1. High Resolution Assimilation of Radar Data for Thunderstorm Forecasting on OSCER Keith A. Brewster, Kevin W. Thomas, Jerry Brotzge, Yunheng Wang, Dan Weber and Ming Xue CAPS and School of Meteorology University of Oklahoma 2007 OSCER Supercomputing Symposium Norman, Oklahoma, October, 2007

2. CASA & NEXRAD Radars CASA NetRad NSF ERC: Collaborative Adaptive Sensing of Atmosphere –X-Band Dual-Polarization Radars –30 km nominal range –Dual-Pol attenuation correction –Clutter filter –Collaborative, Adaptive Scanning –Fill-in below coverage of NEXRAD –Toward phased-array panels – low-cost! NEXRAD –S-Band Radars –Current Weather Service Operational Radars –14 covering domain –Data used out to 230 km

3. CASA NetRad IP1 Southwest Oklahoma

4. CASA NetRad KFDR NEXRAD

5. Near Real-Time NWP Runs CASA & NexRad CASA Only NexRad OnlyNo Radars Four Runs in Near Real-time Hourly output graphics posted to web: http://www.caps.ou.edu/wx/casa http://www.caps.ou.edu/wx/casa 9 Weeks in Spring 2007 6-hour 1-km resolution forecasts Use Reflectivity ARPS Model ADAS Analysis with cloud analysis and latent heat adjustment Run on Parallel Linux Boxes OU OSCER (150 proc/2 runs at a time)

6. Forecast Domain  x = 1 km nx=603 ny=543 nz=53

7. Assimilation Strategy 22102220223004220000 IAU 010203 5.5-hour Forecast40-min Assimilation IAU 2150 Forecast Model started manually when storms enter the network or with initial development within the network

8. Computing - Networking CASA Radars NexRad Radars Other Weather Observations Other Numerical Weather Model Data CASA SOCC CAPS Ingest Cluster IDD (Internet) Internet NFS OSCER Top Dawg HPC CAPS Webserver DataFile Archive

9. Computing – Number Crunching TopDawg Total 1024 processors Pentium4 Xeon EM64T 3.2 GHz GigE Network (976 Mbps, latency 9  s) ARPS Analysis/Forecast 150 processors per job (75 nodes) 4 forecasts, 2 jobs run at a time Horizontal Domain Decomposition, MPICH2 15x10 processors 6 - 6.5 h wall for each forecast 1 - 1.5 h wall for data processing and analysis ~50 GB of data each day created for archive

10. Building Pseudo-Volume from CASA Scans t 3  3 = 5.0˚ t 6  6 = 11.0˚ t 4  4 = 7.0˚ t 2  2 = 3.0˚ t 1  1 = 2.0˚ t 7  7 = 14.0˚ Low-level 360° Scan Upper-level Sector Scans t 0  0 = 1.0˚ Automated to select scans in middle of 10-min assimilation window. t 5  5 = 9.0˚

11. Summary of 2007 Cases 21 Sets of Runs for 20 Cases Weather Summary: –With Funnels and/or Tornadoes (2) –Squall Lines (8) –Clusters of Cells or Individual Cells (6) –Broken Lines and/or Cells on Boundaries (5) –Precip Area with Synoptic Scale Low (1) Some developing, some intensifying, decaying Wide Range of Time-of-Day

12. Sample Impact of CASA NetRad Data NEXRAD Only NetRad & NEXRAD ADAS Analysis of Hydrometeors Converted to Reflect Hgt=~250 m AGL

13. Cross-Section NEXRAD Only NEXRAD & NetRad NetRad Low-Level Info

14. 01 Z May 09 Forecast Animation

15. Reflectivity All Radars Refl + Vr All Radars KTLX (Verif) 40 min - End of Assimilation Period

16. Reflectivity All Radars Refl + Vr All Radars KTLX (Verif) 40 min - End of Assimilation Period Stronger Vort Improved Refl

17. 2-hour Forecast Reflectivity All Radars KTLX (Verif) Refl + Vr All Radars

18. 2-hour Forecast Reflectivity All Radars KTLX (Verif) Refl + Vr All Radars Stronger, Closer

19. Vorticity During Assimilation

20. 2-3 hr Vorticity Forecasts

21. Triangles: Observed Circulations No Radar Data

22. 2-3 hr Vorticity Forecasts Solid Green: NetRad Only Orange: NexRad Only Red: Reflectivity Both

23. 2-3 hr Vorticity Forecasts Radial Velocity Data Used from All Radars Four Circulations Tracked in this run

24. Accomplishments Completed MPI of ADAS with cloud analysis Able to process CASA radar –Automated Data Selection –Data I/O and Quality Control (ref & vel) –Remapping and hydrometeor assignment –Reflectivities are well-handled in model Baseline Runs Complete

25. Ongoing/Future Work Complete MPI of Vr analysis and run simulations with Vr Detailed verification of model forecasts –Comparison of QPF to NSSL 1-km Q2 –Comparison of location and strength of forecast circulation centers vs. radar-observed circulation centers (April 10, May 8/9) Prepare for 2008 Experiment –Complete MPI of 3DVAR and comparison of results –Integrate NWP in end-user training –Design near real-time evaluation for integration in NWC Hazardous WeatherTestbed