floods in germany : analyses of trends, seasonality and ... · institute of geoecology floods in...
TRANSCRIPT
Institute of Geoecology
Floods in Germany – Analyses of Trends,
Seasonality and Circulation Patterns
Dissertationfor the degree of Doctor of Natural Sciences
(Dr.rer.nat.)
in Geoecology
submitted to the Faculty of Mathematics and Natural Sciences
at the University of Potsdam, Germany
by
Theresia Petrow
Potsdam, June 2009
Published online at the Institutional Repository of the University of Potsdam: URL http://opus.kobv.de/ubp/volltexte/2009/3739/ URN urn:nbn:de:kobv:517-opus-37392 [http://nbn-resolving.org/urn:nbn:de:kobv:517-opus-37392]
Abstract
Flood hazard estimations are conducted with a variety of methods. These include flood fre-quency analysis (FFA), hydrologic and hydraulic modelling, probable maximum dischargesas well as climate scenarios. However, most of these methods assume stationarity of theused time series, i.e., the series must not exhibit trends. Against the background of climatechange and proven significant trends in atmospheric circulation patterns, it is questionablewhether these changes are also reflected in the discharge data.
The aim of this PhD thesis is therefore to clarify, in a spatially-explicit manner, whetherthe available discharge data derived from selected German catchments exhibit trends.Concerning the flood hazard, the suitability of the currently used stationary FFA approachesis evaluated for the discharge data. Moreover, dynamics in atmospheric circulation patternsare studied and the link between trends in these patterns and discharges is investigated.
To tackle this research topic, a number of different analyses are conducted. The firstpart of the PhD thesis comprises the study and trend test of 145 discharge series fromcatchments, which cover most of Germany for the period 1951–2002. The seasonality andtrend pattern of eight flood indicators, such as maximum series and peak-over-thresholdseries, are analyzed in a spatially-explicit manner. Analyses are performed on differentspatial scales: at the local scale, through gauge-specific analyses, and on the catchment-wideand basin scales.
Besides the analysis of discharge series, data on atmospheric circulation patterns (CP) arean important source of information, upon which conclusions about the flood hazard can bedrawn. The analyses of these circulation patterns (after Hess und Brezowsky) and the studyof the link to peak discharges form the second part of the thesis. For this, daily data on thedominant CP across Europe are studied; these are represented by different indicators, whichare tested for trend. Moreover, analyses are performed to extract flood triggering circulationpatterns and to estimate the flood potential of CPs. Correlations between discharge seriesand CP indicators are calculated to assess a possible link between them. For this researchtopic, data from 122 meso-scale catchments in the period 1951–2002 are used. In a thirdpart, the Mulde catchment, a mesoscale sub-catchment of the Elbe basin, is studied in moredetail. Fifteen discharge series of different lengths in the period 1910–2002 are availablefor the seasonally differentiated analysis of the flood potential of CPs and flood influencinglandscape parameters.
For trend tests of discharge and CP data, different methods are used. The Mann-Kendalltest is applied with a significance level of 10%, ensuring statistically sound results. Besidesthe test of the entire series for trend, multiple time-varying trend tests are performed withthe help of a resampling approach in order to better differentiate short-term fluctuations
i
Abstract
from long-lasting trends. Calculations of the field significance complement the flood hazardassessment for the studied regions.
The present thesis shows that the flood hazard is indeed significantly increasing forselected regions in Germany during the winter season. Especially affected are the middlemountain ranges in Central Germany. This increase of the flood hazard is attributed toa longer persistence of selected CPs during winter. Increasing trends in summer floodsare found in the Rhine and Danube catchments, decreasing trends in the Elbe and Wesercatchments. Finally, a significant trend towards a reduced diversity of CPs is found causingfewer patterns with longer persistence to dominate the weather over Europe. The detailedstudy of the Mulde catchment reveals a flood regime with frequent low winter floods andfewer summer floods, which bear, however, the potential of becoming extreme.
Based on the results, the use of instationary approaches for flood hazard estimation isrecommended in order to account for the detected trends in many of the series. Throughthis methodology it is possible to directly consider temporal changes in flood series, whichin turn reduces the possibility of large under- or overestimations of the extreme discharges,respectively.
ii
Zusammenfassung
Hochwasserabschätzungen werden mit Hilfe einer Vielzahl von Methoden ermittelt. Zudiesen zählen Hochwasserhäufigkeitsanalysen, die hydrologische und hydraulische Mo-dellierung, Abschätzungen zu maximal möglichen Abflüssen wie auch Langzeitstudienund Klimaszenarien. Den meisten Methoden ist jedoch gemein, dass sie stationäre Bedin-gungen der beobachteten Abflussdaten annehmen. Das heißt, in den genutzten Zeitreihendürfen keine Trends vorliegen. Vor dem Hintergrund des Klimawandels und nachgewiesenerTrends in atmosphärischen Zirkulationsmustern, stellt sich jedoch die Frage, ob sich dieseVeränderungen nicht auch in den Abflussdaten widerspiegeln.
Ziel der Dissertation ist daher die Überprüfung der Annahme von Trendfreiheit in Abflüs-sen und Großwetterlagen, um zu klären, ob die aktuell genutzten stationären Verfahren zurHochwasserbemessung für die vorhandenen Daten in Deutschland geeignet sind. Zu prüfenist des Weiteren, inwiefern regional und saisonal eine Verschärfung bzw. Abschwächungder Hochwassergefahr beobachtet werden kann und ob eindeutige Korrelationen zwischenAbflüssen und Großwetterlagen bestehen.
Den ersten Schwerpunkt der vorliegenden Dissertation bildet die deutschlandweite Ana-lyse von 145 Abflusszeitreihen für den Zeitraum 1951–2002. Acht Hochwasserindikatoren,die verschiedene Aspekte der Hochwasser-Charakteristik beleuchten, werden analysiertund bezüglich möglicher Trends getestet. Um saisonalen Unterschieden in der Hochwasser-charakteristik der einzelnen Regionen gerecht zu werden, werden neben jährlichen auchsaisonale Reihen untersucht. Die Analyse von Maximalreihen wird durch Schwellenwert-analysen ergänzt, die die Hochwasserdynamik bzgl. Frequenz und Magnitude detailliertererfassen. Die Daten werden auf verschiedenen Skalen untersucht: sowohl für jeden einzelnenPegel wie auch für ganze Regionen und Einzugsgebiete.
