CIMSS/NESDIS-USAF/NRL Experimental AMSU TC Intensity Estimation: Storm position corresponds to AMSU-A FOV 8 [1<--->30] Raw Ch8 (~150 hPa) Tb Anomaly: 5.36 C Raw Ch7 (~250 hPa) Tb Anomaly: 5.34 C AMSU-A MSLP (Ch8): 909.9 hPa RMW value: 24.0 Km Storm is sub-sampled based on RMW and FOV. Bias correction applied is: -15.1 hPa SUPER TYPHOON 19W Thursday 26aug04 Time: 0447 UTC Latitude: 23.79 Longitude: 135.960 Satellite: NOAA-16
TOWARD AN OBJECTIVE SATELLITE-BASED ALGORITHM TO PROVIDE REAL-TIME ESTIMATES OF TC INTENSITY USING INTEGRATED MULTISPECTRAL (IR AND MW) OBSERVATIONS
Christopher Velden, James Kossin, Tim Olander, Derrick Herndon, Tony Wimmers, Howard Berger University of Wisconsin – Cooperative Institute for Meteorological Satellite Studies
Robert Wacker,
United States Air forceJeff Hawkins NRL
at Monterey, CA
Uses pattern recognition techniques to extract TC characteristics in SSM/I
imagery (85 GHz). Bankert and Tag, 2002 (JAM)
“Computer Vision” Approach
Automated intensity estimates from passive microwave imagery
Example: SSMI 85 GHz and Rain Rate features
Using passive microwave. Example: TRMM Microwave Imager (TMI), 85 GHz overpass of Hurricane Isidore between the Yucatan peninsula and Cuba. “PCT” is a weighted difference between vertical and horizontal polarizations that indicates scattering by ice crystals and is a proxy for precipitation.Best track center, white cross; spiral-fitting score field, white contours; optimum spiral center, white square.
Using IR data. Example: GOES IR image of Hurricane Juan; initial guess of TC center based on a forecast, black triangle; spiral-fitting score field, white contours; area used in calculating the score field, gray circle; optimum eye ring, black circle.
Integrated Satellite-Based
TC IntensityEstimation System
AMSU
Microwave ImageryAODT
89 GHz defines eye based on ice scattering in the eyewall
1/nwi (est)iEnsemble Intensity Estimate =
The weights (wi) represent the confidences of the various (n) algorithm estimates (esti). The confidence is based on performance characteristics
of the algorithm as well as any additional factors such as data latency associated with polar orbiting satellite data.
AODT AMSU Consensus
7.0
5.3
0.56
6.16.910.5RMSE
4.85.58.6ABS Error
0.17-0.22-1.0Bias
N=214 Hybrid
AODT Statistics for Version 6.3AODT Statistics for Version 6.3Homogeneous (independent) data sample of 522 cases from 2003Homogeneous (independent) data sample of 522 cases from 2003
9.3311.812.67Op Center
8.08 9.932.40AODT (auto)
Abs. Err.RMSEBiasUnits in (hPa)
Stratified by Post-Eye and Scene TypeStratified by Post-Eye and Scene Type
680.610.80-0.07Curved Band
720.470.55 0.01Irregular CDO
1400.380.52 0.10Embedded Center
2620.580.81-0.12Shear
10630.400.50-0.08All Eye Scenes
10970.460.63-0.04All No Eye Scenes
0.41
0.43
AbsErr
555
2160
Sample
0.57-0.06All Scenes
RMSEBiasUnits in T-Number
0.52-0.04CDO
AODT Statistics for Version 6.3AODT Statistics for Version 6.3
For a complete description of the latest version of theFor a complete description of the latest version of theAdvanced Objective Dvorak Technique (AODT), see theAdvanced Objective Dvorak Technique (AODT), see theabst by Olander, Velden and Kossin, 26abst by Olander, Velden and Kossin, 26thth AMS Hurr Conf AMS Hurr Conf
CIMSS TC Intensity Methods for Hurricane Gabrielle
970
975
980
985
990
995
1000
1005
1010
1015
18:57 18:46 7:17 13:32 13:08 18:26 6:54 12:45 18:16 12:21
12-Sep 13-Sep 14-Sep 14-Sep 15-Sep 15-Sep 16-Sep 16-Sep 16-Sep 17-Sep
recon amsu aodt Hybrid
CIMSSAMSU
Super Typhoon 19W
Introduction and Motivation
Current Satellite-Based TC Intensity Estimation Methods Developed at CIMSS
Several existing or promising satellite-based methods to estimate tropical cyclone (TC) intensity are available to forecasters today. Some of these, such as the Dvorak Technique, have been utilized operationally for over 30 years. Others, such as those based on microwave data, are just emerging as new, more capable, meteorological satellite instruments become operational. Each of the methods by themselves represents or promises significant contributions to TC intensity analysis. However, each technique (or instrument that it is based on) also has its limitations. An effort is underway at CIMSS to build an integrated algorithm that is fully automated and objective, and utilizes a multispectral approach. This system would build on, and take advantage of, the latest science advances in existing (and emerging) methods.
