Key analysis errors and airborne wind lidar observations

Abstract

Inaccurate initial conditions can produce significant forecast failures of numerical weather prediction models. An iterative algorithm that uses the adjoint forecast model and is aimed at minimizing the forecast error leads to the so-called key analysis errors (KAEs). Assuming that forecast error growth is dominated by the analysis error, the KAEs are assumed to represent that part of the analysis error that is primarily responsible for a poor forecast. Thus, KAEs should indicate how to improve an analysis. In addition, analysis errors can be identified by monitoring the differences of observations and analysis fields (analysis departures). The purpose of this study is to gain a further understanding of the structure of KAEs and to investigate the question to what degree KAEs are related to analysis errors. Airborne Doppler wind lidar (DWL) observations over the Northern Atlantic collected during the Atlantic THORPEX Regional Campaign (A-TReC) are analysed to evaluate these considerations. These observations were passively monitored and actively assimilated in experiments using the ECMWF global model to form the basis for the computation of analysis departures and analysis differences. Results confirm that analysis departures and KAEs optimized for both the Northern hemisphere and a predefined forecast domain are not correlated. Qualitative comparisons also show large differences in structure and magnitude of KAEs and analysis departures. Nevertheless, primarily in view of the different magnitude of KAEs and analysis departures, there is no basis for rejecting the hypothesis that KAEs are actually embedded in, and an important fast-growing part of the true analysis errors.