This figure shows how the sigma-point statistics compares to the true statistics. Specifically, we consider the case of a frequency-dependent phase gradient (clock offset, ionospheric effect, or source position offset). The true covariance corresponds to the gray area, while the sigma points appear as dotted lines, together with their associated estimated covariance (thick dashed line). The noise in the data is represented by the dash-dotted line. Qualitatively, the goodness of the description depends on the type of the data representations operator.