Model sensitivity across scales: a case study of simulating an offshore low-level jet

In this study, a seven-member ensemble of mesoscale-to-microscale simulations with varying sea surface temperature (SST) is conducted for a case in which an offshore low-level jet was observed via floating lidar. The performance of each SST setup in reproducing the physical characteristics of the observed low-level jet is compared across the mesoscale and microscale domains. It is shown that the representation of low-level shear, jet-nose height, and hub-height wind speed are generally improved when moving from mesoscale to microscale. Specifically, low-level shear is improved in the microscale by reducing near-surface wind speeds and lowering the jet-nose height to be closer to that observed. Counterintuitively, the sensible heat flux on the mesoscale domains is more negative than on the microscale domains, which would indicate a more stable boundary layer with higher shear; however, the low-level shear in the mesoscale is weaker than that of the microscale domains. This indicates over-mixing of the (planetary boundary layer) PBL scheme in the mesoscale domains and/or the overprediction of surface drag in the microscale domain. We analyze performance considering a real-world scenario in which the computational burden of running an ensemble of large-eddy simulations (LESs) limits a study to performing a mesoscale ensemble to select the best model setup that will drive a single LES run. In the context of this study, the best model setup is subjective and weighs model performance in the physical representation of the low-level jet as well as the model surface forcing through the temperature gradient between air and sea. The expectation of this approach is that the best-performing setup of the mesoscale simulations will produce the best result for the microscale simulations. It is shown that there are large fundamental changes in the characteristics of the low-level jet as well as in the surface forcing conditions between the mesoscale and microscale domains. This results in a non-linear ranking of performance between the mesoscale domains and the microscale domains. While the best-performing mesoscale setup is also deemed to produce the best results on the microscale, the second-best-performing mesoscale setup produces the worst results on the microscale.