A Novel Method to Estimate Sea Ice Correlation Length for NRCS Simulation Using the Advanced Integral Equation Model

Arctic sea ice is a critical component of the global climate system. Synthetic aperture radar (SAR) enables large-scale observation but requires precise surface roughness parameters (e.g., correlation length) for accurate microwave scattering modeling, which traditional methods fail to provide efficiently. To address this gap, a novel method is proposed for estimating correlation length directly from SAR images. First, spatial features of sea ice are extracted using 2-D continuous wavelet transform (CWT) with three wavelet functions (Cauchy, Morlet, Mexican hat), generating peak scales and angles. Second, these features, alongside root mean square (rms) height and incident angle, are input to a Bayesian-regularized neural network (two-layer feedforward architecture) to optimize correlation length for the advanced integral equation model (AIEM). Finally, the calibrated AIEM simulates normalized radar cross section (NRCS) with enhanced accuracy. Results show the Cauchy wavelet achieves optimal performance ( $R^{2}=0.82$ , rms error (RMSE) = 4.99 cm for correlation length), and NRCS simulation errors reduce to 0.66 dB (versus 0.97 dB empirically), validated robustly on independent data (RMSE = 1.17 dB). This method eliminates in situ limitations, advancing scalable sea ice monitoring.