Piecewise log-normal modeling of Sentinel-1 co-cross polarization coherence and application to wind field Bayesian inversion

Accurate estimates of kilometer-scale ocean surface wind fields is critical for many applications, including weather forecasting, marine navigation, offshore wind energy development, and coastal risk management. Synthetic Aperture Radar (SAR) plays a key role in retrieving wind speed by analyzing changes in sea surface roughness, described through Geophysical Model Functions (GMFs). However, the inverse problem of wind retrieval from SAR data is under-determined, requiring ancillary information from Numerical Weather Models (NWP) to resolve ambiguities in the wind direction. The Co-Cross-Polarization Coherence (CCPC) has recently emerged as a valuable SAR-derived observation for improving wind direction estimates. While an initial GMF for CCPC has been proposed, it proved inadequate for high wind speeds. We present PGMF-2, a refined GMF for CCPC, ensuring reliable performance up to approximately 25 m/s by modeling its wind speed components as log-normal functions. The need for crosstalk calibration is mitigated by a burst-based processing. Monte Carlo simulations confirm that the unique odd-symmetry of the CCPC and the different impact of the wind direction modulation on its real and imaginary parts complements the Normalized Radar Cross Section (NRCS) to estimate the wind direction. This combination enables wind direction retrieval without NWP priors, even outperforming NWP-based inversions between 7 and 15 m/s. The Mean Absolute Error on direction reaches as low as 25 ∘ in favorable (in particular between 30 ∘ –90 ∘ and 270 ∘ –330 ∘ of wind direction relative to the satellite antenna, where the NRCS and the CCPC have overlapping wind direction variability) high-wind conditions. Co-observations with the HY-2B scatterometer, supplemented by a case study on Storm Alex (2020), serve as a proof of concept of the Bayesian scheme on real-world observations, successfully retrieving the typical flow organization of the cyclonic system without reliance on NWP priors.