Improving the skill of medium range ensemble rainfall forecasts over India using MoES grand ensemble (MGE)-part-I

Abstract

One of the key attributes of an ensemble prediction system (EPS) is the spread among the members. It plays a crucial role in conveying the uncertainty associated with the forecasted parameters. It is a quantitative measure of forecast uncertainty, provides a range of possible outcomes, and helps in the assessment of risk and decision making. Additionally, the spread can also serve as a diagnostic tool for assessing the reliability and variability among the ensemble members. If the spread is consistently narrow, it may indicate that the ensemble members are not diverse enough and the uncertainties may not be adequately captured resulting in sub-optimal decision making. In this study, the rainfall forecasts from two EPSs over India have been assessed during four monsoon seasons (2019–2022) with an aim to boost the ensemble spread by constructing a ‘Grand Ensemble’. The two high-resolution operational global EPSs of Ministry of Earth Science (MoES) in India are (i) National Centre for Medium Range Weather Forecasting (NCMRWF) EPS (NEPS) which has a 12 km grid, and 23 members and (ii) Global Ensemble Forecast System (GEFS) with a 12 km grid and 21 members. Both EPSs have been used for operational medium range forecasts out to Day-10 since 2018. The MoES Grand Ensemble (MGE) constructed by combining the two EPSs (NEPS & GEFS), features a higher spread and an improved Spread Vs Bias relationship compared to the constituent models. Further, the results indicate lowest CRPS in the MGE compared to the constituent EPSs, over the Indian land region. The improved performance of MGE is also demonstrated for moderate and heavy rainfall events using Brier Skill Score (BSS), Reliability Diagram and ROC curves.