Calibrating Subseasonal to Seasonal Precipitation Forecasts to Improve Predictive Performance

Subseasonal to seasonal (S2S) precipitation forecasts encompassing lead times ranging from two weeks to three months can bridge the gap between weather and seasonal forecasts. For practical applications, calibration is a necessary step to improve predictive performances of raw forecasts from S2S models. This paper illustrates the calibration of the S2S precipitation forecasts by a case study of the European Centre for Medium-Range Weather Forecasts. The Bernoulli-Gamma-Gaussian model and quantile mapping are used to calibrate raw S2S forecasts. The results of three catchments in China show that raw forecasts exhibit reasonable correlations with observed precipitation but suffer from considerable biases and unreliable spreads, leading to negative skill. The Bernoulli-Gamma-Gaussian model overall outperforms the quantile mapping in correcting biases and improving reliability and skill. For S2S precipitation, the Bernoulli-Gamma-Gaussian model can serve as an effective tool to generate skillful ensemble time series forecasts for the early warning of rainfall-induced geohazards.