Using a Bayesian joint probability approach to improve the skill of medium-range forecasts of the Indian summer monsoon rainfall

Abstract

Study region: Ganga, Mahanadi, Godavari, Narmada, and Tapti River basins of India. Study focus: The manuscript focuses on improving skills of the Indian summer-monsoon precipitation forecasts obtained from National Center for Medium-Range Weather Forecasting (NCMRWF) at both sub-basin and gridded scale. A well-established Bayesian Joint Probability (BJP) based statistical post-processing approach, operational in Australia, is used for the first time in India throughout diverse geographical extent. The work evaluates how the post-processor can be used in a summer-monsoon dominated region like India. The study informs whether annual or seasonal precipitation forecasts should be used as the length of data will play crucial role in both the cases. The spread-skill of the ensembles obtained from BJP approach and the NCMRWF is explored. New hydrological insights for the region: Introduction of the BJP-based post-processing approach in India could help the forecast community to implement more robust approach in improving the skills of the forecasts. Our results show that instead of using the data of whole year, only monsoonal precipitation forecasts are adequate to setup the BJP approach. The calibrated forecasts obtained using three years of hindcast and observations data at grids and at the centroid of 177 sub-basins are found to be more skillful. The calibrated forecasts can discriminate between extreme and low precipitation events, and have appropriate ensemble spread to capture precipitation peaks. This study presents a guideline for water managers and forecasters to apply BJP approach to improve the forecasts.

Keywords

• Bayesian Post-processing
• Precipitation forecasts
• Indian summer-monsoon
• Ensembles
• NCMRWF
• Sub-basin scale