@article{ZHANG2025, 
author = {Yun ZHANG and Sheng XIAO and Lifei JIANG and Wanting MENG and Shuhu YANG and Yanling HAN},
title = {Significant wave height retrieval model of CYGNSS based on multivariate machine learning},
year = {2025},
journal = {Journal of Beijing University of Aeronautics and Astronautics},
volume = {51},
number = {5},
pages = {1503-1513},
keywords = {significant wave height, CYGNSS, satellite-based GNSS-R, machine learning, convolutional neural network},
url = {https://www.sciopen.com/article/10.13700/j.bh.1001-5965.2023.0265},
doi = {10.13700/j.bh.1001-5965.2023.0265},
abstract = {The cyclone global navigation satellite system (CYGNSS) provides high-quality global navigation satellite system reflectometry (GNSS-R) data that can be reliably used for the retrieval of significant wave height (SWH). Due to the high dynamic nature of CYGNSS, the received signal is easily affected by environmental factors, and the complexity of sea conditions makes it difficult for simple models to accurately retrieve SWH. To address the above issues, this article proposed an SWH retrieval model based on multivariate machine learning. According to the mechanism of wave formation and the analysis of the correlation between CYGNSS parameters and SWH, relevant parameters were selected, and three training schemes were designed, involving five parameters, nine parameters, and 17 parameters, respectively. Random forest (RF) and convolutional neural network (CNN) were used to train and verify the retrieval model, and the SWH retrieval results were compared with the reference values of the European Centre for Medium-Range Weather Forecasts (ECMWF). The best retrieval model among them was the 17-parameter CNN retrieval model, with root mean square error（RMSE）was 0.184 0 m and  R2= 0.948 5. Compared with the 17-parameter CNN retrieval model, the 9-parameter CNN retrieval model reduced training time by 24% and has minimal accuracy loss. However, the 9-parameter retrieval model performed poorly in terms of generalization evaluation compared to the 17-parameter retrieval model. To improve the generalization ability of the model, wind speed was added as a parameter to the 17-parameter retrieval model, resulting in a 17 + 1-parameter generalization model. The best generalization model among them was the 17 + 1 parameter RF generalization model, with RMSE was 0.497 1 m and  R2= 0.584 6. This effectively proves that the model proposed in this article has good potential in SWH retrieval.}
}