부산대학교 도서관

A new temporally evolving quotient on the Ensemble of Data Assimilations (EDA) technique for estimating background error covariances has been developed for the Copernicus European Regional Re‐Analysis (CERRA). The B‐matrix is modeled on a bi‐Fourier limited area model. Background errors are assumed isotropic, homogeneous and non‐separable. Linearized geostrophic and hydrostatic balances are incorporated as multivariate relationships, coupling vorticity, and geopotential extended to mass‐wind and specific humidity fields via the f‐plane approximation. The EDA comprises two main pools: seasonal and daily. The seasonal component comprises winter and summer EDA forecast differences at reanalysis resolution (5.5 km). The new time quotient function temporally changes the mixture of differences from each season, to make up the seasonal component. The daily component is an 11 km moving 2.5 days average changing in real‐time. Subsequent B‐matrix computation sees the ingestion of forecast differences from both components, with a fixed split of 80%–20% seasonal‐daily, every 2 days. The sourcing of these forecast differences from both seasonal and daily sources is in continuous temporal flux therefore. We consider a case study to illustrate the potential of estimating weather regime change using CERRA‐EDA with varying proportions of seasonal‐daily mixing, while including settings used for CERRA production. Our case study shows that the most influential factors are differences in observation networks between the given years, their spatial distribution across the CERRA domain, and the proportion of seasonal‐daily split. It is shown that our method provides improvement over a static B‐matrix. Plain Language Summary: The Copernicus European Regional Re‐Analysis (CERRA) is a "second look" at the evolution of weather over Europe and North Africa between 1984 and 2021. This "re‐analysis" merges state‐of‐the‐art observations with regional weather model data. The re‐analysis is performed long after the fact, after previous predictions, using data at that time. The quality of the reanalysis is highly dependent on the errors in the previous forecast (B). No computer on Earth is capable of storing the mammoth B‐matrix explicitly. A subset of prognostic variables (specific humidity, temperature and surface pressure, vorticity and divergence) are used to represent the weather, and simplistic assumptions of their error structures are made in order to quantify B. To this end, their statistical tendencies are obtained using forecast differences from an Ensemble of Data Assimilations (EDA). Each EDA member is an independent analysis using the same data but perturbed observations. The structure of B assumes: linearized meteorological balances with respect to the Earth's rotation and vertical motion, and the positional and directional distribution of errors are horizontally uniform (equally spread), while the vertical is governed by the vertical pressure profile. This roughly resembles a deformed tear‐drop. In this paper, we augment the established EDA method in estimating our B matrix with our assumptions to demonstrate its large‐scale regional estimation capability against the EDA method without our new augmentation. Key Points: Newly developed functionality for continuously temporally updated Ensemble of Data Assimilations (EDA) system to estimate B‐matrix for a high resolution European reanalysisB‐matrix successfully estimates weather regime change but observation network and proportions of EDA forecast difference sourcing are vitalDemonstrated improvements in statistical profiles of B‐matrix with some improvements in forecast scores and Copernicus European Regional Re‐Analysis (CERRA) system performance [ABSTRACT FROM AUTHOR]