Direct 4D‐Var assimilation of space‐borne cloud radar reflectivity and lidar backscatter. Part I: Observation operator and implementation

Abstract

The direct assimilation of space-borne cloud radar and lidar observations into a global numerical weather prediction model has not previously been attempted for several reasons. Firstly, the modification of a data assimilation system to handle space-borne profiling observations is a technical challenge. Secondly, the relationship between model-scale control variables and the relatively narrow footprint of the radar and lidar instruments were thought to be unrepresentative and too nonlinear for a variational assimilation system that is based on assumptions of linearity. However, motivation was provided when previous experiments assimilating cloud radar and lidar profiles showed, using an intermediary step to generate pseudo-observations of temperature and humidity, that there was potential for the observations to improve both the analysis and the subsequent forecast. This article presents the developments made to facilitate the direct assimilation of cloud radar and lidar observations into the four-dimensional variational assimilation (4D-Var) system of the European Centre for Medium-Range Weather Forecasts (ECMWF). A review of the observation operators shows how they have been optimised for data assimilation with an emphasis on efficiency, consistency and differentiability. A key part of this work is the specification of a new fully flow-dependent characterisation of the observation error. By taking an error inventory approach, we account for different sources of error in a physical way. Also, to avoid degrading the analysis, careful screening, quality control and bias correction is necessary and is described herein. Finally, an initial assessment of the impact of the observations is achieved through single-analysis tests that show that the analysis fit to the new observations is improved. In-depth results demonstrating the impact of these observations are shown in the second part of this two-part series.

Direct 4D‐Var assimilation of space‐borne cloud radar and lidar observations. Part II: Impact on analysis and subsequent forecast

“…Obviously, increasing the observation errors means decreasing the impact of these observations in the analysis system. These results indicate that perhaps observation errors as defined by Fielding and Janisková (2017,2020) give too much weight to these observations. As a consequence, the analysis might be driven too close to these observations instead of using a well‐balanced combination of information provided by all other observations assimilated in the system.…”

“…First, the impact of scaling the observation error is evaluated (Figures 8 and 9) over a one-month period. For this assessment, cloud radar and lidar observations were used with the observation errors as defined in Fielding and Janisková (2020), as well as 1.5 or 2 times the defined values. Inflating observation errors for cloud radar reflectivity and lidar backscatter to account for the known correlations leads to relatively better FG fits to other assimilated observations.…”

“…Using the full system of regularly assimilated observations at ECMWF, several assimilation experiments have been performed. Short experiments, only over the 12‐hr period (i.e., the current length of 4D‐Var assimilation window at ECMWF) between July 31, 2007 2100 UTC and August 1, 2007 0900 UTC (situation 2007080100 ), are summarized in Fielding and Janisková (2020). The longer experiments, run over the period of one to three months between August 1 and October 31, 2007, are part of this paper.…”

“…Cloud radar and lidar observations averaged over a model grid‐box at approximately 36 km have been used with: 1err – single observation errors as defined in Fielding and Janisková (2017,2020); 1.5err – observation errors one and half times the defined value; 2err – observation errors twice the defined value. …”

“…This paper is the second of two papers investigating the feasibility and impact of direct assimilation of observations from space-borne cloud radar and lidar to NWP. In the first part (Part I; Fielding and Janisková, 2020), the details of the necessary adaptions to the assimilation system at the European Centre for Medium-Range Forecasts (ECMWF) are described. In this part (Part II), the new developments are used to assimilate several months of CloudSat radar reflectivity and CALIPSO lidar backscatter and to assess their longer-term impact on both the analysis and subsequent forecast.…”