Detecting tropical cyclone formation from satellite imagery

Elizabeth A. Ritchie Miguel F. Piñeros J. Scott Tyo S. Galvin

University of Arizona

Acknowledgements: Office of Naval Research Marine Meteorology ProgramTRIF – image processing fellowshipHFIP – Hurricane Forecast Improvement Program

90 W 75 W 60 W 45 W

30 N

20 N

10 N

The Deviation-Angle Variance Technique (DAV-T)Hurricane Beta (2005)

350 km

-- Produce the map of variance values-- produce a time series of the minimum variance value for a cloud system-- the lower the variance, the more organised the system

Correlation: Total time series -0.93From 25 kts on -0.92

Hurricane Wilma 2005

33 ktNHC first warning

Low deviation-angle variances at early stages

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

0.00 0.10 0.20 0.30 0.40 0.50 0.60 0.70 0.80 0.90 1.00

False Positive Rate

1700 1750 1800 1850 1900 1950 2000

Variance Thresholds1550

1500

1600

1400/1450

1350

1650

Tru

e P

osi

tive

Rat

eROC curve for IR imagery

-40

-30

-20

-10

0

10

20

30

40

50

1350 1400 1450 1500 1550 1600 1650 1700 1750 1800 1850 1900 1950 2000

Mean

Median

Variance Threshold [deg2]

Tim

e [h

]2004 & 2005 Time to Detection – “TD”

TPR = 93%FPR = 22%

TPR = 96%FPR = 40%

What Next?What Next?

1.1. Improve the True-Positive Rate while reducing the False-AlarmsImprove the True-Positive Rate while reducing the False-Alarms

- Add more information – water vapor channel- Add more information – water vapor channel - Test sensitivity to some of the parameters in the technique - Test sensitivity to some of the parameters in the technique

- Brightness threshold- Brightness threshold

2.2. Increase the time-to-detectionIncrease the time-to-detection

- Add more information- Add more information- Test sensitivity to some of the parameters in the technique- Test sensitivity to some of the parameters in the technique

3.3. Increase the database to ensure statistical stability (ongoing)Increase the database to ensure statistical stability (ongoing)

4.4. Develop a probabilistic Forecasting Protocol (Summer/10)Develop a probabilistic Forecasting Protocol (Summer/10)

5.5. Making the process of cloud tracking more efficient!!Making the process of cloud tracking more efficient!!

- the Cloud-Tracking Interface- the Cloud-Tracking Interface

Water Vapor Imagery … and the “Cloud-Tracking Interface”

Study:- Aug-Sep 2005

* Test sensitivity to Brightness Threshold

•* Cloud-Tracking Interface

•* Is a gui designed by a (desperate) undergraduate student to make the process more efficient (less boring)

•* displays satellite imagery – can be stepped through in time

•* stores the first time the minimum variance value meets a threshold in a database

•* links to the NHC best-track database to display tracks and assign actual TCs to cloud clusters.

Test with 2 months of water vapor imagery:

Sensitivity to Brightness Temperature Threshold:

Sensitivity to Brightness Temperature Threshold:

Tru

e P

osi

tive

Rat

e

ROC curve Aug/Sep 2005 Inter-comparison

17001800

1700

1800

1700

1800

False Positive Rate

Threshold Minimum Variance Values

Tim

e (h

)

Mean Time to Detection

-30

-20

-10

0

10

20

30

40

50

60

1300 1400 1500 1600 1700 1800 1900 2000

IR

WV - 0.6

WV - 0.52

PoD: 85%FAR: 13%

PoD: 92%FAR: 26%

PoD: 100%FAR: 43%

Conclusions

• A completely objective technique to determine whether a cloud A completely objective technique to determine whether a cloud system will develop into a TC.system will develop into a TC.

• Water Vapor imagery gives similar PoD vs FAR results to IRWater Vapor imagery gives similar PoD vs FAR results to IR

• Mean detection times can be improved by use of WV imagery and by Mean detection times can be improved by use of WV imagery and by tuning some of the parameterstuning some of the parameters

• Best potential is using a combination of the two types of imagery – IR Best potential is using a combination of the two types of imagery – IR provides better PoD to FAR, WV provides earlier detectionprovides better PoD to FAR, WV provides earlier detection

• Future Work:- Future Work:- • Continue to test the sensitivity to parametersContinue to test the sensitivity to parameters• Develop a Forecaster protocol as per NHC needsDevelop a Forecaster protocol as per NHC needs

Thank youThank you

Piñeros, M. F., E. A. Ritchie, and J. S. Tyo 2008: Objective measures of tropical cyclone structure and intensity change from remotely-sensed infrared image data. IEEE Trans. Geosciences and remote sensing. 46, 3574-3580.

Piñeros, M. F., E. A. Ritchie, and J. S. Tyo 2010: Detecting tropical cyclone genesis from remotely-sensed infrared image data. (IEEE Trans. Geosciences and remote sensing. In Review)