Corresponding author: [email protected]
AODT AMSU
For a complete description of the latest version of the CIMSSAdvanced Microwave Sounding Unit (AMSU) algorithm,
see the abstract by Herndon and Velden, 26th AMS Hurr Conf
Overall Performance
Multi-SensorInformation Sharing
Improving Center-Fix Methods
Satellite Estimates of RMW
TC Intensity Estimation: Integrated Approach
Basic Consensus (2 CIMSS Methods)
Preliminary Results
Weighted Ensemble (Multiple Methods)
Situational Performance
Other Methods as Potential Candidates for the Ensemble
SSMI/TMI/AMSRE
Empirical ApproachCorrelates patterns in SSMI imagery
with Dvorak-like patterns.Edson and Lander, 2002 (Proc. Of
25th AMS Hurricane Conf.)
High Confidence
Storm core well defined Nadir FOV FOV matches storm center
Multiple storm ‘cores’
Poor Confidence
Near limb FOV FOV offset from storm center
AMSU Confidence Scenarios
Accurate sub/over-samplingcorrections
Wrong choice of RMWcan lead to large estimate error
FOV captures all of warming
FOV captures fraction of warming
Colors represent confidence (green high, red low). The colored barsindicate ‘probabilities’ based on climate/persistence. The final estimate
is a weighted blend with error bars (black).
All units in hPa. The ‘hybrid’ uses an additional predictor,which is the estimate spread of the 2 members in the consensus
Two methods based on the study summarized in Wimmers and Velden,
26th AMS Hurr Conf
Empirical method employed at JTWCUsing SSMI and TRMM/TMI
CIMSS AMSU algorithm performance for storms from 2001-2004 using latest algorithm logic
6.35.0Mean Error
8.97.8RMSE
1.41.0Bias
DvorakCIMSS AMSUMSLP (hPa)
333333N
Summary
As part of an R&D effort at CIMSS to develop improved TC intensity estimation from satellites, existing methods to estimate intensity from different satellite platforms/sensors are being employed to create a more robust and reliable integrated approach. Taking advantage of the single method characteristics and situational tendencies, the final TC intensity estimate at a given analysis time will be obtained by employing a weighted consensus, decision tree, or “expert system” technique to blend/resolve the independent estimates. The algorithm will output both TC intensity parameters and confidence indicators.
This work is being sponsored by the Office of Naval Research, Program Element (PE-0602435N), the Oceanographer of the Navy through the program office at the PEO C4I&Space/PMW-180 (PE-
0603207N), and the Naval Research Laboratory-Monterey. T-Number relates to TC Vmax via the Dvorak relationship.
T-Number increments give a more realistic representation of actual intensity change due in part to the nonlinear relationship between MSLP and Vmax
Channel 8 Tb Anomaly
Average IR-calculated Eye Size (km)
Air
craft
Measu
red
RM
W (
km)
MSLP (hPa)
Best Guess
IR Estimate ATCF
Bias 1.6 -0.5 5.1
Absolute Error
5.4 6.8 8.3
RMSE 7.5 8.7 10.6
N 50 50 50
AMSU Intensity estimates using IR RMW method perform better than using ATCF RMW on independent
cases verified against Atlantic recon.
RMSE = 6.16 kmR2 = 0.60
Relationship between eye size, as measured by IR, and aircraft-measured RMW, for clear-eye Atlantic TC cases (AODT now provides these RMW estimates for clear-eye scenes).
Existing Method – Microwave-Based - Subjective
New Method – IR-Based - Objective
Integrated Satellite-Based
TC IntensityEstimation System
AMSU
Microwave Imagery
AODT
The weights (wi) represent the confidences of the various (n) algorithm estimates (esti). The confidences will be based on performance characteristics of the algorithm as well as any additional factors such as data latency associated with polar orbiting satellite data.
Ensemble Intensity Estimate =1/nwi (est)i
Colors represent confidences (green high, red low). The colored barsindicate ‘probabilities’ based on climate/persistence. The final estimate
is a weighted blend with error bars (black).
Approach