Nicht nur die Analyse der Abflussdaten bietet die Möglichkeit, Bewertungen für diezukünftige Hochwasserabschätzung abzuleiten. Auch Großwetterlagen bilden eine bedeu-tende Informationsquelle über die Hochwassergefahr, da in der Regel nur ausgewählteZirkulationsmuster die Entstehung von Hochwasser begünstigen. Die saisonal differenzierteUntersuchung der Großwetterlagen und die Prüfung einer Korrelation zu den Abflüssen an122 mesoskaligen Einzugsgebieten bilden deshalb den zweiten Schwerpunkt der Arbeit.Hierzu werden tägliche Daten der über Europa dominierenden Großwetterlage (nach Hessund Brezowsky) mit Hilfe verschiedener Indikatoren untersucht. Analysen zum Hoch-wasserpotential der einzelnen Wetterlagen und weiterer Einflussfaktoren werden für dasmesoskalige Einzugsgebiet der Mulde in einer separaten Studie durchgeführt. Für dieseDetail-Studie stehen 15 Abflusszeitreihen verschiedener Länge im Zeitraum 1909–2002 zurVerfügung.
iii
Zusammenfassung
Um die Daten von Abflüssen und Großwetterlagen bezüglich vorhandener Trend zu testen,werden verschiedene Methoden genutzt. Der Mann-Kendall Test wird mit einem Signifi-kanzniveau von 10% (zweiseitiger Test) angewendet, was statistisch sichere Bewertungenermöglicht. Neben der Prüfung der gesamten Datenreihe werden multiple zeitlich-variableTrendanalysen mit Hilfe eines Resampling-Ansatzes durchgeführt. Darüber hinaus werdenräumlich differenzierte Analysen durchgeführt, um die saisonale Hochwassercharakteristikeinzelner Regionen besser zu verstehen. Diese werden durch Tests zur Feldsignifikanz derTrends ergänzt.
Mit der vorliegenden Arbeit kann gezeigt werden, dass die Hochwassergefahr für ein-zelne Regionen im Winterhalbjahr signifikant steigt. Davon sind insbesondere Gebiete inMitteldeutschland betroffen. Die Verschärfung der Hochwassergefahr durch eine längerePersistenz ausgewählter Großwetterlagen konnte ebenfalls für das Winterhalbjahr nachge-wiesen werden. Sommerhochwasser zeigen zwar ebenfalls steigende, aber auch fallendeTrends, die räumlich geclustert sind. Im Elbe- und Weser-Einzugsgebiet sinken die Abflüssesignifikant, im Donau- und Rheineinzugsgebiet steigen sie nachweisbar. Darüber hinaus isteine signifikante Abnahme der Anzahl verschiedener Großwetterlagen sowohl im Sommerals auch im Winter zu verzeichnen. Bzgl. der Studie zum Mulde-Einzugsgebiet konnteein zweigeteiltes Hochwasserregime nachgewiesen werden. In den Wintermonaten tretenhäufig kleine Hochwasser auf, die auch die Mehrheit der jährlichen Maximalwerte bilden.Sommerhochwasser sind seltener, können aber extreme Ausmaße annehmen. Ein Vergleichder geschätzten Jährlichkeiten mit verschiedenen Zeitreihen zeigt die Notwendigkeit derBerücksichtigung saisonaler Aspekte für die Bemessung von Hochwassern.
Aufgrund der Ergebnisse müssen die bisher genutzten stationären Verfahren als nichtmehr geeignet bewertet werden. Es wird daher die Nutzung instationärer Verfahren zurAbschätzung von Extremhochwasser und der damit verbundenen Bemessung von Schutz-maßnahmen empfohlen, um den teilweise vorliegenden Trends in den Daten Rechnungzu tragen. Durch diesen Ansatz ist es möglich, zeitlich dynamische Veränderungen imHochwassergeschehen stärker zu berücksichtigen. Darüber hinaus sollten saisonale Aspektedes Einzugsgebietes Eingang in die Gefahrenabschätzung finden.
iv
Contents
Abstract i
Zusammenfassung iii
List of Abbreviations ix
List of Figures xi
List of Tables xv
1 Introduction 1
1.1 Purpose & objectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51.2 Study areas . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61.3 Outline of the thesis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2 Flood trends in Germany 9
2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102.2 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142.3 Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 172.4 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.4.1 Results for the gauges Cologne/Rhine and Donauwoerth/Danube . 192.4.2 Spatial distribution of significant trends . . . . . . . . . . . . . . 202.4.3 Scale-dependency . . . . . . . . . . . . . . . . . . . . . . . . . 24
2.5 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 242.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
3 Changes in the flood hazard through changing circulation patterns 31
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 323.2 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
3.2.1 Discharge Data . . . . . . . . . . . . . . . . . . . . . . . . . . . 343.2.2 Circulation patterns . . . . . . . . . . . . . . . . . . . . . . . . . 35
3.3 Trend detection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 383.4 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
3.4.1 Trends in flood data . . . . . . . . . . . . . . . . . . . . . . . . . 403.4.1.1 Region West . . . . . . . . . . . . . . . . . . . . . . . 413.4.1.2 Region East . . . . . . . . . . . . . . . . . . . . . . . 42
v
Contents
3.4.1.3 Region South . . . . . . . . . . . . . . . . . . . . . . 433.4.2 Identification of flood triggering circulation patterns . . . . . . . 433.4.3 Trends in daily CP data . . . . . . . . . . . . . . . . . . . . . . . 43
3.4.3.1 Number of days . . . . . . . . . . . . . . . . . . . . . 463.4.3.2 Number of events . . . . . . . . . . . . . . . . . . . . 463.4.3.3 Mean persistence . . . . . . . . . . . . . . . . . . . . 48
3.4.4 Trend development of selected circulation patterns . . . . . . . . 493.4.4.1 Winter . . . . . . . . . . . . . . . . . . . . . . . . . . 493.4.4.2 Summer . . . . . . . . . . . . . . . . . . . . . . . . . 503.4.4.3 Correlation between seasonal MAXF and combinations
of CP . . . . . . . . . . . . . . . . . . . . . . . . . . . 513.5 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
3.5.1 Region West . . . . . . . . . . . . . . . . . . . . . . . . . . . . 523.5.2 Region East . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 543.5.3 Region South . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
3.6 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
4 Aspects of seasonality and flood generating circulation patterns 57
4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 584.2 Study area and data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
4.2.1 Study area . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 594.2.2 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
4.2.2.1 Discharge data . . . . . . . . . . . . . . . . . . . . . . 604.2.2.2 Precipitation Data . . . . . . . . . . . . . . . . . . . . 624.2.2.3 Atmospheric circulation patterns . . . . . . . . . . . . 62
4.3 Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 624.3.1 Flood frequency analysis . . . . . . . . . . . . . . . . . . . . . . 624.3.2 Spatial distribution of flood characteristics . . . . . . . . . . . . . 644.3.3 Relationship between precipitation maxima and discharge maxima 644.3.4 Circulation pattern and flood generation . . . . . . . . . . . . . . 65
4.4 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 664.4.1 Testing for trends in the flood AMS . . . . . . . . . . . . . . . . 664.4.2 Seasonal occurrence and magnitude of floods . . . . . . . . . . . 664.4.3 Spatial distribution of flood characteristics . . . . . . . . . . . . . 674.4.4 Landscape characteristics . . . . . . . . . . . . . . . . . . . . . . 714.4.5 Relationship between precipitation AMS and discharge AMS . . 714.4.6 Circulation pattern and flood generation . . . . . . . . . . . . . . 72
4.5 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
5 Summary and Conclusions 77
5.1 Main findings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 775.2 Consequences for the flood hazard estimation . . . . . . . . . . . . . . . 80
vi
Contents
5.3 Data constraints . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 815.3.1 Flood data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 815.3.2 Data on circulation patterns . . . . . . . . . . . . . . . . . . . . 83
5.4 Influence of the selected time period . . . . . . . . . . . . . . . . . . . . 845.5 Future research prospects . . . . . . . . . . . . . . . . . . . . . . . . . . 85
Bibliography 87
Appendix 97
Acknowledgements 101
List of Publications 105
vii
List of Abbreviations
For abbreviations of the circulation patterns see Table 3.1.
