Changes in precipitation and temperature extremes in Central

INTERNATIONAL JOURNAL OF CLIMATOLOGY

Int. J. Climatol. 25: 139–166 (2005)

Published online in Wiley InterScience (www.interscience.wiley.com). DOI: 10.1002/joc.1135

A NEW INSTRUMENTAL PRECIPITATION DATASET FOR THE GREATERALPINE REGION FOR THE PERIOD 1800–2002

INGEBORG AUER,a,* REINHARD BOHM,a ANITA JURKOVIC,a ALEXANDER ORLIK,a ROLAND POTZMANN,a

WOLFGANG SCHONER,a MARKUS UNGERSBOCK,a MICHELE BRUNETTI,b TERESA NANNI,b MAURIZIO MAUGERI,c

KEITH BRIFFA,d PHIL JONES,d DIMITRIOS EFTHYMIADIS,d OLIVIER MESTRE,e JEAN-MARC MOISSELIN,e

MICHAEL BEGERT,f RUDOLF BRAZDIL,g OLIVER BOCHNICEK,h TANJA CEGNAR,i MARJANA GAJIC-CAPKA,j

KSENIJA ZANINOVIC,j ZELJKO MAJSTOROVIC,k SANDOR SZALAI,l TAMAS SZENTIMREYl and LUCA MERCALLIm

a ZAMG–Central Institute for Meteorology and Geodynamics, Vienna, Austriab Istituto ISAC-CNR, Bologna, Italy

c Istituto di Fisica Generale Applicata, Universita di Milano, Milan, Italyd CRU–Climatic Research Unit, University of East Anglia, Norwich, UK

e Meteo France, Toulouse, Francef MeteoSchweiz, Zurich, Switzerland

g Masaryk University, Brno, Czech Republich SHMU, Bratislava, Slovakiai HMZS, Ljubljana, Slovenia

j DHMZ, Zagreb, Croatiak FMZ, Sarajevo, Bosnia and Herzegovina

l HMS, Budapest, Hungarym SMI, Torino, Italy

Received 30 March 2004Revised 25 October 2004Accepted 25 October 2004

ABSTRACT

The paper describes the development of a dataset of 192 monthly precipitation series covering the greater alpine region(GAR, 4–18 °E by 43–49 °N). A few of the time series extend back to 1800. A description is provided of the sometimeslaborious processes that were involved in this work: from locating the original sources of the data to homogenizing therecords and eliminating as many of the outliers as possible. Locating the records required exhaustive searches of archivescurrently held in yearbooks and other sources of the states, countries and smaller regional authorities that existed atvarious times during the last 200 years. Homogeneity of each record was assessed by comparison with neighbouringseries, although this becomes difficult when the density of stations reduces in the earliest years. An additional 47 serieswere used, but the density of the sites in Austria and Switzerland was reduced to maintain an even coverage in spaceacross the whole of the GAR. We are confident of the series back to 1840, but the quality of data before this date mustbe considered poorer. Of all of the issues involved in homogenizing these data, perhaps the most serious problem isassociated with the differences in the height above ground of the precipitation gauges, in particular the general loweringof gauge heights in the late 19th century for all countries, with the exception of Italy. The standard gauge height in theearly-to-mid 19th century was 15–30 m above the ground, with gauges being generally sited on rooftops. Adjustments tosome series of the order of 30–50% are necessary for compatibility with the near-ground location of gauges during muchof the 20th century. Adjustments are sometimes larger in the winter, when catching snowfall presents serious problems.Data from mountain-top observatories have not been included in this compilation (because of the problem of measuringsnowfall), so the highest gauge sites are at elevations of 1600–1900 m in high alpine valley locations. Two subsequentpapers will analyse the dataset. The first will compare the series with other large-scale precipitation datasets for thisregion, and the second will describe the major modes of temporal variability of precipitation totals in different seasonsand determine coherent regions of spatial variability. Copyright 2005 Royal Meteorological Society.

KEY WORDS: monthly precipitation time series; homogeneity; instrumental period; greater alpine region

* Correspondence to: Ingeborg Auer, Central Institute for Meteorology and Geodynamics, Hohe Warte 38, A-1190 Wien, Austria;e-mail: [email protected]

Copyright 2005 Royal Meteorological Society

140 I. AUER ET AL.

1. INTRODUCTION

The reconstruction of climate variability during the instrumental period has, to date, progressed via two mainapproaches: (1) an intended focus on the ‘continental to global’ scale, e.g. the development of gridded datasetssuch as those of the Climatic Research Unit, UK (Hulme, 1994; Jones and Moberg, 2003), and (2) a focus ata more ‘national’ scale, where various individual national datasets are developed in isolation (e.g. Auer et al.,2001). Complete and accurate global datasets are the ideal basis for climate research, but, invariably, large-scale data compilations are likely to have some deficiencies, such as low spatial resolution and varying dataquality. So, although national studies can achieve better data quality (where data homogenization is based on ahigher density of stations and detailed station histories) they are artificially limited by national borders that donot necessarily coincide with coherent climatic regions. The first steps in creating supra-national precipitationdatasets for the Alps and their surroundings had been taken, for example, by Schmidli et al. (2001, 2002).They concentrated on data collection and analysis and worked with samples of 20 to 100 years. As the datapotential in the region extends further back and as the homogeneity problems had yet to be completely solved,further investments into data improvement, mainly in terms of length and quality, seemed eligible.

Some years ago, we began a joint effort to develop instrumental datasets for the ‘greater alpine region’(GAR), encompassing the Alps and their surroundings from 4° to 18 °E and from 43° to 49 °N), thus tryingto bridge the gap between global-to-continental and national scales, as well as to overcome the still existingshortcomings of homogeneity and length. The GAR is an interesting region for a number of reasons. Itprovides a density and length of long-term climate data not easily attainable in many other regions. It is atransitional region, lying between at least three climatic zones (Atlantic Ocean, Mediterranean Sea, Europeancontinent) and additionally influenced by elevational effects. The climate data for the GAR originate frommore than 15 different data providers. Although this complicates data collection and integration, it helps inthe detection of systematic biases that might be due to specific national, observational practices. The recentlydeveloped database system HISTALP (Ungersbock et al., 2003) was employed to hold, test, correct, adjustand study original and homogenized monthly instrumental climate time series and metadata from stationhistories in the GAR. The ultimate goal is to extend HISTALP to include all climate elements appropriatefor homogenization (see the regional subset of nine climate elements given by Auer et al. (2001)). A firstpublication (Bohm et al., 2001) describes the variability of monthly mean temperatures in the GAR forthe period 1760 to 1998 (a reanalysis and update is currently in preparation). Currently, four other climateelements are being analysed: air pressure, sunshine, cloudiness and snow cover.

This paper describes the development of monthly precipitation series from 1800 to 2002. Unlike temperature,which is easier to homogenize because of its greater spatial coherence, precipitation data require much greatereffort, as their variability is more spatially complex. The content of this paper is restricted to a description ofdataset homogeneity and development. Two subsequent companion papers will describe the results of detailedanalyses of the data.

2. HISTORICAL REVIEW

The GAR has been subject to intensive historical development during the past two centuries. The climate datain the region have always been produced and managed by a number of regional or local authorities, by singlescientific institutes and later by national weather services. Thus, a historical review of the two centuries ofthe ‘instrumental precipitation period’ may help to understand better the difficulties that had to be solved toproduce a coherent dataset. Figure 1 shows the number of stations available for each year of the study period1800–2002. Figure 2 illustrates the history of changing political boundaries for eight subperiods with stablecountry borders, together with the respective station network in each subperiod.

Although a prominent early attempt to establish a real climatic measuring network — that of the ‘SocietasMeteorologica Palatina’ (Anon., 1783; see also Kington (1974)) — had already ended in 1792, some of its

Copyright 2005 Royal Meteorological Society Int. J. Climatol. 25: 139–166 (2005)

PRECIPITATION DATASET: EUROPEAN GREATER ALPINE REGION 141

020406080

100120140160180200

1800

1820

1840

1860

1880

1900

1920

1940

1960

1980

2000

num

ber

of s

erie

s

Figure 1. Development of the network of precipitation gauges still recording and homogenizable in the GAR

stations and general procedures (the standardization of instruments, regular observing hours, etc.) survivedand provided the core of climatic information during the first part of the 19th century. The first 15 years of theinstrumental period (for precipitation starting in 1800/earlier, series exist but are too scarce for homogeneitytesting and adjusting) were unsettled, strongly influenced by the Napoleonic wars. On the one hand, therewas much progress in those times towards the development of a modern scientific outlook (a general open-mindedness towards new ideas, the metric system, freedom of thought). On the other hand, the simple factthat sporadic warfare was commonplace for much of the time hindered a regular and steady developmentof sciences like climatology that rely to a large extent on the development of standardized techniques andadoption of general concepts, such as the use of common instruments and the continuity of observing networks.For the first two decades of the 19th century, precipitation series (five in 1800, 16 in 1815) existed mainlydue to the ongoing activities of a number of physical institutes and/or astronomical observatories (e.g. Torino,Bologna, Milano-Brera, Padova, Marseille, Strasbourg, Hohenpeissenberg, Regensburg, Karlsruhe).

From 1815 (Vienna-Congress) until 1859 (Battle of Solferino) the political borders in the region remainedfixed (see the first map of Figure 2). It was in this period that regional precipitation networks began to develop.They were maintained by scientific and/or economic societies and led to an expansion of the precipitationnetwork in the region to 26 series by 1850 (typical examples were the agricultural societies of Moravia and ofCarinthia). From 1850 to 1859 the number of stations more than doubled (to 56), mainly due to the activitiesof the first large-scale weather service in the region (the Austrian ‘K.K. Central-Anstalt fur Meteorologieund Erdmagnetismus’, founded in 1851). The period from 1859 to 1866–67 (maps 2 and 3 of Figure 2) wascharacterized in political terms by the birth of the Italian nation (‘risorgimento’) involving the unification(in the GAR) of Piedmont, Tuscany, Modena, Parma, Lombardy–Veneto, and parts of the Church State tothe new Italy. In terms of management of the climate network, the traditional Italian observatories remainedlargely autonomous individual bodies with a high degree of scientific freedom, though, with some limits ofregional, national or international coordination. In 1863, the foundation of the Swiss meteorological network(now maintained by ‘MeteoSwiss’) set a new benchmark and was responsible for a significant increase in theavailable precipitation series from 56 to 85. Switzerland has always been a stable part of the GAR region.In political terms it is notable that it has had no change of borders during the entire study period, and it hasoperated a consistent network of climatological stations. The Italian and the Swiss approaches are, in someaspects, counterparts. Switzerland has followed a philosophy of centralized network management, whereas inItaly the observatories have followed their own individual paths. Neither approach can be considered ‘best’ interms of providing data for the reconstruction of climatic time series. Although a higher homogeneity in themethods of observation can be an obvious useful feature, it does present major problems. Sudden synchronizedchanges in the whole network can occur, leading to inhomogeneities that are difficult to eliminate. Another‘monolithic block’ in the GAR used to be the Austrian Empire, whose weather service set the standards fornearly half of the study region in the mid-19th century. The standardization survived the separation of themonarchy into Austrian and Hungarian parts in 1867 (in administrative terms, a Hungarian weather servicewas founded in 1871, see map 4 in Figure 2), and also exerted an influence on the weather services of thesuccessor states after 1918 (map 6, Figure 2).


