The Study On Oceanic Wind Field Retrieve Technique Based On Neural Networks Of Microwave Scatterometer

Ocean surface wind field is an important parameter of ocean environment. Spaceborne microwave scatterometer is an active microwave sensor which measures the backscatter coefficient of ocean and terrain surfaces. It can achieve a large scale measurement in all weather and all time conditions. The scatterometer has been playing a very important role in global ocean wind vector inversion due to its large swath and different azimuth angle combinations. The products have become very important data for analysis and prediction of global weather and ocean current.The scatterometer products can usually be classified to two levels. The level one data is about the backscatter coefficient and the level two data is mainly about the wind vectors.The correspondence relationship between backscatter coefficient and sea surface wind field established a semi-empirical geophysical model function(GMF) of microwave scatterometer. Traditional wind retrieval methods are based on GMF. Under high wind conditions, the backscattering coefficient prone to saturation, so it is difficult to describe the correspondence between the backscattering coefficient and the wind field.The neural networks are used for microwave scatterometer data processing to retrieve wind fields, especially for data gained by the scatterometer onboard HY-2A satellite(HSCAT) under high wind speed conditions. The retrieval of wind speeds are based on Back Propagation(BP) neural network, while multiple solutions of wind direction inversions are realized by Mixture Density Network(MDN) neural network. During the process, Gaussian kernel function is employed. The wind fields used in network training is from corresponding European Centre for Medium-Range Weather Foresting(ECMWF).It is proved that wind fields retrieved in this paper could get results meeting the accuracy requirement for HSCAT by comparison with ECMWF wind fields. Results are also compared with the L2 B wind field products distributed by the National Satellite Oceanic Application Service, it is shown that the method in this paper gave results with closer values to the wind field of ECMWF than L2 B products.