Preliminary and final long range forecast for seasonal monsoon rainfall over India

Abstract

India was the first country to begin systematic work on Long Range Forecasting (LRF) of seasonal rainfall, particularly the south-west monsoon rainfall during June to September which accounts for 80% of the country's annual rainfall and has a dominant impact on its socio-economic environment. The first operational LRF for seasonal monsoon rainfall over the country as a whole was issued by the India Meteorological Department (IMD) on 4 June 1886. In India, the correlation method was introduced in the first decade of this century, as an objective LRF technique. Subsequently, several forecast techniques were developed for improving the accuracy of the forecast. Presently, four LRF models, namely Parametric, Power Regression, Multiple Regression and Dynamic Stochastic Transfer are being used to issue a rainfall forecast over the whole country. The first three models use correlation as the basis while the fourth assumes that the atmosphere is a dynamic system which transfers various inputs into output such as rainfall. These models generally use different combinations of 16 land–ocean–atmospheric parameters (predictors) which appear to have physical links with the evolution of subsequent monsoon over India.For India as a whole, the monsoon rainfall forecast is issued in two stages. In the first stage, a preliminary forecast is issued in qualitative form towards the end of March or early April. In the second stage, the final forecast is issued in quantitative form towards the end of May. In this paper, LRF models which are used for issuing the final monsoon rainfall forecast are discussed in detail. For the first time, stability of the monsoon relationship with different parameters is analysed. It shows that in order to overcome epochal fluctuations in the relationship, the forecast model should be revised every year. An attempt is also made to analyse the influence of individual parameters on model forecast. Final quantitative forecasts, based on these models, have all proved to be accurate during the past successive 4 years (1989–1992).An attempt is also made to develop a few LRF models that can be used for issuing a preliminary monsoon rainfall forecast, in qualitative form, at least 2 months ahead of the final forecast. For this purpose, eight parameters out of the 16 which are available by the end of March have been utilised for developing preliminary Parametric, Power Regression, Multiple Regression and Dynamic Stochastic Transfer models. Results obtained from all the four preliminary models are analysed. Recognising the limitations of statistical models for producing very early monsoon forecast, it has been found appropriate that the preliminary monsoon forecasts are issued to users in qualitative form and not in quantitative form. While formulating a qualitative forecast, inferences obtained from different models are given due weightage. Preliminary forecasts issued based on these models have proved accurate during the past 5 successive years (1988–1992) and have become popular among different kinds of users.