WESTERN REGION TECHNICAL ATTACHMENT
NO. 97-18
June 3, 1997
RESULTS OF THE WESTERN REGION EVALUATION
OF THE ETA-10 MODEL
Mike Staudenmaier, Jr. - WRH SSD/NWSFO Salt Lake City, UT
and
Jon Mittelstadt - WRH SSD, Salt Lake City, UT
Introduction
During the period of January through April 1997, Western Region National Weather
Service (WR NWS) participated in a formal evaluation of the experimental Eta-10 model.
This model is a higher resolution version of the currently operational Eta-29 model, with
similar physics and parameterizations. More information regarding these physics and
parameterizations can be found in Staudenmaier, 1996. This evaluation was an important
step in preparing for high resolution numerical modeling in the National Weather Service.
The National Centers for Environmental Prediction (NCEP) plans to make higher resolution
versions of the Eta model operational in the future. In order for the Eta model to run at
higher resolution over a national domain, a next generation supercomputer is required.
This Technical Attachment summarizes the results which were discovered by the field and
relayed to the model developers at the NCEP.
Basis Of The Evaluation
In December of 1996, an evaluation form was designed and sent out to Western Region
field offices, in preparation of the evaluation which would begin in early January. The form
was designed so that the forecast problem of the day' could be investigated to see if the
Eta-10 model added any value to the forecast process. The evaluator would write down
the investigated phenomena, whether he/she thought the Eta-10 should capture the event
and why, and then verified what impact the model output had on the overall forecast of the
event. Through the four months of the evaluation, 104 distinct phenomena were
investigated by Western Region.
Summary Of The Results
As with any evaluation, results were mixed, with some evaluations showing significant skill
with the Eta-10 model, and other results showing potential problems with the model. This
section will discuss the main points of the evaluations, broken down into positive and
negative results. The Eta-10 model was compared only against other operational
numerical guidance, not against any locally run mesoscale models.
Positive Results
The Eta-10 model appears to do better than the operational numerical models with arctic
air masses, delineating which valleys they will move into and the timing of their movement.
This is likely due to the better representation of topography and the increased vertical
resolution. For example, Fig. 1a shows an arctic front that was correctly blocked from
entering western Montana in the Eta-10 simulation. This blockage did not occur in the Eta-29 (Fig. 1b), nor in any other NCEP model.
Rainfall is better delineated, especially in areas with orography. Rain shadows now occur
due to downslope flow over single mountains, as opposed to only occurring in the lee of
entire mountain ranges. Bull's-eyes of precipitation are much better placed on mountain
slopes when compared with reality. Figure 2 compares Eta-10 three-hour accumulated
precipitation with observed six-hour accumulated precipitation valid over northern
California. Note that many of the Eta-10 maxima line up with locations of heaviest
observed precipitation.
The Eta-10 model generally does a better job delineating winds in larger valleys and near
the coastline, where the better resolved topography plays an important role. Some of this
improvement near the coastline may be due to the higher vertical resolution of the model
near sea level.
The Eta-10 model does well with moisture which tends to get trapped on mountain slopes
by mountain ranges. This shows up best in the explicit cloud scheme which is
incorporated in the model. The Eta models are the only models with an explicit cloud
scheme, so only direct comparisons between differing resolutions of the Eta models are
possible. However, based on relative humidity from other numerical models, it can be
seen that the Eta-10 is the only model to consistently trap moisture against mountain
ranges. With the explicit cloud scheme in the Eta model, forecasters have, for the first
time, the opportunity to see model-derived clouds, as opposed to inferring them from
planer views of relative humidity. Figure 3 shows a low-level cloud fraction field derived
from the explicit Eta-10 forecast of cloud water and ice. Low clouds are trapped against
the model terrain in northern Arizona.
The 2-meter temperatures in the Eta-10 model are generally more accurate than at 29 km
resolution. This is especially true near coastal areas, where the higher vertical resolution
has much more impact on the derivation of 2-meter fields. However, the cold bias over the
Intermountain Region, which was noted in the Eta-29 model, continues in the Eta-10
model.
The Eta-10 model often shows precipitation in much more realistic looking structures, like
banding, than is seen in other operational numerical models. This may be due to the
higher horizontal/vertical resolutions, and a better representation of topographical
influences on the synoptic pattern.
The Eta-10 model develops accurate looking lee-side pressure troughs (and windward
pressure ridges) along mountain ranges, as demonstrated by Fig.4. Mesoscale detail
can be seen in the ridge-trough patterns located over the Coastal Range Mountains and
the Sierra Nevada Mountains (terrain not shown). These types of features have a positive
impact on low-level wind fields, and the location of orographically forced precipitation
maxima and minima. Additionally, some field stations have noted that pressure gradients
appear to be much more realistic, and are now useful for forecasting such phenomena as
gap flow through the Columbia River Basin, Sundowner Winds, and Santa Ana events.
