Gap-free GNSS-R wind field reconstruction: A neural mapping scheme and initial validation

Spaceborne Global Navigation Satellite System Reflectometry (GNSS-R) has been widely demonstrated as an effective method for ocean wind speed retrieval. This study explores the feasibility of using track-wise GNSS-R wind products to generate gap-free wind fields. A physics-informed neural mapping sc...

ver descrição completa

Detalhes bibliográficos
Autores: Du, Hao, Fablet, Ronan, Nguyen, Thi Thuy Nga, Li, Weiqiang, Cardellach, Estel, Chapron, Bertrand
Tipo de documento: artigo
Estado:Versão publicada
Data de publicação:2026
País:España
Recursos:Consejo Superior de Investigaciones Científicas (CSIC)
Repositório:DIGITAL.CSIC. Repositorio Institucional del CSIC
OAI Identifier:oai:digital.csic.es:10261/425000
Acesso em linha:http://hdl.handle.net/10261/425000
https://api.elsevier.com/content/abstract/scopus_id/105030115102
Access Level:Acceso aberto
Palavra-chave:Global navigation satellite system reflectometry (GNSS-R)
Neural mapping scheme
Ocean wind speed
Spatiotemporal evolution
Tropical cyclones
Descrição
Resumo:Spaceborne Global Navigation Satellite System Reflectometry (GNSS-R) has been widely demonstrated as an effective method for ocean wind speed retrieval. This study explores the feasibility of using track-wise GNSS-R wind products to generate gap-free wind fields. A physics-informed neural mapping scheme, 4DVarNet, is adapted to reconstruct wind fields. Results indicate that the root mean square errors (RMSEs) of the 1-hour, 3-hour, and 6-hour 4DVarNet winds are 1.13 m/s, 1.16 m/s, and 1.24 m/s compared to European Center for Medium-Range Weather Forecast (ECMWF) ERA5 wind products, while 1.40 m/s, 1.41 m/s, and 1.48 m/s are referred to Advanced Microwave Scanning Radiometer-2 (AMSR2) all-weather winds. Spatial and temporal error analyses further confirm the robustness of 4DVarNet-derived winds, with daily RMSEs remaining below 1.6 m/s. Error decomposition reveals discrepancies between ECMWF and GNSS-R winds, which may support future recalibration of GNSS-R wind products or enhancements to ECMWF forecasts. A case study of Super Typhoon Surigae proves that 4DVarNet winds closely align with the International Best Track Archive for Climate Stewardship (IBTrACS) track data. The reconstructed winds detect the peak intensity temporally consistent with IBTrACS data, whereas ECMWF forecasts exhibit a two-day lag. Moreover, asymmetries in Tropical Storm Kompasu are observed, with the radius of maximum wind (R<inf>max</inf>[jls-end-space/]) over the Northeast quadrant 38% larger than that over the Northwest quadrant. Despite the absence of background wind field inputs, 4DVarNet effectively learns wind patterns from ECMWF data and integrates GNSS-R observations to generate gap-free wind mappings, exhibiting strong agreement with ECMWF wind fields. The reconstruction performance is degraded at high winds due to the underestimation of referenced ECMWF ERA5 winds and the small quantity of observations. This limitation could be alleviated through denser GNSS-R observations from multiple missions such as Fengyun-3, Tianmu-1, recently launched HydroGNSS, etc., and other training references with more high winds for improving the representation of 4DVarNet at high winds.