AMAXF annual maximum discharge seriesAMS annual maximum discharge seriesAO Arctic oscillationASMAXF summer maximum discharge seriesAWMAF winter maximum discharge seriesCDF cumulative distribution functionCI confidence intervalCP circulation patternFFA flood frequency analysisGEV Generalized Extreme Value distributionGL General Logistics distributionHQT probability of a flood quantileMK test Mann-Kendall testN NumberNAO North-Atlantic OscillationPOTxF peak-over-threshold series,
annual number of discharge frequencyPOTxM peak-over-threshold series,
discharge peaks above magnitude of x events per yearSL significance levelSMS summer maximum discharge seriesTFPW trend free pre-whiteningSPOTxF peak-over-threshold series,
number of discharge frequency in summerWPOTxF peak-over-threshold series,
number of discharge frequency in winter
ix
List of Figures
1.1 Study areas and location of discharge gauges: left: 145 discharge gaugesin Germany (grey circles indicate gauges of meso-scale catchments); right:The Mulde catchment in south-eastern Germany with 15 discharge gauges 6
2.1 Location of the analyzed gauges, main rivers, large river basins and eleva-tion above sea level (in m) . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.2 Observations and linear regression trends in the flood indicators given inTable 2.2. Solid lines indicate significant trend (10% significance level),dotted lines indicate no trend. Left column: gauge Cologne; right column:gauge: Donauwoerth. From top to bottom: AMAXF, AWMAXF, ASMAXF,POT1M, POT3M, POT3F, WPOT3F, SPOT3F . . . . . . . . . . . . . . . 18
2.3 Observations and linear regression trends in AMAXF and in the threetime series that resulted from the stratification of POT3M in the upper,middle and lower third (upper panel). Lower panel: Boxplots of the foursamples (Red line: median; box: interquartile range; whiskers: minimumand maximum value; crosses: outliers) . . . . . . . . . . . . . . . . . . . 20
2.4 Spatial distribution of trends in annual maximum daily mean streamflow -AMAXF (left) and in peak-over-threshold magnitude with on average oneevent per year - POT1M (right); (Upward arrows: significant increasingtrend; downward arrows: significant decreasing trend; circles: no significanttrend; size of arrows: relative change within 52 years; Mann-Kendall test,2-sided option; 10% significance level) . . . . . . . . . . . . . . . . . . . 22
2.5 Spatial distribution of trends in seasonal maximum series - AWMAXF(winter, left map), ASMAXF (summer, right map); (Upward arrows: sig-nificant increasing trend; downward arrows: significant decreasing trend;circles: no significant trend; size of arrows: relative change within 52 years;Mann-Kendall test, 2-sided option; 10% significance level) . . . . . . . . 23
2.6 Spatial distribution of trends in the peak-over-threshold frequency - POT3F(Upward arrows: significant increasing trend; downward arrows: signifi-cant decreasing trend; circles: no significant trend; size of arrows: relativechange within 52 years; Mann-Kendall test, 2-sided option; 10% signifi-cance level) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
xi
List of Figures
2.7 Spatial distribution of trends in seasonal peak-over-threshold frequency -WPOT3F (winter, left map), SPOT3F (summer, right map); (Upward arrows:significant increasing trend; downward arrows: significant decreasing trend;circles: no significant trend; size of arrows: relative change within 52 years;Mann-Kendall test, 2-sided option; 10% significance level) . . . . . . . . 25
2.8 Relative change [%] as function of basin area. Triangles indicate significantchanges at the 10% significance level. From top to bottom: AMAXF,AWMAXF, ASMAXF, POT1M, POT3M, POT3F, WPOT3F, SPOT3F . . 27
2.9 Percentages of gauges with significant trends per catchment and floodindicator. Dark grey bars show percentage of upward trends, light grey barsshow percentages of downward trends; abbr. for the catchments are D -Danube, R - Rhine, W - Weser, E - Elbe . . . . . . . . . . . . . . . . . . 28
3.1 Discharge gauges of 122 meso-scale catchments and their assignment toone of the three regions. Catchment borders illustrate the spatial coverageof the dataset. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
3.2 Composite maximum discharge time series with trends per region and floodindicator (note: in Region West the trend lines of AMAXF and AWMAXFare almost the same and therefore not easily distinguishable) . . . . . . . 39
3.3 Significance level of trends of different flood indicators for varying timeperiods for Region West (first row), Region East (second row), and RegionSouth (third row). Results between 95 and 100 indicate upward trends,results between 0 and 5 downward trends. . . . . . . . . . . . . . . . . . 41
3.4 Mean frequencies of flood causing CPs in AMAXF series for the RegionWest, East and South . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
3.5 Comparison of CP frequencies for the decades 1951–1960 and 1991–2000 473.6 Number of days per year in selected summer and winter daily CP series . 483.7 Mean duration of CPs per year in selected summer and winter daily CP series 493.8 Significance level of trends in the number of days and mean duration during
winter of the circulation patterns WZ, WS and NWZ. Results between 95and 100 indicate upward trends, results between 0 and 5 downward trends. 50
3.9 Significance level of trends in the frequency and mean duration duringsummer of the circulation patterns WZ, NWZ and TRM. Results between95 and 100 indicate upward trends, results between 0 and 5 downward trends. 51
3.10 Comparison of seasonal MAXF data and combinations of CPs (number ofdays) for Region West (first row) and Region East (second row) . . . . . . 53
4.1 Study area Mulde catchment: left: discharge gauge locations (numberedaccording to Tab.4.1) and the digital elevation model; right: geographicallocation in Germany . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
xii
List of Figures
4.2 Locations of the 49 precipitation stations in and around the study area . . 624.3 Regional trend test based on discharge data of 15 stations. . . . . . . . . . 654.4 Monthly distribution of the number of discharge AMS, summed up over
the 15 gauges for all AMS floods and for the 20% largest events . . . . . 674.5 Variation of return periods for six different floods (SD = standard deviation) 684.6 Estimated return periods (GEV, L-Moments) for the floods in 1954, 1958,
2002 (period 1929–2002 (above)) and the corresponding precipitation fields(below). Note that for a better illustration of the spatial distribution theclasses of discharge return periods and precipitation amounts differ . . . . 69
