1. UW-CIMSS Tropical Cyclone Research : Current Progress and Developments Tim Olander, Chris Velden, Jim Kossin, and Derrick Herndon 2004 Meteorological Satellite Coordinators Conference Pearl Harbor, HI Research and development funding provided by : Naval Research Laboratory - Monterey, CA Funding contract N00173-01-C-2024 through Space and Naval Warfare Systems Command (PMW-155) under PE 0603207N

2. Advanced Objective Dvorak Technique (AODT) Advanced Objective Dvorak Technique (AODT) Tropical cyclone Intensity Estimate (TIE) Model Tropical cyclone Intensity Estimate (TIE) Model AMSU AMSU Environmental Shear Estimates Environmental Shear Estimates UW-CIMSS Tropical Cyclone Research Briefing Overview

3. AODT – Latest Improvements Added new scene type classifications Added new scene type classifications – Separate classifications for eye and surrounding cloud regions Eye: Clear, Pinhole, Large, Ragged, Obscured, None Eye: Clear, Pinhole, Large, Ragged, Obscured, None Cloud: Uniform, Embedded Center, Irregular CDO, Curved Band, Shear Cloud: Uniform, Embedded Center, Irregular CDO, Curved Band, Shear – Scenes more closely mimic Dvorak Technique classifications – Convective symmetry parameter used to classify cloud region Expanded for tropical storm and weaker systems Expanded for tropical storm and weaker systems – Utilize Curved Band analysis as defined in Dvorak EIR Technique to estimate cloud curvature signatures – Irregular CDO scene covers “transition” scenes Removed Rapid Flag / modified time averaging scheme Removed Rapid Flag / modified time averaging scheme – Final T# changed from 12-hour to 6-hour weighted average – New technique still allows for rapid intensification

4. Modified land interaction rule Modified land interaction rule – Analysis suspended while storm is over land (keyword override) – Rule 9 application stopped after storm is over land for >12 hours Added Dvorak EIR Technique Rule 8 Added Dvorak EIR Technique Rule 8 – Constrains growth/decay rate of storm intensity over time Position of maximum Curved Band analysis determined Position of maximum Curved Band analysis determined – Location provided in addition to manual position values AODT – Latest Improvements

5. New Eye Scene Classifications Clear EyePinhole Eye Obscured Eye Ragged Eye Large Eye

6. New Cloud Scene Classifications Shear Irregular CDOCurved Band Uniform Embedded Center

7. Latitude bias adjustment applied to CI# value Latitude bias adjustment applied to CI# value – A bias was found in Dvorak EIR Technique related to latitude – The bias is caused by the latitudinal variation of eyewall cloud-top temperature, resulting from the latitudinal variation of tropopause temperature – Bias adjustment increases/decreases intensity for storm positions north/south of approximately 22ºN latitude – Adjustment based on linear regression and applied to CI# value – Adjusted and unadjusted CI# values stored in AODT history file – Reduces MSLP estimate error by about 10% in test cases – This bias should also be applied to conventional Dvorak method AODT – Latest Improvements (continued)

8. The overall bias is always between ±1.5mb and appears benign, but the latitude-dependent bias is as large as –10mb at 10°N and +14mb at 40°N. The bias explains 15% to 20% of the variance of error. Latitude- dependent bias in Dvorak Enhanced IR (EIR)technique applied to Atlantic TC’s.

9. Removal of the bias decreases RMSE 9% to 11%

10. Physical explanation: IR-measured eyewall cloud-top temperature is a fundamental input parameter of the Dvorak EIR technique (via a nomogram) and depends on latitude much more strongly than intensity does. 47% of the variance of cloud-top temperature is explained

11. Application to other oceanic basins: Mean tropopause temperature (NCEP reanalysis). Averaged over AUG SEP OCT (FEB MAR APR) in Northern (Southern) hemisphere. WMO-RMSC regions

12. AODT-v6.3 – Results Including Latitude Bias Adjustment All Tropical Cyclone Intensities (MSLP) (errors in millibars) Bias RMSE Sample AODT adj 0.09 9.501630 AODT unadj 0.4210.721630 Op Centers 0.2210.651630 All Tropical Cyclone Intensities (MSLP) (errors in millibars) Bias RMSE Sample AODT adj 0.09 9.501630 AODT unadj 0.4210.721630 Op Centers 0.2210.651630 Homogeneous Comparisons with Aircraft Reconnaissance (26 storms between 1995 and 2002) Note: Positive error indicates underestimate (e.g. AODT minus Recon)

