Close Window

University of Wisconsin-Madison: "Operational prediction of meso-ß-scale weather in the vicinity of the western Great Lakes using the UW-NWS"

Final Report



During the third year of this project the UW-NMS (University of Wisconsin Nonhydrostatic Modeling System) was run twice daily at the University of Wisconsin-Madison. James Riffel, a graduate student at the University of Wisconsin-Madison, continued to work as liaison between the UW-NMS group and the Sullivan office. Riffel performed research to evaluate and compare the statistical performance of the operational UW-NMS to the statistical performance of the ETA. Riffel completed his Master's degree in December of 2001. Over the period of the third year several model improvements have been under development.

Operational UW-NMS Product

The UW-NMS was run operationally two times per day. The operations are automated as much as is possible. A rather complex chain of automated procedures coordinated between five computers at the UW, the suite of computers at the National Center for Environmental Prediction (NCEP) in Washington DC, Central Region Computers in Kansas City and finally the Milwaukee/Sullivan WSFO make this product possible.

A 4 processor SGI Origin-2000 that runs the model anchors the UW computers. This is an $110K platform that has been funded by other research projects. We also employ an SGI INDIGO workstation that we use to acquire the ETA model grids from NCEP and place them on the Origin. We use another IBM RISC workstation to scrub graphics files from the SGI as it runs and make plots for the model Web Page. Finally, we use a PC given to us by the Central Office of the USWS as a machine to scrub the UW-NMS GRIB files from the Origin as the model runs. The scrubber sends the GRIB (AWIPS compatible) files via the Internet to the frame relay network at the USWS Central Region Office where the Milwaukee/Sullivan USWSFO Office can access the files. The PC also operates a Web Page that places the model products on the Web for alternate USWSFO or public access. This PC was returned to the Milwaukee/Sullivan office during the summer of 2001. Since that time the web page has been kept up on a dec-alpha machine.

The operational system runs twice a day, initialized at 0000z and 1200z. The UW-NMS is run with two grids. The outer grid was run at a 60km resolution. The inner grid is typically run at 20km resolution. The inner grid was normally centered over Minneapolis, MN in order to have fine resolution over Wisconsin and the surrounding states, especially those up stream of Wisconsin. The inner grid was occasionally moved to accommodate special weather situations. During the 2001-2002 winter, the inner grid was expanded to encompass the entire Great Lakes region, in order to provide high-resolution lake-effect snow forecasts for NWS offices in the Great Lakes.

The PI spends an average of 4 hours a week keeping the operational model working. Most of the instances when the UW-NMS did not run were the result of initialization files from NCEP not being available in a timely enough fashion for the operational run. Riffel spent several hours a week archiving the data from the model runs, and statistically evaluating the performance of the operational UW-NMS.

Model Development

In order to have a quality operational product, continued operational development is necessary. The UW-NMS is supported as a research tool for mesoscale weather studies, but operational applications expose the model to new and exciting challenges. One of the focuses of continued model development is the vegetation scheme. A new vegetation scheme based on the ALEX model of Professor Norman, is ready for operational implementation. This new vegetation scheme will better capture the local production of humidity from the Corn Belt. This new vegetation component has more soil layers, and thus hopefully better surface heat fluxes. The new vegetation scheme will also allow for more accurate canopy heights to be used in computing boundary layer turbulence.

There are several other areas of the UW-NMS under development. A non-local turbulence scheme is under development to improve the boundary layer scheme. This new scheme will replace standard K-theory, in that is will allow for counter-gradient mixing in the boundary layer. The UW-NMS has started to explicitly predict the height of the boundary layer top as well. Work done on the UW-NMS for the United States Army has resulted in many new and interesting forecasted variables. The VIS5D files for the operational runs now provide variables for pilots such as areas where aircraft icing, and clear air turbulence are likely. The microphysics and cumulus parameterization are areas that remain under continual development.

Model Verification

Jim Riffel evaluated model verification statistics. Riffel reported the verification results to Sullivan through personal contact as well as the UW-NMS web page. Riffel performed a yearlong statistical comparison between the UW-NMS and the ETA forecast model. The 24 and 36-hour 850mb temperatures, as well as daily precipitation amounts and coverage, were evaluated. The xsau5 verification program, developed by Bob Aune of the Space Science and Engineering Center at the University of Wisconsin-Madison, was used for completing the model verifications of the year 2000.

