Transcript Document

Analysis of High-Resolution WRF Simulations During
A Severe Weather Event
Jason A. Otkin*
Cooperative Institute for Meteorological Satellite Studies, University of Wisconsin–Madison
3. RESULTS
1. INTRODUCTION
Cloud microphysical data from several high-resolution simulations of a severe weather event are
used to examine the general characteristics of the microphysical schemes currently implemented in
the WRF model. The inclusion of additional ice species in more sophisticated schemes, along with
variations in how certain microphysical processes are parameterized, may lead to substantial
differences in the simulated cloud field. The purpose of this study is not to evaluate the accuracy of
a given microphysical scheme but rather to simply compare the model-generated cloud structures.
The severe weather outbreak examined here occurred over the Northern Plains during the
evening of 24 June 2003. Over 100 tornadoes were reported across the region, including the
devastating F4 tornado that completely destroyed the town of Manchester, SD. This event was
characterized by the development of numerous supercell thunderstorms within a very moist and
unstable airmass extending from central Nebraska northeastward into central Minnesota. Stratiform
clouds with embedded convection were also present to the northwest of this region. The complex
cloud structure associated with this event represents an ideal opportunity to examine the behavior
of the microphysical schemes during a severe weather event.
Total model precipitation from 12
UTC 24 June to 06 UTC 25 June
for each simulation.
Visible satellite image valid at 0015
UTC on 25 June 2003.
WSR-88D radar summary valid at
0015 UTC on 25 June 2003.
2. MODEL CONFIGURATION
Simulated atmospheric fields were generated using version 2.0.3.1 of the WRF model. Each
model simulation was initialized at 1200 UTC 23 June 2003 using 1° GFS data and then run for 42
hours on a single 300 x 300 grid point domain with 4 km horizontal grid spacing and 50 vertical
levels. The geographical region covered by this domain is shown below.
Microphysical schemes employed by this study include the relatively complex WRF SingleMoment 6-class graupel (WSM6) and Purdue Lin schemes, which include prognostic variables for
cloud water, rain water, ice, snow and graupel; the less complex WRF Single-Moment 5-class
(WSM5) scheme which is similar to the WSM6 scheme but does not include graupel; and the
relatively simple Ferrier scheme that only includes a prognostic variable for cloud water. Aside from
the different microphysical schemes, each simulation employed identical model configurations:
• YSU planetary boundary layer
• RRTM/Dudhia radiation
• Explicit cumulus convection
• NOAH land surface model
Geographical
region
covered by the single
300x300 grid point domain
used for the WRF model
simulations.
Horizontal
grid spacing for this
domain was 4 km.
Prior studies have shown that a model’s
microphysical scheme can strongly influence
the magnitude of surface precipitation. The
total model precipitation for each simulation
clearly illustrates this sensitivity. Although each
simulation reasonably predicted the region of
maximum precipitation over eastern South
Dakota and southern Minnesota, the total
rainfall varied widely with the cloud-water only
Ferrier scheme producing much less rainfall
than the schemes that include ice and snow
processes. The similar rainfall distributions in
the WSM6 and Lin simulations also indicates
that graupel strongly modulates the location of
surface precipitation. The presence of heavier
rainfall concentrated over a narrower region is
likely due to the faster fall speed of graupel,
which tends to produce a more compact
precipitation core and also reduces the
evaporation loss due to a shorter residence
time.
4. GRAUPEL SENSITIVITY RESULTS
A major weakness of many single-moment microphysical schemes is the need to specify the
density (ρ) and slope intercept (n) parameters used to determine the size distribution of a given
hydrometeor species. In this section, the graupel parameters in the WSM6 scheme (ρ5n6) are
modified to represent values characteristic of large hail (ρ9n3) and small graupel (ρ4n8) in order to
examine the influence that these values have on the domain-averaged vertical profiles.
