Deep learning for inversion of significant wave height based on actual sea surface backscattering coefficient model

Abstract

Ocean waves are complex systems with the contributions of wind waves and swells. The study on interaction mechanism between electromagnetic wave and actual sea surface is of significant importance in ocean remote sensing and engineering application, which is also helpful in the prediction and inversion of wave information. In this paper, an efficient model for estimating backscattering coefficient is built, considering the characteristics of the wind-wave regime based on the inverse wave age. The backscattering coefficient results have been verified by comparing with the data collected in Lingshan Island during the period of October and November 2014 at low grazing angles and the Ku-band measurements at moderate grazing angles. The results indicate perfect agreement (within about 2 dB) with field data. Deep learning is an excellent method that can be used not only for classification but also for inversion and fitting of non-linear functions. In order to simulate the application of actual radar detection and inversion technology, the inversion of significant wave height from actual sea surface backscattering coefficients train data sets has been performed by using deep learning technology. The accuracy of 99.01% has been achieved under the condition of three hidden layers and iterating 100 times. The root mean square errors of the test data sets are less than 0.10, which indicates that deep learning is available in the inversion of significant wave height.