A Probabilistic Pattern Recognition Method for Detection of Overshooting Cloud Tops Using Satellite Imager and Numerical Weather Analysis Data
A new GOES-R Risk Reduction Research Program project seeks to improve upon the existing algorithm through the use of sophisticated pattern recognition techniques and enhanced use of numerical weather analysis data. The net result is a probabilistic OT detection product that can be 1) generated using data from any polar-orbiting or geostationary imager and 2) applied uniformly across any geographic region or season. The detection method has been tested on GOES 1 km/pix imagery in 0.65 um VIS band and the 4 km/pix in the 10.7 um BT band, as well as on MODIS imagery resampled to the same resolution. The algorithm uses a series of pixel-level tests each producing a relative confidence rating with the final rating being a total accumulated sum. From the BT imagery, the most important indictors of an OT include the lowest BT within an OT region, the OT shape and prominence compared to the anvil, the BT distribution within the anvil, and the roundness of the anvil's boundary. Visible (VIS) imagery is used during the daytime as a supplementary field mainly because of the often insufficient OT contrast relative to bright cirrus anvil near solar noon and the requirement for the algorithm to also operate during the night time. The VIS branch of the algorithm includes the tests for highest reflectance, localized reflectance gradients (i.e. texture), and identification of shadow patterns induced by the OT upon the surrounding anvil. The satellite-based confidence rating is combined with level of neutral buoyancy and tropopause information derived from numerical weather analysis data to generate the final probabilistic OT detection product. This presentation will provide an overview of the new algorithm and associated validation results, and examples of OT detection product applications.