13. Improved automated center location determination scheme Improved automated center location determination scheme – Operational Center forecast used as first guess JTWC or NHC forecasts used in conjunction with JTWC or NHC forecasts used in conjunction with 6 and 12 hour old positions from AODT history file Current position interpolated using polynomial interpolation scheme Current position interpolated using polynomial interpolation scheme – Laplacian Analysis of cloud top region performed Analysis region within 75 km radius of interpolated forecast position Analysis region within 75 km radius of interpolated forecast position Identifies pixels with strong gradients in cloud top temperature field Identifies pixels with strong gradients in cloud top temperature field – Confidence factors derived for each methodology Forecast : Time difference from “Initial/Warning” position Forecast : Time difference from “Initial/Warning” position Laplacian : Density and scatter of large temperature gradients Laplacian : Density and scatter of large temperature gradients – Laplacian typically only used where strong but concentrated temperature gradient fields are identified (e.g. Eye Scenes) AODT – Latest Improvements (continued)

14. AODT Auto-Positioning Comparison Hurricane Floyd AODT Auto-Position Clear Eye  6.2 T# NHC Interpolated Forecast Position Uniform Cloud Region  4.5 T#

15. Dependent Data Sample (26 storms between 1995-2002) (errors in millibars) Bias RMSE Sample AODT Manual0.09 9.501630 AODT Automated1.7010.041630 Op Centers0.2210.651630 Dependent Data Sample (26 storms between 1995-2002) (errors in millibars) Bias RMSE Sample AODT Manual0.09 9.501630 AODT Automated1.7010.041630 Op Centers0.2210.651630 AODT-v6.3 – Results Automated AODT Homogeneous Comparisons with Aircraft Reconnaissance Independent Data Sample (5 storms in 2003) (errors in millibars) Bias RMSE Sample AODT Automated2.40 9.93 522 Op Centers2.6711.81 522 Independent Data Sample (5 storms in 2003) (errors in millibars) Bias RMSE Sample AODT Automated2.40 9.93 522 Op Centers2.6711.81 522 Note: Positive error indicates underestimate (e.g. AODT minus Recon)

16. Statistical Breakdown by AODT scene type (errors in T# units) Bias RMSE AAE Sample All Points 0.11 0.67 0.51 3735 All Eye Scenes -0.08 0.50 0.40 1063 Clear Eye -0.02 0.42 0.32 445 Clear Eye -0.02 0.42 0.32 445 All other Eye -0.12 0.55 0.46 618 All other Eye -0.12 0.55 0.46 618 All Non-Eye Scenes 0.19 0.73 0.56 2672 Uniform CDO 0.17 0.64 0.51 1093 Uniform CDO 0.17 0.64 0.51 1093 Embedded Center 0.19 0.66 0.46 193 Embedded Center 0.19 0.66 0.46 193 Irregular CDO 0.11 0.70 0.55 278 Irregular CDO 0.11 0.70 0.55 278 Curved Band 0.17 0.73 0.60 324 Curved Band 0.17 0.73 0.60 324 Shear 0.25 0.85 0.64 784 Shear 0.25 0.85 0.64 784 Statistical Breakdown by AODT scene type (errors in T# units) Bias RMSE AAE Sample All Points 0.11 0.67 0.51 3735 All Eye Scenes -0.08 0.50 0.40 1063 Clear Eye -0.02 0.42 0.32 445 Clear Eye -0.02 0.42 0.32 445 All other Eye -0.12 0.55 0.46 618 All other Eye -0.12 0.55 0.46 618 All Non-Eye Scenes 0.19 0.73 0.56 2672 Uniform CDO 0.17 0.64 0.51 1093 Uniform CDO 0.17 0.64 0.51 1093 Embedded Center 0.19 0.66 0.46 193 Embedded Center 0.19 0.66 0.46 193 Irregular CDO 0.11 0.70 0.55 278 Irregular CDO 0.11 0.70 0.55 278 Curved Band 0.17 0.73 0.60 324 Curved Band 0.17 0.73 0.60 324 Shear 0.25 0.85 0.64 784 Shear 0.25 0.85 0.64 784 AODT-v6.3 – Results Homogeneous Comparisons with Aircraft Reconnaissance Note: Positive error indicates overestimate (e.g. AODT minus Recon)

