Prediction of Indian Summer Monsoon Rainfall by Phase-Space Reconstruction Model

Abstract:

The prediction of the Indian summer monsoon rainfall at intraseasonal time scale is investigated in this study. The summer monsoon exhibits intraseasonal oscillations (ISOs) with different periods. The leading ISO, with a period of 45 days, is presumably related to the Madden-Julian Oscillation. The ISOs have large-scale spatial structure and propagate northeastward and northwestward. A prediction model, based on some basic results of nonlinear dynamical systems theory, is constructed to predict the monsoon rainfall. An equivalent phase space of reduced dimension can be reconstructed from a long time-series of a single or a few variables of the dynamical system. In such a phase space, the trajectory of the dynamical system can be examined to search for nearest neighbors. An ensemble of such nearest neighbors and their subsequent evolution are used to construct the prediction model. In some respects, this method is similar to Lorenz’s analog method. The reduced phase space is reconstructed by using a limited number of eigenmodes obtained from multi-channel singular spectrum analysis of the rainfall over the monsoon region. For this purpose, the daily gridded rainfall over India for the period 1901-2010 is used. These eigenmodes represent the ISOs and seasonally persistent modes. The prediction of the monsoon rainfall by this model is compared with the retrospective forecasts made by NCEP CFSv2 and other S2S models.