A multimodel ensemble (MME) technique for cyclone track prediction over the North Indian Sea

Abstract

A multimodel ensemble (MME) technique for predicting track of tropical cyclones over the North Indian Sea has been proposed. The technique is developed applying multiple linear regression procedure. Parameters of the ensemble technique are determined from the forecast datasets on the tracks of tropical cyclones over the North Indian Sea during the year 2008-2009. The parameters selected as predictors are: forecast latitude and longitude positions at 12-hour interval up to 72-hours forecast of five operational numerical weather prediction models. The dynamical models included for development of the ensemble technique are: (i) forecasts from the European Centre for Medium-Range Weather Forecasts (ECMWF), (ii) the National Centers for Environmental Prediction Global Forecast System (NCEP), (iii) the MM5 model, (iv) the Quasi-Lagrangian model (QLM) and (v) the model of Japan Meteorological Agency (JMA). A collective bias correction is included in the ensemble technique in which a multiple linear regression based minimization principle for the model forecast position against to the observed position is applied. These bias factors are described by separate weights at every 12-hours interval up to the 72-hour forecasts for each of the member model. When the technique is tested with the independent samples, forecast skill of the MME technique is found to be reasonably good. The average error ranges from of the order of 74 km to 290 km for forecasts up to 72-hour. Performance of the MME technique shows that there are skill improvements up to 30 km for the position errors over the best model at 72-hour forecast. The forecast skill of the MME technique for forecasts up to 72-hour also shows an improvement as compared to the forecasts from member models and the simple ensemble mean (ENM).