84th AMS Annual Meeting

Monday, 12 January 2004: 4:45 PM
Verification of the FSL ensemble of mesoscale models used for a winter weather application
Room 619/620
Paul Schultz, NOAA/ERL/FSL, Boulder, CO
Poster PDF (87.0 kB)
The LAPS group at FSL has built an ensemble of mesoscale models that runs in real time in support of field projects and demonstrations. One of these projects is sponsored by the Federal Highways Administration and is focused on winter weather. This paper discusses the ensemble design and how the emphasis on winter weather influenced the design as it was deployed during a demonstration during late winter 2003. Verification statistics will be presented, and it will be shown how the results of the verification led to changes in the ensemble configuration for next winters demonstration.

Last winter, the ensemble consisted of six members: three mesoscale models (MM5, RAMS, and WRF) with two lateral bounds sources (Eta and AVN, as provided by NCEP). The models were run out to 27 h to provide a 24-h forecast service. The grid configuration is the same for all models, and for this test the grid increment was 12 km, with no convective parameterizations because of the focus on winter weather. The execution schedule was driven by the update frequency of the NCEP models; thus, all six members were run four times per day, upon receipt of the NCEP model grids.

Two factors led to a change in the execution plan for next winters demonstration. First, the client has concluded that forecast quality in the earlier stages (3-12 h) would be more valuable than the previously emphasized 12-24 h range. And second, the verification activity has revealed that model pairs that differed only in their lateral bounds showed very little difference. In other words, variety in lateral boundaries did not provide much dispersion in the forecasts (an unexpected result).

In light of these considerations, the ensemble will be constructed differently for next winters demonstration. We are currently in the process of determining exactly how the ensemble will be configured, but we will likely run fewer members, initialized more often, with time-lagged ensemb le techniques applied. The final configuration will be running by October, and real-time demonstrations will be given at the conference.

Supplementary URL: