Ross Bannister, National Centre for Earth Observation

Properties of three methods of representing non-Gaussian background error distributions for moisture

David Livings, Marek Wlasak


Assimilation of humidity-related information (e.g. specific humidity, $q$) has raised some interesting and challenging issues. One such issue is its non-Gaussianity, which is an inevitable property given the bounded nature of specific humidity ($0\leq q\leq q_{\mathrm{sat}}$). There have been various attempts in the last decade to account for this non-Gaussianity in the background errors of $q$ in operational data assimilation systems. Here we compare the merits of three of them, namely (i) $\ln q$, (ii) the 'symmetrising' transform of H\'{o}lm et al. (2002) - later adapted by Ingleby et al. (2013), and (iii) Gaussian anamorphosis. Emphasis is placed on assimilation in convective-scale models.

File available here