气候变化领域集成服务门户(CCPortal): Estimating sea surface swell height using a hybrid model combining CNN, ConvLSTM, and FCN based on spaceborne GNSS-R data from the CYGNSS mission

CCPortal

DOI	10.1007/s10291-024-01671-1
	Estimating sea surface swell height using a hybrid model combining CNN, ConvLSTM, and FCN based on spaceborne GNSS-R data from the CYGNSS mission
	Bu, Jinwei; Wang, Qiulan; Ni, Jun
发表日期	2024
ISSN	1080-5370
EISSN	1521-1886
起始页码	28
结束页码	3
卷号	28 期号:3
英文摘要	Compared to traditional swell height measurement methods, spaceborne global navigation satellite system reflectometry (GNSS-R) has many advantages, including remote sensing capabilities, global coverage, real-time monitoring, etc. It can provide wave observation data with high spatiotemporal resolution and is not limited by time, weather, and other conditions. Spaceborne GNSS-R provides a very effective method for estimating swell height, which can monitor and measure wave changes over a large area of the ocean surface in real time. This is of great significance for understanding the marine environment, climate change, and weather forecasting. However, there is relatively little research on the estimation of swell height using this technology, especially in the retrieval model of swell height. For this purpose, the article proposes a global ocean swell height retrieval method based on the convolutional neural network (CNN), convolutional long short-term memory (ConvLSTM) and fully connected network (FCN) hybrid deep learning model (i.e., CNN-ConvLSTM-FCN) for spaceborne GNSS-R. CNN-ConvLSTM-FCN model not only uses CNN to extract spatial features around specular points (SPs) from a two-dimensional (2-D) matrix of a single image (bistatic radar scattering cross-section (BRCS), effective scattering area, or power delay-Doppler map (DDM), but also uses ConvLSTM network to infer feature relationships and FCN to output estimated swell heights. The hybrid model improves its retrieval ability by simultaneously considering feature information related to time and space. The performance of the CNN-ConvLSTM-FCN model in retrieving swell height was tested using ERA5 and WaveWatch III swell height as reference data. The results show that when ERA5 data is used as a reference, compared to the empirical model method based on DDM average (DDMA) observable, the proposed CNN-ConvLSTM-FCN swell height retrieval model improves root mean square error (RMSE), correlation coefficient (CC), and mean absolute percentage error (MAPE) by 50.76%, 26.28%, and 29.63%, respectively. When WaveWatch III data is used as a reference, improvements in RMSE, CC, and MAPE are 51.09%, 25.35%, and 44.21%, respectively. The CNN-ConvLSTM-FCN model can demonstrate the ability of high-precision and high-resolution ocean swell height retrieval on a global scale, providing a new reference method for spaceborne GNSS-R ocean swell height estimation.
英文关键词	Global navigation satellite system-reflectometry (GNSS-R); Delay-Doppler maps; Observables; Swell height; Deep learning model
语种	英语
WOS研究方向	Remote Sensing
WOS类目	Remote Sensing
WOS记录号	WOS:001233943700001
来源期刊	GPS SOLUTIONS
文献类型	期刊论文
条目标识符	http://gcip.llas.ac.cn/handle/2XKMVOVA/306009
作者单位	Kunming University of Science & Technology; Yunnan University
推荐引用方式 GB/T 7714	Bu, Jinwei,Wang, Qiulan,Ni, Jun. Estimating sea surface swell height using a hybrid model combining CNN, ConvLSTM, and FCN based on spaceborne GNSS-R data from the CYGNSS mission[J],2024,28(3).
APA	Bu, Jinwei,Wang, Qiulan,&Ni, Jun.(2024).Estimating sea surface swell height using a hybrid model combining CNN, ConvLSTM, and FCN based on spaceborne GNSS-R data from the CYGNSS mission.GPS SOLUTIONS,28(3).
MLA	Bu, Jinwei,et al."Estimating sea surface swell height using a hybrid model combining CNN, ConvLSTM, and FCN based on spaceborne GNSS-R data from the CYGNSS mission".GPS SOLUTIONS 28.3(2024).