A Predictability study using geostationary satellite wind observations during NORPEX

There is much debate over the design and mix of observing systems that will provide the maximum benefits for numerical weather forecasts. Evidence suggests that the early stages of error growth in most forecasts are dominated by relatively few growing structures, and that preferentially reducing analysis errors that project onto these structures can lead to significant improvements in forecast skill. Mathematically, these structures tend to be highly correlated with the locations of the leading singular vectors (SVs) of the linearized forecast model.

The North Pacific Experiment (NORPEX-1998) provides the first opportunity to assess the impact of both in-situ objectively targeted and experimental remotely sensed observations on numerical weather forecasts over western North America. High-density GOES/GMS geostationary satellite wind data, provided at 6-hr intervals during NORPEX, have become an important new contributor to the observing network over oceanic regions. In addition to the benefit to operational forecasts, these data provide an excellent resource for studying error growth, predictability and the design of future observing systems and data assimilation strategies.

Experiments with the Navy Operational Global Atmospheric Prediction System (NOGAPS) confirm that the large positive impact of the geostationary satellite winds results mainly from repeated reduction of analysis errors that project onto the leading SVs in successive assimilation cycles. These errors account for only a small fraction of the total analysis error and, during NORPEX, were confined mostly to the middle and lower troposphere with localized maxima over the central Pacific. These errors do not necessarily coincide with the locations of the largest analysis errors. Experiments in which the satellite information is retained only at prescribed vertical levels in the analysis confirm that observations in the middle and lower troposphere account for most of the forecast impact. Implications of these results for the design of future observing systems, including strategies for targeted observing, are discussed.