142 I. AUER ET AL.

Figure 2. Historical political maps of the study region from 1815 to 2002. Dots: HISTALP precipitation sites at the end of the respectiveperiod. Numerical country codes: (1) France, (2) Switzerland, (3) Piedmont and Sardinia, (4) Parma, (5) Modena, (6) Tuscany, (7) ChurchState, (8) Austria, (9) Ottoman Empire, (10) Bavaria, (11) Wurttemberg, (12) Baden, (13) Italy, (14) Germany, (15) Alsace–Lorraine,(16) Hungary, (17) Czechoslovakia, (18) Yugoslavia, (19) Czech Republic, (20) Slovakia, (21) Slovenia, (22) Croatia, (23) Bosnia and

Herzegovina



Also in 1871, the Franco-German war changed the borders between France and the new German state (map4, Figure 2). In 1878, the foundation of the ‘Bureau Central de Meteorologie’ initiated another sudden increasein station numbers with well-organized and high-quality time series in the French part of the GAR adding tothe existing traditional sites of Marseille, Orange, Lyon, Dijon, Nancy and Strasbourg (the latter being underGerman administration from 1871 to 1918). Also in 1878, Austria occupied Bosnia and Herzegovina (Congressof Berlin) and installed a regional weather service there (map 5, Figure 2). Thanks to a recently reinstatedcooperation with the now independent Bosnia and Herzegovina, nine precipitation series starting in the 1870sand 1880s could be developed and now constitute the southeast part of the GAR network. Also, despite thedeclaration of a united German empire in 1871, the regional weather services of Baden, Wurttemberg, Bavariaand Alsace–Lorraine remained relatively independent bodies with some individual evolution of instruments,network management and producing their own yearbooks that continued to be separately published up to the1930s (see Table II). The 1880s, 1890s and the pre-World War I period until 1914 saw no new border changesin the GAR (map 5, Figure 2) and the precipitation network almost reached its maximum density with 192series in this period.

The war of 1914–18 caused fundamental political changes and was a period of great difficulty with respectto the continuation of climatic time series. Map 6 in Figure 2 (compare this with map 5) shows the well-known fundamental changes of political borders. It took a number of years until the new authorities couldestablish well-organized climatic networks again. Problems at this time mainly affected the territory of theformer Austro-Hungarian monarchy, although treaties with the successor states (St Germain, Trianon) alloweda continuity in handling of all data, metadata and instruments. In practice, this did not really work. Extensivegaps in many series occurred until the mid 1920s. These could not be entirely filled. Similar problems werecaused by World War II, but these were associated with the years of warfare, 1939–45, and the post-warrecovery of the climate network was much more rapid and was achieved by the late 1940s.

The most recent political interference in the GAR happened in the early 1990s (maps 7 and 8). Theysaw the peaceful separation of Czechoslovakia, with no influence on the archiving of climatic series. Incontrast, the militant separation of the former Yugoslavia into (in the GAR) Slovenia, Croatia and, Bosniaand Herzegovina led to severe problems with the climate series of Bosnia and Herzegovina. Large amountsof missing data since 1992 have not yet been filled by the national weather service, although attempts areunder way to address this issue.

3. DATA SOURCES

Three main data sources were used to establish the precipitation dataset in the GAR. The first comprises thealready digitized data supplied by the national and subnational data holders (Table I). Roughly 70% of all datawere available in this form, but with metadata only for some rare exceptions (15 series with extensive digitalmetadata from ZAMG, and limited electronic background information from Meteo France). Some (less than10) series could be electronically completed using data from NOAA’s GHCN databank (Vose et al., 1992).The greatest part of the missing 30% and the majority of all metadata had to be collected and digitized frompublished data in yearbooks (Table II), from compendia of printed data (Millosevich, 1882, 1885; Liznar,1886; Mohorovicic, 1902; Eredia, 1908, 1919, 1925; Maurer et al., 1909; Schuepp, 1964; Attmannspacher,1981; SMA, 1981; Penzar et al., 1992; Katusin, 1994; Luksic, 1996; Moisselin et al., 2002; DHMZ, 1998,2002) and also from unpublished original sheets and station history files. For the recent 50 years there is nearly100% coverage with electronic data. Further back, the ‘digital to non-digital ratio’ decreased considerably. Inmany cases the early parts of series could be recovered and digitized. These significantly increase the valueof the GAR database, enabling the study of two entire centuries of precipitation variability in the region.

All data were expressed as monthly totals in millimetres. As not all totals (especially the earlier data fromyearbooks) were available in tenths of millimetres, all totals were rounded to the nearest whole millimetre.During most of the 19th century, precipitation was originally measured in non-metric units (regionally differentkinds of inches and lines; in most cases, even outside France, the ‘ligne de Paris’ was the most commonmeasure of length). The transfer to metric units was undertaken by the data holders. It should be mentioned


144 I. AUER ET AL.

Table I. Data and metadata source 1: electronic data supplied by national and subnational data holders

ZAMG, Vienna Complete series for the recent Austria, most of them prehomogenized; 15 series alsowith metadata in electronic form (Auer, 1993; Auer et al., 2001)

MeteoSwiss, Zurich All Swiss series back to 1901, some prehomogenized (Aschwanden et al., 1996a,b;Begert et al., 2003)

DWD, Offenbach All German series back to the 1940s, not homogenized; two prehomogenized seriesback to the 1870s (Herzog and Muller-Westermeier, 1998)

Meteo France, Toulouse Most of the French series in complete length, some prehomogenized, most series alsowith metadata in electronic form (Moisselin et al., 2002)

CNR-ISAC, Bologna Most of the Italian series in full length and quality checked (Buffoni and Chlistovsky,1992; Bellume et al., 1998; Buffoni et al., 1999; Brunetti et al., 2000a,b, 2001) plussome Croatian data from the Italian period between the two world wars

SMI, Torino All Italian series from Piedmont in full length and quality checked (Romano andMercalli, 1994; Di Napoli, 1996)

Hydrology Service of theProvince of Bolzano/Bozen

Three series in full length (W. Rigott, personal communication)

HMZS, Ljubljana All Slovenian series back to 1871DHMZ, Zagreb Most of the Croatian series in full length plus metadata informationFMZ, Sarajevo All series from Bosnia and Herzegovina back to the 1940sHMS, Budapest Some complete and quality-controlled Hungarian series (S. Szalai and T. Szentimrey,

personal communication)SHMU, Bratislava All Slovakian series, in full length and prehomogenizedMasaryk University, Brno Brno series in full length plus metadata

here that even such apparently easy calculations turned out not to be that trivial in a number of cases. Wefound, for example, original data sheets where it was not clear whether 1 inch was subdivided into 10 or 12lines. It was not possible in each case to solve the problems of measuring units, so some early parts of serieshad to be truncated.

Some of the series had already been preprocessed using different methods of homogeneity testing andadjustment (see Table I). To obtain a uniform database, the original versions were reconstructed for all butnine series. The expression ‘original’ will be used henceforth for quality-controlled but not homogenized andinfilled data. All of the 183 original and nine ‘prehomogenized’ series were then further processed to providehomogenized series as described in the following sections.

4. GENERAL REMARKS ON THE HOMOGENEITY PROBLEM

There is general agreement among climatologists that only a part of the variability of an instrumentalclimate time series reflects real or ‘true’ climate variability. There is always a certain part that is due toa number of non-climatic factors (station relocations, changes of the surroundings, instrumental incuracies,poor installation, observational and calculation rules, etc.). A recent WMO publication (Aguilar et al., 2003)clearly states that ‘all of these inhomogeneities can bias a time series and lead to misinterpretations of thestudied climate. It is important, therefore, to remove the inhomogeneities or at least to determine the possibleerror they may cause’. Most climatologists are aware of the homogeneity problem. Good reviews on methodsand practical experience are given by Peterson et al. (1998) and by the homogeneity seminars in Budapest atregular intervals (Hungarian Meteorological Service, 1997, 2001; Szalai et al., 1999; Szalai and Szentimrey,2004), but it is still not yet a rigid rule for all studies of climate variability to rely on a ‘clean’ and clearlydefined database in terms of homogeneity. It is often argued that homogenizing should be avoided in ordernot to disturb physical laws (e.g. independently homogenized temperature and relative humidity series thatmay produce problems with thermodynamic laws). A second argument against homogenizing is the dangerof distant climatic signals from one (or a few) reference series being transferred to many homogenized



Table II. Data and metadata source 2: published data in yearbooks

1. Jahrbucher der K. k. Central-Anstalt fur Meteorologie und Erdmagnetismus (Meteorological Yearbooks ofthe former Austria), 1848 to 1903

2. Jahrbucher der Zentralanstalt fur Meteorologie und Geodynamik (Meteorological Yearbooks seriescontinued under new name, until 1916 for the former Austria, since 1917 for the modern Austria) 1904 to2002

3. Beitrage zur Hydrographie Osterreichs, Heft 10, Lieferung I, II und III (Compendium of early (1792–1905)precipitation data from Switzerland, Bavaria and the former Austria), Wien, 1913–14

4. Jahrbucher des K. k. hydrographischen Central-Bureau (Hydrographical Yearbooks of the former Austria)1893 to 1912

5. Hydrographische Jahrbucher von Osterreich (Hydrographical Yearbooks of Austria), 1913–20026. Meteorologische Beobachtungen an den Landesstationen in Bosnien und Herzegovina (Meteorological

Yearbooks of Bosnia and Herzegovina), 1892 to 19127. Podaci meteoroloskih opazanja u Bosni i Hercegovini u godini 1913 (Meteorological Yearbook of Bosnia

and Herzegovina), 19138. Godisnje izvjesce, Kr. zem. zavoda za meteorologiju i geodynamiku (Geofizickog zavoda) u Zagrebu za

godine 1914–22 (Meteorological Yearbooks of the former Yugoslavia), godista XIV–XXII, dio IIIBeograd, 1939

9. 1924–40 Izvjestaj o vodenim talozima, vodostajem i kolicinama vode za 1923 god, Kraljevina Jugoslavija,Ministarstvo gradjevina, Hidrotehnicko odelenje-, (Hydrological compendium for the kingdom ofYugoslavia) Beograd, Sarajevo, 1924

10. Padavine u Jugoslaviji, Rezultati osmatranija za period 1925–40 (Precipitation in Yugoslavia, 1925–40),Beograd, 1957

11. Izvestaj o padavinama za 1941–49 (Precipitation reviews of Yugoslavia), Nova Gradiska, Beograd,1952–86

12. Rocenka povetrnostnich pozorovani meteorologickych stanic (Czechoslovakian meteorological yearbooks),1925–64

13. Annales du Bureau Central Meteorologique (Meteorological Yearbooks of France), 1878–192014. Resume Mensuel du Temps en France (Monthly climate bulletin of France), 1929–8815. Bulletin Climatique (Monthly climate bulletin of France), 1989–200216. Meteorologische Beobachtungen im Grossherzogthum Baden (Meteorological Yearbooks of Baden),

1869–193317. Wurttembergisches Meteorologisches Jahrbuch (Meteorological Yearbooks of Wurttemberg), 1867–193318. Beobachtungen der meteorologischen Stationen im Konigreich Bayern (Meteorological Yearbooks of

Bavaria), 1879–193419. Ergebnisse der meteorologischen Beobachtungen im Reichsland Elsass-Lothringen — Annuaires

Meteorologiques d’Alsace et de Lorraine (Meteorological Yearbooks of Alsace and Lorraine), 1890–191820. Annuaires (Annales) de l’Institut de Physique du Globe (Meteorological Yearbooks of Alsace and

Lorraine), 1919–4721. Deutsches Meteorologisches Jahrbuch (Meteorological Yearbooks of Germany), 1934–2002 (1946–87

GDR-excluded)22. Meteorologiai es Foldmagnessegi — Intezet — Evkonyvei (Meteorological Yearbooks of Hungary),

1871–198923. Meteorologia Italiana (Meteorological Yearbooks of Italy, series 1), 1865–7424. Annali dell’ Ufficio Centrale di Meteorologia e Geofisica Italiano (Meteorological Yearbooks of Italy,

series 2), 1879–192525. Servizio Idrografico, (1924 and following years): Annali, Ist. Poligrafico dello Stato, Rome26. Schweizerische Meteorologische Beobachtungen — Annalen der Schweizerischen Meteorologischen

Zentralanstalt (Swiss Meteorological Yearbooks), 1864–200227. Schweizerische Meteorologische Beobachtungen — Supplementband I, Zurich, 1886 (Compendium of

pre-1864 Swiss data in this volume and in the first 12 volumes of the ‘Annalen’)


146 I. AUER ET AL.

series, with the final result that existing spatial variability becomes extensively smoothed. The most commonargument against homogenizing is based on the hope that, for a large number of series, the inhomogeneitiesbecome random and can thus be neglected.

Our own experience in the field of homogenizing leads us to a point of view that extends beyond thatexpressed in the WMO statement (Aguilar et al., 2003). We believe (and will show in detail in this study withexamples from the dense and long-term precipitation network) that all series exceeding a few decades are‘infected’ by non-climatic information. Further, for larger regions (e.g. entire national networks) systematicbiases are possible, but we will also show that with certain precautions a great part of those inhomogeneitiescan be removed. We will also show this for time series of monthly resolution only. Our limited experiencewith homogenizing daily series has shown severe problems, mainly due to the rapid decorrelation of dailyclimate series with increasing station separation (Scheifinger et al., 2003). We believe that there still needsto be a methodological breakthrough in the field of homogenizing daily series, not so much concerning thedetection of ‘breaks’ but much more with the problem of how to adjust the series.

