Slide1: Lessons learned during eight years of real-time mesoscale modeling at the University of Utah
Jim Steenburgh, Ken Hart, Will Cheng,
Daryl Onton, and Andy Siffert
NOAA Cooperative Institute for Regional Prediction
Department of Meteorology
University of Utah
Motivating questions: Motivating questions
Does decreasing grid spacing below 12 km produce measurable forecast improvements?
How can we produce better point-specific and gridded forecasts over complex terrain?
Is WRF any better than MM5?
Models employed: Models employed Olympic MM5 (Jan-Mar 2002) WRF (Jun-Aug 2003)
Does decreasing grid spacing help?: Does decreasing grid spacing help? For temperature: No, except at mountain sites
Why not for temperature?: Why not for temperature?
Persistent cold pool example: Persistent cold pool example
Nocturnal cold pool example: Nocturnal cold pool example Temp BE (°C) Temp MAE (°C)
What about wind?: What about wind?
Precipitation gains more substantial: Precipitation gains more substantial Bias Score Equitable Threat Score False Alarm Rate Probability of Detection
24-h Precipitation Bias Scores – 12 km: 24-h Precipitation Bias Scores – 12 km Underforecasted over Stansbury, Oquirrh and portions of Wasatch mountains
Overforecasted in mountain valleys east of Wasatch
Poor terrain representation led to strong association between bias score and model elevation bias
Overforecast (≥ 160%) Underforecast (≤ 80%)
24-h Precipitation Bias Scores – 4 km: 24-h Precipitation Bias Scores – 4 km Local minima along and immediately to the lee of major mountain crests, but generally more than 12 km
Local maxima along eastern (windward) bench
Better terrain representation mitigated strong correlation between bias score and model elevation bias Overforecast (≥ 160%) Underforecast (≤ 80%)
How can we do better?: How can we do better? Wasatch Front
Mountain Valley
Mountain How do statistical approaches (MOS) compare to high res guidance?
Major improvements for temperature and wind: Major improvements for temperature and wind
What about WRF?: What about WRF?
Must work on the land surface!: Must work on the land surface!
Lessons: Lessons Native model sensible weather forecasts are “not good”, even at high resolution
Persistent and nocturnal cold pools are a major Achilles heal for MM5 and WRF with “MM5 physics”
Eta not great either
Improvements possible with better land-surface model/initialization
Over fine-scale Intermountain orography, model skill does improve at grid spacing is decreased to 4 km
Impact of greatest for precipitation
Wind to a lesser degree
But, MOS more beneficial than resolution for T, RH, Wind
Recommendations: Recommendations The community must look beyond simply running local mesoscale modeling systems
Land surface model/initialization improvements are needed
More emphasis should be placed on developing integrated (numerical and statistical) predictive approaches
Use our heads rather than computer cycles
See next talk
A broad, directed effort should be undertaken to improve the simulation of stable boundary layers in WRF
Issues: Numerics (e.g., diffusion), initialization, parameterization, land surface, optimal vertical and horizontal resolution
Slide18: Hart et al. 2004: An evaluation of mesoscale
model based model output statistics (MOS) during
the 2002 Olympic and Paralympic Winter Games.
Wea. Forecasting, 19, 200-218.
Cheng and Steenburgh 2004: Evaluation of surface sensible weather forecasts by the WRF and Eta models over the western United States. Submitted to Wea. Forecasting.
Hart et al. 2004: Model forecast improvements with decreased horizontal grid spacing over fine-scale Intermountain orography during the 2002 Olympic Winter Games. Submitted to Wea. Forecasting.
Available at www.met.utah.edu/jimsteen/publications.html