Development of a single "all-weather" neural network algorithm for estimating ocean surface winds from the special sensor microwave imager

Brightness temperatures from the Special Sensor Microwave lmager(SSM/I) are being used routinely to estimate surface wind speeds over the open ocean. Recently, Stogryn, Butler and Bartolac(SBB) developed an SSM/I wind speed retrieval algorithm using neural networks (NNs) as a basis. As a starting point for the present work, we first reproduce the results of SBB for "clear" and "cloudy" conditions using the same data sets and identically-constructed NNs. We then proceed to extend the results of SBB by constructing and training a single NN which can be applied to both clear and cloudy conditions. This single network not only eliminates the problem of bridging the gap between the clear and cloudy regions, but it also provided the basis for developing a network which could be extended to atmospheric conditions where higher levels of moisture exist. As a result, NNs were trained to cover adverse atmospheric conditions considered by SBB to be beyond the region where useful retrievals could be obtained. Finally, an "all-weather" network was constructed and trained which yields a bias of -0.05 m/sec and an rms error of 1.65 m/sec over the entire range of conditions encountered in the data.