Improving Retrieval of Sea Surface Roughness From Landsat-8 OLI and Sentinel-2 MSI Imagery by Eliminating for Ambiguity Effects

Abstract

Sea surface roughness (SSR) is a common metric used to characterize the condition of sea surfaces. Sun glitter (SG) is the reflected sunlight off the fluctuating sea surface, with its distribution and intensity markedly influenced by SSR. However, the estimation of SSR through single-angle SG images may produce ambiguity, due to the assumptions applied in the underlying statistical model describing the orientation of the sea surface facets as a function of wind speed. In this study, we investigate the causes of SSR ambiguity and its distribution characteristics via simulation experiments. Based on these findings, we propose a method for identifying critical points of ambiguity and eliminating them for SSR estimation, leveraging single-angle SG images. This approach enables pixel-scale adaptive quantitative inversion of SSR. To validate our model, publicly available global high-resolution SG images (near-infrared (NIR) and shortwave infrared (SWIR) bands) captured by the Landsat-8 Operational Land Imager (OLI) and Sentinel-2 MultiSpectral Instrument sensors are utilized for SSR estimation. Subsequently, the modal values of SSR are expressed as wind-speed values for accuracy assessment. The estimated wind speeds exhibit good agreement with reanalysis wind data from the European Centre for Medium-Range Weather Forecasting (ECMWF) and buoy wind data from the National Data Buoy Center (NDBC). Meanwhile, the comparison results of the two bands are also very consistent, indicating that the method is stable for application on NIR and SWIR bands. The model’s efficacy is demonstrated using typical oceanographic scenarios from two regions: the Timor Sea and the English Channel. The resulting SSR images depict detailed oceanographic features with minimal noise, showcasing quantitative SSR changes at the pixel level. These outcomes underscore the effectiveness of the proposed method in resolving SSR ambiguity inherent in widely used optical remote sensing images such as Landsat-8 and Sentinel-2, thereby enhancing SSR retrieval accuracy.