FORECASTING THE SPATIAL VARIABILITY OF THE INDIAN MONSOON RAINFALL USING CANONICAL CORRELATION

✍ Scribed by PRASAD, K. D.; SINGH, S. V.

• Publisher: John Wiley and Sons
• Year: 1996
• Tongue: English
• Weight: 771 KB
• Volume: 16
• Category: Article
• ISSN: 0899-8418
• DOI: 10.1002/(sici)1097-0088(199612)16:12<1379::aid-joc98>3.0.co;2-j

✦ Synopsis

The application of a canonical correlation model to the long-range forecast of the spatial variability of the Indian monsoon (Juneeptember) rainfall has been demonstrated. The predictands used in the model are the summer monsoon rainfall of 29 contiguous meteorological subdivisions of India and the predictors are the 500 hPa ridge axis position over India for April, the Darwin surface pressure tendency (April-January), the sea-surface temperature of the central and eastern equatorial Pacific for the five successive months preceding the monsoon (January to May) and the rainfall of the southernmost subdivision of India (Kerala) for April. The model is developed on 30 years (19354968) of data and tested on 16 independent years thereafter.

The model demonstrates positive skill for the large contiguous meteorological subdivisions of India using the first canonical mode (found significant). The root-mean-square error and the absolute error between the observed and the predicted rainfall for different meteorological subdivisions are of the order of 1 cm. The high skill score (20.3) is found particularly for the meteorological subdivisions lying in west-central India.

The performance of the model appears to be better than that of the multiple regression model developed earlier by Prasad and Singh. The combined model (containing the first and the second canonical modes) appears to perform even better than that of the single model. These results, therefore, seem to be important in view of the long-range forecast of the spatial variability of the Indian monsoon rainfall.