4.7 Skewness (A) and coefficient of variation (B) of the discharge AMS for the15 gauges . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
4.8 Monthly distribution of AMS discharges at Golzern and the assigned circu-lation pattern . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
4.9 Histogram of the circulation patterns at the gauge Golzern that generatedAMS discharges between 1911 and 2002 (abbr. see Table 4.2) . . . . . . 73
4.10 Flood potential of different circulation patterns to cause a flood of a certainreturn period . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
5.1 Annual maximum discharge series of gauge Rockenau/Neckar catchment(top) and flood frequency functions for the stationary and instationary GEVmodel (bottom) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
xiii
List of Tables
1.1 Research aspects of the thesis . . . . . . . . . . . . . . . . . . . . . . . . 7
2.1 Summary of recent studies on flood trends in observational data (Annualmaximum daily mean streamflow (AMAXF), Peak over threshold (POT)) 11
2.2 Flood indicators studied for all gauges . . . . . . . . . . . . . . . . . . . 162.3 Percentages of gauges showing significant trends; bold numbers indicate
field significance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
3.1 Classification of the circulation form and its specific pattern after Hess andBrezowsky (1952) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
3.2 Result of MK test (10% SL) of different flood indicators for each region(bold numbers indicate field significance) . . . . . . . . . . . . . . . . . 42
3.3 Relative change within 52 years in % (bold numbers indicate trends (MKtest, 10% SL); grey rows show flood relevant CPs) . . . . . . . . . . . . . 45
4.1 Analyzed discharge gauges in the study area (* stations with one year ofmissing values) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
4.2 Classification of the form of circulation and its specific pattern (* indicatescirculation patterns which are relevant for AMS discharges in the Muldecatchment) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
4.3 Monthly relative frequency of discharge AMS (in percent) . . . . . . . . 664.4 Percentages of the dominating landscape characteristics . . . . . . . . . . 704.5 Percentages of agreement between precipitation AMS (precipitation sums
of 24h, 48h and 72h) and discharge AMS . . . . . . . . . . . . . . . . . 71
5.1 Deviation of mean and maximum daily discharges for selected flood eventsat gauge Golzern (Mulde catchment) . . . . . . . . . . . . . . . . . . . . 83
xv
1 Introduction
Germany has a well documented floodhistory dating back more than 1000 years.Chronicles across Germany report devas-tating damages, numerous fatalities andthe outbreak of famine and serious dis-eases for a large number of flood events.Written documents for German catchmentsare available in archives spanning fromCologne to Dresden and from northernItaly to Hamburg (Weikinn, 1958; Pfister,1999). Numerous floods have occurred inthe course of the centuries, some locally,some regionally and a few countrywide.During the last two decades, several de-structive flood events in the Rhine, Elbeand Danube catchments led to the publicimpression of an overall growing flood haz-ard.
In order to prevent and mitigate floods,first prevention measures such as dikeswere constructed as early as the 12th cen-tury. To gain more detailed informationon the discharge regime and flood gener-ation along rivers, measurements of dis-charge and water level were implementedin the early 19th century (Pohl, 2004). Inthe beginning, measurements were under-taken only during flood events. Later, adetailed network of gauging stations wasestablished, which now comprises morethan 1000 stations with daily recordingsthroughout Germany. At the majorityof gauges, discharge and water level arerecorded automatically at regular intervals,
from which daily means are calculated.From these measurements, series of e.g.annual maximum discharges are derived,which are used for the estimation of theflood hazard.
Flood hazard estimation
A large variety of approaches is currentlyavailable for the estimation of flood haz-ards, upon which flood protection mea-sures are designed. Some popular meth-ods are the estimation of maximized dis-charges, the use of model chains and floodfrequency analysis (FFA). As an exam-ple, the FFA approach will be describedin more detail.
FFA is a tool for estimating the magni-tude of a flood for a given return period.For this, the exceedance probability of aflood is estimated. It is defined to be theprobability of a discharge or water levelexceeding a defined value within a giventime period (Merz, 2006). The recipro-cal of this probability is defined to be thereturn period. Thus, it is the statisticallyestimated time frame between two floodsof a prescribed extent.
The statistical distribution functionsused for FFA are simple representationsof all hydrological processes behind theseries. Two- to four-parametric functionscan be fitted to observed data (Hosking andWallis, 1997). The most common methodsfor FFA use as input data annual maximum
1
1 Introduction
series (AMAXF) and peak over thresh-old (POT) series (Institute of Hydrology,1999). In addition, seasonal maximum se-ries or seasonal POT series can be used,for instance, if one flood process type (e.g.,flash floods) is of special interest.
Independence, homogeneity and station-arity are required characteristics of thedata to legitimate flood frequency anal-ysis (Stedinger, 2000; Kundzewicz andRobson, 2004). However, changes in cli-mate, land use or the vulnerability of thestudy area question the existence of sta-tionary conditions (Merz, 2006). This iscritical since stationarity of the data is of-ten assumed (Kundzewicz et al., 2005).The presence of trends in the data canhave an important impact on the resultsof flood estimation as they can lead to over-or underestimations of the return period(Khaliq et al., 2006). In turn, this canfalsify estimations of discharge and waterlevel for a given return period. As floodprotection measures are based on theseresults, testing the data for trends is cru-cial for ensuring valid estimates. If trendsare detected in the data, flood estimationprocedures have to be adjusted by consid-ering changing regimes, e.g. by introduc-ing time-varying parameters of the floodfrequency distribution (e.g. Strupczewskiet al., 2001a; Zhang et al., 2004; Khaliqet al., 2006).
