Improving the Validation and Prediction of Tropical Cyclone Rainfall

1. Improving the Validation and Prediction of Tropical Cyclone Rainfall Timothy Marchok NOAA / GFDL Robert Rogers NOAA / AOML / HRD Robert Tuleya NOAA / NCEP / EMC / SAIC Additional Collaborator: Manuel Lonfat, Risk Management Solutions This project was funded by the Joint Hurricane Testbed (JHT)

2. Goals • Develop a set of rainfall validation schemes specifically designed for TCs • Produce model QPF error statistics for a set of historic U.S. landfalling storms. • Develop a forecasting tool based on R-CLIPER that utilizes vertical shear forecast data and the effect of topography.

3. Outline • Models & storms • Development of TC QPF validation techniques • 1998-2004 base sample vs. 2005 season • Skill indices based on new techniques • New forecasting tool based on R-CLIPER

4. Models included in this study NCEP/GFS Global T384 (~0.4o) 64 levels Rainfall- CLIPER NCEP/NAM Regional 12 km 60 levels Climatology-based parametric model GFDL Regional 1/2o, 1/6o (2-nest) 42 levels 2003 version

5. U.S. Landfalling Cases for Model Evaluation:1998-2004 Base Sample 1998 1998 1999 1999 2000 2000 2001 2001 2002 2002 2003 2003 2004 2004 Bonnie Bonnie Bret Bret Gordon 55 Gordon 55 Allison Allison Bertha Bertha Bill Bill Bonnie Bonnie 45 45 95 95 100 100 45 45 35 35 50 50 Charley 40 Charley 40 Dennis Dennis Helene Helene Barry Barry Edouard Edouard Claudette Claudette Charley Charley 125 125 60 60 65 65 60 60 35 35 75 75 Earl Earl Floyd Floyd Gabrielle 60 Gabrielle 60 Fay Fay Grace Grace Frances Frances 95 95 70 70 90 90 50 50 35 35 Frances Frances Harvey Harvey Hanna Hanna Henri Henri Gaston Gaston 65 65 45 45 50 50 45 45 30 30 Georges Georges Irene Irene Isidore Isidore Isabel Isabel Ivan Ivan 110 110 90 90 70 70 55 55 90 90 Hermine Hermine Kyle Kyle Jeanne Jeanne 105 105 35 35 35 35 Lili Lili Matthew Matthew 40 40 85 85

6. U.S. Landfalling Cases for Model Evaluation:2005 Season • Ophelia • Rita • Tammy • Wilma • Arlene • Cindy • Dennis • Katrina (Florida) • Katrina (Louisiana)

7. Outline • Models & storms • Development of TC QPF validation techniques • 1998-2004 base sample vs. 2005 season • Skill indices based on new techniques • New forecasting tool based on R-CLIPER

8. Parameters describing skill of TC QPF forecasts • Rainfall patterns • Rainfall volume • Extreme amounts • Sensitivity to track errors

9. Rainfall patterns 2005 1998-2004 Equitable threat score comparison

10. Rainfall volume Example: Tropical Storm Cindy (2005) Stage IV (Observed) GFS

11. Rainfall volume Comparison of rain volume bias by model 2005 1998-2004

12. Rainfall volume: “Rain flux” and track-relative analyses Observed rain flux PDF for all 1998-2004 storms in selected bands surrounding best track

13. Rain volume: Rain flux in select bands GFS, R-CLIPER GFDL, NAM 0–100 km 300-400 km

14. Extreme amounts: Comparison of top 5% of rain flux

15. Sensitivity to track error Eta shifted Eta unshifted Example of grid-shifting of rain field Lili Stage IV r increased from 0.36 (unshifted) to 0.85 (shifted)

16. = Original = Shifted Sensitivity to track error Impact of grid shift on pattern correlations 1998-2004 2005

17. Outline • Models & storms • Development of TC QPF validation techniques • 1998-2004 base sample vs. 2005 season • Skill indices based on new techniques • New forecasting tool based on R-CLIPER

18. Matrix of TC QPF Skill Indices

19. Skill Indices: Pattern Matching 1998-2004 2005 • GFS performs the best in both samples • All models have skill relative to R-CLIPER

20. Skill Indices: Volume 1998-2004 2005 • R-CLIPER significantly better in 2005 season • GFS worse in 2005 due to over-forecast bias

21. Skill Indices: Extreme Amounts 2005 1998-2004 • GFDL worse in 2005 due to core region over-forecast bias • GFS performs best despite lowest resolution

22. Skill Indices: Sensitivity to track error 2005 1998-2004

23. Outline • Models & storms • Development of TC QPF validation techniques • 1998-2004 base sample vs. 2005 season • Skill indices based on new techniques • New forecasting tool based on R-CLIPER

24. Building on R-CLIPER: Inclusion of vertical shear forecast data & topography Rtot = RR-CLIPER + RShear mod + Rtopo Topography modification Standard R-CLIPER Shear modification Formulation Rtot(r,θ) = a0(r) +

25. Example of shear footprint: Hurricane Ivan b) Contribution from Wavenumbers 1,2 a) Contribution from Wavenumber 0 c) Shear footprint is “stamped” on a lon/lat grid every 15 minutes, providing a contribution to storm total accumulation

26. Stage IV R-CLIPER R-CLIPER + Shear + Topog R-CLIPER + Shear

27. Examples of R-CLIPER / SHRAPS validations Equitable Threat Score Rain flux CDF

28. Summary • Developed QPF validation schemes specific for unique characteristics of TC rainfall. • Developed TC QPF skill indices to allow for objective year-to-year comparisons of operational TC rainfall forecasts. • Used the new techniques to provide TC QPF statistics for a baseline 1998-2004 sample as well as for the 2005 season. • Developed a forecasting tool based on R-CLIPER that includes the effects of vertical shear and topography. Additional work… • Work with TPC to automate R-CLIPER forecasts and allow for transmission of forecast data via NWS AWIPS network. • Implement “SHRAPS” version with both shear & topography • Streamline the TC QPF validation system to facilitate easier end-of-season TC rainfall verification.

30. Extra slides…..

31. Rainfall volume

32. Rain volume: GFDL & NAM rain flux in select bands 0 – 100 km 300-400 km 1998 - 2004 2005