Augmented Flow-Dependent Perturbations to Mitigate Sampling Errors: Experiments for a Regional Application of the NOAA Unified Forecast System

Abstract Ensemble data assimilation in modern regional weather prediction models often faces challenges in managing sampling errors due to small ensemble size and model errors. Increasing the ensemble size is not often feasible because of the computational resources needed for implementing models with large, high-resolution domains. The current study introduces a strategy for mitigating issues of sampling error in operational data assimilation systems by supplementing ensemble-estimated error covariance needed for data assimilation with perturbations sourced from a global model. This approach resembles hybrid data assimilation methods that use a weighted sum of two background error covariances to mitigate sampling deficiency from ensembles. Specifically, we enhance the NOAA Hurricane Analysis and Forecast System (HAFS) by incorporating an ensemble Kalman filter (EnKF) with augmented perturbations that utilize flow-dependent perturbations from the Global Data Assimilation System (GDAS) to reduce sampling errors. Additionally, we implement a localized particle filter (LPF) with augmented perturbations, which is not part of the original HAFS data assimilation system, and conduct a comparative analysis of the EnKF with augmented perturbations, the LPF with augmented perturbations, and a hybrid filter that combines the two methods. Experiments that rely on augmented perturbations from GDAS for updating 40-member ensembles are found to produce substantial improvements over benchmark experiments. The new approaches are evaluated over multiweek cycling data assimilation experiments focusing on Hurricanes Laura and Marco from August 2020.