IMD Forecast For Monsoon rainfall (June-September 1999) 
(Issued on the 25 May 1999)

Around this time, the India Meteorological Department(IMD) issues long range seasonal forecast for the ensuing southwest monsoon rainfall spanning the four month period (June to September) over the country as a whole. IMD has developed models like Parametric, Power regression, Multiple regression, and Dynamic Stochastic Transfer models for issuing long-range seasonal forecasts.

FORECASTS FOR MONSOON RAINFALL OVER THE COUNTRY AS A WHOLE

Parametric Model

This model provides qualitative forecast and utilizes the signals from 16 antecedent global and regional land-ocean-atmospheric parameters. Analysis of the signals from these parameters indicates that this year,
69% of them are favourable for a normal monsoon.  The seasonal rainfall or the country as a whole is thus expected to be normal (defined as  90 to 110% of the long period average value).

Power Regression Model

This model provides quantitative forecast and is based on the physical relationship of monsoon rainfall with 16 different individual parameters.  According to this model, the total amount of rainfall over the country as a whole during the monsoon season from June to September 1999 is likely to be 108% of its Long Period Average (LPA), with an estimated model error of  ± 4 %.

Multiple Regression Model

This model provides quantitative forecast and is based on correlations between monsoon rainfall and certain global and regional meteorological parameters. The model utilizes 8 predictors and the forecast based on this model indicates that the monsoon rainfall during June to September 1999 is likely to be  110% of  its LPA.

Dynamic Stochastic Transfer Model

This model considers the atmosphere as a dynamic system. It utilizes regional parameters and climatic forcings as the inputs and yields monsoon rainfall over the country as a whole as output. The forecast
based on this model indicates that the monsoon rainfall over the country as a whole during June to September 1999 is likely to be 104% of its LPA.

Power Transfer Model

The model has been developed based on Dynamic Stochastic Transfer and Power Regression techniques. The model uses 12 regional and global parameters as the inputs. The forecast based on this model indicates that the monsoon rainfall during June to September1999 over the country as a whole is likely to be 110% of its LPA.

Principal Component Regression Model

The model has been developed based on the Principal Component Analysis technique and makes use of 8 regional and global parameters. The forecast from this model indicates that the monsoon rainfall during June to September 1999 over the country as a whole is likely to be 113% of its LPA.

Hybrid Artificial Neural Network Model

The model has been developed based on the Principal Component Analysis of 8 regional and global parameters and Artificial Neural Network technique. The forecast from this model indicates that the monsoon rainfall during June to September1999 over the country as a whole is likely to be 110% of its LPA.

Model Performance

In recent years, long range monsoon rainfall forecast based on the Power Regression Model has been found to be comparatively more accurate. Therefore the result of the Power Regression Model has been given weightage in this forecast.

Summary of the Long range forecast for 1999 monsoon over the country as a whole

The long range forecast for the 1999 Southwest Monsoon formulated on the basis of Parametric and Power Regression Model is given below:

a) In 1999, the Southwest Monsoon rainfall over the country is likely to be normal, (the normal being defined as rainfall within ( 10 % of its long period average value) making 1999 the eleventh normal monsoon year in succession.

b) Quantitatively, the  rainfall for  the  country  as  a  whole for the entire monsoon  season  (June to September) is likely to be 108 % of its long period average value with the model error limit of ± 4 %.

FORECASTS FOR MONSOON RAINFALL OVER THE HOMOGENEOUS REGIONS OF INDIA

 This year, in addition to the long range forecasts of monsoon rainfall over the country as a whole, IMD is issuing long range forecasts for three homogeneous regions of India, viz., Northwest India, Peninsula and
Northeast India.

Forecasts for Northwest India

 For the purpose of this forecast, Northwest India is taken to consist of Uttar Pradesh, Haryana, Chandigarh and Delhi, Punjab, Himachal Pradesh, Jammu & Kashmir and  Rajasthan.

