AMBIGUITY REMOVAL AND ASSIMILATION OF SCATTEROMETER DATA

The ERS-1 scatterometer has proved to be a source of high quality ocea n surface wind data, but a problem remains, namely the dual directiona l ambiguity of the solutions. An ambiguity removal scheme, called PRES CAT, is described based on our experience (a) that information on wind direction retrieval skill is an important input to ambiguity removal, (b) that wind-vector filtering is beneficial compared to wind-directi on filtering, and (c) that already meteorological forecast information enables us to remove correctly approximately 95% of all ambiguities. The performance of the ambiguity filter is very good compared with oth er operational ambiguity removal schemes. Furthermore, a statistical i nterpolation analysis system called 'buddy' check is used effectively to identify and remove the few (approx. 0.1%) wrongly selected solutio ns. Assimilation of scatterometer winds has a beneficial impact on ana lyses and short-range forecasts, probably mainly from improvements on the subsynoptic scales. On the wider temporal and spatial scales, scat terometer winds were also found beneficial, but only in the absence of satellite temperature soundings (SATEMs). In assimilation experiments in which the latter were included, the scatterometer did not provide a beneficial impact on the medium-range forecast. Moreover, the conven tional observations, including SATEMs are shown to have adverse effect s on the surface-wind analysis. We believe that both the redundancy an d the adverse effects on the surface-wind field are explained by the r igid formulation of the 6-hour-forecast error structure; the forecast error is assumed to be now-independent, and information on the special meteorological conditions in the atmospheric planetary boundary layer is lacking. To make observational systems more useful and complementa ry for numerical weather prediction the effects of the structure funct ions have to be investigated more precisely In an adaptive four-dimens ional variational assimilation scheme the effect of the assumptions on forecast-error structure will be less. We show that, in a variational framework, scatterometer backscatter measurements are difficult to as similate directly. Instead, we derive and illustrate an alternative pr ocedure to assimilate retrieved winds rather than backscatter measurem ents.