Inversion Of GNSS-R Sea Surface Wind Speed Based On Neural Network Model

The sea surface wind field is the main energy source for upper parts of the ocean movement,which is an important physical parameter of oceanography.And it has a great impact on regional and global climate change,so accurate prediction of the sea surface wind field is conducive to the development of global weather forecasting and the prevention of ocean waves.Therefore,it is very meaningful to carry out research and monitoring of the sea surface wind field.For the wind speed in the range of 3-18 m / s,the RMSE of the current inversion method based on the GNSS-R technology is about 2.2m / s.However,there are shortcomings of the method such as incomplete extraction of physical quantities,complicated models and so on.In order to shorten the time consumption and reduce the RMSE,deep learning neural algorithms and GNSS-R technology are combined to invert the sea surface wind speed in this investigation.In this paper,the original sample set is obtained by space-time matching of TDS-1(Tech Demo Sat-1)satellite data and ECMWF(European Center for Medium-Range Weather Foresting)data.The sampling algorithm and the normalization algorithm are used to process the original data set,and then divide it into a training set and a test set.After training and testing the BP model and CNN model with the above two data sets,we obtain two prediction models with DDM data map as input and sea surface wind speed as output.The RMSE of the BP model is in the range of 1.75-2.01m/s while the RMSE of the CNN model is in the range of 1.50-1.82m/s.Compared with the RMSE of two traditional sea surface wind speed inversion methods—empirical function method and waveform matching method,and the RMSE of two machine learning sea surface wind speed inversion models—random forest and SVR,the results of the neural network models are much better in inversion accuracy,which are improved by about 20%.The results of this paper verify that the neural network models can be used for the sea surface wind speed inversion based on the GNSS-R technology.What is more,it is verified that the CNN model has the advantages in modeling time,efficiency of inversion and the accuracy of the prediction,which will promote the application of deep learning technology in the field of sea surface wind speed inversion based on the GNSS-R technology and contribute to the research of the GNOSII on the Fengyun3 E satellite to be launched by China soonly.