A new climatology of maximum and minimum temperature (1951-2010) in the Spanish mainland: a comparison between three different interpolation methods

This study presents a new climatology of monthly temperature for mainland Spain (1951-2010), performed with the highest quality and spatially dense, up-to-date monthly temperature dataset available in the study area (MOTEDAS). Three different interpolation techniques were evaluated: the Local Weighted Linear Regression (LWLR), the Regression-Kriging (RK) and the Regression-Kriging with stepwise selection (RKS), a modification of RK. The performances of the different models were evaluated by the leave-one-out validation procedure, comparing the results from the models with the original data and calculating different error measurements. The three techniques performed better for Tmax than for Tmin, and for the cold, rather than warmer months, also at lower altitude than highland areas. The best results were achieved with LWLR applied for the first time on temperatures in the Spanish mainland. This method improved the accuracy of the temperature reconstruction with respect to RK and RKS. We present a collection of Tmax and Tmin monthly charts, using the same temperature legend to prevent any visual bias in the interpretation of the results. The dataset is available upon request.