Addressing horizontal correlations of SEVIRI observation errors in a simplified framework of the AROME‐France data assimilation system

Abstract

In variational data assimilation (DA), high‐resolution satellite observation errors exhibit significant inter‐channel, temporal, and horizontal correlations. The vast number of observations and their uneven distribution complicate modeling of the observation‐error covariance matrix R . Consequently, strategies like thinning, variance inflation, and superobbing are commonly employed to mitigate horizontal correlations. These techniques lead to a sharp decrease in the number of observations assimilated into operational systems. However, discarding observations that hold valuable information results in a suboptimal analysis. To support high‐resolution DA, we propose to take into account horizontal observation‐error correlations in the convective‐scale model Application de la Recherche à l'Opérationnel à Méso‐échelle (AROME‐France). To address this issue, we choose to focus on infrared spectrum observations from the Spinning Enhanced Visible and InfraRed Imager (SEVIRI), which are well‐known to exhibit strong horizontal correlations. We model those correlations using a double exponential model and we incorporate it in a simplified framework that uses the three‐dimensional variational (3DVar) version of the DA system of the AROME‐France model, performing the analysis at a lower horizontal resolution and using a reduced set of observations. Experiments were conducted by eliminating thinning and introducing horizontal error correlations. The impact of these changes is assessed by examinating numerical performances and analysis of increments in a specific case study. Incorporating horizontal error correlations does not degrade the numerical performance of the Lanczos minimization algorithm, which remains capable of converging within a finite number of iterations. This approach also results in consistent increments in both spatial structures and amplitudes, which are enhanced at smaller spatial scales and reduced at larger scales, aligning with the anticipated changes due to the modification in R . Consequently, this study illustrates that high‐resolution DA, alongside the integration of horizontal correlations in SEVIRI observation errors, can extract small‐scale information effectively from dense observations.