Representation of model error in a convective-scale forecast ensemble
Bannister, R.N.1; Migliorini, S.2; Rudd, A.C.3; Baker, L.H.3

Errors in weather forecasts originate from a number of sources, especially (i) the forecast's initial conditions, (ii) the boundary conditions and (iii) the model formulation, which all follow from and influence the data assimilation schemes used. Meso-to-convective-scale data assimilation and forecasting present new challenges because at these scales model errors are thought to become dominant. Here we investigate one means of simulating model error which we expect to have an impact on forecast skill at convective scale. This method makes stochastic fluctuations to physics parameters (specifically associated with microphysics and turbulent boundary layer processes) in ensembles of convective-scale forecasts. This technique is known as the random parameters (RP) scheme. The forecast model that the RP scheme is applied to is an experimental 1.5 km resolution (convection-permitting) version of the UK Met Office's Unified Model with 24 ensemble members (the 1.5km-EPS). We assess the ability of the RP scheme to affect ensemble spread and study how it affects forecast comparisons with observations. We determine the sensitivity of diagnostics, such as forecast error variances and correlation length scales, to the size and structure of the model errors. By switching on and off the different sources of forecast error (initial condition error, lateral boundary condition error and model error) we attempt to disentangle the effect of model error from initial condition and boundary condition errors.