Methodology to evaluate numerical weather predictions using large‐aperture scintillometry sensible heat fluxes: Demonstration in London

Abstract To evaluate numerical weather predictions (NWP) using observations, the horizontal, vertical and temporal resolution characteristics need to be considered to ensure consistency. Here, large‐aperture scintillometry (LAS) is used to derive turbulent sensible heat fluxes ( Q H ), as the source area extents are comparable to NWP with horizontal grid box resolutions of order 1 km. We demonstrate our methodology using LAS observations undertaken in central London and the Met Office's operational forecast model for the United Kingdom (UKV). In the horizontal, we ensure consistency between the LAS source area locations and extent, as they vary with wind direction and stability, and therefore the NWP grid boxes that should be selected for comparison. In the vertical, the appropriate model level needs to be selected relative to the changing effective measurement height ( z f ) of the observations. The LAS fluxes time and space averaging allows Q H to be obtained for time periods similar to the model timestep (e.g., 1 min) but the UKV does not explicitly represent higher‐frequency turbulent motions. During two spring days, three to five UKV grid boxes (1.5 km resolution) fall within the LAS source area (60%). However, with the model land cover data lacking realistic spatial variability, modelled effective surface Q H spatial variability is small, leaving a central grid box on the LAS path representative. Larger differences occur between modelled Q H at the surface and closest to z f , despite both theoretically being within the inertial sublayer. The modelled Q H has better agreement with observed fluxes when averaged over long periods (10+ min cf . 1 min). As LAS‐derived Q H has larger source areas than eddy covariance (EC) measurements, there are benefits to assessing NWP model performance with LAS. With multiple paths, spatial variability in response to changes in surface cover could be assessed at shorter times scales than possible with EC.