About this document 

Based on the evaluation of high-resolution simulations and its comparison with observations and with results from coarser models, parametrizations of boundary-layer depth, convective mixing, and surface fluxes will be assessed, providing recommendations for improved formulations to be tested during the application phase.

Work package in charge

WP6

Executive summary

Storm-resolving models provide unprecedented realism and new opportunities for comparison to observations, but some processes remain unresolved or partially resolved and need parameterization. This is the case for boundary-layer processes. We focus on the tropical marine boundary layer because of its relevance in cloud radiative feedbacks and validate model data with measurements from the EUREC4A field campaign. The simultaneous consideration of ICON and IFS allows us to compare two different strategies. The former simplifies parameterizations to better understand process interactions, sacrificing degrees of freedom to tune the model. The latter considers more sophisticated parameterizations, which allow for better tuning. The results of this study show the value of both. The Smagorinsky model in ICON, despite the simplicity and arguably inappropriate range of application in the kilometer scale, provides boundary-layer properties that are comparable to IFS, which includes a shallow convection scheme, a mass-flux component, and a longer tuning experience. IFS tends to provide better cloud-layer properties, but it depends on the choices of the convective parameterization. Stability correction functions and surface parameterizations can be improved in both models. Both IFS and ICON show greater sensitivity to parameterizations than to grid spacing, which is promising towards certain grid independence in SRMs.

Conclusion & Results

The spatio-temporal numerical representation of boundary-layer properties in nextGEMS simulations was validated against measurements during the EUREC4A field campaign, which includes data from the Barbados Cloud Observatory (BCO). This campaign took place east of Barbados island, a region that is representative of the trade-cumulus convection regime. This regime is one of the key elements in the climate system because of its role in Earth’s radiative balance, and it is, therefore, pertinent to ascertain how well the new storm-resolving models (SRMs) represent the atmosphere in this region, particularly the planetary boundary layer where shallow clouds are rooted. For this aim, surface heat and momentum fluxes, boundary-layer profiles, and near-surface properties were analyzed. The comparison during the first phase has focused on data from the BCO, and we have started the comparison with data from the ocean region.

The first main result is that both ICON and IFS models represent well the profiles of wind velocity, temperature, moisture, and liquid water content observed at the BCO. This includes the various horizontal resolutions and turbulence models considered in the numerical experiments. It is particularly remarkable that ICON with the Smagorinsky model provides such a good estimate without tuning and using coefficients from large-eddy simulations.

We also find some discrepancies between the model representation and the observations. The simulations represented a colder vertical profile than the observations, consistent with global values reported in other projects. The magnitude of the horizontal wind also evolved differently in both models compared to observations, with accumulated biases of about 2–3 m s−1 after 4 weeks. The ICON represented a drier cloud layer between 1–2 km and a moister layer above it, which is attributed to too much vertical mixing across the top of the cloud layer and suggests some revision of the stability correction function. The IFS model represented this region better than ICON in cycle 2, which was expected because IFS uses a shallow convection scheme, which allows better control of this region. However, the IFS model simulations in cycle 3 were drier close to the surface and moister between 2 and 3 km compared to cycle 2 simulations, deviating more from the measurements. This was caused by a change in the convection scheme. This sensitivity of results to the boundary-layer parameterization was also observed in ICON, where the change in the turbulence closure model from Smagorinsky to Total Turbulent Energy led to stronger winds, which affected the surface fluxes and vertical profile. On the other hand, both models showed a smaller sensitivity to changes in the grid spacing between 10 and 5 km. This constitutes a second main conclusion, namely, that changes in model physics in this range of kilometer-scale resolution are more important than changes in the horizontal grid spacing. This suggests some degree of grid convergence in some properties.

One question that needs to be further studied is the representation of the near-surface properties, and in particular, the diurnal cycle. Compared to BCO observations, simulations show a diurnal cycle of 2-m air temperature over land that is too strong. If simulation data was sampled over the ocean upstream of the BCO, the agreement with observations improves, which suggests that the surface parameterization in the models is too local, whereas, in reality, horizontal advection seems to play a more important role. How to improve surface parameterizations (Monin-Obukhov similarity theory) seems to remain a challenge in these SRMs, at least in regions of high heterogeneity like islands.