A geostatistics-assisted approach to the deterministic approximation of climate data

We propose a nonconventional application of variogram analysis to support climate data modelling with analytical functions. This geostatistical technique is applied in the theoretical domain defined by each model variable to detect the systematic behaviours buried in the fluctuations determined by other driving factors and to verify the ability of candidate fits to remove correlations from the data. The climatic average of the atmospheric temperature measured at 387 European meteorological stations has been analysed as a function of geographical parameters by a step-wise procedure. Our final model accounts for non-linearity in latitude with a local-scale residual correlation that decays in approximately ten kilometres. The variance of the residuals from the fitted model (approximately 3% of the total) is mostly determined by local heterogeneity in transitional climates and by urban islands. Our approach is user-friendly, and the support of statistical inference makes the modelling self-consistent.