Trend detection in discharge data
Numerous test statistics are available fortrend detection (e.g. Kundzewicz and Rob-son, 2004; Svensson et al., 2005). Exam-ples for widely used tests for detecting stepchanges in the data are the Mann-Whitneytest and the Student’s t test. Tests for grad-
ual trends include, e.g., the test statisticfor linear regression and the Mann-Kendalltest. Most tests define certain requirementsfor the data. Hydrologic datasets are usu-ally not normally distributed, which partic-ularly applies to flood data. This reducesthe number of suitable trend tests. TheMann-Kendall test is a robust test, whichsets no requirements with respect to thedistribution of the data. For this reason,the test is suitable for the assessment oftrends in extremes (Kendall, 1975). It isdescribed in detail by Douglas et al. (2000)and Yue et al. (2002b).
Besides testing the entire data set fortrends, it is important to pay attention todifferent time periods and their respectivetrend behaviour. For instance, in a sub-series of a dataset a visual increase mightnot be significant at a given significancelevel. However, if a few years are addedto the series, this only visually detectableincrease might result in a significant trend.McCabe and Wolock (2002) present mul-tiple time-varying trend tests that calcu-late trends for all possible combinationsof time lengths in a series. Through thismethodology it is possible to distinguishrecent changes from trends that are stableover longer time periods. Moreover, cy-cles of upward and downward periods maybecome apparent.
The significance of trends across a re-gion can be assessed through the test forfield significance (Douglas et al., 2000;Burn and Hag Elnur, 2002; Svensson et al.,2006). Here, the number of observed sig-nificant trends is compared with the num-ber expected within a given region. Basedon this test, it is possible to draw conclu-sions on the significance of changes for theentire region in question.
2
Deciding which significance level to setis crucial for the trend results. A trendtested with a significance level of 20% onlyindicates a potential direction, whereas atrend based on a significance level of 5% isrobust and reliable (e.g. Belz et al., 2004;Kundzewicz and Robson, 2004).
A review of literature reveals that manystudies have been conducted worldwideon trends in hydrological variables suchas streamflow, water quality or snowmelt.In contrast, studies that focus on trends infloods are more seldom and often focuson catchments in North-America. Large-scale flood trend studies focusing on theEuropean continent are rarely found. Rob-son et al. (1998) and Robson (2002) inves-tigate flood trends in the UK, Lindströmand Bergström (2004) in Sweden. For Ger-many, flood trend studies are not yet avail-able for the entire country. Recent studiesare found only for parts of the Rhine catch-ment such as Lammersen et al. (2002), Pfis-ter et al. (2004a) or Pinter et al. (2006). In-vestigations on parts of the Elbe and Wesercatchments were carried out by Helms et al.(2002) and Mudelsee et al. (2006). How-ever, it is not possible to draw a comprehen-sive picture of flood trends for the entirecountry.
Atmospheric circulation patterns
Besides discharge series, data on prevail-ing atmospheric circulation patterns (CP)are an important source of information onthe flood hazard, since only selected CPsfavour the development of floods in Ger-many (e.g., Caspary and Bárdossy, 1995).Thus, the investigation of CPs provides rel-evant insights into the relationship betweenthe atmosphere and flood generation and
facilitates the assessment of the impact ofclimate change on the flood hazard.
Information about CPs across Europeis available from a number of differ-ent sources. They are either derivedby 500 hPa geopotential height fieldsfrom reanalyses data or from classificationschemes, which can be of manual or au-tomated mode. Well-known classificationschemes are provided by Hess and Bre-zowsky (1952) for all of Europe and Lamb(1972) for north-west Europe with a fo-cus on the British Isles. Lamb (1972) in-troduces a classification scheme known as"Lamb’s daily weather types (DWT)" thatis derived on a daily basis from flow direc-tions over the British Isles. This systemhas been automated by an objective deter-mination method for the circulation pat-terns (Jones et al., 1993). The deficiency ofthe Lamb model is the small spatial scale,which concentrates on areas over and nearto the British Isles (James, 2007). There-fore, it is difficult to apply this scheme toGermany.
The "Catalogue of Großwetterlagen inEurope 1881–2004" by Hess and Bre-zowsky (1952) provides information on thedaily dominant circulation pattern acrossEurope. It is derived from pressure systemsover Europe as well as from location offrontal zones. The catalogue distinguishes30 different CPs. They comprise zonal,mixed and meridional circulation forms(for details see Table 3.1). This widelyused classification scheme is currently theonly one available that captures the large-scale European pattern, while still focusingon local details (James, 2007). Althoughthe entire daily European atmospheric situ-ation is described by only one pattern, thedataset is very valuable with respect to the
3
1 Introduction
analysis of atmospheric variability and cli-mate change analyses (Bárdossy and Filiz,2005; Werner et al., 2008).
The influence of CPs on floods has notyet been studied in detail. Kingston et al.(2006) review studies on the connectionbetween climate, streamflow and atmo-spheric circulations (esp. North-AtlanticOscillation (NAO) and the Arctic Oscil-lation (AO)). Bouwer et al. (2008) corre-late four atmospheric indicators and win-ter flood discharges at 11 gauges in Cen-tral Europe. They find annual maximumdischarges to react more sensitive to vari-ability in atmospheric variables than meandischarges. Correlations between trends inthe NAO index and floods across Europeare conducted by Svensson et al. (2006).McKerchar and Henderson (2003) foundchanges in several hydrological variablesin New Zealand which are consistent withchanges in the Interdecadal Pacific Oscilla-tion.
For Germany, the link between changesin atmospheric circulations and floodtrends has only been analyzed in selectedregions (e.g. Pfister et al., 2004a; Mudelseeet al., 2006; Pinter et al., 2006). Belz et al.(2007) find increases in wet CPs, areal pre-cipitation and discharge in the Rhine catch-ment area for the period 1951–2000. How-ever, they do not correlate these variables.For south-western Germany, an increasein the westerly cyclonic pattern in winter,leading to a dramatic increase in the floodhazard, is detected by Caspary and Bár-dossy (1995). Regions in eastern Germanyare investigated by Mudelsee et al. (2004).Downward flood trends in the Elbe andOdra catchments are detected in winter,but no significant changes during summer.Moreover, floods are correlated with CPs.
All of these studies are limited to se-lected regions in Germany. Thus, a coun-trywide picture of the link between trendsin floods and atmospheric patterns cannotbe drawn based on current knowledge.
