Surface-Dependent Correlated Infrared Observation Errors And Quality Control In The Fv3 Framework

Data assimilation relies on the accurate representation of both observation- and background-error statistics. In this study, the Desroziers diagnostic is used to estimate error covariance matrices for CrIS and IASI brightness temperature observations over land and sea surfaces. Infrared observations have different error characteristics over different surface types, which warrant separate treatments. Over land, correlations between surface-sensitive channels tend to be much larger than over sea, and there may even be strong anti-correlations between certain channel groups. After reconditioning the matrices, they are used in several forecast experiments in the National Centers for Environmental Prediction's Finite Volume Cubed-Sphere Dynamical Core Global Forecast System (FV3GFS). With the incorporation of the new covariance matrices, upgrades to the infrared quality control procedures are re-examined. In particular, stricter cloud detection which relies on the new, smaller, observation errors is studied. It is found that for IASI, accounting for error correlations generally has a positive forecast impact which is enhanced by the new cloud detection scheme. For CrIS, successfully accounting for error correlations necessitates stricter cloud detection. In the next implementation of the FV3GFS (version 16), correlated error for IASI and CrIS will become operational.