17. AODT – Future Directions Integrate regression-based TIE Model with AODT – Use 2003 TC results to guide development of combined method Examine additional geostationary channels Examine additional geostationary channels – IRW-WV channel difference (Velden and Olander, 1998) – Additional Dvorak methods using visible and/or shortwave infrared Investigate additional satellite information Investigate additional satellite information – AMSU analysis using technique developed at UW-CIMSS – SSM/I and TRMM analysis based on work by NRL-MRY/R. Edson GOAL Develop an advanced multi-sensor objective technique Develop an advanced multi-sensor objective technique – Fuse results/output from different instruments and analysis techniques into an expert system

18. Integrated Satellite-Based TC Intensity Estimation System IRW-WV channel difference AMSU Microwave imagery Current AODT

19. AODT – Availability The latest version of the AODT is currently available to any/all interested users for McIDAS platforms from the AODT webpage (along with the updated Users’ Guide) The latest version of the AODT is being integrated into the TeraScan system for use at JTWC during the 2004 TC season. The code has also been adapted for N-AWIPS at NOAA sites and Mark-IVB at AFWA An X-Window/Motif version of the AODT has also been developed and will is available for UNIX and Linux based systems (also available at the AODT webpage) An X-Window/Motif version of the AODT has also been developed and will is available for UNIX and Linux based systems (also available at the AODT webpage) AODT webpagehttp://cimss.ssec.wisc.edu/tropic/aodt

20. The Tropical cyclone Intensity Estimation (TIE) model: Developed to formally explore the relationships between TC intensity and features in geostationary satellite infrared (IR) imagery, without the constraints of the Dvorak-based methods and their attendant rules.Developed to formally explore the relationships between TC intensity and features in geostationary satellite infrared (IR) imagery, without the constraints of the Dvorak-based methods and their attendant rules. Applicable at all stages of TC lifetime.Applicable at all stages of TC lifetime. Completely objective; no scene-typing.Completely objective; no scene-typing.

21. 63% of the MSLP variance is explained by the regression. TIE-model: Predictor (sig > 99.9%) Normalized Coefficient Warmest eye temperature – 0.64 Mean eyewall temperature + 0.43 Symmetric organization + 0.34 SST-based parameter + 0.26 Latitude – 0.16 }IR-derived multivariate linear regression TC central pressure (MSLP) measured from aircraft reconnaissance is regressed onto 5 predictors

22. TIE-model performance: Errors relative to recon. TIE-model errors based on independent jackknife analysis.

23. The good: The bad: TIE-model performance:

24. TIE-model R&D: Serves as a natural platform for testing additional parameters for correlation with TC intensity.Serves as a natural platform for testing additional parameters for correlation with TC intensity. Indices derived from synoptic fields (analogous to SHIPS) are being presently tested.Indices derived from synoptic fields (analogous to SHIPS) are being presently tested. Further development will include indices derived from microwave imagery; these can include indices related to eyewall replacement cycles.Further development will include indices derived from microwave imagery; these can include indices related to eyewall replacement cycles. TIE model estimates will be available in conjunction with AODT estimates on TeraScan system this coming season.TIE model estimates will be available in conjunction with AODT estimates on TeraScan system this coming season.

25. AMSU Background What does AMSU really observe? What does AMSU really observe? AMSU-A passive microwave radiometer AMSU-A passive microwave radiometer Flown on NOAA-15/16/17 and Aqua polar orbiting satellites Flown on NOAA-15/16/17 and Aqua polar orbiting satellites Up to 6 warm core observations daily (NOAA birds) Up to 6 warm core observations daily (NOAA birds) Senses terrestrial radiation near 55GHz region Senses terrestrial radiation near 55GHz region Challenges Challenges Scattering prohibits hydrostatic integration Scattering prohibits hydrostatic integration Variability in peak warming altitude Variability in peak warming altitude Variable horizontal resolution Variable horizontal resolution FOV 1 FOV 30

