Seasonal forecasts of Indian summer monsoon rainfall using local polynomial based non-parametric regression model

Abstract

In this paper, details of new statistical models for forecasting southwest monsoon (June-September) rainfall over India (ISMR) and for northwest India summer monsoon rainfall (NWISMR) are discussed. These models are based on the local polynomial based non-parametric regression method.  Two predictor sets (SET-I & SET-II consisting of 4 and 5 predictors respectively) were selected for developing two separate models for making predictions in April and late June respectively. Another predictor set (SET-III) was selected for developing model for monsoon rainfall over NW India (NWISMR). Principle Component Analysis (PCA) of predictor data set was done and the first two principal components were selected for model development. Data for the period 1977-2005 have been used for developing the model and the Jackknife method was used to assess the skill of the model. Both the models showed useful skill in predicting ISMR and showed better performance than the model based on pure climatology.  The Hit scores for the three category forecasts during the verification period by April and June models are 0.65 and 0.66 respectively. Root Mean Square Error (RMSE) of these models during the verification period is 5.99 and 6.0% respectively from the Long Period Average (LPA) as against 10.0% from the LPA of the model based on climatology alone.  RMSE of the Northwest India model during the independent period is 11.5% from LPA as against 18.5% from the LPA of the model based on the climatology alone. Hit score for the three category forecast for NW India during the verification period is 0.55.