Prediction of Tropical Cyclogenesis Using Scatterometer Data

Abstract

In the present work, a new technique based on the data mining approach has been discussed for the prediction of tropical cyclogenesis using the ocean wind vectors derived from the sea-winds scatterometer on QuikScat satellite. The technique is based on similarity of given wind pattern with wind vector signatures of developing systems, available from the past observations. A database has been formed in this work using the QuikScat-observed vector wind patterns associated with the early stages of tropical cyclones that developed in the North Indian basin during the years 2000-2008. The prediction of possibility of cyclogenesis, in a given scene, is determined by matching it with all the archived scenes in a database using vector block matching algorithm. The system is predicted as developing, if its matching index exceeds the predetermined threshold value (0.5). The prediction of cyclo genesis can be made 24-96 h prior to its classification as a tropical storm by the Joint Typhoon Warning Centre. The probability of detection of the technique is determined using the equivalent of the Jackknife approach in the database as 0.93. The algorithm is tested with a continuous wind data of years 2007-2009 for active cyclone months (excluding the scenes archived in a database). All the 14 tropical disturbances that developed into tropical storms during the test period were predicted using the discussed method, and two false alarms were found.