26. Suitability of the AMSU 929hPa Hurricane Inez 1966 8 7 6 5 Hurricane Floyd 1999

27. CIMSS AMSU-based TC Intensity Estimation Algorithm Extrapolate storm position for AMSU pass time based on latest warning Find max AMSU Tb in ch7 and 8 within 100km of storm position (this is the TC upper-level warm core) Average environmental Tbs from 4 surrounding points and subtract from core value to determine thermal anomaly Estimate MSLP for ch7 and ch8 anomalies based on statistical regression. Use lower of the two values. Apply FOV bias correction based on RMW or... Call channel 7 retrieval if: Raw Ch7 Tb anomaly > Raw Ch8 Raw Ch7 Tb anomaly > Raw Ch8 Raw Ch7 anomaly > 1.0K Raw Ch7 anomaly > 1.0K Storm is located near the edge of the satellite swath Storm is located near the edge of the satellite swath X X 10 930 1013 0 AMSU MSLP ?

28. 2002 Results (MSLP) Statistics for Atlantic (N= 60) in hPa CIMSSDvorak Mean Error 3.29 4.52 Std Dev 2.30 3.40 Bias -0.08 -0.90 RMSE 4.00 5.62 Avg MSLP of sample 1002.4 Statistics for Atlantic (N= 60) in hPa CIMSSDvorak Mean Error 3.29 4.52 Std Dev 2.30 3.40 Bias -0.08 -0.90 RMSE 4.00 5.62 Avg MSLP of sample 1002.4 Statistics for Pacific (N = 10) in hPa CIMSSJTWC Mean Error 8.00 11.16 Std Dev 7.25 8.79 Bias 4.79 6.10 RMSE 10.57 13.93 Avg MSLP of sample 964.4 Statistics for Pacific (N = 10) in hPa CIMSSJTWC Mean Error 8.00 11.16 Std Dev 7.25 8.79 Bias 4.79 6.10 RMSE 10.57 13.93 Avg MSLP of sample 964.4

29. MSLP

30. 2002 Atlantic Basin Results AMSU 81% within +/- 0.5T NOAA/NESDIS Satellite Analysis Branch (SAB) vs.AMSU (N=34, +/- 2hrs of aircraft reconnaissance) SAB 78% within +/- 0.5T

31. Sources of Error Radius of Maximum Winds Used as a proxy for eye size in algorithm Used as a proxy for eye size in algorithm Especially important for small intense storms Especially important for small intense storms How is it determined? How is it determined? Position First Guess comes from Warning Message First Guess comes from Warning Message Search algorithm locates max anomaly Search algorithm locates max anomaly Sheared systems Sheared systems

32. AAFB 937mb X

33. X

34. Future Work Improve Technique Bias correction when storm is near satellite limb Bias correction when storm is near satellite limb Separate treatment of pinhole eye storms using new coefficients Separate treatment of pinhole eye storms using new coefficients Time averaging / smoothing Time averaging / smoothing Add channel 8 retrieval Add channel 8 retrieval Remove precipitation contamination (Bob Wacker) Remove precipitation contamination (Bob Wacker) Addition of AMSU on Aqua (2 more passes per day) Addition of AMSU on Aqua (2 more passes per day) Basin specific coefficients (need sufficient “observations”) Basin specific coefficients (need sufficient “observations”) Confidence Indicator ? Give forecasters a measure of estimate confidence Give forecasters a measure of estimate confidence Based on FOV, eye size (RMW), anomaly height Based on FOV, eye size (RMW), anomaly height

35. Summary and Conclusions AMSU provides unique tropical cyclone perspective Can see beneath upper level cloud cover Can see beneath upper level cloud cover 55 Ghz region radiances can quantify inner core thermal 55 Ghz region radiances can quantify inner core thermal structure / changes. structure / changes. Temperature anomaly strength directly associated with tropical Temperature anomaly strength directly associated with tropical cyclone intensity cyclone intensity A unique addition to the forecaster “tool kit” A unique addition to the forecaster “tool kit” AMSU-generated MSLP estimate accuracy improved in 2002 Addition of second AMSU channel and advanced logic Addition of second AMSU channel and advanced logic Skill comparable to, and in some cases better than, the Dvorak estimates Skill comparable to, and in some cases better than, the Dvorak estimates Future changes will improve performance Future changes will improve performance

