1 / 25

Evaluation of Mesoscale Numerical Weather Prediction Models

Explore characteristics, limitations, and sensitivity test approaches of mesoscale models for weather prediction. Discuss verification metrics, validation, and Pacific Data Void challenges.

lisasims
Télécharger la présentation

Evaluation of Mesoscale Numerical Weather Prediction Models

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript

  1. Evaluation of Mesoscale Numerical Weather Prediction Models Roland Stull Professor, UBC Earth & Ocean Sciences Dept. Director: Geophysical Disaster Computational Fluid Dynamics Center rstull @ eos.ubc.ca 604-822-5901 NW-AIRQUEST Annual Meeting Portland, OR 6-7 Oct 2003 NW-AIRQUEST

  2. http://weather.eos.ubc.ca/wxfcst Topics • UBC. Who we are, & what models we run. • Mesoscale model characteristics & limitations • Sensitivity test approaches (& ensembles) • Verification metrics, validation, requirements • Credibility, accuracy, Pacific Data Void NW-AIRQUEST

  3. http://weather.eos.ubc.ca/wxfcst Who we are, and what models we run. • University of British Columbia (UBC) • Geophysical Disaster Computational Fluid Dynamics Center • Team: Director (Prof. Stull), 1 Sr. Scientist (Terry Clark), 6 PhD students,1 MS student, 4 undergrad assistants, 4 computer scientists, & 1 volunteer. • Hardware: 256 processor IBM linux cluster, 24 processor Beowulf cluster, 8 processor Itanium, 8 processor SGI Origin 2000, 4 processor Origin 300, numerous servers & workstations • Mission: pure & applied NWP research for coastal, mtn terrain • Real-time, daily NWP forecasts since 1976. NW-AIRQUEST

  4. Geophysical DisasterComputational Fluid Dynamics Centre Director: Prof. Roland Stull, CCM Using large computers to forecast weather-related disasters • snowfall -> avalanches• forest firestorms• rain -> floods• cyclones & windstorms• air-pollution episodes Okanagan Park fire - 20 Aug 03 photo by Steve Devries NW-AIRQUEST

  5. Real-time Forecasts Produced Daily,and tailored to the needs of industry. 256 processors NW-AIRQUEST

  6. The GeoDisaster CFD Centre also gathers real-time weather observations. We maintain a Real-Time, On-Line Database of Surface Weather Observations, calledEmergency Weather Net - BC We also develop new sensors,including theRocketsonde Buoy System 2302 weather stationsreporting NW-AIRQUEST

  7. http://weather.eos.ubc.ca/wxfcst • The Models • MC2 • Pro: Fast; allows higher resolution over larger domains • Con: Less accurate in steep topography; doesn’t have 2-way nesting; on-again/off-again support from EC • MM5: • Pro: NCAR community support; 2-way nesting; more accurate; large worldwide experience • Con: slower; can cover medium domains; will be phased out by NCAR • WRF: • Pro: NCAR/NCEP; designed for distributed/shared mem. platforms (fast); will replace MM5 • Con: in beta-test; unknown characteristics • NMS: • Pro: two-way nesting; more accurate • Con: very slow; smallest domains; no user support • WFIS (Clark model): • Pro: ultra-fine grid spacing (20 m); optimized for steep terrain • Con: not real time NW-AIRQUEST

  8. http://weather.eos.ubc.ca/wxfcst Your entry point: http://weather.eos.ubc.ca/wxfcst NW-AIRQUEST

  9. 108 km MM5 MC2 WRF NMS NW-AIRQUEST

  10. MC2 36 km 12 km 2 km 4 km NW-AIRQUEST

  11. 800% Zoom of MC2 2 km pdf NW-AIRQUEST

  12. Traditional trajectory methods: NW-AIRQUEST

  13. NW-AIRQUEST

  14. http://weather.eos.ubc.ca/wxfcst NW-AIRQUEST

  15. http://weather.eos.ubc.ca/wxfcst NW-AIRQUEST

  16. http://weather.eos.ubc.ca/wxfcst Mesoscale model characteristics & limitations • Forecast skill in Pac NW decreases to 60% in 24h in summer • Reason for poor skill is Pacific Data Void !!!! (no amount of ensembles nor finer grid resolution, and no tweeking of sigma-y, sigma-z, emission rates, physics params., or PBL depth can compensate) • No effective limit to smallness of horiz. grid size; NWP models are becoming more like LES models NW-AIRQUEST

  17. Clark’s WFIS Model Simulation of forest fire in Rocky Mtns.showing T (color) & w (contours),at 0.25 km horiz. grid spacing NW-AIRQUEST

  18. Sensitivity test approaches (& ensembles) • Because the Pacific Data Void as the weakest link in the AQ/ Met. forecast chain, the most appropriate sensitivity tests are:Observing System Simulation Experiments (OSSEs). • All other sensitivity tests (advection; sub-grid; turb closure; param schemes, etc.) are irrelevant for the Pac NW. • MM5 runs of 7 winter cyclones & 5 summer cyclones 2001-2002. • Experimented with effects of buoy locations • Found N. Amer. paid 20 - 35% penalty in fcst accuracy due to the Pacific Data Void. • Optimum RBS: 6 buoys in cross (see Fig ->) 6 km altitude, 12Z each day, only Fall-Spr. • Penalty reduced to 5 - 15% with optimum RBS Spagnol & Stull, 2003 NW-AIRQUEST

  19. Multi-model, multi-grid size Ensembles (with each model Kalman filtered before ensemble averaging) NW-AIRQUEST

  20. Verification metrics & validation, requirements Merritt, BC For the operational wx fcsts at UBC, we do:mean error, mean absolute error, RMSE, error variance, correlation, slope. Will soon do equitable threat score for precip. NW-AIRQUEST

  21. Credibility, accuracy, Pacific Data Void • No credibility at any time, anywhere in Pac NW. • To address the Pacific Data Void problem is THe Observing System Research and Predictability Experiment (THORpex). It is endorsed by WMO, World Weather Research Program (WWRP), US Weather Research Program (USWRP), Canadian Weather Research Program (CWRP). • Recommendation: Regional AQ managers via NW-AIRQUEST should solidly support THORpex initiatives. (Good leverage/synergy, because many countries and national gov’ts are funding.) NW-AIRQUEST

  22. Winter Storm Recon. Program. 3 yrs dropsondes from aircraft. Begs question: If the favored sounding locations are always in the same part of NE Pacific, why not deploy an array of fixed sounding systems, rather than use expensive manned aircraft? NW-AIRQUEST

  23. NW-AIRQUEST

  24. Rocketsonde Developmentat UBC. NW-AIRQUEST

  25. Evaluation of Mesoscale Numerical Weather Prediction Models Roland Stull Professor, UBC Earth & Ocean Sciences Dept. Director: Geophysical Disaster Computational Fluid Dynamics Center rstull @ eos.ubc.ca 604-822-5901 Summary: • UBC GDCFDC. Run MC2, MM5, WRF, NMS, WFIS. • Mesoscale models becoming like LES models. • Use OSSEs & ensembles (but Pac. Data Void cancels) • All verification metrics possible. Building Rocketsonde. • No Credibility, poor accuracy, due to Pacific Data Void Recommendation: Plan strategically. Support THORpex. NW-AIRQUEST

More Related