As the present study is devoted to one climate element only, we will not deal here with the possibleviolations of laws of atmospheric physics through homogenization, but we plan to explore this issue in thefuture through the use of regional climate modelling in the GAR.

5. PROCESSING THE DATA (HOMOGENEITY TESTS, OUTLIERS, METADATA,ADJUSTMENTS, GAPS)

The nucleus of relative homogeneity testing (e.g. Peterson et al., 1998) is the comparison of two neighbouringtime series of sufficient similarity. In an early ‘classic’ paper on homogenization, Craddock (1979) testedpairs of stations and concluded that best results were obtained by the use of station pairs with the minimumcoefficient of variation of the ratio of the two series. Schonwiese (1985) demanded a sufficiently highcorrelation between the two stations. Our own experience indicates that a minimum value of 50% commonvariance (r2) is required. A value less than 0.5 allows the potential discontinuities in series to disappearinto statistical noise. Spatial correlation between climatic time series depends on various factors that governregional climate patterns (topography, land use, etc.), on the chosen climate element, and on the time resolutionof the series. Scheifinger et al. (2003) studied the variability in spatial structure of air temperature andprecipitation at different time resolutions in the GAR and at the European scale. Figure 3 illustrates one oftheir results: the mean decorrelation (expressed in terms of r2 falling below 0.5) of the two climate elementsfor daily, monthly, seasonal and annual means (or totals respectively). Figure 4 extends this by comparingthe decorrelation distances with the actual interstation distances in the GAR since 1800. The comparison ofFigures 3 and 4 shows that, for annual precipitation series, the critical test criterion (r2 > 0.5) is generallymaintained, on average, after 1840, for seasonal series from 1857, and for monthly series since 1863. Inthis study, homogeneity tests were performed on seasonal and annual series (adjustments were applied to themonthly series). Therefore, before the year 1840, homogeneity testing difficulties become severe. Althoughthe difficulties in the early instrumental period were to some extent reduced by the use of early metadatainformation, it must be stressed that the reliability of the pre-1840 parts of the series is not equal to that forlater times.

Discussion of relative homogeneity testing of daily precipitation series is beyond the scope of this paper.The necessary network density of 42 km is not attainable for longer time series. Though this is not a problemfor the work described here, it is for the increasing number of studies that attempt to describe trends ofshort-term extremes and any other variables that can only be calculated using daily observations.

During the past 10 years the HOCLIS system, a software package for the homogenization of climatologicaltime series, has been developed at ZAMG (Auer et al., 2001b). It allows the use of three different testprocedures, i.e. the CRADDOCK test, the MASH test and the SNHT (Craddock, 1979; Alexandersson, 1986;Alexandersson and Moberg, 1997; Szentimrey, 1997, 1999, 2001), supported by an intensive use of metadata.Our practical experience in homogenizing several hundred single series of different climatic elements (Aueret al., 1999, 2001; Bohm et al., 2001) leads to the conclusion that the choice of any one specific homogeneity



mean decorrelation (km)

42

105

120

149

533

722

765

993

0 100 200 300 400 500 600 700 800 900 1000

precip-daily

precip-monthly

precip-seasonal

precip-annual

temperature-daily

temperature-monthly

temperature-seasonal

temperature-annual

Figure 3. Average decorrelation distances (r2 decreasing below 0.5) for two climate elements in four time resolutions. Samples: dailyvalues for all of Europe; monthly, seasonal and annual values for the GAR

0

50

100

150

200

250

300

350

400

1800

1820

1840

1860

1880

1900

1920

1940

1960

1980

2000

mea

n di

stan

ce (

km)

1800: 381km

1840: 153km

1820: 201km

since 1927: 61km

1900: 63km

1880: 74km

1860: 113km

Figure 4. Temporal evolution of the mean distance between precipitation gauges in the GAR network since 1800

test is of minor importance. There are a number of tests available that provide similar results (e.g. see thereview in Peterson et al. (1998)). More important than the choice of a specific test method is the strictadherence to the seven principles listed below:

1. Ignore any previous homogeneity work undertaken for any of the series (i.e. start from the beginning,assuming all series contain potential breaks).

2. Test in small, well-correlated subregions (a maximum of 10 series tested against each other results in a10 × 10 decision matrix, which enables most breaks detected to be assigned to a most likely candidateseries).

3. Choose the most appropriate reference series with a non-affected subinterval for the adjustment of eachbreak detected (i.e. different reference series can be used for each break detected in a candidate series).

4. Avoid erratic monthly precipitation adjustments by smoothing the annual course of adjustment factors.5. Detect outliers and ‘overshooting adjustments’ using spatial comparisons (by mapping precipitation values

both in absolute and relative units) for each month of the study period.6. Attempt to determine support for homogeneity adjustments when few metadata are available (i.e. contact

data providers for more information in difficult cases).


148 I. AUER ET AL.

Figure 5. The precipitation network in the GAR (dots) with the subregions for regionally independent homogeneity testing and adjustment

7. Give preference to good metadata rather than mathematical methods in all cases, especially whereadjustment factors can be calculated directly from sufficiently long series of parallel measurements.

Details for the attribution (principles 1 and 2) of a single break (only detects ‘relative breaks’ with a pairof stations) to the determination of the errant series are well described by Auer et al. (2001), following theprinciple of the subregional decision matrix (Mestre, 1999). Figure 5 shows the subregions in which theregional testing was performed in most cases (with the exception of increasing metadata support and thenecessary extension across subregional borders for the early decades).

Principle 3 is a consequence of principle 1: not only for testing, but also for adjustment, no single referenceseries is used for several candidate stations. For each pair of subperiods, before and after one break, a non-affected subperiod of a comparative series is used for the calculation of the adjustment factors, assumingthat the factor between the two series remains constant in the case of homogeneity. The earlier subperiod isalways adjusted to the more recent one, resulting in an unchanged recent period and all earlier subperiodsbeing adjusted to it (this allows for easier updating).

Principle 4 is of special importance for strongly variable climate elements like precipitation. For thehomogenization of the GAR temperature series (Bohm et al., 2001) it could be neglected. Temperatureadjustments (additive) were calculated separately for each single month, resulting in more or less pronounced,but rather steady annual courses of the adjustment differences. The monthly adjustment factors for precipitationseries, in contrast, showed an irregular annual course in many cases. Their application produced unacceptablechanges of the annual course of the adjusted series. The conclusion based on previous homogeneity studies(e.g. Moisselin et al., 2002) is that one annual mean adjustment factor for all 12 months should be used.Comparative series of nearby or on-site parallel measurements, however, clearly show that constancy ofmonthly ratios between two series is very rare. Examples are shown later. Therefore, another method toinvestigate the problem was chosen, one that did not eliminate the seasonality of the adjustment factors:monthly factors are calculated separately for each break first, and these are subsequently smoothed. A 7 monthGaussian low-pass filter was used, which allows for one- or bi-modal annual variation and damps all higherfrequency variation. Figure 6 shows three examples for unfiltered and filtered monthly adjustment factors.The smoothing obviously also reduces another typical homogeneity problem. Often, subintervals between thedetected breaks are too short to allow the calculation of adjustment factors. The smoothing potentially increasesthe sample size by adding information from neighbouring months, e.g. an adjustment value of August alsoincludes the information of May, June and July, as well as of September, October and November. This allowsus to adjust a great number of short subperiods that would otherwise have to be left unchanged without



0.6

0.8

1.0

1.2

1.4

0.6

0.8

1.0

1.2

1.4

0.6

0.8

1.0

1.2

1.4

JAN

MA

R

MA

Y

JUL

SE

P

NO

V

JAN

MA

R

MA

Y

JUL

SE

P

NO

V

JAN

MA

R

MA

Y

JUL

SE

P

NO

V

Lyon (FR) 1841-1864 Venezia (IT) 1837-1866 Luzern (CH) 1882-1900

Figure 6. Three examples of smoothed (bold) and unsmoothed (thin) monthly adjustment factors (smoothing by a 7 month Gaussianlow-pass filter)

22m

/1m

(%

)

0

10

20

30

40

50

60

70

80

90

100

0

10

20

30

40

50

60

70

80

90

100

JAN

MA

R

MA

Y

JUL

SE

P

NO

V

JAN

MA

R

MA

Y

JUL

SE

P

NO

VHohenpeissenberg (DE)

14.5

m/1

.3m

(%

)

Pula (HR)

0

20

40

60

80

100

120

140

160

180

JAN

MA

R

MA

Y

JUL

SE

P

NO

V

Admont (AT)

over

catc

h er

ror

(%)

Figure 7. Mean monthly ratios of parallel on-site measurements or highly correlated comparative series for (from left to right)Hohenpeissenberg (DE, 1800–78, 22 m rooftop versus near to ground installation), Pula (HR, 1873–97, 14.5 m measuring platform

versus 1.3 m) and Admont (AT, 1991–2002, two different locations and orifices on site in summer and winter)

smoothing. Differences in the adjustment factors based on non-smoothed and smoothed data have beencompared for all series. With very few exceptions, the seasonal course of precipitation remained unchangedafter adjustment with smoothed factors.

Figure 7 shows three examples for cases with long-term parallel measurements (a good illustration of theadvantage of having strong metadata support). The graphs underline the necessity of allowing a degree ofannual variation of the adjustment factors. The Hohenpeissenberg case shows the most extreme adjustmentsin the entire network. The extreme undercatch was caused by the well-known combined effects of wind andsnow on precipitation measurements on the exposed hilltop in the Bavarian Pre-Alps. The Pula examplehas similar causes (exposed versus shaded exposure), but the adjustments are less dramatic and the annualvariation is less pronounced. Located on the Adriatic coast, Pula has a much lower ratio of solid to liquidprecipitation, which reduces the undercatch of the exposed tower installation. The Admont case illustratesthat strange things can happen, but we can still produce the annual cycle of adjustments. Here, the observerhad the habit (unknown by the weather service for 12 years!) of removing the upper part of the gauge (thus


150 I. AUER ET AL.

0

20

40

60

80

100

120

140

0

20

40

60

80

100

120

140

0

20

40

60

80

100

120

140

JAN

MA

R

MA

Y

JUL

SE

P

NO

V

JAN

MA

R

MA

Y

JUL

SE

P

NO

V

JAN

MA

R

MA

Y

JUL

SE

P

NO

V

Padova (IT) 1841-1850 Rovereto (IT) 1941-1950 Firenze (IT) 1911-1920

Figure 8. Annual course of decadal mean monthly precipitation (mm) of three Italian sites for original data (bold), and for adjusted databy the old (not smoothed) system (dotted) versus the new (smoothed) adjusting system (thin)

producing a larger orifice) and moving it in winter from the mounting pile to a site near to the house forconvenience (thus producing a remarkable overcatch, especially in winter, compared with the official, morewind-exposed site in the garden).

The three examples of Figure 8 illustrate the overwhelming majority of cases for which the smoothingmethod considerably reduced the problem of biased annual precipitation course caused by erratic annualcycles in the adjustment factors. The problem is more severe in the Mediterranean climate (with strongerprecipitation variability), but it also exists north of the Alps. Hence, the smoothing method was applied toall series for the GAR.

Principle 5 assesses overshooting adjustments, real outliers and false dry months. Although the smoothingmethod also has the positive side-effect of reducing the problem of ‘overshooting adjustments’ it cannotcompletely avoid them. Overshooting, again, is a special problem of highly variable climate elements likeprecipitation. It happens when adjustment factors (derived as the mean ratio of a number of pairs of monthlytotals of two series) are applied to single months that considerably exceed the long-term mean. It can happenthat ‘apparent outliers’ are produced with unrealistically high monthly totals. Together with the fact that, inthe original data, a greater number of questionable high (and low) values already existed, the overshootingproblem had to be solved together with that of real outliers. After trying some less effective ways (isolationof excessive values using monthly frequency distributions, for example) it was finally decided that it wasnecessary to follow a time-consuming method of controlling the spatial precipitation fields of all single monthsfrom 1800 to 2002. A modified version of ZAMG’s routine procedure GEKIS (Potzmann, 1999) for assessingreal-time data allowed a quick look at analysed monthly GAR precipitation fields. An accentuated spline-kriging of each of the relative homogenized (deviations from 1961 to 1990 average as a percentage) data, theabsolute homogenized (millimetres) data, and of the original values, enabled rapid decisions as to whethera value was a potential outlier or not. For the great majority of months and subregions the relative analysesbrought better results. Only for dry summer months in the Mediterranean was the spatial distribution smoother(and, therefore, real outliers easier to detect) for absolute values than for the strongly varying relative values(a July value of 70 mm for the Riviera site Imperia, for example, means a 600% excess over the long-termnormal of 10 mm — not necessarily an outlier).

