Adjoint-based forecast sensitivity applied to observation-error variance tuning

Abstract

This article deals with the estimation of observation errors for Infrared Atmospheric Sounding Interferometer (IASI) radiances. We investigate the possibility of combining an established method based on diagnosing errors from innovation statistics (the so-called Desroziers method) with guidance obtained from adjoint sensitivity tools (which aim to minimise short-range forecast error). In a test version of the European Centre for Medium-range Weather Forecasts (ECMWF) 4D-Var assimilation system which uses insitu observations and IASI as the only source of satellite data, it is found that tuning the IASI observation errors with a combined approach is beneficial (compared to using the innovation-based method alone). Fits to data within the analysis are improved and forecasts initiated from the retuned analyses also show a moderate increase in skill.