#16: Automated cloud-drift winds from GOES images


C. Kambhamettu, K. Palaniappan, and A. F. Hasler

SPIE Proc. on GOES-8 and Beyond, Volume 2812, pgs. 122--133, 1996

motion, visualization, parallelization, stereo, cloud, big data, remote sensing

PlainText, Bibtex, PDF, URL, DOI, Google Scholar

Abstract

Estimation of the atmospheric wind field based on cloud tracking using a time sequence of satellite imagery is an extremely challenging problem due to the complex dynamics of the imaging instruments and the underlying non-linear phenomena of cloud formation and weather. Cloud motion may involve both partial fluid motion and partial solid motion, which we model as semi-fluid motion. Motion algorithm with subpixel accuracy using differential geometry invariants of surfaces was developed to track clouds. The motion model is general enough to include both physical and geometrical constraints. Typically, a polynomial displacement function is used to model the local deformation behavior of a surface patch undergoing semi-fluid motion. The cloud tracking algorithm recovers local cloud surface deformations using a sequence of dense depth maps and corresponding intensity imagery, that captures the time evolution of cloud-top heights. Either intensity or depth information can be used by the semi-fluid motion analysis algorithm. A dense disparity or depth map that can be related to cloud-top heights is provided by the Goddard Automatic Stereo Analysis module for input to the motion analysis module. The results of the automatic cloud tracking algorithm are extremely promising with errors comparable to manually tracked winds. Experiments were performed on GOES images of Hurricanes Frederic, Gilbert and Luis, and a temporally dense 1.5 minute time interval thunderstorm sequence covering Florida region. Future work involves using multispectral information, incorporating robustness, cloud motion segmentation and adaptive searching for improving operational cloud-tracking performance.