Assessment of parallel temperature measurements network

Long historical climate records usually contain non-climatic changes that can influence the observed behaviour of meteorological variables. The availability of parallel measurements offers an ideal occasion to study these discontinuities as they record the same climate. The transition from manual to automatic measurements has been analysed in this study. The dataset has been obtained from two independent climate networks in the Piedmont region, in Northwest Italy and in Québec, Canada. The first selection of pairs of stations was based on three parameters, that is, the overlapping period, the difference in elevation and the distance. The second step was to evaluate the exposure of the selected stations and their characteristics, the type of instrumentation, the neighbours and the general conditions. Therefore, the dataset contain 16 pairs of stations with up to 15 years of overlapping. On average, the overlapping period is 12 years with 8760 daily data matched for a pair of stations. The dataset was divided in three groups by the instruments utilized. The first the pairs of stations have a manual thermometer and a thermograph, the second two thermograms, old type and new type, and the third a termistance and a temperature probe. In order to be able to make a direct comparison between the daily temperature series, any values that were missing in one series were also set to be missing in its counterpart before the monthly statistics were computed. Non-parametric tests were applied to the daily values to evaluate the preliminary relationships between the pairs of series. The root mean square error (RMSE) was used to identify the mean difference between the two series, while the Spearman method was used to evaluate the correlation coefficient. The Kolmogorov–Smirnov test (KS) was applied to determine whether two datasets could have come from the same distribution, while the Wilcoxon rank sum test (W) was considered to establish whether two samples had identical population medians. A p = 5% significance level was used for all the tests. The thresholds by percentile were calculated on a daily scale to identify the different temperature types. Five classes of temperature were established: extremely cold, cold, medium, heat and extremely heat. The number of events and the mean values of temperature were calculated for each class and for each pair of series. The transition between the networks has highlighted important differences in the temperature values. In the first and in the third group there are not a clear relationship between the instruments. The behaviour between the pairs of stations depend to the peculiarity of the locations. In the second group in the extreme classes, extreme cold and extreme heat, the new type of thermograms recorded a greater number of events. These differences produce a spurious change in the temperature of the analysed area, thus showing the importance of having a homogeneous dataset to identify real climate variations.