The southern New England snow storm of
25 February:
The challenge of multiple models
By
Richard H. Grumm
With contributions from:
John L. LaCorte
And
David J. Nicosia
Introduction:
On 25 February 1999, a strong oceanic cyclone produced heavy snow north and west of the surface cyclone center. The heavy snow impacted southeastern New England and eastern Long Island. With the exception of Boston, MA, all the major metropolitan areas of the eastern seaboard were missed by this storm.
This storm represented several forecast challenges to the operational weather community. First, longer range model guidance, in the 5 day range suggested a potential East Coast storm. Forecasts from 20 February suggested this storm would impact the major metropolitan areas from near Richmond northward to Boston. The models quickly backed off this solution.
As the event began to fall into the realm of the short-range models (0-48 hours), the cyclone of 20 February's forecasts began to reveal themselves, with a cyclone too far offshore to be a significant factor. However, the stepped terrain Eta model suggested heavy snow was possible for New York City, Long Island, much of Connecticut, Rhode Island, and Massachusetts.
As the event drew near, the Eta forecasts slowly converged on a solution moving the surface cyclone farther to the east, as forecast by the AVN. Despite this correct trend, heavy snow fell over much of southeastern New England and eastern Long Island, NY.
The purpose of this page is to show how the Eta forecasts slowly moved toward those of the AVN in relation to its position of the surface cyclone. Despite this more eastward forecast, the strength of the upper level low and the low-level circulation produce conditions favorable for heavy snow west of the surface low. These bands appeared on radar and produced a significant portion of the heavy snow.
Methods
Currently, all data are from the NCEP data feed which is converted to GEMPAK format. Scripts were used to show the forecast trends (dProg/dt) for various model runs. Only Eta and AVN model forecasts are shown. All dProg/dt charts show the forecast cycle (990223/0000F48 would be the 48-h forecast from 23 Feb 1999 at 0000 UTC )
In addition to dProg/dt charts, several equivalent potential vorticity (EPV) and frontogenesis (FGEN) charts have been produced in an attempt to explain roughly why and where the snow fell. All EPV was computed using the geostrophic wind
In addition to the Eta forecasts, one run of the MM5 model is available in VIS5D and GRADS format. The MM5 was initialized off the 1200 UTC 23 February Eta. Despite the initialization off the Eta, the MM5 simulation appeared to produce a more accurate depiction of the southern New England snowstorm. The Pennsylvania State University Numerical Weather Prediction Group provided MM5 forecasts. Some MM5 visualization graphics can be obtained here. These graphics represent isosurfaces of equivalent potential temperature with other fields superimposed, including mean-sea-level pressure.
The Case
NCEP, UKMET, and ECMWF runs, from as early as Saturday, 20 February 1999 showed the potential for a significant East Coast cyclone. However from 21 February onward, the forecasts continued to show a weaker, more eastward forecast track of this cyclone. Forecasts from 20 and 21 February showed the threat of a potential major winter storm along the entire eastern Seaboard by mid-week from the Carolina's northward. Ensemble forecasts suggested great variation in both the storm track and intensity. These forecasts will not specifically be discussed. But they are instructive in that they showed a critical envelope of solutions, which as will be shown, the models also showed in the shorter forecast ranges.
The AVN forecasts in the 72 to 36-h range, all valid at 25/0000 UTC are shown in Figure AVN5. These forecasts show that the AVN forecast a relatively weak cyclone, which would remain well offshore.
A series of dProg/dt forecasts are shown beginning from 72 hours out from the 22/1200 UTC forecast cycle. All of the following forecasts are valid at 25/1200 UTC, about the height of the storm. The longer-range forecasts are shown in Figure AVN4. These forecasts show a relatively well developed cyclone forecast for the morning of 25 February. Most of the strong winds and sensible weather of interest would be confined to the western Atlantic with only a suggestion of some unsettled weather in extreme southeastern New England. However, the trend appeared to be a steady increasing threat to southeastern New England.
