Search

In order to correctly detect climate signals and discard possible instrumentation errors, establishing coherent data records has become increasingly relevant. However, since real measurements can be inhomogeneous, their use for assessing homogenization techniques is not directly possible, and the study of their performance must be done on homogeneous datasets subjected to controlled, artificial inhomogeneities. In this paper, considering two European temperature networks over the 1950–2005 period, up to 7 artificial breaks and an average of 107 missing data per station were introduced, in order to determine that mean square error, absolute bias and factor of exceedance can be meaningfully used to validate the best-performing homogenization technique. Three techniques were used, ACMANT and two versions of HOMER: the standard, automated setup mode and a manual setup. Results showed that the HOMER techniques performed better regarding the factor of exceedance, while ACMANT was best with regard to absolute error and root mean square error. Regardless of the technique used, it was also established that homogenization quality anti-correlated meaningfully to the number of breaks. On the other hand, as missing data are almost always replaced in the two HOMER techniques, only ACMANT performance is significantly, negatively affected by the amount of missing data. [ABSTRACT FROM AUTHOR]