Published in the 12th Conference on
Numerical Weather Prediction, Phoenix, Arizona, American Meteorological
Society, 320-323; 11-16 January 1998.
NEAR OPERATIONAL SHORT-TERM FORECASTS FROM
TWO MESOSCALE MODELS
Ed Szoke1,2 John A. McGinley, Paul Schultz, and
John S. Snook
NOAA Forecast Systems Laboratory
Boulder, Colorado 80303
1Corresponding Author
address:
Ed Szoke, 325 Broadway, R/E/FS1, Boulder, CO 80303.
Email: szoke@fsl.noaa.gov
Atmosphere, Colorado State University, Fort Collins, CO.
1. INTRODUCTION
The
NOAA Forecast Systems Laboratory (FSL) has been testing two mesoscale
models in real-time for several years, with an initial goal of implementing
one of the models for real-time use on a trial basis at the
Denver
National Weather Service Weather Forecast Office (WFO). The plan is
to integrate the model within the FSL-developed WFO-Advanced workstation
being tested at the Denver WFO as the prototype for the
Advanced
Weather Interactive Processing System (AWIPS), for deployment as part
of the NWS modernization.
Within the possibilities of
the modernization is running local models at a WFO. Along these lines,
the
FSL Local Analysis and Prediction
System (LAPS, McGinley et al., 1991) was developed a number of years
ago. The analysis portion of LAPS has been successfully supporting operations
at the Denver WFO for several years, and more recently in a number of other
NWS offices. The prediction portion of LAPS would come from a mesoscale
model such as one of the models being tested at FSL, or possibly one that
is tied to a cooperative effort with a nearby university or research laboratory.
Details of how this might occur are discussed by Schultz at this conference
(Paper 14.3).
Two mesoscale models have
been used for testing; one a version of RAMS (CSU Regional Atmospheric
Modeling System) and the other of MM5 (Penn State/NCAR Mesoscale Model
Version 5). Similarities and differences exist between the models, as further
detailed by Snook et al. (1998; also at this conference, Paper 1.2) and
Cram and Snook (1996). Both have 10-km horizontal grid resolution covering
the area shown in Fig. 1, and 25 layers from the surface to approximately
100 mb (a terrain-following
z
system for RAMS,
p for MM5).
Other differences include explicit cloud physics for RAMS compared to a
combination of explicit microphysics and a cumulus parameterization scheme
for MM5, and different radiation schemes and treatment of the soil layer.
Both models were run for 12 h twice a day beginning at 0000 and 1200 UTC
and initialized for the area shown in Fig. 1 using LAPS, with boundary
conditions from the previous operational Eta model run.

Figure 1: Model domain and county outlines
on a topography image (feet), and METAR sites.
Both models were run on most
days at FSL, with output often available for display on the WFO-Advanced
workstation at FSL, though not at the same workstation in Denver. There
forecasters could look at selected fields located on the FSL
homepage at address http://www.fsl.noaa.gov/wthr/fsl-weather.html.
In an attempt to familiarize the forecasters with the performance of the
models, a number of cases from this past winter and summer were collected
in hardcopy form, categorized, and subjectively compared to observations
as well as to performance of other models when possible, and brought down
to the WFO.
Quantitative verifications of
RAMS and MM5 have been presented previously (Cram and Snook, 1996; Schultz
and Snook, 1996), with an updated version elsewhere in this volume (Snook
et al., 1998). The purpose of this paper is to present a qualitative look
at model performance, and to look at a few operational issues regarding
how such model output might be used most effectively.
2. OVERVIEW OF MODEL PERFORMANCE
Quite a variety of cases were
compiled (Table 1), especially during the 1996-97 winter, which was characterized
by a number of smaller scale events. These provided ideal scenarios to
test a mesoscale model and see how useful it might be for mesoscale events
that were often very difficult to forecast even in the short term (6 to
12 hours).
