Options for dealing with clouds in PCA space

Clouds have always posed a rather fundamental challenge to the full exploitation of infrared sounding observations. Clouds are extremely common and can have a radiative impact of many tens of kelvin - significantly larger than the magnitude of the signals from temperature and composition contained within the data.

Considerable progress has been made over the years in accounting for the presence of clouds in radiance assimilation - from the recovery of clear channels in cloud affected scenes to the assimilation of cloudy – or even fully overcast spectra.

This talk considers to what extent the useful developments adopted for the assimilation of cloudy radiances can be applied to the the use of cloud affected principal component data (in particular PC scores). The Jacobians of PC scores (describing the sensitivity of the observation to different levels in the atmosphere) differ from those of radiance measurements in that they are strongly non-local. However, it will be argued that - while not straightforward - similar approaches can indeed be used in PC space.