Using SAR Satellite Imagery, HF Radar, and Machine Learning to Estimate Significant Wave Height Along the New Jersey Coast

Abstract:

The ESA’s Sentinel synthetic aperture radar (SAR) satellites are capable of measuring sub-mesoscale phenomena with high resolution (<10 m) and with large-scale coverage of the earth’s surface. Sentinel-1, which takes images of the New Jersey Coast every 12 days, is used in the present study to collect a rich database of the nearshore ocean dynamics. In this study, we analyze the pixel data of the images (measurements of the sea surface backscatter) to estimate significant wave height in the coastal area of New Jersey. We focus on the methods of linear regression, kriging, machine learning regression, and convolutional neural network (CNN) to translate the sea surface roughness data into the significant wave height. High-frequency (HF) radar data of the ocean (aka CODAR) is used as ground truth to calibrate and validate the wave height estimate. The results show that kriging is comparable to the deep learning technique reported in other literature and CNN has a potential to further improve the accuracy of wave height estimates. More details will be reported in the presentation.

This study developed a reliable algorithm to enhance the current capability to process the satellite data and demonstrate a new possibility to monitor the coastal environment. The collected data will help further our understanding of the wave spectrum in a coastal environment and the data can support other research in the related topics, e.g. the interaction of waves and ice sheets, wetlands, shorelines, wind farms and aquaculture.