Development of F-ResNet for Spaceborne GNSS-R Sea Surface Height Measurement From CYGNSS

The diversity and complexity of the environment, such as wind speed, wave height, sea clutter, etc., are the main reasons for the current limitation in the spaceborne Global Navigation Satellite System-Reflectometry(GNSS-R) sea surface height(SSH) accuracy. This letter proposes a modified ResNet architecture for sea surface altimetry, named fractional Fourier transform-ResNet (F-ResNet), using filtered Delay-Doppler Maps(DDMs) and other information from Cyclone Global Navigation Satellite System(CYGNSS) Level 1 data. The DDMs are filtered by fractional Fourier transform(FrFT) to keep the doppler and delay information in the five maps according to the different fractions. F-ResNet is based on convolutional layers for the filtered DDMs, along with fully connected layers for processing ancillary technical and higher-level input parameters. The validity of our proposed method is demonstrated using the Jason-3 data set for the whole year 2020. The RMSE of the SSH using F-ResNet is improved by 15.2% compared to other machine learning methods, which can achieve 1.17 meters. And the influence of different signal-to-noise ratios on prediction accuracy is analyzed.