Table 1. Number and type of
events collected
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
While the overall number of
cases is not sufficient for making any definitive conclusions, there were
some trends of model behavior that were observed. One was for the Denver
Cyclone (or DCVZ, Szoke et al., 1984), a mesoscale flow feature that develops
from southeasterly low-level flow over the topography of the eastern plains
of northeastern Colorado (see Fig. 1). Its scale (a circulation of about
50 to 100 km in diameter, or a convergence zone of similar length) makes
it a good test for a mesoscale model. Both RAMS and MM5 were able to simulate
the zone with relative success, but had some difficulty in developing summertime
convection along the zone. The horizontal resolution of the MesoEta was
not quite fine enough to simulate the DCVZ. Another topographically forced
low-level flow feature of roughly similar scale, the Longmont (LGM) Anticyclone
(Wesley et al., 1995), was also simulated by the finer resolution models
and to some extent by the MesoEta (Black, 1994).
Both the models, as well as
the MesoEta, were able to simulate the positioning of a larger-scale dryline
type boundary that develops near the eastern Colorado border, but again
thunderstorm development (location, timing, etc.) along the boundary was
not as well forecast and also varied among the models. In general, from
tracking model performance over a fairly long period when forcing was weak
and diurnal upslope trends were the principal mechanism to force thunderstorm
development along the Front Range, the RAMS and MM5 models did a credible
job with the timing, and to a lesser extent, the location of the first
storms of the day. The resolution is not sufficient, however, to model
the interactions that might take place after a number of storms develop
(isolated supercells might be an exception to this).
For the type of winter events
that produced snowfall, the model performance was encouraging but also
somewhat mixed. Orographic snows, for most of our cases having moist west
or northwest flow, were handled fairly well by RAMS and MM5 in terms of
indicating the location of expected snowfall relative to the impinging
flow. The predicted amounts under these conditions were roughly similar
for each model, though there were some nonsystematic differences. When
compared to actual snow that fell the results were mixed, and we hypothesize
that the errors might lie in having the lateral boundary conditions supplied
from a 12 h old run of the operational Eta model.
A number of the cases with
northwesterly flow aloft and orographic mountain snows also had low level
upslope on the plains forcing snow that increased closer to the Front Range
foothills, as well as organized bands of snow well out onto the plains
in association with forcing from an upper level jet streak in the presence
of CSI (Convective Symmetric Instability). When the upslope was relatively
shallow and the overlying northwest flow fairly strong both MM5 and RAMS
underestimated the snowfall on the plains, sometimes missing the banding
entirely. This may be related to excessive sinking and drying downstream
of the higher terrain under such conditions with a vertical grid that is
still too coarse. On most occasions though the models did retain the cold
air on the plains, in contrast to the typical behavior of the Nested Grid
Model (NGM). Unfortunately, we were not able to save MesoEta forecasts,
but it is clearly superior to the NGM with low level cold air on the plains.
3. CASE STUDY: 20 FEBRUARY
1997
The case of 20 February 97
also involved CSI and banded distribution of snowfall, but occurred with
southwesterly flow aloft. A mesoscale band of heavy snow approximately
60 km wide and 100 km long fell with amounts ranging from 5 to 10 cm on
the plains close to the foothills to none farther east (2 cm fell in eastern
Denver), but up to 40 cm over the higher terrain west of Denver and Boulder.
Figure 2. Eta 500 mb heights (dm) with upper air and profiler plots
on IR image (white=coldest) for 2100 UTC.
As seen in Fig. 2, the southwesterly flow at midlevels was ahead of
a shortwave trough at 500 mb moving southeastward. Another trough with
more extensive precipitation, all just east of Colorado, was lifting northward
out of Texas. At the surface, high pressure built east across southern
Wyoming, sending a surge of northeast winds past Denver at 1500 UTC. The
band of snow developed before 1800 UTC from west of Denver northward past
Boulder, then expanded with time (Fig. 3).
Figure 3. Radar 1 km AGL
composite reflectivity image with county map background and METAR plots
for 2300 UTC 20 Feb 97.
Forecasts of accumulated precipitation and near-surface
winds (at a height of 150 m for RAMS and about 50 m for MM5) valid for
the same time as Fig. 3 are shown for RAMS (Fig. 4) and MM5 (Fig. 5). In
Fig. 6 are the surface winds, MSL pressure, and 3-h accumulated precipitation
from the 1500 UTC run of the MesoEta valid at 0000 UTC 21 Feb.
Figure 4. RAMS 11-h forecast of near
surface wind and total
precipitation (inches) valid for 2300 UTC 20 Feb 97.
