University of Washington Mesoscale Ensemble (UWME) Information

General Information: A group in the Department of Atmospheric Sciences at the University of Washington is running a mesoscale, short-range (0-2 day forecasts) ensemble with the fifth-generation Penn State/NCAR mesoscale model (MM5) twice per day (at the 0000 UTC and 1200 UTC cycles) in order to produce skillful probabilistic mesoscale meteorological forecasts for the Pacific Northwest. This effort, supported by a consortium of local, state, and federal agencies, as well as the Department of Defense Multi-disciplinary Univerisity Research Initiative (DoD MURI), has several goals:

• To test the applicability and effectiveness of multianalysis short-range ensemble forecasting (SREF; 0-48 hours lead time) at mesoscale ( O(10km) ) grid resolutions.
• To determine whether mesoscale SREFs have more or less value in regions of large orography and weak convection (northwestern U.S.) than in regions with mostly flat topography and more frequent, deep convection.
• To determine whether the mesoscale SREF mean possesses greater skill than any of its individual members.
• To determine whether the mesoscale SREF spread can be used as a predictor of both deterministic and probabilistic forecast error.
• To determine whether the mesoscale SREF can be used to produce skillful probabilistic forecast guidance for surface weather parameters, especially for temperature, winds, and precipitation.
• To determine the relative skill of the mesoscale SREF members in real-time and during the recent past to potentially derive appropriate weightings, that when applied to each ensemble member, would result in more accurate deterministic and probabilistic weather forecasts.
• To determine whether model error plays a significant role in the prediction of surface weather parameters for a region subject to large topographic influences.
• To determine whether incorporating model uncertainty into a mesoscale SREF is worthwhile, even for a region where the mesoscale forecast uncertainty is often dominated by the synoptic-scale uncertainty.
• To determine whether the multianalysis ensemble size can be expanded using linear combinations of the synoptic-scale ICs in order to improve the chance of encompassing truth and to provide more skillful probabilistic forecast guidance.

Model Configuration: The MM5 mesoscale ensemble forecasts currently feature an outer grid (151x127) of 36 km horizontal grid spacing that covers much of western North America and the northeastern Pacific and a nested grid (103x100) of 12 km grid spacing that covers Oregon, Washington, and southern British Columbia. The model utilizes 33 vertical sigma levels, and is run in non-hydrostatic mode in order to limit pressure gradient force errors in the complex terrain. An upper-radiative boundary condition is used to allow gravity waves to radiate through the model top without being reflected. The standard sub-grid scale parameterizations used for the non-physically perturbed members (ie- only in the UWME system) include:

• MRF planetary boundary layer scheme as implemented in the NCEP MRF model
  (see Hong and Pan, 1996)
• Explicit moisture scheme including simple ice physics but no mixed phase processes (see Dudhia, 1993)
• Kain-Fritsch cumulus parameterization (see Kain and Fritsch, 1993)

Detailed terraind and land use information for each domain was derived from the 1-km U.S.G.S. digital database.

Initial Condition Selection Strategy: MULTIANALYSIS
Initial conditions and lateral boundary conditions for the MM5 mesoscale ensemble are currently generated by interpolation of separate global/synoptic-scale model analysis and forecast fields obtained from several operational weather prediction centers worldwide. A current list of these models includes:

• The National Center for Environmental Prediction (NCEP) Global Forecast System (GFS)
• The Canadian Meteorological Centre (CMC) Global Environmental Multiscale (GEM) model
• The National Center for Environmental Prediction (NCEP) Eta model
• The Bureau of Meteorology (Australia) Global AnalysiS and Prediction (GASP) model
• The Japan Meteorological Agency (JMA) Global Spectral Model (GSM)
• The Fleet Numerical Meteorology and Oceanography Center (FNMOC) Navy Operational Global Analysis and Prediciton System (NOGAPS)
• The Central Weather Bureau (CWB) (Taiwan) Global Forecast System
• The United Kingdom Met Office (UKMO) Unified Model (UM)

See the summary of members.

Hardware and Timing: Forecasts are computed on three computers:

1. A 8-node (16 AMD Athlon 2.4GHz processors) Linux PC Beowulf cluster
2. A 4-node (8 Intel Xeon 2.8GHz processors) Linux PC Beowulf cluster
3. A 16-node (32 AMD Athlon 1.8GHz processors) Linux PC Beowulf cluster

TIMING
Each 36/12 km 48-hour forecast finishes in a different amount of wallclock time depending on the PC cluster and the MM5 physics package used. For the fixed configuration used in UWME, each MM5 ensemble member takes approximately the same amount of wallclock time. Times (HH:MM) are summarized below:

1. 00:40 8-node Linux PC Beowulf cluster
2. 00:57 4-node Linux PC Beowulf cluster
3. 00:37 16-node Linux PC Beowulf cluster

All eight UWME members plus the centroid run (CENT) are usually completed by 1030/2230 UTC (0230/1430 PST) provided all initializations are in on time. Ensemble mean and spread calculations are made available at that time. UWME+ members lag only slightly behind the UWME members and are usually complete before the next cycle begins.

Future Improvements: The ensemble modeling system is constantly being improved, so expect changes and delays. Among the changes we hope to make in the future are:

• Adding graphics of near real-time performance and historical verification over extended periods to the UWME Verification & Performance page.
• Adding an array of post-processed forecast products to the website, including calibrated (via Bayesian Model Averaging) probability plots and data range products.