For the 850mb temperature verifications the root mean square (RMS) error and temperature bias were explored. The RMS error estimates the common within-group standard deviation, and therefore a perfect forecast would produce a RMS error of zero. The UW-NMS 24-hour forecast RMS error for 850mb temperatures generally ranged from 1.25 to 2.25 degrees Celsius through the entire year. This was very similar to the RMS error of the ETA. The greatest errors were during the winter and early spring. The lowest errors were in the late summer and fall months. The temperature bias showed a tendency for the UW-NMS to be too cold. The 36-hour forecast showed similar verification results. The RMS error for the 36-hour forecast was slightly greater (1.5-2.5oC) than that of the 24-hour forecast, which is to be expected with the greater amount of uncertainty in a longer range forecast. Once again, the ETA and UW-NMS were comparable, as both models had a cold bias. The 36-hour RMS error was greatest in winter and early spring.

The Equitable Threat Scores (ETS) for 24-hour precipitation showed the ETA to have a better handle on precipitation over the entire United States, most likely due to the smaller grid size of the ETA during the analysis runs (20km versus the 60km of the UW-NMS). However, in the upper Midwest, where the UW-NMS inner grid is run, the UW-NMS had ETS values comparable to the ETA. The ETS values for the UW-NMS ranged from 0.2 to 0.5. The UW-NMS tended to slightly overestimate precipitation during the fall and winter months. During the spring and summer months the UW-NMS showed an underestimation of precipitation. The underestimation of precipitation in spring and summer is most likely due to the convective nature of precipitation events in spring and summer, and supports the need to continue development on the UW-NMS cumulus parameterization.

Interaction with Sullivan NWS office

The Sullivan office of the NWS used the NMS model forecasts on a routine basis for its daily forecasts. The results were somewhat mixed, ie., the NMS would seem to handle certain situations superior to the NCEP forecast products and fall short in other situations. The greatest benefit of having the ANMS available was the alternative look at various weather situations that it provided.

Problems did arise, however, having to do with the inherent unreliability of the NMS in many situations stemming from the basic weaknesses of a University run operation staffed with graduate students on the learning curve and based on a research model under constant development. We were unable to staff an operation 24/7 . Instead we could set up the model to run automatically, as long as a long list of things went as planned. Given the many opportunities for failure, the percentage of time that the model did run without a hitch was impressive. Nevertheless, the failure rate did seem to exceed acceptable levels for NMS operations in Sullivan and this frustrated NWS personnel and UW personnel alike.

Our attempts to build a lake water wave prediction system also met with frustration. Since we were not performing basic research, we relied on a Masters student to implement this system. The student was asked to perform the verification studies for this project and to set up this new system. Because it was implementing an existing system, there was little interest in the project from more qualified students who are oriented more toward basic research endeavors. As a result, the programming task exceeded the ability of the graduate student personnel willing to spend time on the project. In retrospect, this project should have been given to a professional programmer.


Graduate Student Interaction

Jim Riffel met with Milwaukee Sullivan Office personnel on several occasions to describe verification results. In addition to direct communication, Riffel provided verification results on the UW-NMS web page on a monthly basis.

Model Updates

Model development changes are written up and communicated to the Sullivan Office. Once a year, UW personnel provide a presentation on the model to the Sullivan Office.

Interaction with NCEP

The UW-NMS is the only operational nonhydrostatic model using a stepping coordinate system. There are many problems with the formulation of the ETA coordinate, in the NCEP ETA model that have prevented them from being able to achieve the successes demonstrated with the UW-NMS. As a result, Tripoli has been consulting with Fedor Messinger and Geof DiAmigo of NCEP on solutions to the ETA model problems. There is no formal support for this at this time, but this collaboration is possible at least in part because of the operational product that is supported by COMET.

Private Industry

Over the past 6-8 years, our operational work with COMET has spawned a large and growing private industry in local modeling. Beginning with former graduate students bringing the NMS model to Weather Central Inc. over 5 years ago, the NMS is now used as the basis for animated television forecasts in over 150 television markets across the nation. In the past 5 years, NMS has been repackaged into a Windows product run locally at over 50 television stations on PCs. Literally, tens of millions of Americans see the NMS model prediction on their televisions each day as part of their evening news. Television stations even routinely spawn fine nested grids to follow interesting phenomena such as hurricanes, in real time and present the animated forecasts to the public, often outforecasting NCEP products themselves. In addition, two new companies, and, spun off from weather Central are preparing custom NMS forecasts for individuals and industry interests. In this sense, the COMET initiative has spawned a revolution in the way forecasts are delivered to the public and the role which government forecast agencies play.


There are numerous benefits to the university and the community as a result of this project. These include: