Spectral correlation

On the last page, we examined error correlations over the spatial and temporal dimensions. However, it should be noted that error correlation in the spectral dimension is also an important consideration, especially for users of the data whose retrievals involve the combination of data from different spectral bands.

The correlation between channels is given in FIDUCEO through a correlation coefficient matrix. Note that an effect can be fully correlated between channels, e.g. if you were to know the error in one channel you could predict the error in another channel, and have a different uncertainty for different channels. For example, a temperature error on an internal warm calibration target acting as a pseudo blackbody would affect all spectral channels, but would cause a larger error for shorter wavelength channels than for longer wavelength channels.

this recipe, we’ve introduced the concept of a correlation structure and briefly introduced three correlation dimensions: spatial, temporal and spectral. We’ll continue our examination of correlation in the next recipe, where we’ll see how we can evaluate correlation.