LAUSR

The mesoscale enhancement of surface turbulent fluxes at the air–sea interface is driven by the mesoscale surface wind‐speed variability, especially the gustiness velocity and the mesoscale wind‐speed standard variation. This study proposes a parametrization of these two variables. A large dataset based on the operational 2.5‐km AROME convection‐permitting model is used in a coarse‐graining framework, to quantify various quantities that are subgrid at the scale of a 100‐km resolution global circulation model grid cell. This provides a learning dataset to help build the parametrization. The analysis of two case studies of intense wind‐speed mesoscale variability, combined with a literature review, provides a physically based set of 12 potential predictors, accounting for the convection activity and the large‐scale dynamics. The least absolute shrinkage and selection operator then frames a penalized multivariate linear regression approach to identify the most relevant predictors objectively. Five predictors are selected for predicting the gustiness velocity: the updraft mass flux at the lifting condensation level, the density‐current spreading velocity, the large‐scale horizontal shear and divergence, and the large‐scale wind speed. The parametrization of the mesoscale wind‐speed standard deviation requires an additional predictor, namely the cold‐pool object aggregation index. The proposed parametrization performs significantly better than the previously published parametrizations and is able to capture 80, 99, and 93 %$$ \% $$ of the mesoscale enhancement of the momentum, sensible heat, and latent heat fluxes, respectively. From the perspective of a global circulation model implementation, in which some predictors may be unavailable, simpler versions of the parametrization, that is, involving fewer predictors, are also discussed.