Ensemble error diagnosticsof hybrid data assimilation in the Korea integrated model

Numerical weather prediction models inevitably produce forecast errors due to structural errors from discretization and limitations in physical parameterization processes. Accurately representing and incorporating these forecast errors is a key element for improving data assimilation performance. The background error covariance is operated in a hybrid form, combining a static climatological component with a flow-dependent ensemble component. The appropriate combination of these two components, and how well they reproduce the actual error characteristics, serves as a critical prerequisite for enhancing forecast accuracy.

This study aims to improve the hybrid data assimilation system of the Korea Integrated Model KIM by advancing ensemble-based background error diagnostics and enhancing its application to data assimilation performance. To this end, the study seeks to overcome the limitations of static climatological background error covariance and to develop diagnostic and adjustment techniques that can effectively reflect the spatiotemporal variability and uncertainty of forecast errors. In particular, ensemble error diagnostic information will be utilized to analyze error structure characteristics, adjust weighting factors by region and altitude, and conduct observation sensitivity analysis, thereby enabling adaptive optimization of the data assimilation system. Ultimately, this study aims to establish a practical technological foundation that can improve the efficiency of observation utilization in operational environments and contribute to enhancing the accuracy of short-and medium-range forecasts.

The forecast error characteristics and ensemble spread of the Korea Integrated Model KIM and the ECMWF operational model IFS were analyzed through intercomparison experiments. The root-mean-square error RMSE and ensemble spread were calculated based on the differences between the forecast fields and observations, and these results were used to diagnose the background error covariance (B).
 The background error covariance was constructed by separately calculating the static component (Static B) based on climatological statistics and the flow-dependent component (Ensemble B) derived from ensemble forecasts. These two components were combined using optimal weighting factors to generate the hybrid background error covariance Hybrid B. In this study, a series of experiments were conducted using the KIM-based hybrid data assimilation system, including sensitivity tests on the weighting factors, error diagnostics.

 

Key Word : Hybrid data assimilation, Background error covariance, Ensemble spread