Abstract
The global navigation satellite system reflectometer (GNSS-R) can improve the observation and inversion of mesoscale by increasing the spatial coverage of ocean surface observations. The traditional retracking method is an empirical model with lower accuracy and condenses the Delay-Doppler Map information to a single scalar metric cannot completely represent the sea surface height (SSH) information. Firstly, to use multi-dimensional inputs for SSH retrieval, this paper constructs a new machine learning weighted average fusion feature extraction method based on the machine learning fusion model and feature extraction, which takes airborne delay waveform (DW) data as input and SSH as output. R2-Ranking method is used for weighted fusion, and the weights are distributed by the coefficient of determination of cross validation on the training set. Moreover, based on the airborne delay waveform data set, three features that are sensitive to the height of the sea surface are constructed, including the delay of the 70% peak correlation power (PCP70), the waveform leading edge peak first derivative (PFD), and the leading edge slope (LES). The effect of feature sets with varying levels of information details are analyzed as well. Secondly, the global average sea surface DTU15, which has been corrected by tides, is used to verify the reliability of the new machine learning weighted average fusion feature extraction method. The results show that the best retrieval performance can be obtained by using DW, PCP70 and PFD features. Compared with the DTU15 model, the root mean square error is about 0.23 m, and the correlation coefficient is about 0.75. Thirdly, the retrieval performance of the new machine learning weighted average fusion feature extraction method and the traditional single-point re-tracking method are compared and analyzed. The results show that the new machine learning weighted average fusion feature extraction method can effectively improve the precision of SSH retrieval, in which the mean absolute error is reduced by 63.1 and 59.2% respectively, and the root mean square error is reduced by 63.3 and 61.8% respectively; The correlation coefficient increased by 31.6 and 44.2% respectively. This method will provide the theoretical method support for the future GNSS-R SSH altimetry verification satellite.
Highlights
Sea Surface Height (SSH), as an important ocean parameter, plays an important role in establishing global ocean tide models, observing large-scale ocean circulation, and monitoring global sea level changes (Zhang et al, 2020)
This paper first evaluates the sea surface height (SSH) retrieval performance of regression methods commonly used in machine learning, such as linear regression model {Lasso regression (Zou 2006), Ridge regression (Hoerl and Kennard 2000), Support Vector Machine regression (Keerthi et al, 2014) (SVR) and ensembled tree regression model [XGBoost (Luo et al, 2020), LightGBM (Luo et al, 2020), Random Forests (Liu et al, 2020)]}
Random Forests, XGBoost and Ridge models with better SSH retrieval performance and lower correlation are used for model fusion, which further improve the SSH retrieval accuracy
Summary
Sea Surface Height (SSH), as an important ocean parameter, plays an important role in establishing global ocean tide models, observing large-scale ocean circulation, and monitoring global sea level changes (Zhang et al, 2020). The prediction result is compared with the sea surface height SSH provided by the DTU15 validation model, using Mean Absolute Difference (MAD), Root Mean Square Error (RMSE) and Pearson Correlation Coefficient (CC) (Garrison, 2016) evaluate the effectiveness of the model. The new GNSS-R SSH retrieval model based on machine learning fusion model and feature optimization used the information of the entire delay waveform for height inversion. The application of the new machine learning weighted average fusion feature extraction method effectively improves the accuracy of SSH retrieval, in which the mean absolute error (MAD) is reduced by 63.1 and 59.2% respectively, and the root mean square error (RMSE) is reduced by 63.3 and 61.8% respectively; The correlation coefficient (CC) increased by 31.6 and 44.2% respectively
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a callDisclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.
