| The complex relationship between electromagnetic waves backscattered from an ocean surface with geophysical variables such as surface wind stress, significant wave height, wave slope and age, and sea surface temperature is not well understood. Since radar scatterometers are being employed to remotely measure ocean surface wind speeds and directions from space, a relationship is needed between the measured normalized radar cross section (sigma-0) and inferred ocean wind vectors. All scatterometers to date have used empirically derived "geophysical model functions" (GMF's) to relate measured sigma-0's to simultaneous neutral stability wind vectors at 19.5 m or 10 m height.; In this work, we apply a multilayer perceptron neural network to empirically derive a C-band GMF based on the European Remote Sensing Satellite, ERS-1 scatterometer measurements with coregistered wind vector components produced by a global numerical weather prediction model.; The derived neural net based GMF, NN-CMOD, is shown to estimate the ERS-1 measured sigma-0's more accurately than the operational C-band GMF, CMOD4, for all ranges of wind speeds, wind directions, and incidence angle. Further, the NN-CMOD sigma-0 estimates are shown to be unbiased with small random errors when compared to the measured sigma-0's by ERS-1. Use of this GMF should significantly improve wind retrievals from the ERS scatterometer. |