Flood seasonality in German
catchments
The flood regime in Germany is stronglydetermined by the seasonality. Winterfloods dominate the flood regime in mostcatchments. Only the southern tributariesof the Danube river are dominated bysummer floods. Changes in the flood-generating processes can have a signifi-cant impact on the seasonality of floods,as shown by Kundzewicz (2008) for manyEuropean regions. For instance, floods dueto snow melting are expected to decreasein Eastern and Central Europe. This in turncan have a profound impact on the dom-inance of flood process types in a region.In the following, a brief overview is givenabout the flood seasonality and triggeringCPs in the large German river basins.
Large parts of the Weser and Rhinecatchments are dominated by winter floods,which are often caused by westerly, south-westerly and north-westerly circulationtypes (Beurton and Thieken, 2009). Highpressure systems are rarely responsible forfloods in this region. Floods occur predom-inantly during mild and wet episodes inwinter. The flood generation due to snowmelting plays a particularly important rolein the upper parts of the Rhine catchment,i.e. in Switzerland and the middle moun-tain ranges of the Black Forest. The riversystem of the Weser is as a typical middle-mountain river system with flash floodsdue to the fast concentration in the steep
4
1.1 Purpose & objectives
valleys (Staatliches Amt für Umwelt undArbeitsschutz, 2005).
The Elbe and Odra catchments are alsomainly dominated by winter floods; how-ever, summer flooding does occur. Therunoff seasonality of both river basinsis strongly influenced by the middle-mountain ranges in the Czech Republic,Poland and Germany. Small but frequentfloods dominate during winter and thesnow melting season, which are largelytriggered by westerly and north-westerlycyclonic patterns. Summers are often drywith low flows. A larger percentage of highpressure systems, especially during fall andwinter, and the occurrence of Vb-weatherregimes distinguish the region from thewestern part of the country (Beurton andThieken, 2009). Torrential storms andlong-lasting rains can cause extreme floodevents, such as along the Odra in 1997 oralong the Elbe in 2002 (Grünewald et al.,1998; DKKV, 2003).
The flood seasonality of the Danube dif-fers depending on the region. The upper-most section as well as the tributaries com-ing from the north are dominated by winterand spring floods associated with snow-melting, which are mainly caused by west-erly patterns of varying sub-directions. Incontrast, the southern tributaries have analpine runoff regime with maxima duringthe summer months. High pressure sys-tems dominate. However, the Vb-weatherregime plays also an important role in caus-ing extreme floods.
This short overview illustrates the ne-cessity of conducting studies that considerseasonal aspects for each region in Ger-many.
1.1 Purpose & objectives
The aim of the PhD thesis is thereforeto clarify, in a spatially-explicit manner,whether the available discharge data of Ger-man catchments exhibit trends. Besidesthe analysis of annual series, a possible in-crease or decrease in the flood hazard isevaluated on different spatial scales for thewinter and summer seasons. The suitabil-ity of the currently used stationary FFAapproaches is evaluated for the studied dis-charge data. Moreover, indicators derivedfrom atmospheric circulation patterns arestudied and tested for trends. Finally, theirlink to flood discharges is investigated formost of the country.
In sum, two main research questions areextracted from this overall research topic:
• Can significant flood trends be de-tected in Germany?
• Is there a link between trends in CPsand trends in flood discharges?
Since both questions are tackled in a spa-tially and seasonally-differentiated manner,no separate research question is posed withrespect to seasonality.
A number of different analyses is per-formed in order to find appropriate answersto these questions. Series of daily meandischarge across Germany are studied andtested for trends for each single station andfor aggregated regions. The seasonalityand trend patterns of eight flood indicators,such as maximum series and peak-over-threshold series, are analyzed in a spatially-explicit manner for varying time periodsbetween 1951 and 2002. The significanceof trends for entire regions is assessed withthe help of the field significance. In order
5
1 Introduction
to study the possible link between CPs anddischarge, these data are tested for trendsfor varying time periods. The flood poten-tial of selected CPs is assessed for differentreturn periods. Moreover, CP data are cor-related with discharge data.
1.2 Study areas
Investigations for this PhD thesis are car-ried out on different spatial scales, namelygauge-specific and on the catchment-wideand basin scales. Two different study ar-eas are investigated in the thesis. The firstarea covers most of Germany. The large
drainage basins of Weser, Rhine, Elbe,Ems, Odra and Danube are represented al-together by 145 gauge stations. Due tolimited data availability, the drainage basinof the Odra is only represented by two sta-tions. Unfortunately, information from thecatchments of the Baltic and North Seasis completely lacking. Figure 1.1 gives anoverview about the location of the investi-gated regions and the location of the gaugestations.
The second study area is the Muldecatchment, a sub-catchment of the Elberiver basin in south-eastern Germany(Fig. 1.1, right). The German-Czech bor-
!.
!.!.!.
!.
!.
!.!.!.
!.
!.
!.!.
!.!.
!.
!.
!.
!.
!.!.!.
!.
!.
!.
!.
!.
!.
!.!.
!.!.!.!.!.
!.!.!.
!.
!.
!.
!.
!.
!.!.
!.
!.
!.
!.
!.
!.!.!.!.
!.
!.!.
!.
!.!.!.
!.!.
!.
!.
!.!.
!.
!.
!.!.
!.!.
!.!.
!. !.!.
!.!.
!.!.!.
!.
!.
!.
!.!.
!.!.
!.
!.
!.
!.
!.!.
!.!.!.
!.!.!.
!.
!.
!.!.
!. !.!.!.
!. !.
!.!.
!.
!.!.
!.
!.!.
!.
!.
!
!!
!
!
!!
!
!!
!
!
!
!
!
!!
!
!
!
!
!
!!
!
!
!
!
!
!
!!
!!
!
!!
!
!!
!
!
!
!
!
!
!
!
!
!!
!
!
!
!
!
!!
!!
!
!
!
!!
!
!!
!
!
!
!
!!
!
!
!!
!
!
!
!
!
!
!!
! !
!
!!
!
!
!
!
!
!
!
!!
!!
!
!
!
!
!!
!
!
!
!
!!
!!!
!
!
!
!!
! !!
!
! !
!
!
!
!
!!
!
!
!
!
!
!
