一种改进海面风速反演的分类神经网络方法

AN IMPROVED ALGORITHM FOR SEA SURFACE WIND SPEED RETRIEVAL OF A CLASSIFIED NEURAL NETWORK

为了提高使用SSM/I资料反演全球海面风速的精度,发展了一个新型的神经网络方法。在这个方法中,使用高风速、中、低风速状态和天气状态分类的方法分别训练神经网络,然后根据其类别的不同使用不同的神经网络计算风速。此方法较好地去除了由于高风速和云天天气状态下训练样本数据的缺少所产生的误差,改进了在高风速状态下反演风速值比实际风速偏低的情况,使得反演的高风速值被校正到了正常位置。本方法反演海面风速的值与浮标实测风速值之间的均方根误差达到1.60m/s。

A new neural network algorithm is developed to improve the retrieval precision of the global sea surface wind speed from the SSM/I brightness data. At first, the data in different conditions, such as high-speed and low-speed winds, and clear and cloudy weather, are used to train different neural networks. Then these neural networks are used in- dependently to retrieve the sea surface wind speed. Compared with the buoy wind, the RMS （root mean square） error of the retrieving is about 1.60m/s. This method reduces the bias resulted from the lack of quality data in high-speed wind, and cloudy weather on the neural network algorithm.