Accuracy of homogenization results

The accuracy of homogenization methods is tested by synthetically developed test datasets. The use of such test datasets is indispensable, since the true inhomogeneity properties are known only for synthetically developed datasets. Test datasets imitate the statistical properties of observed time series. On large test datasets, only automatic homogenization methods can be tested. The accuracy of homogenization is characterized by the residual root mean square error and residual mean absolute trend bias for networks of time series and for individual time series. For individual time series, the ACMANT and Climatol methods give the most accurate results, while the network mean errors are reduced most by ACMANT and PHA. ACMANT and Climatol can be successfully applied to the homogenization of several climate variables, and these methods produce fairly accurate homogenization results both for monthly and daily data.