36. UW-CIMSS/NESDIS wind shear products in Hurricane Lili (2002) Deep-layer Tendency Mid-level

37. TROPICAL STORM KYLE 18:00UTC 07October2002 UW-CIMSS Experimental Vertical Shear TC Intensity Trend Estimates Current Conditions : Current Conditions : Latitude : 32:19:27 N Latitude : 32:19:27 N Longitude : 70:50:45 W Longitude : 70:50:45 W Intensity (MSLP) : 1005.0 hPa Intensity (MSLP) : 1005.0 hPa Max Pot Int (MPI) : 971.7 hPa Max Pot Int (MPI) : 971.7 hPa MPI differential (MSLP-MPI) : 33.3 hPa MPI differential (MSLP-MPI) : 33.3 hPa CIMSS Vertical Shear Magnitude : 2.8 m/s CIMSS Vertical Shear Magnitude : 2.8 m/s Direction : 215.0 deg Direction : 215.0 deg Outlook for TC Intensification Based on Current Env. Shear Values Outlook for TC Intensification Based on Current Env. Shear Values Forecast Interval : 6hr 12hr 18hr 24hr Forecast Interval : 6hr 12hr 18hr 24hr F F N N F F N N Legend: VF - Very Favorable F - Favorable N - Neutral U – Unfavorable VU - Very Unfavorable U – Unfavorable VU - Very Unfavorable -- Mean Intensity Trend (negative indicates TC deepening) -- -- Mean Intensity Trend (negative indicates TC deepening) -- 6hr 12hr 18hr 24hr 6hr 12hr 18hr 24hr VF < -3.0mb/ 6hr < -6.0mb/12hr < -9.0mb/18hr < -12.0mb/24hr VF < -3.0mb/ 6hr < -6.0mb/12hr < -9.0mb/18hr < -12.0mb/24hr F -3.0 - -1.5 -6.0 - -3.0 -9.0 - -4.5 -12.0 - -6.0 F -3.0 - -1.5 -6.0 - -3.0 -9.0 - -4.5 -12.0 - -6.0 N -1.5 - +1.5 -3.0 - +3.0 -4.5 - +4.5 -6.0 - +6.0 N -1.5 - +1.5 -3.0 - +3.0 -4.5 - +4.5 -6.0 - +6.0 U +1.5 - +3.0 +3.0 - +6.0 +4.5 - +9.0 +6.0 - +12.0 U +1.5 - +3.0 +3.0 - +6.0 +4.5 - +9.0 +6.0 - +12.0 VU >+3.0 >+6.0 >+9.0 >+12 VU >+3.0 >+6.0 >+9.0 >+12 CIMSS experimental vertical shear product (Gallina and Velden 2002)

39. SUMMARY During the 2002 TC season, a fully automated algorithm based partially on quantitative satellite data estimates of environmental wind shear was used to derive short-term TC intensity outlooks in real time. The experimental output products were distributed by CIMSS via email to selected Tropical Analysis Centers. Preliminary analysis and forecaster feedback suggests these products can be quite useful as indicators of favorable/unfavorable tropospheric environments precluding TC deepening/filling trends. Quantitative relationships between shear and short-term TC intensity trends were developed based on an exhaustive statistical analysis of shear fields produced at CIMSS using satellite-derived wind information. The resulting prognostic algorithm shows preliminary evidence of skill over SHIPS and other methods in some cases.

40. RESULTS 1) Qualitatively, the new shear products are showing promise to aid the TC intensity forecast process. Initial assessment and feedback from Tropical Analysis Centers has been quite favorable. 2) In general, intensity trends are related to changes in the environmental shear as depicted by the CIMSS fields and outlook products. The short term (24-hr) intensity trend outlook (deepening or filling) was correct 93% of the time. 3) Case studies illustrate situations when shear thresholds and trends in shear can be indicators of TC intensity changes. 4) The more quantitative, experimental outlook product based on a statistical analysis of the CIMSS shear estimates vs. observed TC intensity was skillful relative to SHIPS and other guidance in Lili, but not in Kyle. Further analysis is underway.