The vast majority of the suspected outliers were caused by overshooting adjustments. Overshooting wasimproved considerably by the graphical test procedure — replacing the overshooting value by the one resultingfrom kriging of the homogenized anomaly field under exclusion of the value to be corrected. All remainingoutlier suspects (a great number of excessive precipitation values, but also a smaller number of zero valueswere found) were sent to the data holders for more intensive quality checks. About half of the suspectvalues turned out to be true measurements. The remaining errors were then corrected either by inserting



0100200

300400500600

700800900

1800

1825

1850

1875

1900

1925

1950

1975

2000

Oct 1889:805mm489%

Kornat (AT)

mm

/mon

th

0100200

300400500600

700800900

1800

1825

1850

1875

1900

1925

1950

1975

2000

Oct 1872:776mm387%

Genova (IT)

mm

/mon

th

0100200

300400500600

700800900

1800

1825

1850

1875

1900

1925

1950

1975

2000

Aug 2002:431mm457%

Freistadt (AT)

mm

/mon

th

0100200

300400500600

700800900

1800

1825

1850

1875

1900

1925

1950

1975

2000

Jan 1951:296mm679%

Lienz (AT)

mm

/mon

th

0100200

300400500600

700800900

1800

1825

1850

1875

1900

1925

1950

1975

2000

Jul 1956:89mm740%

Arles (FR)

mm

/mon

th

0100200

300400500600

700800900

1800

1825

1850

1875

1900

1925

1950

1975

2000

Aug 1880:319mm877%

Hvar (HR)

mm

/mon

th

Figure 9. Six examples of homogenized and outlier checked monthly precipitation time series (mm) with ‘excessive’ precipitation:Kornat and Genova (excessive in terms of absolute values); Freistadt and Lienz (excessive flooding and avalanche damage); Arles and

Hvar (excessive in terms of relative values — percentage of long-term mean)

the correct measured values (in the cases of typing errors), by interpolation with data from national climatenetworks (usually denser than the long-term HISTALP network) or by using the values derived from thekriging interpolation of the test procedure. Although the outlier elimination comprised the greater part ofthe whole homogenizing work (2400 single monthly precipitation fields were checked) it was consideredessential. The errors produced by overshooting adjustments were greatly reduced and excessive values thatcan greatly reduce the value of any analysis of extreme events based on less intensively quality-controlleddata were detected. Hence, these data are not only fit for long-term trend analysis, but also for all kinds ofextreme event analysis down to a time resolution of 1 month. Similarly, more confidence can be placed on theremaining statistical outliers (like the examples shown in Figure 9). The six monthly time series in Figure 9also underscore again the difficulties in the use of the terms ‘excessive’ or ‘outlier’. They can be used forhigh absolute monthly precipitation amounts that exceed 800 mm sometimes in the GAR (examples 1 and2, e.g. see Maugeri et al. (1998)). They can be defined by the damage they cause, as is shown in example 3


152 I. AUER ET AL.

for the large flooding event in August 2002 in parts of northern Austria, Czech Republic and Germany (seeRudolf and Simmer (2003)); or in example 4 for the avalanche catastrophe in 1951 that destroyed entire partsof alpine villages and also caused great losses of life (Fliri, 1998). Finally, sometimes high relative values(examples 5 and 6) may only be caused by relatively insignificant absolute amounts for regions and seasonswith low averages.

The last two principles, 6 and 7, refer to the use of metadata from station history files. Firstly, the inclusionof metadata provides a very useful support. When homogenizing single series, statistical break detectionis never a simple ‘black-and-white’ decision. There are always grey zones concerning either the strengthof the necessary adjustment, or concerning the precise date of the break. Secondly (unfortunately in rarecases only, as for example those shown in Figure 7), mathematical testing can be completely avoided ifparallel measurements have been performed. Thirdly (an aspect we regard as most important if study regionsare identical with the domains of national weather services), metadata about simultaneous changes in entirenetworks are the only way to detect their consequences; relative homogeneity tests provide zero information insuch cases. A typical example is the recent trend towards automation, which quickly changes whole networkssometimes with remarkable decreases in measured precipitation amounts. Another example is provided bythe simultaneous change of instruments in the 1880s in the precipitation network in the state of Baden. Thisproduced abrupt discontinuities with magnitudes of the order of 30%. A third example, not a sudden break,but a very effective long-term trend, resulted from the general evolution, over the entire GAR, from an oldphilosophy preferring rain gauge installation on high and exposed places to the recent one which tends toavoid the confounding influence of wind through the use of near-to-ground gauges (Figure 10).

The two graphs in Figure 10 show that, in spite of a great variability from site to site (left graph), therehas been a significant decrease of the mean height above ground of rain gauge orifices from 27 m in 1800to 2.7 m 200 years later. In the early years of the study period, platform, roof-top or tower installations upto 50 m above ground were the standard. Nowadays, the WMO recommends 1 to 2 m above ground. Theright graph in Figure 10 indicates that the change from higher to lower heights occurred mainly between1850 and 1880 and for all weather services in the region. In September 1873 (Vienna Congress), it wasstrongly recommended that circular rain gauge orifices only be used, to install rain gauges between 1 and1.5 m above the surface and to provide the information on rain gauge heights in publications (Authorityof the Meteorological Committee, 1874). Only in Italy, with its highly individual station management, didmany of the traditional and independent university and observatory sites not follow the general trend. Themean height of the Italian GAR series remains at 7.5 m; the maximum height is 45 m (Genova Observatory).The special situation in Italy is explained by the fact that many of the traditional observations, from historiccity centres (Buffoni and Chlistovsky, 1992; Romano and Mercalli, 1994; Di Napoli, 1996; Bellume et al.,1998; Brunetti et al., 2001), have been maintained at their historical locations. The 25 m of the Milano-Breraobservatory, for example, does not exceed the surrounding urban canopy and the wind-induced precipitation

0

10

20

30

40

50

60

1800

1800

1820

1840

1860

1880

1900

1920

1940

1960

1980

2000

met

ers

abov

e gr

ound

met

ers

abov

e gr

ound

0

5

10

15

20

25

30

35

1820

1840

1860

1880

1900

1920

1940

1960

1980

2000

IT

BA-SI-HRAT

DE

CH

HU

182 series

Figure 10. Development of the installation height of rain gauges in the GAR 1800 to 2002. Left: all 182 sites with respective metadatainformation (thin) and mean height above ground of all sites (bold). Right: national means (thin) and GAR-mean (bold)



loss is not that of an isolated tower. The general trend towards lower installations can be considered as theprimary cause of the most severe large-scale systematic changes that had to be applied to the GAR data. Onaverage, the early precipitation series had to be increased by up to 10%, in good agreement with the averagedevelopment of installation heights (more details are provided in Section 6).

Lastly, missing months were replaced by estimates derived from highly correlated comparative series. Thisinfilling uses the same method as the adjustment (based on the assumption of the constancy of ratios). Notethat this infilling was the last of the steps in the process of converting from original to homogenized data.It is not advisable to infill before homogeneity testing, as test signals will become less clear. It may also beargued (in terms of strict mathematics) that gaps should not be infilled at all. We believe that the compellingargument of the advantages for further analysis, which require complete series, outweigh this consideration.In certain cases (if intersite ratios series are analysed, for example) the (documented) gaps can be restored.

6. QUANTITATIVE ANALYSIS OF BREAKS, OUTLIERS AND GAPS

The system of data quality improvement described above was successfully applied to 239 long-termprecipitation series in the study region, but the spatial coverage of the 239-sites-dataset was biased by muchhigher densities in Switzerland and Austria compared with the other subregions. Therefore, 24 series inAustria and 23 series in Switzerland were excluded, resulting in an almost even spatial distribution, as shownin Figure 5. For reasons of measurement problems, one recognizable subgroup, i.e. high alpine summitsites, was a priori not included in the homogenization procedure. The highest sites included in the finalnetwork are at altitudes of 1600 to 1900 m a.s.l. and are high-elevation valley sites, not summit sites. For thelatter, the combination of wind exposure and a high contribution from solid precipitation makes precipitationmeasurement extremely difficult (e.g. see Auer (1992) and Auer and Schoner (2001)). Finally, 192 seriesremained as members of the ‘homogenized’ station version of the HISTALP precipitation dataset (Table III).The 192 series are also held in the database in their ‘original’ forms. The quotes are set here to stress oncemore what the suffixes ‘hom’ and ‘ori’ used in the databank represent: ‘hom’ means homogenized, outlier-and overshooting-corrected and with closed gaps, ‘ori’ means ‘as original as possible’ (contemporary dataquality improvements from the data holders were not removed, and it was also not always possible to receivedefinitive information about the real status of the series received in terms of prehomogenization). What hasbeen changed from the ori to the hom versions was analysed quantitatively. The results of the analysis aredescribed in this section.

Table IV sketches the basic statistical characteristics of the HISTALP precipitation homogenization. A totalof 966 breaks and 529 outliers could be detected in the 26 063 station years of 192 series. On average, onebreak could be detected every 23rd year in a series of 136 years in length. The size of the average breakwas 18% (root-square mean of all positive and negative breaks). The most outstanding breaks had valuesup to 238% — resulting from errors concerning non-metric measuring units, inaccurate measuring devicesand other metadata-supported cases (Figure 11, left panel, shows the respective frequency distribution cutat 100%). The typical break detected was not constant during the year (examples in Figure 12). More orless pronounced mono- or bi-modal annual courses of sometimes strong amplitudes were typical (more thanbimodal courses were damped through the use of the smoothing filter, as already described). The right graphof Figure 11 shows the respective frequency distribution of break-amplitudes.

One of the strong points of the HISTALP dataset is its length. However, the extension into the earlyinstrumental period caused several problems. The analysis of the temporal distribution of the breaks detectedcan be used as one possible means of quality estimation of the early instrumental period. Figure 13 showstime series of annually detected breaks in absolute values (breaks per year) in the left graph, and in relation tothe available number of series in the right graph. Already, the comparison of the absolute number of breakswith the number of available series suggests what can be seen more clearly in the right graph: there is only avery slight tendency in the very early years of the 19th century towards a higher break frequency. In general,there is no apparent trend towards less breaks in recent times. It is a moot point as to whether this constancyof breaks is a quality feature, as only the breaks detected could be analysed. The difficulties of testing whennetwork density decreases might also explain at least a part of the stability in breaks detected.


154 I. AUER ET AL.

Tabl

eII

I.N

ames

,ab

brev

iatio

ns,lo

catio

nsan

dst

artin

gtim

esof

the

hom

ogen

ized

prec

ipita

tion

series

Rec

entfu

llna

me

Cou

ntry

code

aA

cr.

x(°E)

y(°N

)A

ltitu

de(m

)St

art

Rec

entfu

llna

me

Cou

ntry

code

aA

cr.

x(°E)

y(°N

)A

ltitu

de(m

)St

art

Adm

ont(H

ZB

)AT

AD

M14

.46

47.5

770

018

53Liv

noB

ALV

O17

.02

43.8

372

418

87A

ix-E

n-Pr

oven

ceFR

AIX

5.37

43.5

010

618

92Liv

orno

ITLIV

10.2

543

.55

318

57A

ltdor

fC

HA

LT8.

6346

.87

449

1864

Lju

blja

naSI

LJU

14.5

246

.07

316

1853

Are

zzo

ITA

RE

12.0

043

.45

274

1876

Loc

arno

-Mon

tiC

HLO

C8.

7946

.17

379

1876

Arles

-Sal

ins

deG

irau

dFR

AR

L4.

7243

.41

118

82Lug

ano

CH

LU

G8.