The next available model cycle is used in Figure AVN3. The panel "b" in the previous Figure is now panel "a". The forecasts show a trend toward a more northward position and more cyclonic flow into southeastern New England. The next forecast cycle contributed more interesting data as shown in Figure AVN2. The forecast trend for a more northward position continued, with a hint of a slightly more westward track in latter forecasts. The strong low-level cyclonic flow continues to increase with each forecast over southeastern New England.
Finally, by the last available forecast before the snow began, is shown in Figure AVN1. Panel "d" represents the AVN verification at 25/1200 UTC. The model clearly forecast strong low-level cyclonic flow over southeastern New England.
The precipitation forecasts reflect this trend toward more cyclonic flow over southeastern New England. Note the westward progression of the measurable precipitation in Figure AVNPRECIP1.
One method to improve upon the sensible weather forecast is to examine model trends with time. This is often referred to as dProg/dt, and such a function is available for practical use in real-time in AWIPS. The MSLP forecasts, all valid at 25/0000 UTC are shown in Figure x. The key to these figures is that each successive Eta run showed the surface low to be weaker and farther west then previous runs. The trend was toward the forecasts forecast by the AVN (not shown-YET).
The next forecast cycle set is shown in Figure xx. All the forecasts are valid at 25/0000 UTC. Note that panel d is basically the Eta's verification for the specified time (Forecast length = 0). Again, it is interesting to note how much weaker and farther east the surface low was observed relative to longer-range forecasts.
The forecasts valid at 25/1200 UTC are shown in Figure xxx. After the forecast from 24/0000 UTC, the Eta track and intensity forecasts became consistent, with a slightly weaker and more eastward tracking low-pressure system. The Eta initial analysis at this time suggest that 12 and 24-h forecasts were slightly too high with the pressure of the center.
Forecast of the 12-h accumulate precipitation (mm) are shown in Figures yy and zz. Contours in millimeters and the contour interval is in powers of 2 mm. It is interesting to note that the 48-h forecast from the 23/1200 UTC Eta forecast in excess of 16 mm of snow over NYC and Boston valid at 25/1200 UTC (Fig yy panel a). The next forecast cycle (panel b) the model backed off the amount of precipitation over NYC and Boston. Anyone choosing the Eta as the model of the day at 1200 UTC 23 February would have had to forecast heavy snow for NYC, Long Island and southern New England. Successive forecasts continued to back off on the forecast precipitation for NYC and southern New England.
Forecast from the model run on 25 February showed a reversal of this trend, despite the consistent more eastward cyclone track. Note the resurgence of the 16 mm isopleth over eastern Long Island, Rhode Island, and Connecticut in panel C (25/0000 UTC forecast valid at 26/0000 UTC. It is theorized that the model produced the necessary mesoscale signature to produce heavy precipitation well west of the cyclone center as the forecast length decreased.
The 850 and 700 MB frontogenesis forecasts from the 25/0300 UTC Eta area shown in Figures F1 and F2 respectively. Note the strong north-south area of frontogenesis, maximized mainly below 700 MB. A cross section through the frontogenesis (not shown) showed two strong low-level areas of frontogenesis.
A cross section was taken through the baroclinic zone to find areas of negative EPV, which might favor banded precipitation. This cross section showed negative EPV over Cape Cod (CHH) and aloft over NJ. Note the very positive area of EPV between these two regions. Based on this cross section, EPV in the layer from 900 to 700 MB is shown in plan view space. We hope to convert these data to VIS5D format to show isosurfaces of negative EPV.
The plan view EPV forecasts valid from 25/0600 UTC through 25/1500 UTC is shown in Figure EP1. Note the westward propagation of the area of negative EPV over southeastern New England with time and the second area along the coast. These two areas matched up well with radar imagery during the event. Hopefully, some good imagery from KOKX and KBOS will be available in the near future. The next set of three-hour forecasts is shown in EP2. Note the westward progression of the one area and general weakening with time. The forecasts were not perfect, but they suggested the potential for banded snowfall over southern New England slowly moving eastward toward eastern New York State.
