Bias correction of seasonal ensemble rainfall forecasts from global dynamical seasonal forecast systems for Mekong river delta

Abstract

    In recent years, seasonal forecasting using dynamical models that simulate the coupled atmosphere, ocean and land surface system has become common in operational weather forecasting centres around the world. The dynamical forecasting models not only provide forecasts with a longer lead time (up to 9 month in advance), but also provide an ensemble of forecasts instead of a single-value forecast. However, spatial resolutions of the forecasts are typically coarse, and the forecasts often suffer from substantial systematic biases as compared to observations. Therefore, this study evaluates the potential use of raw (without bias correction) and statistically calibrated seasonal ensemble rainfall forecasts using empirical quantile mapping bias correction (QM) approach. The evaluation is illustrated for onemonth lead seasonal rainfall forecasts obtained from the European Centre for Medium-Range Weather Forecasts (ECMWF) seasonal forecast system 5 for forecasting daily rainfall in 1-month lead time at stations across the Mekong river delta. The reforecast data based on 24 years with 21 ensemble members are used for the evaluation purpose. Evaluation results are conducted in a cross-valiation setting based on several deteministic and probabilistic verification metrics. The results showed that the statistically calibrated reforecasts using QM approach significantly improve upon the raw reforecasts.