Spatial gridding of daily maximum and minimum 2 m temperatures supported by satellite observations

The usefulness of two remotely sensed variables, land surface temperature (LST) and cloud cover (CC), as predictors for the gridding of daily maximum and minimum 2 m temperature (T min/T max) was assessed. Four similar gridding methods were compared, each of which applied regression kriging to capture the spatial variation explained by the predictors used; however, both methods differed in the interpolation steps performed and predictor combinations used. The robustness of the gridding methods was tested for daily observations in January and July in the period 2009–2011 and in two different regions: the Central European region (CER) and the Iberian Peninsula (IP). Moreover, the uncertainty estimate provided by each method was evaluated using cross-validation. The regression analyses for both regions demonstrated the high predictive skills of LST for T min and T max on daily and monthly timescales (and lower predictive skills of CC). The application of LST as a predictor considerably improved the gridding performance over the IP region in July; however, there was only a slight improvement over the CER region. CC reduced the loss of spatial variability in the interpolated daily T min/T max values over the IP region. The interpolation skill was mainly controlled by the station density, but also depended on the complexity of the terrain. LST was shown to be of particular value for very low station densities (1 station per 50,000 km2). Analyses with artificially decreasing station densities showed that even in the case of very low station densities, LST allows the determination of useful regression functions.