INTERNATIONAL JOURNAL OF CLIMATOLOGY
Int. J. Climatol. (2015)
Published online in Wiley Online Library (wileyonlinelibrary.com) DOI: 10.1002/joc.4265
Ensemble homogenization of Slovenian monthly air temperature series
Gregor Vertacˇnik,* Mojca Dolinar, Renato Bertalanicˇ, Matija Klancˇar, Damjan Dvoršek and Mateja Nadbath
Climatology Section, Meteorology Office, Slovenian Environment Agency, Ljubljana, Slovenia
ABSTRACT: This paper presents an attempt to obtain high-quality data series of monthly air temperature for Slovenian stations network in the period from 1961 to 2011. Intensive quality control procedure was applied to mean, maximum and minimum air temperature datasets from the Slovenian Environment Agency. Recently developed semi-automatic homogenization tool HOMER (HOMogenisation softwarE in R) was used to homogenize the selected high-quality datasets.
To estimate the reliability of homogenized datasets, three to six experts independently homogenized the same datasets or their subsets. Different homogenization parameter settings were used by each of the experts, thus comprising ensemble homogenization experiment. Resulting datasets were compared by break statistics, root-mean-squared-difference (RMSD) of monthly and annual values, and RMSD of the long-term trend. This semi-automatic homogenization approach based on metadata gave more reliable homogenization results than a fully automatic approach without metadata. While the network-wide linear trend of the dataset did not change after semi-automatic homogenization was applied, the distribution of the trends of individual stations became spatially more uniform. The arithmetic mean of the homogenized datasets of three experts was assigned as a reference homogenized dataset and it was compared with some publicly available homogenized datasets. The calculated linear trend on an annual level for Slovenia is strongly positive in all datasets, though the trend values are significantly different between the datasets. We conclude that the warming trend of near-surface air temperature in Slovenia in 1961–2011 is significant and unequivocal in all seasons, except for autumn. Mean, maximum and minimum temperature series indicate linear trend of around 0.3–0.4 ∘Cdecade–1 on an annual level.
KEY WORDS air temperature; Slovenian time series; subjective homogenization; HOMER
Received 25 February 2014; Revised 18 December 2014; Accepted 23 December 2014 1. Introduction
Climate change and variability have become important topics among climatologists as well as the public due to the increasing human footprint on the climate system (Le
Treut et al., 2007). High-quality climate data is needed for a robust assessment of climate change. To obtain such data, several important steps should be done frommeasurements to data analysis, including quality control (QC) and homogenization. Many examples of climate time series homogenization on a regional to global scale exist in scientific literature (e.g. Hansen et al., 1999; Auer et al., 2005;
Beger et al., 2005; Menne et al., 2010). However, until 2013 only incomplete attempts, missing intensive homogenization and QC procedures have been made to get a high-quality climate series for Slovenia. To fill the gap, the
Slovenian Environment Agency (ARSO) launched the climate variability in Slovenia (CVS) project in November 2008. The aim of the project was to deliver comprehensive analysis of climate change and variability in Slovenia since 1961. *Correspondence to: G. Vertacˇnik, Climatology Section, Meteorology Office, Slovenian Environment Agency, Vojkova cesta 1b, SI-1000
Ljubljana, Slovenia. E-mail: firstname.lastname@example.org
Quality of long-term climate analysis depends on the homogeneity of the underlying time series. A homogeneous climate time series is defined as onewhere variations are caused only by variations in climate (Aguilar et al., 2003). In reality, most time series also reflect changes of instruments, observing practices, station locations, station environment, formulae used to calculate their values (e.g. ‘daily mean temperature’) and so on. Some of these changes result in abrupt shifts or breaks in time series (e.g. location change), while others (e.g. urbanization) lead to gradual bias or trend. These spurious signals, called inhomogeneities, can greatly affect climate analysis and should be removed from the time series to the highest possible degree. To meet this criterion, a variety of homogenization methods have been developed in recent decades (Reeves et al., 2007; Costa and Soares, 2009;
Domonkos et al., 2012).
Performance of homogenization methods has been evaluated in the COST action HOME (Advances in homogenization methods of climate series: an integrated approach), which was launched in 2007 (Venema et al., 2012). Homogenization methods were validated against a realistic benchmark dataset of monthly temperature and precipitation. The focus of validation was on a surrogate dataset, where inhomogeneities were inserted in an © 2015 Royal Meteorological Society
G. VERTACˇNIK et al. artificially homogenous dataset, resembling the statistical properties of a real homogenized dataset (Venema et al., 2012, and references therein). Based on the analysis of homogenized datasets the authors strongly recommend the use of direct homogenization algorithms, such as
ACMANT, Craddock, MASH, Prodige and USHCN.
Using these methods, the quality of homogenized temperature datasets was in general considerably improved compared to the input in homogeneous datasets.
Following recommendations of the COST action, new software has been developed. HOMogenisation softwarE in R (HOMER) represents a synthesis of the best aspects of some of the most efficient methods (Mestre et al., 2013). It is an interactive semi-automatic method, primarily designed for homogenization of monthly and annual time series of temperature and precipitation. A user can take advantage of metadata in the iterative process of detection and correction of a time series. At the time of submission of this paper, only a few applications of