Domain-averaged vertical profiles of graupel
mixing ratio are shown to the left. It is clear that the
graupel density and slope intercept parameters
strongly influence the shape of the graupel profile.
For instance, the abundance of small graupel
characterized by a relatively slow average fall
speed in the ρ4n8 simulation increases the amount
of graupel in the upper troposphere since the
longer residence time and greater surface area
tends to enhance graupel growth. The much faster
fall speed of large hail during the ρ9n3 simulation,
however, efficiently removes graupel from the
upper troposphere. The faster fall speed also
preserves more low-level graupel since the shorter
residence time in the relatively warm lower
troposphere results in less graupel loss due to
melting and evaporation.
The domain-averaged ice mixing ratio
profiles for each simulation are shown to the
right. It is evident that both WSM schemes
generate a much greater amount of ice mass
than the Lin scheme.
The presence of
substantially greater ice in the middle
troposphere could be due to the ice initiation
formula, which tends to produce more ice at
warmer temperatures than the formula
employed by the Lin scheme. When compared
to the WSM profiles, the rapid decrease of ice
mass below 325 hPa in the Lin profile may be
due to the ice to snow autoconversion formula,
in which nearly all existing ice crystals are
converted to snow for temperatures warmer
than -27º C. The WSM schemes use a formula
that is not temperature dependent and,
therefore, tend to preserve more ice at warmer
temperatures.
Domain-averaged vertical profiles of ice and
snow mixing ratio are shown to the left. Although
the small graupel case is characterized by nearly
twice as much ice in the upper troposphere as the
large hail case, it appears that the ice field is
relatively insensitive to changes in the graupel
parameters. It is readily apparent, however, that
much larger sensitivities exist in the snow field. In
fact, the large hail case is characterized by nearly
an order of magnitude more snow than the small
graupel case. Most likely, the enhanced snow
content is due to the greatly diminished graupel
mixing ratio in the upper-troposphere, which could
serve to increase both the snow growth rate and
the magnitude of the ice-to-snow autoconversion
since more cloud water would be available for ice
and snow growth processes.
The domain-averaged vertical profiles of
snow and graupel mixing ratio are shown to the
left. Although the WSM6 scheme generates
more snow and graupel than the Lin scheme, it
is clear that the two graupel schemes behave in
a similar manner. The absence of graupel in
the WSM5 scheme, however, significantly
impacts the snow distribution, which is
characterized by 5 to 10 times more snow than
the Lin and WSM6 profiles. The maximum
snow mixing ratio for the WSM5 profile is
located between the maximum snow and
graupel mixing ratios of the two graupel
schemes. In essence, the snow processes in
the WSM5 scheme include both snow and
graupel effects that would be included in a more
sophisticated graupel scheme. The greatly
enhanced snow mass could be due to a slower
average fall speed, which results in more time
for growth processes to occur, or to enhanced
accretion of cloud and rain water by snow in the
absence of graupel.
Domain-averaged vertical profiles of cloud and
rain water mixing ratio are shown to the left. It is
evident that the control and small graupel
simulations exhibit similar profiles. Melting of the
smaller
graupel particles in each of these
simulations produced smaller rain droplets that
were relatively easy to evaporate, which resulted in
decreasing rain water with decreasing height in the
lower troposphere. A vastly different rain water
profile, characterized by increasing rain with
decreasing height, occurred during the large hail
simulation. Slower melting of the relatively large
graupel particles likely caused the increasing rain
water mixing ratio. It also appears that the lack of
upper-level graupel during the large hail simulation
may have resulted in enhanced cloud water due to
diminished cloud water scavenging.
ACKNOWLEDGEMENTS
This work was sponsored by the Office of Naval Research under MURI grant N00015-01-1-0850 and by
NOAA under GOES-R grant NA07EC0676.
*Contact: Jason A. Otkin • Address: 1225 W. Dayton Street • Madison, WI 53706 • Phone: 608/265-2476 • Email: [email protected]