A probabilistic approach for SeaWinds data assimilation: an improvement in the nadir region

Scatterometer sea-surface wind observations are being successfully assimilated into Numerical Weather Prediction (NWP) models. However, the impact of such observations often critically depends on their quality. In this respect, the quality of the winds retrieved from the new SeaWinds scatterometer (onboard QuikSCAT) depends on the subsatellite cross-track location. In particular, the poor azimuth separation or diversity between views (beams) in the nadir region results in poor quality winds.

The standard wind retrieval procedure consists of considering the Maximum Likelihood Estimator (MLE) cost function local minima as the potential (ambiguous) wind solutions that are used by the Ambiguity Removal (AR) procedure to select the observed wind. In the QuikSCAT nadir region, where the cost function minima are broad, the use of the standard procedure results in arbitrary and inaccurate winds. A scheme, which allows more ambiguous wind solutions when the retrieval results in broad cost function minima, i.e., a multiple solution scheme (MSS), is proposed as alternative to the standard procedure. The probability of every ambiguous solution of being the “true” wind is empirically derived and used in the AR procedure to make the scheme flexible enough to accept many wind solutions. The AR scheme uses National Centre for Environmental Prediction (NCEP) 24-hour forecasts as NWP background.

A comparison between the standard wind retrieval and the MSS procedures at 100-km resolution is then performed, using the European Centre for Medium-range Weather Forecast (ECMWF) First Guess at Appropriate Time (FGAT) model winds for validation. The MSS turns out to be more in agreement with ECMWF than the standard procedure, especially at nadir. Moreover, it shows more spatially consistent and realistic winds by more effectively exploiting the information content of the observations. In fact, AR results in winds with generally higher a priori probability and generally good agreement between a priori probability and AR selection. As such, the MSS concept is potentially beneficial for QuikSCAT data assimilation purposes in NWP. Finally, the lack of an effective Quality Control (QC) at 100-km resolution, essential for assimilation purposes, is discussed and several methods are recommended for further investigation.