Significant Wave Height Retrieval Using XGBoost from Polarimetric Gaofen-3 SAR and Feature Importance Analysis

Empirical algorithms have become the mainstream of significant wave height (SWH) retrieval from synthetic aperture radar (SAR). But plentiful features multi-polarizations make selection input for empirical model a problem. Therefore, XGBoost models are developed and evaluated SWH polarimetric Gaofen-3 mode imagettes using SAR different polarization combinations, then importance each feature on is further discussed. The results show that reliability independently confirmed based collocations SAR-buoy SAR-altimeter. Moreover, combined-polarization achieve better performance than single-polarizations. In addition, to inversion not same. For example, normalized cross section (NRCS), cutoff wavelength (λc), incident angle (θ) more decisive contributions other features, while peak (λp) direction (φ) almost no contribution. Besides, NRCS cross-polarization has substantial effect, λc hybrid stronger one models.