9746

.00

276

1861

Aro

saC

HA

RO

9.68

46.7

818

4718

90Luz

ern

CH

LU

Z8.

3047

.04

456

1861

Aug

sbur

g-St

.Ste

phan

DE

AU

G10

.93

48.4

246

318

12Lyo

n-B

ron

FRLY

O4.

9445

.72

198

1841

Bad

Ble

iber

gAT

BB

L13

.66

46.6

290

718

74M

acon

-Aer

opor

tFR

MA

C4.

8046

.30

216

1887

Bad

Gas

tein

AT

BG

A13

.13

47.1

211

0018

58M

aliLos

inj

HR

MLO

14.4

744

.53

5318

81B

adG

leic

henb

erg

AT

BG

L15

.90

46.8

730

318

79M

anto

vaIT

MA

N10

.75

45.1

520

1840

Bad

Isch

lAT

BIL

13.6

347

.72

469

1858

Mar

ibor

SIM

AB

15.6

546

.57

270

1876

Bal

me

ITB

AL

7.21

45.3

114

3219

14M

arie

nber

g/M

onte

mar

iaIT

MA

I10

.49

46.7

413

2318

58B

anja

Luk

aB

AB

LU

17.2

244

.78

153

1881

Mar

igny

-Le-

Cah

ouet

FRM

LC

4.46

47.4

631

018

80B

ardo

necc

hia

ITB

AR

6.70

45.0

813

4019

14M

arse

ille-

Mar

igna

gne

FRM

AR

5.23

43.4

45

1800

Bas

el-B

inni

ngen

CH

BA

S7.

6047

.60

316

1861

Mila

noIT

MIL

9.00

45.4

712

218

00B

elfo

rtFR

BFT

6.85

47.6

442

218

95M

illst

att

AT

MST

13.5

846

.80

791

1896

Bel

luno

ITB

EL

12.2

546

.12

404

1875

Mon

tmor

od-L

ons

leSa

unie

rFR

MM

O5.

5146

.69

280

1866

Ber

nC

HB

ER

7.43

46.9

556

518

56M

oson

mag

yaro

var

HU

MO

S17

.27

47.8

812

118

59B

erns

tein

AT

BST

16.2

647

.35

600

1859

Mos

tar

BA

MTR

17.8

043

.35

9918

80B

esan

con

FRB

ES

5.99

47.2

530

718

85M

unch

en-S

tadt

DE

MU

N11

.55

48.1

852

518

48B

iel

CH

BIE

7.26

47.1

343

418

83N

ancy

-Ess

eyFR

NA

N6.

2248

.68

217

1811

Bih

acB

AB

IH15

.88

44.8

224

618

89N

aude

rsAT

NA

U10

.50

46.9

013

6018

96B

jelo

var

HR

BJE

16.8

545

.90

141

1872

Neu

chat

elC

HN

CH

6.95

47.0

048

518

56B

olog

naIT

BO

L11

.25

44.4

860

1813

Nic

e-C

apFe

rrat

FRN

IF7.

3043

.68

138

1900

Boz

en/B

olza

noIT

BO

Z11

.33

46.5

027

218

56N

ice-

Aer

opor

tFR

NIC

7.20

43.6

54

1870

Bra

ITB

RA

7.87

44.7

029

018

63O

bers

tdor

fD

EO

BS

10.2

747

.38

810

1886

Bra

tisla

vaSK

BR

L17

.10

48.1

728

018

57O

dere

nFR

OD

E6.

9847

.92

450

1890

Bre

genz

AT

BR

E9.

7347

.50

424

1874

Ora

nge

FRO

RA

4.85

44.1

453

1817

Brixe

n/B

ress

anon

eIT

BR

X11

.65

46.7

256

918

78O

sije

kH

RO

SK18

.67

45.5

591

1899

Brn

o-Tur

any

CZ

BR

N16

.70

49.1

624

118

05O

vada

ITO

VA

8.64

44.6

418

719

14B

ruck

/Mur

AT

BM

U15

.26

47.4

148

218

76Pa

dova

ITPA

D11

.75

45.4

014

1800

Bud

apes

t-Lor

inc

HU

BU

L19

.22

47.4

513

018

42Pa

pa-P

anno

nhal

ma

HU

PAP

17.4

847

.32

147

1857



Cas

ale

Mon

ferr

ato

ITC

AS

8.50

45.1

311

318

70Pa

rma

ITPA

R10

.25

44.8

057

1833

Cel

jeSI

CEL

15.2

746

.23

234

1853

Pavi

aIT

PAV

9.25

45.1

775

1812

Cer

esol

eR

eale

ITC

ER

7.25

45.4

315

7919

27Pa

yern

eC

HPA

Y6.

9446

.82

450

1900

Cha

ncea

uxFR

CH

A4.

7247

.52

462

1880

Pazi

nH

RPA

Z13

.93

45.2

329

118

85C

hang

ins

CH

CH

G6.

2346

.40

430

1901

Pec

sH

UPE

C18

.17

46.0

615

018

54C

haum

ont

CH

CH

T6.

9947

.05

1073

1864

Peru

gia

ITPR

G12

.50

43.1

052

018

36C

luny

FRC

LU

4.67

46.4

324

018

79Pes

aro

ITPE

S13

.00

43.8

711

1866

Col

mar

FRC

OL

7.36

48.0

819

018

91Pi

acen

zaIT

PIA

9.75

45.0

250

1872

Crikv

enic

aH

RC

RK

14.7

045

.19

1018

91Po

uilly

-En-

Aux

ois

FRPO

U4.

5647

.25

408

1880

Cun

eoIT

CU

N7.

5044

.40

536

1877

Poze

gaH

RPO

Z17

.68

45.3

315

218

83D

elem

ont

CH

DEL

7.35

47.3

641

619

01Pr

ozor

BA

PRZ

17.6

243

.83

800

1892

Deu

tsch

land

sber

gAT

DLB

15.2

246

.83

410

1893

Pula

HR

PUL

13.8

544

.87

3018

65D

ijon-

Lon

gvic

FRD

IJ5.

0847

.27

227

1831

Rad

stad

tAT

RA

D13

.45

47.3

884

518

96Ebn

at-K

appe

lC

HEB

N9.

1247

.27

629

1880

Rau

ris

AT

RA

U12

.99

47.2

293

418

76Ein

sied

eln

CH

EIN

8.76

47.1

391

018

64R

egen

sbur

gD

ER

BG

12.1

049

.03

366

1800

Eis

enka

ppel

AT

EIS

14.5

946

.49

623

1886

Reg

gio

Em

iglia

ITR

EM

10.7

544

.70

6218

67Elm

CH

ELM

9.17

46.9

296

218

78R

etz

AT

RET

15.9

548

.77

242

1895

Eng

elbe

rgC

HEN

G8.

4146

.82

1035

1864

Rhe

infe

lden

CH

RH

E7.

8147

.57

271

1883

Feld

kirc

hAT

FEL

9.62

47.2

744

018

76R

ied

AT

RIE

13.4

848

.22

435

1872

Ferr

ara

ITFE

R11

.50

44.8

215

1865

Rije

ka-K

ozal

aH

RR

IJ14

.45

45.3

310

418

69Fi

renz

eIT

FIR

11.3

043

.80

7518

60R

iva

ITR

IV10

.83

45.8

873

1870

Form

azza

Pont

eIT

FOR

8.44

46.3

813

0019

13R

osen

heim

DE

RO

S12

.12

47.8

744

418

79Fr

eibu

rg/B

reis

gau

DE

FBG

7.83

48.0

030

018

69R

over

eto

ITR

OV

11.0

545

.87

206

1864

Frei

stad

tAT

FRE

14.5

048

.52

548

1878

Rov

igo

ITRV

G11

.75

45.0

59

1877

Gap

-Em

brun

FRG

AP

6.09

44.5

775

018

66Sa

intM

artin

d’H

eres

FRSM

H5.

7645

.20

212

1893

Gen

eve

CH

GN

V6.

1546

.20

405

1826

Sain

t-Pa

ul-L

es-D

uran

ceFR

SPA

5.76

43.7

129

618

82G

enov

aIT

GO

V9.

0044

.40

5318

34Sa

lzbu

rg-F

lugh

afen

AT

SAL

13.0

047

.80

450

1839

Gla

rus

CH

GLA

9.07

47.0

451

518

72Sa

med

anC

HSA

M9.

8846

.53

1705

1861

Gos

pic

HR

GO

S15

.38

44.5

357

318

73Sa

raje

wo

BA

SAR

18.4

343

.87

630

1880

Gra

ndfo

ntai

neFR

GFO

7.12

48.5

251

518

90Se

ckau

AT

SEK

14.7

847

.28

874

1891

Gra

z-U

nive

rsitat

AT

GR

A15

.45

47.0

837

718

37Si

beni

kH

RSI

B15

.92

43.7

377

1890

Hal

lau-

Loh

nC

HH

AL

8.46

47.7

045

018

64Si

onC

HSI

O7.

3746

.23

482

1861

Hei

ligen

blut

AT

HEI

12.8

547

.03

1242

1896

Sopr

onH

USO

P16

.60

47.6

823

418

65H

ohen

peiß

enbe

rgD

EH

OP

11.0

247

.80

986

1800

StG

alle

nC

HST

G9.

4047

.43

779

1864

(con

tinu

edov

erle

af)


156 I. AUER ET AL.Ta

ble

III.

(Con

tinu

ed)

Rec

entfu

llna

me

Cou

ntry

code

aA

cr.

x(°E)

y(°N

)A

ltitu

de(m

)St

art

Rec

entfu

llna

me

Cou

ntry

code

aA

cr.

x(°E)

y(°N

)A

ltitu

de(m

)St

art

Hur

bano

voSK

HU

R18

.20

47.8

712

418

71St

Pol

ten

AT

SPO

15.6

148

.18

285

1894

Hva

rH

RH

VR

16.4

343

.17

2018

58St

Seba

stia

nAT

SSB

15.3

147

.79

872

1884

Impe

ria

ITIM

P8.

0243

.87

5418

76St

a.M

aria

/Mus

tair

CH

STM

10.4

346

.60

1390

1901

Inns

bruc

k-U

nive

rsitat

AT

INN

11.3

847

.27

609

1858

Stra

sbou

rg-E

ntzh

eim

FRST

R7.

6448

.55

150

1803

Inte

rlak

enC

HIT

L7.

8746

.67

580

1890

Stut

tgar

t-Sc

hnar

renb

erg

DE

STU

9.20

48.8

331

418

07Iv

rea

ITIV

R7.

9145

.46

267

1837

Szom

bath

ely

HU

SZO

16.6

347

.27

221

1865

Jajc

eB

AJA

J17

.27

44.3

543

118

92Ta

msw

egAT

TAM

13.8

047

.13

1012

1880

Kal

sAT

KA

L12

.63

47.0

013

5018

96To

rino

ITTO

R7.

7545

.05

275

1803

Kar

lova

cH

RK

VA

15.5

545

.50

112

1872

Torino

Mon

calie

riIT

TO

M7.

7045

.00

267

1864

Kar

lsru

heD

EK

AR

8.35

49.0

311

218

01To

ulon

-La

Mitr

eFR

TO

U5.

9343

.11

2418

70K

eszt

hely

HU

KSZ

17.2

546

.75

115

1861

Tra

vnik

BA

TR

A17

.68

44.2

358

118

81K

lage

nfur

t-Fl

ugha

fen

AT

KLA

14.3

346

.65

459

1814

Tre

nto

ITTRT

11.1

246

.07

199

1864

Koc

evje

SIK

OC

14.8

745

.63

461

1872

Tries

teIT

TR

I13

.75

45.6

567

1841

Kol

lers

chla

gAT

KO

L13

.84

48.6

172

518

87Tuz

laB

ATU

Z18

.70

44.5

530

518

80K

onst

anz-

Mee

rsbu

rg-

Frie

dric

hsha

fen

DE

KM

F9.

