[3-3]

Covariance modelling developments for convective-scale data assimilation at the Met Office

G.W. Inverarity (Met Office)

The Met Office (the UK's National Weather Service) is currently developing an hourly cycling 4D-Var capability for convective-scale numerical weather prediction. In preparation, the covariance model is being changed to reverse the order of the horizontal and vertical control variable transforms so that the parameter transform is first applied to covariance training data followed by the horizontal transform and then the vertical transform. A full horizontal power spectrum is also retained instead of the second-order autoregressive correlation model currently used in our operational three-hourly cycling 3D-Var. These changes also require the vertical adaptive grid transform, used to better model correlations near temperature inversions and introduce a degree of flow-dependence to the covariance model, to be retuned to reflect different wind covariances. The impact of these changes on three-hourly cycling 3D-Var experiments will be described.