!!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
Elevation
< 0 m
0 - 100 m
100.1 - 200 m
200.1 - 300 m
300.1 - 400 m
400.1 - 600 m
600.1 - 800 m
800.1 - 1,200 m
> 1,200 m
! Discharge gauges
!. meso-scale gauges
Mulde catchment
Rivers
Large river basins0 100 200 300 40050
Kilometers±
Mulde catchment
Elbe
Rhine
Weser
Odra
Danube
Figure 1.1: Study areas and location of discharge gauges: left: 145 discharge gauges in Germany
(grey circles indicate gauges of meso-scale catchments); right: The Mulde catchment in south-eastern
Germany with 15 discharge gauges
6
1.3 Outline of the thesis
Aspect Chapter II Chapter III Chapter IVFlood trends Trends in Regionalin Germany floods & circu- Analysis
lation patternsStudied time period 1951–2002 1951–2002 1910–2002Temporal resolution daily data X X
annual series X X Xseasonal series X X Xpeak over threshold X
Spatial resolution basin scale X Xcatchment based X X Xgauge specific X X X
Topics seasonality X X Xflood statistics X X Xflood trends X Xcirculation patterns X Xlandscape parameters X
Table 1.1: Research aspects of the thesis
der marks the southern boundary. Thecatchment has a size of 6171 km2 (at thegauge Bad Düben). The river disemboguesinto the Elbe River north of the city ofDessau. The steep mountain ranges in thesouth lead to fast runoff generation in thetributaries, whereas runoff generation ismuch slower in the lowlands. This catch-ment was chosen for its very good database of long discharge series and the directavailability of digital information about dif-ferent catchment characteristics.
1.3 Outline of the thesis
This introductory part is followed by threechapters. They were written as stand-alone manuscripts that have either beenpublished or are awaiting publication ininternational peer-reviewed journals. Thefull reference of each article (i.e. chapter)is provided below its abstract. Given the
overlap of data and topics, it is unavoid-able that small excerpts or tables occur inmore than one chapter. Table 1.1 gives anoverview of the data and aspects coveredin the thesis.
Chapter 2 investigates, seasonally dif-ferentiated, the presence of trends in dis-charge data from all gauges presented inFig. 1.1 (left). The size of the catch-ments ranges from 500 km2 to 159300km2. Eight flood indicators, which fo-cus on different flood characteristics, arederived from daily mean discharge series.Conclusions are drawn for the large riverbasins of Elbe, Rhine, Danube and Weserconcerning trends in winter and summerfloods, respectively.
Based on the findings in Chapter 2,Chapter 3 investigates the relationship be-tween trends in CPs and flood trends for122 meso-scale catchments with areas from500 km2 to 27088 km2. The catchments
7
1 Introduction
represent a subset of the dataset in Chap-ter 2. In Fig. 1.1 (left), all gauges with agrey border represent this subset of data.In both chapters, a common period of 52years (1951–2002) is used. Four indicatorsare derived from daily CP data reflectingthe frequency and persistence of these pat-terns across Europe. These are tested fortrends. Analyses to extract flood trigger-ing CPs are performed for three large re-gions in Germany. In addition, multipletime-varying trend tests are conducted forflood and CP indicators. Finally, dischargeseries and CP frequencies are correlated.Conclusions are drawn on the link betweenatmospheric circulation patterns and floodsfor Germany.
A detailed analysis of the Mulde catch-ment in Saxony (6171 km2) is presentedin Chapter 4, which focuses on flood trig-gering catchment characteristics and cir-
culation patterns. Datasets of 15 gaugeswith different lengths (42 to 92 years) arestudied (Fig. 1.1, right). Note that some ofthese gauges are not part of the datasets in-vestigated in the Chapters 2 and 3. The in-fluence of the seasonality and different in-dicators such as precipitation, soil or land-use on flood generation is studied. In ad-dition, flood frequency analyses are per-formed for all data, and the flood poten-tial is estimated for different return periodsand CPs. Finally, conclusions on flood fre-quency analysis are presented, concerningthe use of seasonally differentiated maxi-mum series for such estimations.
Overall conclusions, arising from the re-search findings in the three chapters, arepresented in Chapter 5. Finally, recom-mendations and an outlook with furtherresearch foci are presented.
8
2 Trends in flood magnitude, frequency and
seasonality in Germany in the period
1951–2002
Abstract
During the last decades several destructive floods in Germany led to the impression thatthe frequency and/or magnitude of flooding has been increasing. In this study, flood timeseries are derived and analyzed for trends for 145 discharge gauges in Germany. A commontime period of 52 years (1951–2002) is used. In order to obtain a country-wide picture, thegauges are rather homogeneously distributed across Germany. Eight flood indicators arestudied, which are drawn from annual maximum series and peak over threshold series. Ouranalysis detects significant flood trends (at the 10% significance level) for a considerablefraction of basins. In most cases, these trends are upward; decreasing flood trends are rarelyfound and are not field-significant. Marked differences emerge when looking at the spatialand seasonal patterns. Basins with significant trends are spatially clustered. Changes inflood behavior in northeast Germany are small. Most changes are detected for sites in thewest, south and center of Germany. Further, the seasonal analysis reveals larger changes forwinter compared to summer. Both, the spatial and seasonal coherence of the results andthe missing relation between significant changes and basin area, suggest that the observedchanges in flood behavior are climate-driven.
Published as: Petrow, Th., and B. Merz. 2009. Trends in flood magnitude, frequency
and seasonality in Germany in the period 1951–2002. Journal of Hydrology, 371, 129–
141.
9
2 Flood trends in Germany
2.1 Introduction
During the last decades several severefloods in different river basins in Germa-ny (e.g., 1993 and 1995 Rhine; 1997Odra; 1999, 2002 and 2006 Danube; 2002and 2006 Elbe) caused extensive inunda-tions and high flood damages (Grünewaldet al., 1998; Disse and Engel, 2001; DKKV,2004; Thieken et al., 2006). In the societyand the media the impression grew that theflood situation is worsening in Germany,and that this perceived accumulation oflarge floods cannot be explained by naturalvariability. In view of the current debate onclimate change, the worry that floods arebecoming more severe or more frequentis rapidly gaining ground in the Germanpublic.
Flood estimation and flood design aretraditionally based on the assumption thatthe flood regime is stationary. In particu-lar, flood frequency analysis requires theflood data to be homogeneous, indepen-dent and stationary. Trends can have aprofound effect on the results of flood fre-quency estimation and can undermine theusefulness of the concept of return period(Khaliq et al., 2006). If trends are present,flood estimation procedures have to allowfor changing flood regimes, e.g., by assum-ing time-varying parameters of the floodfrequency distribution (e.g. Strupczewskiet al., 2001a,b).
There are many studies worldwide thatanalyze trends in different hydrologicalvariables. Examples are the studies ofAdamowski and Bocci (2001) on annualminimum, mean, and maximum stream-flow, Kunkel et al. (1999) on extreme pre-cipitation events, McNeil and Cox (2007)on stream salinity and groundwater levels,
and Johnson and Stefan (2006) on lake icecover, snowmelt runoff timing and streamwater temperatures. However, only fewstudies can be found which focus on floodtrends.