1847

.67

443

1827

Udi

neIT

UD

I13

.20

46.0

051

1803

Kor

nat

AT

KO

R12

.89

46.6

910

4718

71U

lm-G

ieng

enD

EU

LM

9.95

48.3

856

718

22K

rem

sAT

KR

M15

.60

48.4

019

018

67Val

lom

bros

aIT

VA

L11

.00

43.7

295

518

72K

rem

smun

ster

AT

KR

E14

.13

48.0

538

918

20Ven

ezia

ITV

EN

12.2

545

.43

118

36K

rize

vci

HR

KR

I16

.55

46.0

315

518

73V

illac

hAT

VIL

13.8

746

.62

495

1887

Kuf

stei

nAT

KU

F12

.17

47.5

849

519

05V

isp

CH

VIS

7.84

46.3

064

019

01La

Cha

ux-D

e-Fo

nds

CH

CD

F6.

8047

.09

1018

1900

Vix

FRV

IX4.

5447

.91

205

1880

La

Cot

eSt

And

reFR

LC

A5.

2645

.36

340

1893

Wai

dhof

en/Y

bbs

AT

WA

I14

.76

47.9

542

118

96Lan

deck

AT

LA

D10

.57

47.1

378

518

87W

ien-

Hoh

eW

arte

AT

WIE

16.3

548

.22

209

1841

Lan

dshu

tD

ELH

U12

.10

48.5

339

318

79W

iene

rN

eust

adt

AT

WR

N16

.21

47.8

128

518

57Lan

gen

AT

LA

G10

.12

47.1

312

1818

81Zad

arH

RZA

D15

.22

44.1

35

1865

Lan

gnau

i.E.

CH

LIE

7.79

46.9

469

519

00Zag

reb-

Gric

HR

ZA

G15

.98

45.8

216

218

62Lan

gres

FRLG

R5.

3447

.85

467

1877

Zel

lam

See

AT

ZEL

12.7

847

.32

766

1875

Lem

ie-C

.le.

ITLEM

7.31

45.2

394

019

22Zer

mat

tC

HZER

7.75

46.0

316

3818

87Lie

nzAT

LIZ

12.8

146

.83

659

1854

Zur

ich

CH

ZU

R8.

5747

.38

569

1830

Lin

zAT

LIN

14.2

848

.30

263

1852

Zw

ettl-

Stift

AT

ZW

E15

.20

48.6

250

018

83

aC

ount

ryco

de,fu

llna

me

(no.

ofse

ries

).AT,A

ustria

(43)

;B

A,B

osni

aan

dH

erze

govi

na(9

);C

H,Sw

itzer

land

(31)

;C

Z,C

zech

oslo

vaki

a(1

);D

E,G

erm

any

(12)

;FR

,Fr

ance

(28)

;H

R,C

roat

ia(1

5);H

U,H

unga

ry(7

);IT

,Ital

y(4

0);SI

,Sl

oven

ia(4

);SK

,Sl

ovak

ia(2

).G

AR

tota

l:19

2.



Table IV. Outline statistics of breaks, outliers and gaps

Series 192

Available data (incl. filled gaps) (years) 26 063Available data (incl. filled gaps) (months) 312 756Mean length of series (years) 135.7

Breaks detected (total) 966Mean homogeneous subperiod (years) 22.7Mean breaks per station per year 0.037Square mean break (%) 17.9Maximum break (%) 238Mean break amplitude (%) 23.4

Corrected overshooting adjustments 1861

Real outliers detected 529

Filled gaps 14 927Mean gap rate (%) 4.8

0

500

1000

1500

2000

2500

3000

3500

class borders of break magnitudes (%)

no. o

f cas

es (

966*

12)

-80

-60

-40

-20 0

20 40 60 80 100

-100

0

20

40

60

80

100

120

140

160

180

200

no. o

f cas

es (

966)

class borders of amplitudes (%)

0 >0

10 5030 4020 60 70 80

>85

Figure 11. Frequency distributions of break magnitudes (left) and of break amplitudes (right). Both refer to the 966 detected and adjustedbreaks (later minus earlier as a percentage of earlier). Break amplitude is maximum monthly break minus monthly break of one event

The situation is different concerning outliers (Figure 14). The relative proportion of outliers with respect tothe number of stations shows a clear trend: starting with a high outlier frequency of every 3 to 10 years duringthe first 30 years, a smaller frequency of every 10 to 30 years during 1830 to 1900, and a low frequency ofone detected outlier approximately every 50 to 100 years after 1900. As the outlier detection capability of theprocedure also depends on the network density, a decreasing quality of the series, especially in terms of singlevalues, should be taken into account when analysing extreme events in the early decades of the 19th century.

Some outliers for individual months can exceed even several hundred percent. Most of the 529 real outliershad excessively high monthly totals, but 119 of them were wrong ‘zero-precipitation months’. They sometimesenter a dataset through a misinterpretation of a missing month as a dry month. Both categories of outliers, ifnot detected, would have severely affected any extreme-event statistics.

In 1861 (0.6% cases), homogenization produced overshooting adjustments (apparent outliers; see Section 5).It was possible to readjust these errors during the outlier detection and elimination procedure.

After having finished all corrections concerning breaks, overshooting effects and outliers, 4.8% of the312 756 individual monthly totals had to be reconstructed from nearby stations. The analysis of the missing


158 I. AUER ET AL.

-40

-20

0

20

40

60

Jan

Mar

May Ju

l

Sep

Nov

ARE IT 1930ARE IT 1954ARE IT 1974%

-40

-20

0

20

40

60

Jan

Mar

May Ju

l

Sep

Nov

BLU BA 1935BLU BA 1959BLU BA 1966%

-40

-20

0

20

40

60

Jan

Mar

May Ju

l

Sep

Nov

BRX IT 1914BRX IT 1925BRX IT 1929BRX IT 1939BRX IT 1949BRX IT 1960%

-40

-20

0

20

40

60

Jan

Mar

May Ju

l

Sep

Nov

ELM CH 1931ELM CH 1947ELM CH 1960ELM CH 1979%

-40

-20

0

20

40

60

Jan

Mar

May Ju

l

Sep

Nov

INN AT 1865INN AT 1873INN AT 1890INN AT 1905INN AT 1948INN AT 1970%

-40

-20

0

20

40

60

Jan

Mar

May Ju

l

Sep

Nov

GAP FR 1942GAP FR 1970GAP FR 1989%

Figure 12. Six examples illustrating typical annual courses of breaks (later minus earlier as a percentage of earlier) in the series ofArezzo (ARE), Banja Luka (BLU), Brixen/Bressanone (BRX), Elm (ELM), Innsbruck (INN) and Gap (GAP)

0

18016014012010080604020

200

1800

1820

1840

1860

1880

1900

1920

1940

1960

1980

2000

1800

1820

1840

1860

1880

1900

1920

1940

1960

1980

2000

02468101214161820

0.00

0.200.180.160.140.120.100.080.060.040.02no

. of b

reak

s pe

r st

atio

n

no. o

f sta

tions

no. of breaks

Figure 13. Time series of breaks detected per year. Left: absolute (versus annual number of series). Right: in relation to the respectivenumber of series available. Grey line and scale is the number of stations with data per year

data (Figure 15) shows for the first time (unlike Figures 13 and 14) the difficulties arising from politicalinstabilities. The first seven decades of the 19th century are characterized by a typical missing data ratenear 10%. Afterwards, missing data rates decreased in the late 19th and early 20th centuries (the time whenweather services reached their full capacity and Europe saw a long peaceful period). The two strong 20thcentury peaks of missing data of up to 16% (1920) and up to 23% (1945) are clearly linked to the two worldwars. The unfortunate increase in the 1990s is due mostly to the wars in the territory of the former Yugoslavia(see the historical review). The very recent increase is due to typical updating problems that arise in anyattempt at supranational data collection.



0

20

40

60

80

100

120

140

160

180

200

1800

1820

1840

1860

1880

1900

1920

1940

1960

1980

2000

1800

1820

1840

1860

1880

1900

1920

1940

1960

1980

2000

0

5

10

15

no. o

f sta

tions

no. of outliers

0.0

0.1

0.2

0.3

0.4

no. o

f out

liers

per

sta

tion

Figure 14. Time series of detected outliers per year. Left: absolute (versus annual number of series). Right: in relation to the respectivenumber of series available. Grey line and scale is the number of stations with data per year

0

50

100

150

200

1800

1820

1840

1860

1880

1900

1920

1940

1960

1980

2000

0

100

200

300

400

500

600

missing m

onthsno. o

f sta

tions

0

5

10

15

20

25

1800

1820

1840

1860

1880

1900

1920

1940

1960

1980

2000

annu

al g

ap r

ate

(%)

Figure 15. Time series of filled gaps. Left: absolute (versus annual number of series). Right: in relation to the respective number ofseries available (gap rate in percent)

An overall view of the effect of the whole process of quality improvement of the GAR precipitation seriesis represented in Figures 16 and 17. They show time series of the annual hom/ori ratios — the annual meanstogether with the +1 standard deviation curves and the absolute range. For the entire GAR (Figure 16), thestandard deviation curves span a range of approximately +0.1 (or 10%) around the curve of means in the20th century, with only a slightly increasing tendency in the 19th century. The full range of all hom/ori ratiosgoes from less than 0.6 up to 1.6 in extreme years. The absolute ranges of the smaller subregional (national)samples are slightly smaller. A closer look at the mean ratio curve of Figure 16 shows the existence of threemain systematic biases in the network that had to be adjusted.

Starting from the recent two decades (which constitute the reference period to which all earlier periodswere adjusted), all earlier periods back to the 1880s tend to be slightly lower than ‘1’. Hom/ori ratios below1 indicate that the original pre-1980s data generally had to be reduced by several percent to make themcomparable to the recent decades. The effect is strongest and almost continuous in the national subsample ofItaly, and smallest in Austria, Switzerland and France. The French subsample shows a long period from 1850to 1910 with systematically low original precipitation sums. In the SI-HR-BA subsample, the post World WarII period had to be reduced more strongly (5 to 10%) to make it comparable to recent decades. The majorityof the earlier years had negatively biased original data. Germany shows ratios <1 back to the 1930s. Fromthe 1890s to the 1920s the average ratio curve tends towards 1.0 again.

One possible explanation for the low pre-1980 ratios may be the recent trend towards automation (whichgenerally leads to larger measuring losses than manual methods). It can, however, only partially justify the


160 I. AUER ET AL.

0.4

0.6

0.8

1.0

1.2

1.4

1.6

1800

1820

1840

1860

1880

1900

1920

1940

1960

1980

2000

hom

/ori-

ratio

all series

Figure 16. Time series of annual hom/ori ratios (bold: mean over all series; black-thin: +1 standard deviation; grey-thin: total range);sample: all 192 series

ratios curve, as automation was introduced in Italy only in the 1990s. It is also worth noticing that when theeffect of automation was studied quantitatively, as in Switzerland, it was estimated to be near 5% (Begertet al., 2003). Actually, the step around 1980 in the Italian ratios curve is not easy to explain, as Italian recordsunderwent a very complex evolution in the last 30 years (for most records, the data source changed in the1970s from UCEA to Servizio Idrografico), displaying a very high density of inhomogeneities, especiallyin the second half of the 1970s. Unfortunately, when most of the records of a geographical area displayproblems within a short time period, homogenization by intrastation comparison becomes very difficult andthe confidence of the results becomes low. The step around 1980 in the Italian ratios curve requires furtherstudy when national projects such as CLIMAGRI (see www.climagri.it) make available a larger number ofItalian precipitation records.

In Austria, automation began during the 1980s, but all automatic stations of the GAR network wereadditionally equipped with manual gauges in order to eliminate cases with large deficiencies.

Going further back in time, an interesting period arises in the 1860s, 1870s and 1880s, whereby a generalnegative bias of 5 to 10% of the average GAR hom/ori ratios is shown. A look at the national subsamples(Figure 17) indicates that it is mainly the Swiss and German series that are responsible for this effect, andto a much smaller extent the French subsample. In Italy, Austria and in the SI-HR-BA region it does notexist. The systematic reduction of the overly high original data required to make them comparable to the1980s and 1990s was a matter of intensive discussion among the Swiss and the Austrian groups involved inthe study. The provisional decision to treat the effect as real was supported by station history data from theGrand Duchy of Baden. The 1888 yearbook of Baden (Table II, source no. 16) indicates (on pages 45–50)that a complete replacement of all old rain gauges (called ‘Swiss type’) with new instruments of ‘Hellmanntype’ was introduced over the 2 years, 1887 and 1888. Two years of comparative measurements at all sitesin Baden resulted in extremely large differences in the catches of the two types of an order of 20% to morethan 30%, with the old (Swiss) gauge catching more precipitation.

