A probabilistic precipitation nowcasting system constrained by an Ensemble Prediction System.

Abstract

The latest radar observations are widely used as the main source for deterministic precipitation nowcasting techniques, such as the Integrated Nowcasting through Comprehensive Analysis (INCA) system, in operational and research centers. However, this approach does not take into account other sources of information such as Ensemble Prediction Systems (EPS) from the Numerical Weather Prediction (NWP) models such as C‑LAEF (Convection-Permitting Limited Area Ensemble Forecasting). These systems, even though they do not outperform the Lagrangian extrapolation for several hours, can provide useful probabilistic information. INCA has a deterministic Quantitative Precipitation Nowcasting (QPN) module. For including uncertainties and errors in the prediction of precipitation, an ensemble generator is required. The created ensembles have to reproduce the temporal and spatial statistical properties of a real rainfall field. With this aim, a localized scheme known as Short Space Fast Fourier Transform (SSFT) is used. The information provided by C‑LAEF is introduced in the SSFT technique by using the Ensemble Kalman approach at grid scale and a Bayesian weighting to improve the ensemble mean with the latest information. Once the ensembles are generated with the information from C‑LAEF an empirical distribution matching is applied to avoid lose of variance and to keep the high rainfall values. The final EnQPF is verified in a probabilistic way for the period of July 2016. Several scores are used to evaluate the set of ensembles to determine whether the methodology produces sufficient uncertainty while keeping essential properties from the original field and the EPS.

Keywords

• probabilistic
• blending
• nowcasting
• seamless