The MM5 forecasts were run from the 12 UTC 23 Eta model. Similar to the Eta, the MM5 forecast a much deeper cyclone with a more westward track than observed. The 36 and 48-h forecasts of mean-sea-level pressure are shown in Figs MM5-a and MM55-b. The surface pressure was forecast to be 980-982 at the low center by 25/1200 UTC. Based on this low track, the MM5 forecast significant precipitation amounts along the immediate coast of southern New England. The maximum precipitation over a land area was forecast over eastern Long Island (not shown).
The MM5 relative vorticity near 500 MB valid at 25/1200 UTC is shown in Figure MM5-v. It is interesting to note the two-mesoscale relative vorticity bands extending out of the main upper level system. One such band extends north/northwestward over the western Atlantic over Long Island and into south central Connecticut. Another lobe was located over eastern Massachusetts, northwest of the main vorticity center.
Although no imagery is available yet, these two relative vorticity bands corresponded well with precipitation bands on radar. Secondly, these two bands were close to the location of the Eta negative EPV bands shown earlier.
All surface data were archived in real-time on AWIPS and analyzed by John LaCorte. After manual analysis, these charts were scanned in. These images are very large (in excess of 300 KB) so click carefully! The surface plots from 25/0900, 25/1200, 25/1500, 25/1800, 26/0000, and 26/0300 are provided. In addition too standard isobars at 2-MB increments, isallobars are also shown.
These charts show that the central pressure of the surface low was around 999 MB at 25/1200 UTC. The low continued to deepen reaching approximately 990 MB around 26/0000 UTC. The earlier Eta and the single MM5 forecasts had errors greater than 10 MB in at this time period. However, the low continued to deepen, reaching 990 MB, suggesting part of this error may have been due to timing errors in the models relative to the actual time of rapid deepening.
The snowfall map for the event is shown in Figure S1. These data show that the heavy snowfall was confined to southeastern New England and eastern Long Island.
A radar image from the KBOS radar from 25/0716 UTC shows the snowbands moving over coastal sections of southern New England. The radar was in clear air mode at this time. Several distinct snowbands can be seen. Note the western band over Connecticut. A later image from 25/2027 UTC shows the snowbands over the region, including the second, weaker band to the west. These bands lined up nicely with Eta model forecasts of negative EPV. These two primary bands persisted through the evening hours as shown in the 26/0000 UTC radar image.
Conclusions:
Model trends in both the AVN and Eta were properly forecast. The Eta showed continued forecasts for a weaker surface cyclone with a more easterly track. Similarly, the AVN slowly forecast a slightly more northward and westerly track. Like the Eta, the MM5 forecast too deep a surface cyclone too close to the coast.
QPF trends looked good, except once model had the mesoscale detail, the model produced heavy precipitation well to the west of the surface cyclone. It underforecast the precipitation, and it appeared that the model forecast too much (will we ever know) precipitation near the surface cyclone and it attendant fronts. The earlier runs, from Time-48 hours forecast too much precipitation over Long Island. This was true of both the Eta and MM5 forecasts.
EPV and FGEN showed the potential for banded snowfall west of the surface cyclone. This was true in the Eta model forecasts, which showed strong areas of ascent and favorable conditions for mesoscale precipitation bands over southern New England and Long Island at least 36 hours before the arrival of the heavy precipitation. As this signal increased, the model reversed earlier trends and began to move precipitation to the west, despite the more accurate and easterly surface cyclone track.
Similar to the Eta, the MM5 showed signals, which favored mesoscale precipitation bands up to 48 hours in advance. How the model captured these features, despite the cyclone forecast errors is an interesting forecast problem.
With strong upper level cut-off lows, the old rules of thumb about where heavy snowfalls relative to the surface cyclone may now have an explanation.
This was a difficult storm from several aspects, which make it a storm worth more study to improve our forecast abilities. It is interesting how the Eta initially reduced the QPF as it moved the surface cyclone eastward. Than, despite this correct trend, it reversed the trend in the QPF, producing locally significant precipitation amounts over southern New England, well west of the surface cyclone.
