Tracking of clouds provides an estimate of winds at different levels.  This can then be used to warp a sequence of satellite images into a relative view of a strom system.  Selecting the high cloud view point, one can observe the low level inflow of air into dynamic cloud systems.  The wraping compensates for the general translation of the storm.  We add the new feature of adjusting for rotation and expansion in time.

OBJECTIVE
 

There have been a number of studies of storm system where the reference frame has been translated, centered upon the storm.  These have shown unique features of the circulation around the storm like inflows and outflows of clouds and air masses.  In this research we were interested in going one step further, fix many control points on a storm and watch the circulation relative to those reference locations.
 

RESEARCH ACCOMPLISHED
 

There have been may studies and algorithms developed to track clouds in sequences of geosynchronous satellite imagery.  We have implemented a tracking algorithm which will track hand selected objects (clouds or ground features).  Here we discuss using those sequences of locations to deform a storm system so that the selected clouds remain fixed.  In some sense this provides a Lagrangian view of the circulation around the cloud.

The tracking procedure starts with a reference patch of pixels in a satellite image and searches for the location with the maximum correlation in successive images.  At this time we have selected these clouds by hand, but one could imagine an automatic cloud selection procedure.  The matching locations were then feed into an IDL routine called POLYWARP.  This translates the fixed points from their apparent position to the reference positions, then all points in between are transformed in a smooth manor to construct a “warped” image.  This interpolation procedure is equivalent to translation of the storm if only one control point is selected for each image time.
 

Hurricane Case
 

Figure 1, Time sequence of original data.

Figure 1 and 2 show the reference and warped image for our first case study: Hurricane Luis, 1995.  The circles show the locations of the fixed objects and their apparent motions. The warped images shows that the rotion of the storm is not uniform around the edge.  Different parts are rotating at different rates.
 

Second Hurricane Case; George, 1998
 

Mid Latitude Storm Case,  March, 2000

Few Points

Many Points

An other midlatitude case April 105

An other midlatitude case April 103
 
 

Discussion
 

These case studies show that storm circulation is more complicated than just linear translation of the center of the storm.  Different portions of the storm move at different rates.  This of course could be captured in a full numerical simulation of the storm development, but here we attempted to capture those features just with image analysis.  Some other cases will be discussed in the final presentation.
 

CONCLUSIONS AND RECOMMENDATIONS
 

This effort has shown that a different view of storm system does come from more complex transforms of a time sequence than just linear shifts.  The results is very dependent upon the manual selection of cloud references.  Much remains to be learned about this process.
 

Acknowledgements
 

This work was supported by the Department of Defense Center for Geosciences / Atmospheric Research Agreement #DAAL01-98-2-0078.  DuFour was a visiting scientist from France and he will continue his studies at CEA/DSM/LSCE, Gif sur Yvette, France.
 

References
 

Campbell, G.G., 1998, Asynchronous Stereo Height and Motion Analysis:  Applications, Fourth Winds Workshop, WMO.