A Principal Component Dimensional Reduction Involved Fast Prediction Model for Sea Surface Scattering Based on Improved Wen's Spectrum

The electromagnetic (EM) scattering characteristics of sea surfaces are influenced by both radar parameters and marine environmental parameters, resulting in significant complexity and randomness. Existing sea scattering coefficient estimation models, both EM simulation methods and machine learning-based prediction models, usually overlook the impact of wave parameters. In order to consider wave parameters comprehensively, a wind-wave mixed Wen's improved spectrum is proposed in this study, which overcomes the limitations of traditional wind-driven wave spectra. Based on the proposed spectra and the two-scale model (TSM), a multi-parameter sea scattering coefficient dataset comprising wind parameters, wave parameters, and radar parameters was obtained. Further, a novel machine learning-based model that combines principal component dimension reduction with least squares support vector regression (LSSVR) is presented for accurate prediction of the sea scattering coefficient. This model can improve the overfitting issues in the existing fast prediction models that emerged while dealing with multi-parameters of sea scattering coefficients, owing to the high-dimensional coupling among these parameters, significantly enhancing the precision of sea coefficient prediction, because it can effectively capture crucial features from simulation parameters and reduces inter-parameter coupling. Experimental analyses with various dimension reductions demonstrate that appropriately reducing the simulation parameters substantially improves prediction accuracy, with gains ranging from 10% to 35%, while maintaining model interpretability above 99.5%.