Statistical prediction of movement of cyclonic storms and depressions over Bay of Bengal through LOESS technique

Abstract

Accurate cyclone track prediction has always been a challenge to the operational weather forecaster. An attempt has been made in this study for prediction of the cyclone track by employing three statistical techniques, viz., analogue, analogue-cum-regression and Locally weighted Scatterplot Smoothing (LOESS). Track data of cyclonic disturbances which formed and moved in the Bay of Bengal during the period 1961-2008 has been used. A statistical model has been developed for comparison of the accuracy levels of track prediction through these three techniques and results have been discussed. It has been observed that the average track forecast error of 147 km calculated by LOESS technique is minimum compared to those obtained from analogue and analogue-cum-regression techniques. In the case of recurved systems also, the forecast error obtained through LOESS is minimum. Heidke Skill Score, Peirce Skill Score and Proportion Correct have been calculated for Along-Track and Cross-Track components which indicate better accuracy and superiority of LOESS technique over the analogue and analogue-cum-regression techniques. Other skill score indices have also been computed and results presented.