Relative homogenization: Optional tools

Many homogenization methods include iterations, in which one or more steps of the procedure are repeatedly performed for the step-by-step removal of inhomogeneities. However, iterative procedures are ineffective or even unfavorable for the removal of network mean biases. By contrast, an appropriately designed ensemble homogenization improves accuracy both for network mean and station series characteristics. Multivariate detection is recommended when two or more variables often have breaks at same dates. The use of short and fragmented neighbor series is generally useful when a homogenization method is prepared to their treatment, but the time series with frequent and relatively large observation errors are exceptions. Here, we discuss also the role of different kinds of parameterizations, the possible transformation of the probability distribution of the examined climatic element, the special problems of detecting breaks in series of daily resolution, and the combination of homogenization methods.