A new floating model level scheme for the assimilation of boundary layer top inversions: the univariate assimilation of temperature.
Fowler, A., Bannister, R. and Eyre, J. (2012) A new floating model level scheme for the assimilation of boundary layer top inversions: the univariate assimilation of temperature. Quarterly Journal of the Royal Meteorological Society, 138 (664A). pp. 682-698. ISSN 1477-870X
To link to this article DOI: 10.1002/qj.955
The assimilation of observations with a forecast is often heavily inﬂuenced by the description of the error covariances associated with the forecast. When a temperature inversion is present at the top of the boundary layer (BL), a signiﬁcant part of the forecast error may be described as a vertical positional error (as opposed to amplitude error normally dealt with in data assimilation). In these cases, failing to account for positional error explicitly is shown t o r esult in an analysis for which the inversion structure is erroneously weakened and degraded. In this article, a new assimilation scheme is proposed to explicitly include the positional error associated with an inversion. This is done through the introduction of an extra control variable to allow position errors in the a priori to be treated simultaneously with the usual amplitude errors. This new scheme, referred to as the ‘ﬂoating BL scheme’, is applied to the one-dimensional (vertical) variational assimilation of temperature. The ﬂoating BL scheme is tested with a series of idealised experiments a nd with real data from radiosondes. For each idealised experiment, the ﬂoating BL scheme gives an analysis which has the inversion structure and position in agreement with the truth, and outperforms the a ssimilation which accounts only for forecast a mplitude error. When the ﬂoating BL scheme is used to assimilate a l arge sample of radiosonde data, its ability to give an analysis with an inversion height in better agreement with that observed is conﬁrmed. However, it is found that the use of Gaussian statistics is an inappropriate description o f t he error statistics o f t he extra c ontrol variable. This problem is alleviated by incorporating a non-Gaussian description of the new control variable in the new scheme. Anticipated challenges in implementing the scheme operationally are discussed towards the end of the article.