Five different types of models viz., Power Regression with 7 parameters, Multiple Regression with 6 parameters, Dynamic Stochastic Transfer with East Coast India temperature and climatic forcings as
inputs, Principal Component Regression with 8 parameters and Neural Network with 8 parameters have been used.  The forecasts obtained from different models are 122 % of LPA from the Power Regression model, 124 % of LPA from the Multiple Regression model, 103 % of LPA from the Dynamic
Stochastic Transfer model, 116 % of LPA from the Principal Component Regression Model and 111 % of LPA from the Neural Network Model.

 In recent years, long range monsoon rainfall forecast based on the Neural Network model has been found to be comparatively more accurate. Thus based on the Neural Network Model, the monsoon rainfall over Northwest India for the entire monsoon season (June to September), 1999 is likely to be 111 % of its long period average.

Forecasts for Peninsula

For the purpose of this forecast, Peninsula is taken to consist of Gujarat, Madhya Pradesh, Maharashtra, Andhra Pradesh, Karnataka, Kerala, Tamil Nadu and Lakshadweep.

Five different types of models viz., Power Regression with 12 parameters, Multiple Regression with 6 parameters, Dynamic Stochastic Transfer with Central India temperature and climatic forcings as inputs,
Principal Component Regression with 8 parameters and Neural Network with 8 parameters have been used.  The forecasts obtained from different models are 106 % of LPA from the Power  Regression model, 101 % of LPA from the Multiple Regression model, 95 % of LPA from the Dynamic Stochastic Transfer model, 116 % of LPA from the Principal Component Regression Model and 114 % of LPA from the Neural Network Model.

In recent years, long range monsoon rainfall forecast based on the Neural Network model has been found to be comparatively more accurate. Thus based on the Neural Network Model, the monsoon rainfall over
Peninsula for the entire monsoon season (June to September), 1999 is likely to be 114 % of its long period average.

Forecasts for Northeast India

For the purpose of this forecast, Northeast India is taken to consist of Andaman and Nicobar Islands, Arunachal Pradesh, Assam and Meghalaya, Nagaland, Manipur, Mizoram, Tripura, Sikkim, West Bengal, Orissa and Bihar.

Five different types of models viz., Power Regression with 9 parameters, Multiple Regression with 5 parameters, Dynamic Stochastic Transfer with East Coast India temperature and climatic forcings as inputs, Principal Component Regression with 8 parameters and Neural Network with 8 parameters have been used.  The forecasts obtained from different models are 103 % of LPA from the Power  Regression model, 107 % of LPA from the Multiple Regression model, 107 % of LPA from the Dynamic Stochastic
Transfer model, 90 % of LPA from the Principal Component Regression Model and 98 % of LPA from the Neural Network Model.

In recent years, long range monsoon rainfall forecast based on the Neural Network model has been found to be comparatively more accurate. Thus based on the Neural Network Model, the monsoon rainfall over
Northeast India for the entire monsoon season (June to September), 1999 is likely to be 98 % of its long period average.

SUMMARY OF THE  FORECASTS

The long range forecasts for the 1999 Southwest Monsoon over the country as a whole and the three broad homogeneous regions of India, viz., NW India, Peninsula  and NE India are given below.

a) In 1999, the rainfall over the country as a whole for the entire southwest monsoon season (June to September) is likely to be normal thus making the year 1999 eleventh normal monsoon year in succession. The normal is defined as rainfall within ( 10 % of its long period average.

Quantitatively, the rainfall over the country as a whole for the entire southwest monsoon season is likely to be 108 % of its long period average with an estimated model error of ± 4 %.

b) Over the broad homogeneous regions of India, the rainfall for the entire   southwest monsoon season of 1999 is likely to be 111 % of its LPA over NW India, 114 % of the LPA over Peninsula and 98 % of the LPA over NE India with an estimated model error of ± 8 %.

This information was produced by the Indian Meteorological Department and more details can be found by contacting directly:

The Director General
India Meteorological Department
Mausam Bhavan
Lodi Road
New Delhi 110 003
India