Ensembles of retrospective 2-month dynamical forecasts initiated on 1May are used to predict the onset of the Indian summer monsoon (ISM) for the period 1989-2005. The subseasonal predictions (SSPs) are based on a coupled general circulation model and recently they have been upgraded by the realistic initialization of the atmosphere with initial conditions taken from reanalysis. Two objective large-scalemethods based on dynamical-circulation and hydrological indices are applied to detect the ISMonset. The SSPs show some skill in forecasting earlier-than-normal ISM onsets, while they have difficulty in predicting late onsets. It is shown that significant contribution to the skill in forecasting early ISM onsets comes from the newly developed initialization of the atmosphere from reanalysis. On one hand, atmospheric initialization produces a better representation of the atmospheric mean state in the initial conditions, leading to a systematically improved monsoon onset sequence. On the other hand, the initialization of the atmosphere allows some skill in forecasting the northward-propagating intraseasonal wind and precipitation anomalies over the tropical Indian Ocean. The northwardpropagating intraseasonal modes trigger themonsoon in some early-onset years. The realistic phase initialization of these modes improves the forecasts of the associated earlier-than-normal monsoon onsets. The prediction of late onsets is not noticeably improved by the initialization of the atmosphere. It is suggested that late onsets of the monsoon are too far away from the start date of the forecasts to conserve enough memory of the intraseasonal oscillation (ISO) anomalies and of the improved representation of the mean state in the initial conditions. © 2015 American Meteorological Society.

cited By 8