Optimising impact of radar reflectivity and lidar backscatter obs on analysis

This report provides a description of updates to the ECMWF data assimilation system to improve the impact of cloud radar reflectivity and lidar backscatter on the analysis. Updates include a new triple-column approach for accounting for sub-grid cloud inhomogeneity and a model for observation error correlations. The report begins with a brief description of the data assimilation system at ECMWF to be used for direct assimilation of cloud radar and lidar observations. Next, to study and optimize the impact of cloud radar and lidar observations on the analysis, 4D-Var assimilation experiments have been performed using Cloudsat cloud radar reflectivity and CALIPSO lidar backscatter observations. The first experiments were performed using the ECMWF model cycle CY43R1, for which the system of cloud radar and lidar observations was prepared at first. The results obtained indicate that 4D-Var analyses get closer to these observations and the impact on other assimilated observations is mainly neutral. Generally, the impact of the cloud radar reflectivity is larger than of the lidar backscatter. The report then summarizes updates of the data assimilation system for cloud radar and lidar observations in the more recent model cycle CY46R1. As part of the updates, the observation operator is adapted to model sub-grid condensate variability and a framework for accounting for the vertical correlation in observation error is developed. The impact of these observations in the updated system is positive; the updated observation operator improves the analysis fit to the observations. The fit to AMSU-A observations is also improved compared to both CY43R1 and an experiment using the old single-column version of the observation operator