Monsoon Rainfall Prediction using Multi-Model Ensemble based Approach

A multi-model ensemble based approach has been used to develop a model for 1-15 day advance rainfall prediction during the monsoon months in 5 days pentad (i.e 1-5, 6-10 and 11-15 days).

Data used and Methodology: The model outputs from different global weather forecasting centers were obtained from THORPEX Interactive Grand Global Ensemble (TIGGE). The 5 days, 10 days and 15 days accumulated rainfall forecasts from four state-of-the-art operational global models, namely European Centre for Medium-Range Weather Forecasts (ECMWF), National Centre for Environment Prediction (NCEP), China Meteorological Administration (CMA), and United Kingdom Meteorological Office (UKMO) have been used. A database containing 1-15 days rainfall prediction during the monsoon months of the period 2008-2013 and observed rainfall from IMD have been prepared and used for training the model. The training has been performed using the similarity based approach which selects the 25 days of the similar conditions from the database. For each grid different weights were computed for different models using least square minimization technique.

The outputs of the model include the 1-5 days, 6-10 days and 11-15 days rainfall prediction (mm/day) at each grid on Indian land-mask. The anomaly from the last 30 years climatology has also been computed and presented.