The Eta-10 model has much more detail in convectively significant fields, like low-level
helicity, and CAPE. The convective parameterization, with the new changes made a few
months ago, seems to be working better, with convection now occurring over higher
topography. This change will be made in the other versions of the Eta model in the future.
Additionally, the Eta-10 model is showing forecasters many more mesoscale' details than
previously possible. Some examples are:
1) The Catalina Eddy off the coast of Los Angeles (Fig.5)
2) The Snake River Valley Convergence Zone
3) Mountain-valley wind regimes (upslope/downslope)
4) Coastal stratus near Los Angeles
Overall, the Eta-10 model appears to have made a positive impact on the overall forecast
process at many offices in the Western Region. This was especially true after the model,
which originally ran in the 0900 UTC time frame, was moved to a more convenient 0300
UTC time frame. This allowed output to reach the field forecasters before the early
morning forecast package was complete, giving them an additional look at mesoscale
detail for day 1. However, based on the evaluations, some problems with the Eta-10
model also appeared. These will be discussed in the following section.
Negative Results
It appears that precipitation falls too far down on mountain slopes, or over mountain
valleys, as opposed to reality. In some extreme cases, no precipitation falls on or near the
mountain crests, and holes' appear in the 3-hour precipitation output fields where
significant topography exists. In Fig. 6b, several precipitation maxima are evident along
mountain slopes adjacent to precipitation minima over mountain peaks, .e.g. over the
Uintah Mountains of northeast Utah.
The Eta-10, along with the Eta model in general, appears to be too aggressive in
producing precipitation in warm-air advection cases. This appears to be even more of a
problem with the higher resolution and the better representation of topography in the Eta-10 model.
The low-level winds are typically too light, especially inland. This could be due to the fact
that as one goes higher in the model atmosphere, the vertical layers thicken. The
calculation of the low-level winds, especially the 10-meter winds, is dependent on the
thickness of the model layer at that point.
Although the 2-meter temperatures are generally more accurate in the Eta-10 model, they
are still too cold on average, especially inland. Again, this appears to be due to the thicker
model layers which occur as one goes higher in the atmosphere, and the way that 2-meter
variables are defined.
Due to the way the cloud model is initialized (with zero values in the Eta-29), the clouds
can be inconsistent with resolution of the Eta-10, allowing for a major shift in the look' of
the field by three hours into the integration. This is probably true with many fields in the
model, but is most apparent in the cloud field, as entire shields of clouds may disappear
as the model adjusts to the new topography and vertical motions.
There seems to be some discrepancy between the cloud model and the location of
precipitation. On some occasions, moderate to heavy precipitation falls with cloud model
percentages below 50%. An example is evident in Fig. 6 along the Utah/Nevada border.
Also, since the cloud model is not tied to the convective parameterization, you can have
heavy convective precipitation, with little or no clouds apparent.
There appears to be a problem with momentum transfer along mountain slopes in the
model. High winds associated with mountain waves do not occur in the model. Figure 7
compares Eta-10 and MM5 simulations for a mountain wave event west of the Wasatch
Front in northern Utah. The Eta-10 model produces strong subsidence, yet little
momentum is transferred downward, and theta surfaces remain flat. MM5 simulations of
similar resolution clearly developed classic mountain waves in the two cases which were
rerun (for example, Fig. 7b).
Land-sea interactions around the Great Salt Lake are not modeled correctly by the model.
Even though a tight gradient of temperature can develop around the Great Salt Lake, the
wind field does not respond to it. The expected convergence at night and divergence
during the daytime, that is seen in reality, is not evident in the model.
Conclusions
Based on four months of input from Western Region, through a formal evaluation period
which ran from January through April 1997, it was found that the Eta-10 model in general
did add detail to the forecast process. Some problems were discovered with the Eta-10
model, especially in boundary layer processes. Many of these problems are likely
occurring in the other courser resolutions of the Eta model, but have only been discovered
now that higher resolution makes looking at the boundary layer fields a viable option.
Clearly, even with these problems, the Eta-10 model has lived up to its expectation to
produce superior forecasts with sub-synoptic detail. Perhaps it doesn't capture all the
mesoscale events that one could expect from a model of its resolution, but it still
represents a huge step forward in numerical modeling--to generate operational mesoscale
forecasts across a large domain on a daily basis. This experiment has demonstrated the
ability of the NWS to produce quality numerical guidance at higher resolutions. These
results underscore the need for a next generation supercomputer that will allow higher
resolution model runs to become operational across the United States.
Acknowledgments
The authors thank the Mesoscale Modeling Branch of the EMC, in particular Geoff DiMego
and Tom Black, for providing valuable, practical information about the Eta model design
throughout the evaluation period.
References
Staudenmaier, M.J., 1996: A description of the Meso Eta model. WR-Technical
Attachment 96-06.