Elucidating the causes of errors in 2.2 km Met Office Unified Model simulations of a convective case over the US Great Plains

Abstract

Convective‐scale ensemble simulations with perturbed initial and lateral boundary conditions have been performed to investigate the role of compensating errors in the model representation of a US Great Plains severe convective event. The convective‐scale ensembles were generated by nesting a 2.2 km grid‐length domain within the Met Office global ensemble. Within the ensemble framework, two different science configurations (i.e. parametrization set‐ups) were trialled in the 2.2 km model. The variability due to the use of a different driving global ensemble member was significant with differences in the pre‐convective thermodynamic environment and the initiation time of convection. The science changes also influence initiation time as well as the details in the convective structure. Comparison with observed soundings showed that most of the 2.2 km simulations had too little convective inhibition (CIN). The CIN was found to be independent of the science configuration, implying it is determined by the global model, which was also found to have less CIN than the observed soundings. In both 2.2 km ensembles, the member that produced one of the best simulations of the supercells that developed on this day had more CIN than the other members. However, comparisons with surface station data found that this member had a pre‐convective environment that was too cool and too dry. This suggests there are compensating errors in the factors controlling initiation in these models which comparisons with observations can help elucidate.

The Met Office winter testbed 2020/2021: Experimenting with an on‐demand 300‐m ensemble in a real‐time environment

“…This suggests that perturbations applied in the regional ensembles are likely to be of correct magnitude. Larger differences occur in regions of sharp inversions such as the top of the boundary layer and tropopause, where these are not reliably captured in the same location as observed and the model error lies outside the spread (Hanley & Lean, 2021). By later lead times, the spread–skill relationship changes in structure, with the spread remaining more Gaussian around the mean, centred near zero and relatively constant with height, while the mean error grows and becomes less constant with height.…”

Sensitivity of cloud‐radiative effects to cloud fraction parametrizations in tropical, midlatitude, and arctic kilometre‐scale simulations

“…Efforts are under way at the Met Office to develop a grey‐zone convection parametrization. In concert with further advances in the turbulence scheme and careful coupling with the large‐scale CF scheme, this should enable a smoother transition from shallow convective initiation to deep convection (Lean et al ., 2008; Clark et al ., 2016; Bush et al ., 2020; Hanley and Lean, 2021).…”

“…Indeed, it needs to be further investigated whether these too abundant clouds reflect a correct CF scheme response to a more general problem in convection-permitting models of not capturing the timing and amplitude of convective precipitation. As many studies have suggested, this could hint at the need for an improved representation of turbulence, a grey-zone convection parametrization or some representation of stochasticity to properly initiate convection at kilometre-scale resolutions (Lean et al, 2008;Clark et al, 2016;Bush et al, 2020;Hanley and Lean, 2021). Apart from the mean rainfall diurnal cycle, it is worth analysing the intensity distribution of rainfall as well.…”

Stratification of the vertical spread‐skill relation by radiosonde drift in a convective‐scale ensemble

“…Examination of the vertical aspects are particularly useful as the vertical state of the atmosphere is critical for forecasting certain types of weather phenomenon (e.g., convection and precipitation type; Bourgouin, 2000). The vertical state of the atmosphere is regularly considered through observed profiles (e.g., radiosonde ascents) which can be compared against model profiles to examine spread‐skill relations, understand physical processes, and ascertain skill (e.g., Bain et al, 2022; Hanley & Lean, 2021; Schwartz et al, 2014; Woodhams et al, 2022).…”

“…They indicated similar results between the two different simulations and postulated that this could be a function of the small domain size of the hectometric ensemble due to error refresh rates. They also noted that sharp inversions were not captured well in the ensemble, as in Hanley and Lean (2021).…”