According to MeteoSwiss, the old type was also replaced in Switzerland at that time. However, parallelmeasurements performed in Zurich resulted in lower differences. Eventually, it was decided to apply theresults of the parallel measurements to all of the Baden series (Figure 18). The Swiss series were not directlycorrected, but the station history information was kept in mind during the homogenization process and inthe assessment of the station comparisons. Concluding what has been said about the strange 1860s to 1880shom/ori ratios in certain subregions in southwestern Germany, the solution adopted as described can beconsidered as likely, but not certain, to have overcome the problems in the 1860s to 1880s German and Swissdata. Here, as in the very early period, the additional inclusion of proxy climatic information (e.g. from treerings) might provide some improvements in the future (Bohm, 2003), and this concept will be explored as part



AT series

DE series

IT series

CH series

FR series

SI-HR-BA series

0.4

0.6

0.8

1.0

1.2

1.4

1.6

0.4

0.6

0.8

1.0

1.2

1.4

1.6

0.4

0.6

0.8

1.0

1.2

1.4

1.6

0.4

0.6

0.8

1.0

1.2

1.4

1.6

0.4

0.6

0.8

1.0

1.2

1.4

1.6

0.4

0.6

0.8

1.0

1.2

1.4

1.6

1800

1820

1840

1860

1880

1900

1920

1940

1960

1980

2000

1800

1820

1840

1860

1880

1900

1920

1940

1960

1980

2000

1800

1820

1840

1860

1880

1900

1920

1940

1960

1980

2000

1800

1820

1840

1860

1880

1900

1920

1940

1960

1980

2000

1800

1820

1840

1860

1880

1900

1920

1940

1960

1980

2000

1800

1820

1840

1860

1880

1900

1920

1940

1960

1980

2000

Figure 17. Subregional time series of annual hom/ori ratios (bold: mean over all series; black-thin: +1 standard deviation; grey-thin:total range) of regional subsamples for Austria, Switzerland, Germany, France, Italy and the combined Slovenia, Croatia and Bosnia

and Herzegovina (SI-HR-BA)

of the working programme of a current international research project focused on the study region (ALP-IMP2003–2006 Project, 2003).

The third period with systematically biased original precipitation data is that of the first decades of the earlyinstrumental period in the early 19th century. In the first three decades (1800–1830) the annual means hadto be increased by approximately 10% to make them comparable to recent data. After 1830 the adjustmentsdropped, but an average value of roughly +5% is still visible in the GAR mean of Figure 16. The effectis present (although scattered due to the low network density) in all national subsamples (Figure 17) thatextend that far into the past. The adjustments applied to the early data are greater than those of the two


162 I. AUER ET AL.

1800

1600

1400

1200

1000

800

600

400

1800

1820

1840

1860

1880

1900

1920

1940

1960

1980

2000

1800

1820

1840

1860

1880

1900

1920

1940

1960

1980

2000

mm

1800

1600

1400

1200

1000

800

600

400

mm

Freiburg (DE) Karlsruhe (DE)

Figure 18. Two examples from the Grand Duchy of Baden for the artificial ‘1860s to 1880s peak’ in the original annual precipitationseries (bold, grey) versus the homogenized series (thin, black) caused by the use of the ‘old Swiss’ gauge type

0

200

400

600

800

1000

1200

1800

1850

1900

1950

2000

mm

mm

mm

mm

mm

mm

Strasbourg (FR)

1800

1850

1900

1950

2000

0

200

400

600

800

1000

1200Regensburg (DE)

1800

1850

1900

1950

2000

0

200

400

600

800

1000

1200Brno (CZ)

1800

1850

1900

1950

2000

0

200

400

600

800

1000

1200Marseille (FR)

1800

1850

1900

1950

2000

400

600

800

1000

1200

1400

1600Torino (IT)

1800

1850

1900

1950

2000

400

600

800

1000

1200

1400

1600Milano (IT)

Figure 19. Six examples of annual precipitation series of the HISTALP dataset

other periods. A long-term and wide-ranging 10% adjustment is comparable to the decadal-scale climaticvariations and has a strong influence on any medium- and long-term trends. Therefore, it should be checkedcarefully and discussed before taken as credible. Taking into account the difficulties concerning networkdensity versus decorrelation distances (homogenizing principles 2 and 3), the systematic early adjustmentsmay well be artefacts of inadmissible long-range information exchange among less-well-correlated series. Thereason why the final decision in favour of accepting the results from relative homogeneity testing was takencame from the strong support from metadata. The comparison of the early adjustment curve in Figure 16 with



the mean evolution of the height above ground of the rain gauges (Figure 10) shows a strong similarity. Allgauge intercomparison activities (overview in Sevruk (1986)) agree on one point, i.e. measured deficits frommost gauge types increase with increasing wind speed. Also, the parallel measurements for sites in the GAR(examples in Figure 7) clearly support these findings and also show how large the wind-induced bias is forrooftop, platform or tower installations. Therefore, we believe that the strong positive adjustments applied tothe early original series should be taken as the most likely for the time being. In the near future, comparisonswith proxy data (ALP-IMP 2003–2006 Project, 2003) will either further support the decision or not.

7. SOME INITIAL TIME SERIES PLOTS, CONCLUSIONS AND OUTLOOK

Six examples of annual series in Figure 19 (geographically selected from the 36 pre-1850 series) convey afirst feeling for the internal variability of precipitation within a relatively small region like the 724 000 km2 ofthe GAR (0.5% of the Earth’s land surface, 7% of Europe). Unlike temperature (for which Bohm et al. (2001)could show a high degree of uniform long-term trends in the GAR), the precipitation series show considerableinterannual variability, as well as potentially important variability on decadal and even centennial scales. Afurther six examples of seasonal series in Figure 20 (pairs of summer–winter or summer–autumn series)illustrate different, often contrasting, seasonal trends as well.

The purpose of this paper has been to detail all the processes involved in the construction of the 192monthly series that make up the precipitation database for the GAR. Many of the steps were laborious,with several exchanges between the various institutes necessary to determine the reasons for some of thehomogeneity breaks discovered and whether the outliers detected were real or due to incorrect units or

0

600

500

400

300

200

100

1800

1850

1900

1950

2000

mm

mm

mm

6 to 8 12 to 2

Ulm (DE)

1800

1850

1900

1950

2000

0

600

500

400

300

200

100

mm

1800

1850

1900

1950

2000

0

600

500

400

300

200

100

6 to 8

9 to 11

Hvar (HR)

mm

Kremsmünster (AT)

6 to 8 12 to 2

mm

1800

1850

1900

1950

2000

0

600

500

400

300

200

100

Klagenfurt (AT)

6 to 8 12 to 2

1800

1850

1900

1950

2000

0

800

700

600

400

500

300

200

100

Orange (FR)

6 to 8 9 to 11

1800

1850

1900

1950

2000

0

800900

100011001200

700600

400500

300200100

Genova (IT)

6 to 8 9 to 11

Figure 20. Six examples of seasonal precipitation series of the HISTALP dataset (note: reduced scale for two series, also 6–8 is Juneto August, etc.)


164 I. AUER ET AL.

transcription problems. The final database is considered reliable back to 1840, but it is only the best thatcan currently be achieved for the first 40 years, given the density of gauge sites available and the sometimesserious lack of metadata. It is hoped that homogeneity adjustments might be revisited for this early period inthe light of the many proxy records (tree rings, ice cores and historical documents) that are being collectedas part of the ALP-IMP project.

The most serious problem with the network is associated with the rapid change in gauge height aboveground, from the mid-19th century standard of gauges at the top of towers to the 20th century standard ofgauges at near-ground levels. This has meant that homogeneity adjustments of increases up to 30–50%, andsometimes more in winter, have had to be applied to most series. Sometimes, this has had to be done basedsolely on metadata information, as all gauges were almost simultaneously relocated to near-ground positionsin some countries/states during the late-19th century.

All the adjustments applied throughout the 200 years were based on homogeneity studies and detailedmetadata. Adjustments are only appropriate at the monthly and seasonal time scales. We believe that even ifthe daily data for the 192 series in the network could be recovered and digitized, adjustment of the daily seriesfor homogeneity would be difficult, if not impossible, because the network is not dense enough. Adjustments,for example, are likely to be larger if the precipitation were snow than if it were rain, and even when allrain they are likely to be strongly affected by wind strength. Homogenization of daily precipitation series,therefore, requires not only a greater density of gauges, but also much more information about other climaticvariables at the daily time scale. New approaches will have to be developed for this time scale if long-termchanges in extreme precipitation totals over the last 200 years are to be studied.

The dataset is available from the Website of the ALP-IMP project. Two additional papers are in development;these will (1) compare the dataset with existing precipitation datasets at the global, European and alpine scaleand (2) analyse the database to determine coherent regions of precipitation variability across the GAR.

ACKNOWLEDGEMENTS

We express our thanks to the more than 1000 observers involved in measuring the precipitation data overthe last two centuries. Data processing and analysis were supported by funding from a number of past andongoing national and international projects: CLIVALP (Austrian FWF, P1576-N06), CLIMAGRI (Italian CNRSpecial Project 02- 02/05/97- 037681) and ALP-IMP (EU, EVK2-2001-00241). We also want to thank thetwo anonymous reviewers for making helpful comments and suggestions.

REFERENCES

Aguilar E, Auer I, Brunet M, Peterson TC, Wieringa J. 2003. Guidelines on climate metadata and homogenization. World ClimateProgramme Data and Monitoring WCDMP-No. 53. WMO-TD No. 1186, WMO, Geneva.

Alexandersson H. 1986. A homogeneity test applied to precipitation data. Journal of Climate 6: 661–675.Alexandersson H, Moberg A. 1997. Homogenization of Swedish temperature data, part I: a homogeneity test for linear trends.

International Journal of Climatology 7: 1001–1013.ALP-IMP 2003–2006 Project. 2003. Multi-centennial climate variability in the Alps based on instrumental data, model simulations and

proxy data. EU-FP5, EVK2-CT-2002-00 148. www.zamg.ac.at/ALP-IMP (access for the whole project duration).Anon. 1783. Ephemerides Societatis Meteorologicae Palatinae, Historia et Observationes, vol. 1. Typographische Gesellschaftsdruckerey:

Mannheim.Aschwanden A, Beck M, Haberli C, Haller G, Kiene M, Roesch A, Sie R, Stutz M. 1996a. Bereinigte Zeitreihen — Die Ergebnisse des

Projekts Klima90, Band. 1: Auswertungen. In Klimatologie der Schweiz (Jahrgang 1996). Schweizerische Meteorologische Anstalt:Zurich.

Aschwanden A, Beck M, Haberli C, Haller G, Kiene M, Roesch A, Sie R, Stutz M. 1996b. Bereinigte Zeitreihen — Die Ergebnissedes Projekts Klima90, Band. 2: Methoden. In Klimatologie der Schweiz (Jahrgang 1996). Schweizerische Meteorologische Anstalt:Zurich.

Attmannspacher W. 1981. 200 Jahre Meteorologische Beobachtungen auf dem Hohenpeißenberg 1781–1980. Berichte des DeutschenWetterdienstes 155. Selbstverlag des Deutschen Wetterdienstes: Offenbach am Main.

Auer I. 1992. Precipitation measurements in a high alpine region — inter-comparisons of different measuring systems. TECO 92,WMO/TD 462; 251–256.

Auer I. 1993. Niederschlagsschwankungen in Osterreich seit Beginn der instrumentellen Beobachtungen durch die Zentralanstalt furMeteorologie und Geodynamik . In Osterreichische Beitrage zu Meteorologie und Geophysik 7. Zentralanstalt fur Meteorologie undGeodynamik, Wien.



Auer I, Schoner W. 2001. The reliability of precipitation measurements at high mountains — amounts in dependence of instruments andsites. In Proceedings of the Third Seminar for Homogenisation and Quality Control in Climatological Databases, 25–29 September,2000. http://omsz.met.hu/ismeretterjesztes/rendezvenyek/homogen3 (February 2004).

Auer I, Bohm R, Schoner W, Hagen M. 1999. ALOCLIM — Austrian — Central European Long-term Climate — creation of amultiple homogenized long-term climate data-set. In Proceedings of the Second Seminar for Homogenisation of Surface ClimatologicalData. Budapest, 9–13 November 1998, WCDMP 41, WMO-TD 962, WMO; 47–71.

