Abstract
A simple algorithm is developed and tested to derive a regularly spaced wind field in a limited arm from simulated multi-orbit scatterometer data. The data are generated by sampling a time-varying known wind field, the 1000-mb FGGE data, from a simulated scatterometer. A simple assimilation technique is used to derive a regularly spaced wind field representation of two-day averages from the simulated data. This technique averages the generated scatterometer data in time and space and uses a low-pass filter in primarily the zonal direction. The resultant vectors were compared to the known two-day averages calculated from the FGGE data. To test the technique, synthetic noise was added to the generated data to simulate scatterometer inaccuracies in speed and direction.
Three cases were tested. In the first case, the simulated scatterometer data contained no noise or errors. The average magnitude of the difference wind field, known minus resultant, was less than the natural variability of the known windfield. In the second case, random white noise with standard deviation of 2 m s−1 (and then 4 m s−1) about a zero mean were added to each sampled vector component simulating inaccuracies in speed and direction inherent in scatterometer sampling. The added noise made little difference on the resultant wind field representation. In the third case, spatially correlated noise was added to each simulated swath simulating data with noise in both speed and direction to reflect errors due to sampling a wind field containing both synoptic and mesoscale components. The standard deviation of the spatially correlated noise was initially 2 m s−1 in each vector component. The average magnitude of the difference vectors increased slightly. In addition, when the noise was increased to 3 m s−1 in each component, the error did not increase significantly.
To test the results on another time period, a final case was run with a 24-hour time window. When spatially correlated noise and random white noise, each with standard deviations of 3 m s−1, were added to the sampled vector component the error did not increase significantly over the noise-free case.
This assimilation technique provides representations of two-day averages on a 100 km regularly spaced grid, and might therefore be applicable to large-scale ocean or atmospheric models. The results of this technique signify the level of importance of errors resulting from the application of assimilation schemes on scatterometer data. These errors appear to be more significant and limiting to the application of scatterometer data than errors from scatterometer inaccuracy.