Automation and evaluation of the interactive homogenization tool HOMER

Abstract

AbstractThe interactive HOMER is automated by the use of a set of parameters. This approach retains the skill and flexibility of an interactive method, but adds the speed and reproducibility of an automatic method. The automation of the interactive HOMER also enables systematic testing. Its performance is evaluated by the homogenization of the Indecis homogenization benchmark datasets. The overall performance of the interactive HOMER compares well with the methods using the homogenization tools Climatol and ACMANT and surpass the performance of the standard automatic HOMER. All the homogenization methods reduce the initial error. The average residual error and all error percentiles below and including the 99th error percentile do not differ more than 0.3°C between interactive HOMER and the other methods. Interactive HOMER and Climatol report fewer homogeneity breaks than the true number, while standard automatic HOMER and ACMANT report more homogeneity breaks. Across the methods, a higher number of reported homogeneity breaks renders a higher share of the true homogeneity breaks to be detected, but also a higher share of the reported breaks to be false positives. On average the differences in the corrected times series are small between the methods implying that the choice of homogenization method is a matter of preference.