Figure 5. MM5 11-h forecast of near surface
wind and total
precipitation (inches) valid for 2300 UTC 20 Feb 97.
Figure
6. MesoEta 9-h forecast of surface winds, MSL pressure, and 3-h precipitation
valid 0000 UTC 21 Feb 97, on county map background.
All three models show precipitation
over the foothills and nearby plains, but the MesoEta most closely captures
the character of the precipitation band, with it displaced only slightly
west of the actual band. RAMS shows a distinct maximum of precipitation
near the Front Range, but it is displaced too far south (south of Denver
in the forecast). MM5 shows precipitation farther north but weaker and
not so distinct with the precipitation along the Front Range, instead indicating
more light precipitation over the eastern plains. The anomalous maximum
near the eastern border appears to be a band associated with the upper-level
low that was over Texas (Fig. 2; this area of precipitation was in reality
farther east). The turning low level wind field, somewhat of a LGM Anticyclone,
is predicted best by RAMS and MM5. We suspect that better initialization
for the MesoEta (1200 UTC Eta vs. 0000 UTC Eta for RAMS and MM5) may in
part explain the poorer forecasts for this event. However, for another
snowfall event (not shown) that was related to a somewhat smaller scale
midlevel circulation moving east across Colorado, the MM5 and RAMS showed
much better predictions than the MesoEta, making the above reasoning uncertain.
4. SOME OPERATIONAL ISSUES
Briefly, our experiences thus
far indicate a number of issues. Though available to the WFO forecasters
on the Internet, the model output was not widely used because it was not
yet displayable on their WFO-Advanced workstation. Ease of output, ability
to display many fields of choice and overlay actual data, etc., are important
considerations when trying to bring a model to operational use. We are
experimenting with three-dimensional visualization as a means of at least
overviewing the immense output that is available from model grids. Thought
must be given to timing of the model run (when available to the forecaster
and the length of the forecast) relative to when the forecasters produce
their products, while at the same time being able to initialize with the
most recent larger-scale model output. Some initial tests with a version
of RAMS (ExpRAMS in Table 1) run out to 26 h but only once a day, starting
at 0300 UTC to utilize a more current 0000 UTC model run, and nested within
a larger grid having lower resolution, have been very encouraging along
these lines, as well as showing improved predictability, at least for a
limited number of cases.
5. ACKNOWLEDGEMENTS
Thanks to Tracy Smith and
Nita Fullerton for reviews.
6. REFERENCES
Black, T.L., 1994: The new
NMC mesoscale eta model: Description and forecast examples. Wea. Forecasting,
9, 265-278.
Cram, J.F., and J.S. Snook,
1996: A comparison of the real-time performance of two nonhydrostatic models
over the LAPS domain. Preprints, 11th Conf. on Numerical Weather
Prediction, Norfolk, Virginia, American Meteor. Society, 153-155.
McGinley, J.A., S.C. Albers,
and P.A. Stamus, 1991: Validation of a composite convective index as defined
by a real-time local analysis system. Wea. Forecasting, 6, 337-356.
Schultz, P., 1998: Status
and prospects for local data analysis and mesoscale modeling in AWIPS.
Preprints, 12th Conf. on Numerical Weather Prediction, Phoenix,
Arizona, American Meteorological Society, 324.
Schultz, P., and J.S. Snook,
1996: Objective verification of mesoscale model precipitation forecasts.
Preprints, 12th Conf. on Numerical Weather Prediction, Norfolk,
Virginia, American Meteorological Society, J124-J126.
Snook, J.S., P. Schultz,
and J.A. McGinley, 1998: Initialization improvements to a quasi-operational
local-domain mesoscale model. Preprints, 12th Conf. on Numerical
Weather Prediction, Phoenix, Arizona, American Meteorological Society,
5-7.
Szoke, E.J., M.L. Weisman,
J.M. Brown, F. Caracena and T.W. Schlatter, 1984: A sub-synoptic analysis
of the Denver tornadoes of 3 June 1981. Mon. Wea. Rev., 112, 790-808.
Wesley, D.A., R.M. Rasmussen,
and B.C. Bernstein, 1995: Snowfall associated with a terrain-generated
convergence zone during the Winter Icing and Storms Project. Mon. Wea.
Rev., 123, 2957-2977.