Observations and numerical weather forecasts of land-surface and boundary-layer evolution during an unusually dry spring at a site in central England

Analysis of screen-level and land-surface temperature forecasts are known to have long-standing warm nighttime, and cold daytime, temperature biases in regional models. During a record-breaking spring and subsequent summer in 2020, over 100 radiosondes were launched at the Met Office Cardington site under clear skies on ten morning and one evening transitions. We compare observations with operational Met Office UKV forecasts and a standalone land-surface model (JULES). Wind profiles show the UKV nocturnal jet was too high, suggesting too much mixing in the modelled boundary layer. The simulated nighttime surface inversion was too weak and the profile too cold immediately above the inversion. The radiosondes were in addition to comprehensive long-term observations. The evapotranspiration was too large on a seasonal timescale for both the UKV and JULES. For spring and summer, UKV mean screen temperature errors were -0.2 +/- 1.3 circle C during daytime and 1.3 +/- 1.9 circle C at night. The soil temperature diurnal range was too large in both the UKV (by 3.9 degrees C) and JULES (2.9 degrees C), suggesting the surface is too highly coupled to the soil in the simulations. For the spring experimental days, UKV mean maximum screen temperature error was -0.8 +/- 0.4 circle C. The buoyancy-flux crossover times in the morning were slightly too early in the UKV (18 min on average) yet much earlier in JULES (52 min). Observations show the diagnosed boundary layer lifts too early in the UKV with onset of convection occurring on average 67 min too early. The UKV develops summertime boundary layers that are too deep by early afternoon. There was a time lag between the observed screen-skin temperature and buoyancy-flux crossovers in the morning that was not captured by the simulations. The evening buoyancy-flux crossovers in JULES agreed well with the observations, but the UKV crossovers were on average 55 min too early.