14:45–15:00 | ABSTRACT 1278
CHARACTERIZING SPECTRAL NOISE COVARIANCE IN PACE OCI LEVEL-1B IMAGERY
We present a practical framework to estimate the spectral noise covariance matrix of PACE Ocean Color Instrument (OCI) Level-1b top-of-atmosphere (TOA) reflectance. The approach combines homogeneous-patch analysis, first-difference/variogram estimators, and spectral-class conditioning to obtain stable, physically consistent covariance matrices across the hyperspectral range. For each patch, noise covariance is derived from nearest-neighbor differences, which are mathematically equivalent to the variogram at one-pixel lag, yielding robust estimates of both diagonal (variance) and off-diagonal (cross-band) terms. To account for dependence on spectral regime, patches are grouped using unsupervised classification of TOA spectra, and class-specific covariance matrices are computed using soft probabilistic weighting. The resulting matrices are guaranteed to be symmetric positive semi-definite through eigenvalue regularization. Because homogeneous patches do not span the full range of OCI spectral reflectance and viewing geometries, the covariance model is defined over the sampled domain of spectral shape, brightness, and geometry. Covariance matrices are obtained by interpolation within this domain, avoiding unsupported extrapolation while retaining applicability across scenes. Residual dependence on signal amplitude is assessed within each class; when present, covariance matrices are scaled by a brightness-dependent factor while preserving spectral correlation structure. Diagnostics based on trace, band variances, and correlation matrices confirm that spectral class primarily controls covariance structure, while signal level scales the overall covariance magnitude. This framework provides a computationally efficient and statistically robust method to characterize OCI L1b noise, suitable for uncertainty propagation and inversion algorithms, while explicitly accounting for spectral variability, incomplete coverage of observation conditions, and instrument-specific effects.
Jing Tan, Scripps Institution of Oceanography, [email protected]
Robert Frouin, Scripps Institution of Oceanography, [email protected]
