Evaluating the Met Office Unified Model land surface temperature in Global Atmosphere/Land 3.1 (GA/L3.1), Global Atmosphere/Land 6.1 (GA/L6.1) and limited area 2.2 km configurations

A limitation of the Met Office operational data assimilation scheme is that surface-sensitive infrared satellite sounding channels cannot be used during daytime periods where numerical weather prediction (NWP) model background land surface temperature (LST) biases are greater than 2 K in magnitude. The Met Office Unified Model (UM) has a significant cold LST bias in semi-arid regions when compared with satellite observations; a range of UM configurations were assessed with different model resolutions, land surface cover datasets and bare soil parameterisations. UM LST biases were evaluated at global resolution and in a limited area model (LAM) at a 2.2 km resolution over the SALSTICE (Semi-Arid Land Surface Temperature and IASI Calibration Experiment) experimental domain in south-eastern Arizona. This validation is in conjunction with eddy-covariance flux tower measurements. LST biases in the Global Atmosphere/Land 3.1 (GA/L3.1) configuration were largest in the mid-morning with respect to Moderate Resolution Imaging Spectroradiometer (MODIS) Terra (-13.6±2.8 K at the Kendall Grassland site). The diurnal cycle of LST in Global Atmosphere/Land 6.1 (GA/L6.1) showed a significant improvement relative to GA/L3.1 with the cold LST biases reduced to -1.4±2.7 K and -3.6±3.0 K for Terra and Aqua overpasses, respectively. The higher-resolution LAM showed added value over the global configurations. The spatial distribution of the LST biases relative to MODIS and the modelled bare soil cover fraction were found to be moderately correlated (0.61±0.08) during the daytime, which suggests that regions of cold LST bias are associated with low bare soil cover fraction. Coefficients of correlation with the shrub surface fractions followed the same trend as the bare soil cover fraction, although with a less significant correlation (0.36±0.09), and indicated that the sparse vegetation canopies in south-eastern Arizona are not well represented in UM ancillary datasets. The x component of the orographic slope was positively correlated with the LST bias (0.41±0.05 for MODIS Aqua) and identified that regions of cold model LST bias are found on easterly slopes, and regions of warm model LST bias are found on westerly slopes. An overestimate in the modelled turbulent heat and moisture fluxes at the eddy-covariance flux sites was found to be coincident with an underestimate in the ground heat flux.