Official URL: https://doi.org/10.3354/cr01488

Changes in climatic observations (such as station relocations and changes of instrumentation) often affect the spatial and temporal comparability of the data; therefore, an important part of improving the accuracy of observed climate variability is the time series homogenisation of the source data. In undertaking homogenisation, an essential step is the spatial comparison of the data within the same geographical region. To optimise the efficiency of homogenisation, we should know when and to what extent two series are of the same geographical origin from a climatic perspective, and how many partner series should be used. This study presents a number of novel experiments for obtaining objective answers to these questions. Monthly temperature test datasets were homogenised with ACMANT (Adapted Caussinus-Mestre Algorithm for homogenising Networks of Temperature series) by varying the number of partner series and their spatial correlations with the candidate series. First, a homogeneous benchmark is constructed from 2 regional subsets of a simulated surface air temperature dataset from earlier work. Various kinds of inhomogeneities are then inserted into the time series, producing 5 basic types of test datasets for each geographical region. Further variation is introduced by adding additional noise to some datasets, providing more diverse spatial correlations. The results indicate that for the identification and correction of long-lasting biases in the data, the optimal number of partner series is about 30. The optimum is largely independent from the frequency and intensity of inhomogeneities and from the spatial correlation between the candidate series and its partner series. This latter finding is unexpected; hence, its possible causes and the consequences are discussed and explored more fully here.