Assessing the Impact of Wave Age on GNSS-R L1 products and L2 Wind Speed Inversion

    This study delves into the impact of wave age on GNSS-R L1 observations, with a specific emphasis on Delay-Doppler Map (DDM) and Normalized Bistatic Radar Cross Section (NBRCS) signal strength, as well as its consequent effects on L2 wind speed inversion. Conventionally, DDMs have been simplified in wind speed inversion algorithms to solely represent sea surface roughness influenced by wind speed. However, this research underscores the significant, yet often overlooked, role of sea surface roughness induced by wave characteristics in the reflection of microwave signals. Wave age, denoting the delay of waves under wind influence and the proportion of swell to wind waves, is identified as a crucial factor affecting sea surface roughness and, thereby, the scattering properties of ocean surface signals.

    Recognizing the intricate physical correlation between wave age and sea surface wind speed, along with the multi-variable dependency inherent in GNSS-R observation theory, the study employed machine learning techniques to assess the extent of wave age's influence throughout the observation to wind speed inversion process. For this purpose, the Resnet18 deep convolutional neural network was chosen for its adept handling of the complex features present in DDM data, which can be considered as images. This choice was anchored in Resnet18’s robust feature extraction abilities and its proven track record in tasks requiring high-accuracy image classification.

    This study utilized CyGNSS L1 data along with corresponding ECMWF wind speed and sea surface parameter data for specific time and location. To conduct a comparative analysis, two methodologies were used: a traditional geophysical model function (GMF) developed by our self and machine learning. Preliminary testing indicated a marked enhancement in the accuracy of wind speed predictions when incorporating wave age into both GMF and machine learning approaches. The root mean square error notably decreased from approximately 1.8-2 meters per second to about 1.1 meters per second. The study also found a link between wave age and NBRCS intensity distribution, noting that larger inverse wave ages correlate with more signal scattering and weaker signal strength, underlining the vital impact of wave age on NBRCS intensity distribution.