Neural Network Retrieval of Sea Surface Wind Speed from Advanced Microwave Scanning Radiometer-E Data

A neural network wind speed (WS) retrieval algorithm was developed using 6.9 and 10.7, as well as 36.5 GHz for both horizontal and vertical polarized brightness temperature (Tb) of Advanced Microwave Scanning Radiometer-E (AMSR-E) aboard AQUA. The artificial xneural networks (ANN) technique is employed to find the transfer function relating the input AMSR-E six channels Tb and output (WS) parameter. Input data consist of nearly 12 months (January 2005 - December 2005) of AMSR-E observed brightness temperature and surface marine observations of WS from National Data Buoy Center (NDBC) and Tropical Atmosphere Ocean Project (TAO). The performance of the algorithm is assessed with independent surface marine observations. The retrieval results demonstrate that the combination bright temperatures of lower frequencies such as 6.9 and 10.7 GHz, as well as higher frequencies such as 36.5 GHz from AMSR-E as input parameters provides reasonable estimates of wind speed. The root mean square (rms) error between estimated WS from AMSR-E observations and the buoy measurements is 1.53 m/s.