Evaluating Statistical Bias Correction Methods for Improving NCMRWF Unified Model Operational Forecasts in Support of iFLOWS-Mumbai

Abstract

The study addresses systematic errors in the forecasts of a numerical weather prediction (NWP) model, particularly in critical variables such as precipitation that have significant societal implications. Correcting these errors is imperative for enhancing the accuracy of the NWP model in flood risk management decisions. This research evaluates various quantile mapping (QM) bias correction approaches, employing empirical and parametric methods, to rectify precipitation forecasts generated by the National Centre for Medium-Range Weather Forecasting (NCMRWF) Unified Model, with a focus on Mumbai during the southwest monsoon season. The chosen location is important to the Integrated Flood Warning System (IFLOWS), a key program under the Ministry of Earth Sciences, Government of India, providing early warnings and decision support during flooding. The precipitation forecasts, calibrated using various QM techniques over the Mumbai region, demonstrated significant improvements compared to the raw forecasts, especially for higher thresholds. Particularly noteworthy is the better performance of parametric methods, specifically the Generalized Pareto parametric QM, in surpassing raw forecasts, establishing greater effectiveness for regional-scale flood warning applications during extreme rainfall events. This study highlights the efficacy of QM methodologies in treating precipitation forecasts, contributing valuable insights to the advancement of urban flood modelling, and associated decision-making processes.