Auer I, Bohm R, Schoner W. 2001. Austrian long-term climate 1767–2000 — multiple instrumental climate time series from centralEurope. In Osterreichische Beitrage zu Meteorologie und Geophysik 25. Central Institute for Meteorology and Geodynamics, Vienna.

Authority of the Meteorological Committee. 1874. Report of the Proceedings of the Meteorological Congress at Vienna, Protocols andAppendices translated from the official report. E. Stanford, Charing Cross, London.

Begert M, Seiz G, Schlegel T, Musa M, Baudraz G, Moesch M. 2003. Homogenisierung von Klimamessreihen der Schweiz undBestimmung der Normwerte 1961–1990 — Schlussbericht des Projekts NORM90. Veroffentlichungen MeteoSchweiz 67.

Bellume M, Maugeri M, Mazzucchelli E. 1998. Due secoli di osservazioni meteorologiche a Mantova. CO.DI.MA., Mantova.Bohm R. 2003. Instrumental climate data. PAGES-News 11(2–3): 9–10.Bohm R, Auer I, Brunetti M, Maugeri M, Nanni T, Schoner W. 2001. Regional temperature variability in the European Alps 1760–1998

from homogenized instrumental time series. International Journal of Climatology 21: 1779–1801.Brunetti M, Maugeri M, Nanni T. 2000a. Variations of temperature and precipitation in Italy from 1866 to 1995. Theoretical and Applied

Climatology 65: 165–174.Brunetti M, Buffoni L, Maugeri M, Nanni T. 2000b. Precipitation intensity trends in northern Italy. International Journal of Climatology

20: 1017–1031.Brunetti M, Buffoni L, Lo Vecchio G, Maugeri M, Nanni T. 2001. Tre Secoli di Meteorologia a Bologna. Edizioni CUSL: Bologna.Buffoni L, Chlistovsky F. 1992. Precipitatzioni Giornaliere Rilevate all’Osservatorio Astronomico di Brera in Milano dal 1835 al 1990.

Edi-ermes: Milano.Buffoni L, Maugeri M, Nanni T. 1999. Precipitation in Italy from 1833 to 1996. Theoretical and Applied Climatology 63: 33–40.Craddock JM. 1979. Methods of comparing annual rainfall records for climatic purposes. Weather 34: 332–346.DHMZ. 1998. 50 years of work of Meteorological and Hydrological Service of Croatia 1947–1997. DHMZ, Zagreb.DHMZ. 2002. 150 years of meteorological observations in Croatia. DHMZ, Zagreb.Di Napoli G. 1996. Tre secoli di clima a Torino: stato dei lavori. Nimbus 13–14: 30–31.Eredia F. 1908. Le precipitazioni atmosferiche in Italia dal 1880 al 1905. In Annali dell’Ufficio Centrale di Meteorologia, Serie II, 27,

anno 1905, Rome.Eredia F. 1919. Osservazioni pluviometriche raccolate a tutto l’anno 1915 dal R. Ufficio Centrale di Meteorologia e Geodinamica.

Ministero dei Lavori Pubblici, Rome.Eredia F. 1925. Osservazioni pluviometriche raccolate nel quinquennio 1916–1920 dal R. Ufficio Centrale di Meteorologia e

Geodinamica. Ministero dei Lavori Pubblici, Rome.Fliri F. 1998. Naturchronik von Tirol. Beitrage zur Klimatographie von Tirol. Universitatsverlag Wagner, Innsbruck.Herzog J, Muller-Westermeier G. 1998. Homogenitatsprufung und Homogenisierung klimatologischer Meßreihen im Deutschen

Wetterdienst. Berichte des Deutschen Wetterdienstes 202. Selbstverlag des Deutschen Wetterdienstes, Offenbach am Main.Hulme M. 1994. Validation of large-scale precipitation fields in general circulation models. In Global Precipitations and Climate Change,

Desbois M, Desalmand F (eds). NATO ASI Series. Series I: Global Environmental Change, vol. 26. Springer-Verlag: Berlin; 387–405.Hungarian Meteorological Service. 1997. Proceedings of the First Seminar for Homogenisation of Surface Climatological Data , Budapest,

6–12 October 1996. Keszult a MET-DRUCK nyomdaban, Budapest.Hungarian Meteorological Service. 2001. Proceedings of the Third Seminar for Homogenisation and Quality Control in Climatological

Databases , 25–29 September, 2000. http://omsz.met.hu/ismeretterjesztes/rendezvenyek/homogen3 (Accessed February 2004).Jones PD, Moberg A. 2003. Hemispheric and large-scale surface air temperature variations: an extensive revision and an update to

2001. Journal of Climate. 16: 206–223.Katusin Z. 1994. Meteorological stations network in Croatia. Hrvatski Meteoroloski Casopis (Croatian Meteorological Journal) 29:

47–56.Kington JA. 1974. The Societas Meteorologica Palatina: an eighteenth century meteorological society. Weather 29: 416–426.Liznar J. 1886. Uber das Klima von Brunn. Sonder-Abdruck aus dem XXIV. Bande der Verhandlungen des naturforschenden Vereines

in Brunn.Luksic I. 1996. Meteoroloski podaci iz Hrvatske. Meteorologija za Korisnike, Penzar B i suradnici. (Meteorological data from Croatia.

In Meteorology for Users, Penzar B and collaborators.), Skolska knjiga i Hrvatsko meteorolosko drustvo: 225–249.Maugeri M, Bellume M, Buffoni L, Chlistovsky F. 1998. Reconstruction of daily pressure maps over Italy during some extreme events

of the 19th century. Il Nuovo Cimento C 21: 135–147.Maurer J, Billwiller Jr R, Heß C. 1909. Das Klima der Schweiz: Auf Grundlage der 37jahrigen Beobachtungsperiode 1864–1900.

Kommissionsverlag von Huber & Co.: Frauenfeld.Mestre O. 1999. Step by step procedures for choosing a model with change-points. In Proceedings of the Second Seminar for

Homogenisation of Surface Climatological Data. Budapest, 9–13 November 1998, WCDMP 41, WMO-TD 962, WMO; 15–26.Millosevich E. 1882. Sulla distribuzione della pioggia in Italia. In Annali dell’Ufficio Centrale di Meteorologia, Serie II 3, anno 1881,

Rome.Millosevich E. 1885. Appendice alla memoria sulla pioggia in Italia. In Annali dell’Ufficio Centrale di Meteorologia, Serie II 5, anno

1883, Rome.Mohorovicic A. 1902. Meteoroloske postaje u Hrvatskoj i Slavoniji od god. 1853. Godisnje izvjesce Zagrebackog meteoroloskog

observatorija 1901 (Meteorological stations in Croatia and Slavonia from 1853, Annual report of Meteorological Observatory inZagreb 1901), Zagreb; 4–14.

Moisselin JM, Schneider M, Canellas C, Mestre O. 2002. Changements climatiques en France au 20eme siecle. Etude des longues seriesde donnees homogeneisees francaises de precipitations et temperatures. La Meteorologie 38: 45–56.

Penzar I, Juras J, Marki A. 1992. Long-term meteorological measurements at Zagreb: 1862–1990. Geofizika 9: (suppl. Zagreb).


166 I. AUER ET AL.

Peterson TC, Easterling DR, Karl TR, Groisman P, Auer I, Bohm R, Plummer N, Nicholis N, Torok S, Vincent L, Tuomenvirta H,Salinger J, Førland EJ, Hanssen-Bauer I, Alexandersson H, Jones P, Parker D. 1998. Homogeneity adjustments of in situ climatedata: a review. International Journal of Climatology 18: 1493–1517.

Potzmann R. 1999. Data control and visualisation with a desktop GIS. In Proceedings of 2nd International Conference on Experienceswith Automatic Weather Stations ICEAWS 1999, 27–29 September, Vienna, Austria. Osterreichische Beitrage zu Meteorologie undGeophysik 20. Central Institute for Meteorology and Geodynamics, Vienna, Austria (CD-ROM).

Romano F, Mercalli L. 1994. L’Osservatorio meteorologico di Cuneo: dal 1877 “sentinella” della Granda. Nimbus 3: 8–15.Rudolf B, Simmer C. 2003. Niederschlag, Starkregen und Hochwasser. Wissenschaftliche Mitteilungen aus dem Institut fur Meteorologie

der Universitat Leipzig. Sonderheft zum Jahr der Geowissenschaften; 27–43.Scheifinger H, Bohm R, Auer I. 2003. Raumliche Dekorrelation von Klimazeitreihen unterschiedlicher zeitlicher Auflosung und ihre

Bedeutung fur ihre Homogenisierbarkeit und die Reprasentativitat von Ergebnissen. 6. Deutsche Klimatagung, Klimavariabilitat 2003,22–25, September 2003, Potsdam. Terra Nostra, Schriftenreihe der Alfred-Wegener-Stiftung 2003/6: 375–379.

Schmidli J, Frei C, Schar C. 2001. Reconstruction of mesoscale precipitation fields from sparse observations in complex terrain. Journalof Climate 14: 3289–3306.

Schmidli J, Schmutz C, Frei C, Wanner H, Schar C. 2002. Mesoscale precipitation variability in the region of the European Alps duringthe 20th century. International Journal of Climatology 22: 1049–1074.

Schonwiese CD. 1985. Praktische Statistik fur Meteorologen und Geowissenschaftler. Gebruder Borntrager: Berlin.Schuepp M. 1964. 100 Jahre schweizerisches Beobachtungsnetz 1864–1963. Separatdruck aus der Jubilaumsschrift “Hundert Jahre

Meteorologie in der Schweiz 1864–1963”.SMA. 1981. 100 Jahre Schweizerische Meteorologische Anstalt 1881–1981. Festschrift zum hundertjahrigen Bestehen der

schweizerischen Meteorologischen Anstalt. Schweizerische Meteorologische Anstalt, Zurich.Sevruk B (ed.). 1986. Corrections of Precipitation Measurements. WMO/IAHS/ETH Workshop on Correction of Precipitation

Measurements. Zurich, Switzerland, 1–3 Apr. 1985. Swiss Federal Institute of Technology, ETH Zentrum: Zurich.Szalai S, Szentimrey T (eds). 2004. Proceedings of the 4th Seminar for Homogenisation and Quality Control in Climatological Data

Bases , Budapest, 6–10 October 2003. WCDMP 56, WMO-TD 1236. WMO: Geneva.Szalai S, Szentimrey T, Szinell C (eds). 1999. Proceedings of the Second Seminar for Homogenisation of Surface Climatological Data.

Budapest, 9–13 Nov. 1998 , WCDMP 41, WMO-TD 962. WMO, Geneva.Szentimrey T. 1997. Statistical procedure for joint homogenisation of climate time series. In Proceedings of the First seminar for

Homogenisation of Surface Climatological Data, Budapest, 6–12 October 1996. Hungarian Meteorological Service, Keszult a MET-DRUCK nyomdaban, Budapest; 47–62.

Szentimrey T. 1999. Multiple analysis of series for homogenisation (MASH). In Proceedings of the Second seminar for Homogenisationof Surface Climatological Data, Budapest, 9–13 November 1998, WCDMP 41, WMO-TD 962. WMO, Geneva; 27–46.

Szentimrey T. 2001. Multiple analysis of series for homogenization (MASH), seasonal application of MASH (SAM), automatic usingof meta data. In Proceedings of the Third Seminar for Homogenization and Quality Control in Climatological Databases, Budapest,Hungary, 25–29 September, 2000. http://omsz.met.hu/ismeretterjesztes/rendezvenyek/homogen3 (Accessed February 2004).

Ungersbock M, Orlik A, Jurkovic A. 2003. HISTALP — eine Datenbank zur kombinierten Erfassung von historischen Klimazeitreihenund deren Metadaten. Terra Nostra, Schriftenreihe der Alfred-Wegener-Stiftung. 2003/6: 450–452.

Vose RS, Schmoyer RL, Steurer PM, Peterson TC, Heim R, Karl TR, Eischeid J. 1992. The Global Historical Climatology Network:long-term monthly temperature, precipitation, sea level pressure, and station pressure data. ORNL/CDIAC-53, NDP-041. CarbonDioxide Information Analysis Center, Oak Ridge National Laboratory, Oak Ridge, TN.


Documents

Changes in precipitation and temperature extremes in Central