt Drag (KBOS), Jeff McQueen (some big office symbol in DC!) and my good friend Robert Hart each of whom made significant contributions. Brian provided the imagery and comments. Jeff provided many comments during and just after the event. Walt worked the event but responded to my emails and offered excellent insights. Bob, well he raced to Connecticut Sunday morning to see the big snow there. He also wrote the GrADS accumulator program on 2 March to improve our LAF and QPF displays, which was implemented on 3 March and used to make the consensus forecasts on 4 March.

Ludlum, D. M., 1956: "The Great Atlantic Low". Weatherwise, 9, 64-65.