IITM offers new model for more precise rain forecast

While Krishnan had made this information public at the recent Indian Science Congress in Thiruvananthapuram,Dr A Sahai,Scientist E at IITM and programme manager development of extended range prediction system,had given a presentation to this effect at an IMD annual conference last month. Earlier,a paper outlining this technique was published in Journal of Atmospheric Sciences.

Sahai said a new technique known as self-organising map (SOM) has been introduced to study the monsoon intra-seasonal oscillations (ISOs). “Vigorous ISOs in the form of active and break episodes are an integral part of the Indian summer monsoon. Unlike the linear techniques used so far that could identify only the ensemble mean phases of the monsoon ISO,the SOM is capable of identifying various shades of each ensemble mean phase,including their evolutionary history. This feature has opened up the possibility of a nonlinear extended-range prediction of monsoon rainfall.”

The technique was found to be reliable for prediction of rainfall over central India four pentads in advance. Sahai said it was also used to predict the 2009 monsoons and they were able to do for three events but could not do so only for one. According to him,the prediction scheme is based on the following premise: The SOM classification on the training period extracts 15 patterns and their evolutionary history and stores them in the “reference” vectors. Time histories of the patterns are saved on the dates clustered at each node. For prediction on a given date,a ‘forecast’ vector is created with current and past data for nine days for all the large-scale variables.

This essentially contains the pattern and its evolutionary history at the initial time,the hypothesis being that if they could find an analogue of this pattern and its evolution in the past from the ‘reference’ vectors corresponding to different nodes,they could make a four-pentad prediction from the evolutionary history of the analogue.

A major strength of the model is that it can easily be used for real-time extended-range prediction of monsoons. The limitations,however,are that it is better at predicting breaks spells rather than active ones and there is inter-annual variability of predictability of the spells. Sahai said they were now working on removing these limitations as well.