robert demaria. motivation objective data center-fixing method evaluation method results ...
TRANSCRIPT
Finding Tropical Cyclone Centers with the Circular
Hough TransformRobert DeMaria
Motivation Objective Data Center-Fixing Method Evaluation Method Results Conclusion
Overview
Only west Atlantic has routine hurricane hunter aircraft for finding storm centers
Satellite data used subjectively to find centers across the globe
Improvements to accuracy in real-time highly desirable
Motivation
sos.noaa.gov/Education/tracking.html
Geostationary satellites produce Infrared(IR) every 15 Minutes
Forecast produced every 6 hours Due to time constraints, most of these
images are unused
Automatic method for estimating tropical cyclone location is highly desirable
Motivation
Tropical cyclones are roughly circular Use Circular Hough Transform (CHT) to
produce estimate for tropical cyclone location by finding circles in IR imagery
Compare accuracy to National Hurricane Center real-time center-fix
Objective
2D Image of Temperature◦ Created every 15 minutes
Infrared Data
A-Deck: Real-time estimate of position, velocity, wind speed, etc.◦ Updated every 6 hours
Best-Track: Improved a-deck data available after end of season
Storm Track Data
Find a-deck position◦ Given the time an IR image was created, look up
most recent a-deck information and extrapolate position to IR image time
Subset of IR image used◦ Center image on a-deck position ◦ Image reduced to area around storm/area around
eye◦ Background removed from cloud shield using
temperature threshold
Center-Fixing
IR after subsect & thresholding:
Center Fixing Cont.
Laplacian of image performed to find edge pixels
Center-Fixing Cont.
Circular Hough Transform performed for a range of radii on image
Gaussian fit performed on accumulation space to produce center location
Center-Fixing Cont.
Circular Hough Transform
For each time in best-track, find most recent IR image
Estimate if eye is present in image◦ If it is then perform center-fix searching for radii
roughly the size of an eye◦ If not, perform center-fix searching for radii
roughly the size of the entire storm Error calculated as CHT center-fix distance
from best-track location Compare error to that of the a-deck position
Evaluation Method
Eye Detection Examples
Katrina 08/29/00 2005 Earl 09/02/06 2010 Charley 08/13/18 2004
Katrina 08/25/18 2005 Ericka 09/02/18 2009 Sandy 10/19/18 2012
No Eye Cases
Eye Cases
Charley 2004 – Very small but intense hurricane
Katrina 2005 – Classic large, intense hurricane Ericka 2009 – Very disorganized weak tropical
cyclone, did not make it to hurricane strength Earl 2010 – Strong hurricane in higher
latitudes Sandy 2012 – Unusually large but only
moderate strength, non-classical hurricane structure
Hurricane Cases
Mean a-deck error: 42 km Mean CHT error: 91 km
Bias X: 6 km Bias Y: 8.5 km Bias Explained by Parallax
Results
Results by Storm:
Sandy: Earl: Erika Charley: Katrina:0
20
40
60
80
100
120
140
160
180
Error (km)
Variability
Strong Circular Eye Greatly Improves Accuracy◦ Eye Mean Error: 54 km◦ No Eye Mean Error: 127 km◦ Strong circular eyes are fairly rare
Cyclone Eyes
Cloud Shield Shape Usually Not Ideal
Did not improve real-time center fix Rotational center may not be in center of
cloud features: CHT may not be well suited to large-scale images
CHT may be useful when an eye is present
Conclusions:
Use time-series information to improve Combine with information about vertical
shear Improve eye estimation technique
Future Work