The paper by Riishøjgaard et al. (2004) investigates the assimilation and impact of prospective Doppler wind lidar (DWL) line-of-sight (LOS) single-perspec-tive winds in meteorological analysis. It is argued that single-component wind observations are far less effec-tive in reducing wind analysis error than vector wind information. This work has relevance because the pros-pects are good that space-based DWL instruments will provide accurate wind profiles of single-perspective LOS wind profile measurements in the future. Riishøj-gaard et al. rightly argue that the usefulness of such winds needs to be well addressed in the design phase of space missions. The forthcoming European Space Agency Atmospheric Dynamics Mission (ADM), called Aeolus, is referred to in this context. The Riishøjgaard et al. study is carried out in an idealized and very simplified framework. Our concerns are 1) that the simple framework poorly represents the characteristics of a state-of-the-art global data assimi-lation system for numerical weather prediction (NWP) and 2) that the DWL scenarios that are discussed have abundant and unrealistic coverage and quality. As such, their conclusions may be misleading for, and contribute little toward, the critical design consider-ations for an affordable space-based DWL. The results (and the quality of the analyzed wind fields) could be far more realistic and, in our view, far more favorable for LOS winds in a more carefully designed experiment. The NWP analysis problem would be severely under-determined if it were based on the observations alone. To overcome this problem, data assimilation typically combines the information provided by the relatively sparse observations with a short-range forecast on a dense grid (Daley 1991). Because the NWP model state is poorly observed, it is critical that local observation increments are carefully distributed spatially in a wider area. This process is done based on statistical knowl-edge of the background error structures. In a four-dimensional variational data assimilation (4DVAR) analysis system, information on the temporal evolution of the model state is also exploited. Around any local observation, information on the multivariate spatial correlation of the background errors, as represented in the background-error covariance matrix B, is used to provide a spatially coherent update of the model atmo-spheric state. For LOS wind analysis, the B covariance structures are crucial in both spatially interpolating the observed wind component and inferring the spa-tial pattern of the unobserved component of wind as well as the associated temperature and pressure incre-ments. The design of the B matrix and the sampling strategy of the DWL space mission are the two most important factors that determine the impact of the data, both in real application and within the simplified framework of Riishøjgaard et al. In the case in which B is poor, this would generally result in spatially poor analyses, espe-cially when the observations are sparse or when one or several analysis variables are unobserved. In a rela-tively dense observation network, on the other hand, the multivariate spatial structures associated with many observations will overlap and the effect of an imperfect B will diminish (by oversampling). Our specific comments are in two areas. The first is that the Riishøjgaard et al. paper uses a synthetic vortex