University of Washington Ensemble Systems for Probabilistic Analysis
University of Washington Ensemble Systems for Probabilistic Analysis and Forecasting Cliff Mass, Atmospheric Sciences University of Washington
UW Mesoscale Ensemble Systems • An attempt to create end-to-end mesoscale probabilistic guidance. • Two major ensemble systems exploring different approaches to generating initial conditions. • Based on high-resolution (12 -km or 4 -km grid spacing) ensembles.
Two UW Ensemble Systems • UWME: eight members with initializations and boundary conditions from major operational NWP systems. 72 hr, 36 and 12 -km grid spacing. WRF model. • UW En. KF: 60 members, 36 and 4 -km grid spacing. 3 -hr cycling, with 24 -h forecasts once a day. WRF model and DART infrastructure.
UWME Parent Modeling Systems
UWME • Originally MM 5 based, but last year switched to WRF with improved physics options. • Initially applied physics diversity, but not using that now due to computer limitations. • Kain-Fritsch CU, YSU PBL, Thompson microphysics, RRTM LW, Dudhia SW. Noah LSM
Bayesian Model Averaging • Assumes a Gaussian (or other) PDF for each ensemble member. • Assumes the variance of each member is the same (in current version). • Includes a simple bias correction for each member. • Weights each member by its performance during a training period (we are using 25 days) • Adds the pdfs from each member to get a total pdf.
Application of BMA-Max 2 -m Temperature (all stations in 12 km domain) Improves reliability and sharpness
The BMA Site
The Next Challenge: Making Probabilistic Forecasts Accessible to Users • Creating good probabilistic information is only half the challenge—and probably the easier half.
PROBCAST
UW En. KF System • To build a larger, high-resolution ensemble system directed towards data assimilation and short-term forecasting. • Both probabilistic analyses and forecasts. • Originally based on the Torn-Hakim infrastructure, but now uses the NCAR DART system.
UW En. KF System • • • 36 km and 4 km domains Now a 3 -hr analysis cycle. 60 members using the WRF model. Runs out 24 -h once a day. Completely operational and reliable
Mesoscale Covariances 12 Z January 24, 2004 Camano Island Radar |V 950|-qr covariance
36 -km
UW En. KF • Assimilates a variety of data types: sat winds, surface obs, acars, radiosondes. • Tests with radars completed (winds) and will make use of current radars and the new coastal radar. • Major innovations in data selection and bias removal. • Moving to a one-hour analysis cycle. Add physics diversity. • Research needed during next year on vertical localization, improved bias removal, and other issues • Extensive verification, which will be expanded.
UW As a Regional Mesoscale Testbed for Probabilistic Prediction • A fairly large interdisciplinary effort, previously supported by large MURI project, and recently ending AF JEFS and NWS CSTAR funding. • Lack of support threatens the continued viability of our efforts. • Need for better pathways of research from groups such as ours to operations.
The End
- Slides: 27