Date of this Version
Here we present an exploratory statistical analysis aimed at the minimization of the ‘screen bias’ from affected ancient air temperature time series over the Western Mediterranean. Our approach lies in the statistical analysis of about 6 years of daily paired temperature observations taken using the ancient Montsouri shelter and the modern Stevenson screen for daily maximum (Tx) and minimum (Tn) temperature data recorded at two experimental sites: the meteorological gardens of La Coruña and Murcia, Spain (locations under the influence of the Oceanic/Atlantic/Galician and Mediterranean arid and semi-arid climate types, respectively), where ongoing field trials have been carried out. Descriptive statistical analysis of the paired series shows pre-sheltered temperatures tended to induce a strong warm bias in Tx data (of about 1 °C at the annual scale but with a clear seasonal cycle with higher values in summer and lower in winter), while Tn readings have a small (∼0.2 °C, and sustained all year round) cold bias compared to the modern period. Statistical relationships between the screen bias and other related meteorological variables show the highest correlation coefficients between the ‘screen bias’ and Tx, Tn and the diurnal temperature range (DTR) recorded under the replicated ancient shelters at both locations and point to the reliability of these variables as potential predictors of the ΔTx. We generate a parsimonious regression model based on the data from both experimental sites, which takes into account polynomial terms of lower order for the predictor variables (Tx and DTR recorded under the ancient shelter) and harmonic terms, in order to represent the seasonal cycle of the screen bias. The goodness-of-fit of the model is satisfactory, as it explains up to 51.7% of the additional ΔTx variability.