P7 A combined bias-correction and probability calibration  ensemble post-processing scheme for the Army WRF Ensemble- RTFDDA system

Roux, Gregory, Yubao Liu, Jason Knievel, Luca Delle Monache, Tom Hopson, National Center for Atmospheric Research, Scott Halvorson, U.S. Army Research Laboratory

The NCAR mesoscale ensemble data assimilation and prediction system "Ensemble-RTFDDA" (real time four dimensional data assimilation) is a WRF-based multi-physics and multi-perturbation approach ensemble system. A 32-member E-RTFDDA system with three nested domains at grid intervals of 30, 10 and 3.33 km has been running with four FDDA and forecast cycles per day for 2015 and 2016, producing 4D continuous ensemble analysis and 48h forecasts. This study analyzes the E-RTFDDA outputs to assess overall ensemble performance and features, and evaluates the additional skills from an ensemble calibration process based on an analog-based bias-correction and a quantile-regression algorithm, that are part of the operational system.

 The bias correction scheme is applied at each member independently, and consist of running a post-processing algorithm inspired by the Kalman Filter through an ordered set of analog forecasts (ANKF). The analogs are defined as a past prediction that matches selected features of the current forecast. A quantile regression (QR) is then applied on the bias-corrected ensemble members, leading to ensemble forecasts with both good reliability and sharpness. This statistical method is applied for surface temperature, wind speed and relative humidity forecasts during 2015 and 2016 and verified against the observations at selected stations located in Utah, Arizona and New Mexico.