Table 2.1 summarizes the main findingsof recent studies on flood trends. Thesestudies have in common that large regionsand a large number of discharge time series,measured at different gauges, are analyzed.This approach has two advantages. Firstly,it may be possible to identify spatial pat-terns in flood trends, and to distinguishbasins which have been changing fromthose which have been stable. Secondly,the assessment of many gauges within oneregion may improve the signal-to-noise ra-tio. By studying single basins, existingtrends may not be identifiable due to localnoise and anthropogenic influences, suchas river training works. The joint analysisof many basins decreases the influence oflocal noise. If trends can be identified thatare coherent across several basins, thesetrends can be assumed to be a clear sig-nal of change, and not the result of local,possibly random influences.
Climate change is not the only possi-ble driver of change in flood time series.Germany is densely populated and has along history of water resources manage-ment. Its basins and rivers cannot be as-sumed to be pristine. Most of the basins inGermany have undergone widespread landuse changes, significant volumes of floodretention have been implemented in the lastdecades, and many rivers have experiencedriver training works (e.g. Helms et al.,2002; Lammersen et al., 2002; Mudelseeet al., 2004; Pfister et al., 2004a). In partic-ular, the active floodplains of many riversin Germany have been reduced through
10
2.1 IntroductionF
loo
din
dic
ato
r
Stu
dy
Reg
ion
No
of
sta
tio
ns
Ob
serv
ati
on
per
iod
AMAXF
Seasonal
AMAXF
POT
Other
Ma
infi
nd
ing
s
Sve
nsso
net
al.(
2005
)W
orld
wid
e21
44-
100
year
s;av
.68
year
sX
X•
noge
nera
lpa
tter
nof
decr
easi
ngor
incr
easi
ngnu
mbe
rsor
mag
nitu
des
offl
oods
Kun
dzew
icz
etal
.(20
05)
Wor
ldw
ide
195
vary
ing
peri
ods;
min
.40
year
sX
•at
27st
atio
nssi
gnifi
cant
incr
ease
sw
orld
wid
e
•at
31st
atio
nssi
gnifi
cant
decr
ease
sw
orld
wid
e
•at
137
stat
ions
nosi
gnifi
cant
chan
ges
•no
ubiq
uito
uspa
tter
nw
orld
wid
e
Ada
mow
ski
&B
occi
(200
1)C
anad
a24
8;po
oled
into
10re
gion
s19
57–1
997
X•
sign
ifica
ntly
decr
easi
ngtr
ends
inw
este
rn,n
orth
ern
and
cent
ral-
east
ern
Can
ada
•si
gnifi
cant
lyin
crea
sing
tren
dsin
cent
ral
Can
ada
and
the
Pra
irie
s
Bur
n&
Hag
Eln
ur(2
002)
Can
ada
59A
MA
XF
;va
ryin
gfo
rot
her
1950
–199
7,va
ryin
gpe
riod
sX
X•
for
1950
–199
7,in
gene
ral,
decr
easi
ngtr
ends
inA
MA
XF
inth
eso
uth
and
•in
crea
sing
tren
dsin
the
nort
hern
regi
ons
ofC
anad
a
McC
abe
&W
oloc
k(2
002)
US
A40
019
41–1
999
XX
•re
lati
vely
few
site
sw
ith
incr
easi
ng/d
ecre
asin
gtr
ends
inA
MA
XF
•in
crea
ses
inA
MA
XF
(and
min
imum
and
med
ian
stre
amfl
ow)
appe
aras
step
chan
gear
ound
1970
•fr
eque
ncy
ofda
ysw
ith
disc
harg
es>
99th
perc
enti
lesh
ows
incr
easi
ngtr
ends
only
atfe
wsi
tes
Dou
glas
etal
.(2
000)
US
A14
7418
74–1
988
(av.
48ye
ars)
X•
notr
ends
infi
eld
sign
ifica
nce
inal
lth
ree
geog
r.re
gion
s(E
ast,
Mid
wes
t,W
est)
•no
tren
dsin
fiel
dsi
gnifi
canc
ein
all
nine
hydr
olog
icre
gion
s
Fra
nks
(200
2)
New
Sou
thW
ales
(Aus
tral
ia)
40va
ryin
gpe
riod
sw
ithi
n19
10–1
990
X•
step
chan
gein
AM
AX
Far
ound
the
1940
s
•pr
ior
1945
nofl
oods
ofhi
ghm
agni
tude
,aft
er19
45m
arke
din
crea
se
Lin
dstr
öm&
Ber
gstr
öm(2
004)
Sw
eden
43;
pool
edda
tase
ts19
01–2
002
XX
•sl
ight
incr
ease
inA
MA
XF
inno
rthe
rnS
wed
en
•su
mm
eran
dau
tum
nfl
oods
incr
ease
dco
nsid
erab
lybe
twee
n19
70an
dto
day
•no
incr
ease
dfr
eque
ncy
offl
oods
wit
hre
turn
peri
ods
>10
year
sfo
rpo
oled
data
Rob
son
etal
.(1
998)
and
Rob
son
(200
2)
UK
890
for
PO
T,
1000
for
AM
AX
F,po
oled
acro
ssal
lsi
tes
1941
–198
0;19
41–1
990
XX
X•
nosi
gnifi
cant
tren
dsfo
ran
nual
and
seas
onal
floo
dti
me
seri
es
•qu
asi-
cycl
ical
fluc
tuat
ions
over
5–10
year
peri
ods
inP
OT
3an
dA
MA
XF
;co
uld
beli
nked
toan
nual
rain
fall
fluc
tuat
ions
Table 2.1: Summary of recent studies on flood trends in observational data (Annual maximum daily
mean streamflow (AMAXF), Peak over threshold (POT))
11
2 Flood trends in Germany
the construction of dykes. Pfister et al.(2004a) summarized the impacts of landuse change in the Rhine catchment. Al-though the Rhine catchment has experi-enced widespread land use changes, signif-icant effects on flooding could only be de-tected in small basins. There is no evidencefor the impact of land use changes on theflood discharge of the Rhine River itself.These findings are in line with differentstudies, which found little or no influenceof land use on flood discharge (Blöschlet al., 2007; Robinson et al., 2003; Svens-son et al., 2006). Blöschl et al. (2007) ar-gued that the impact of land use changes onfloods is a matter of spatial scale. In smallbasins land use changes can significantlyalter the runoff processes, effecting floodmagnitude and frequency. However, theseeffects are expected to fade with increasingbasin scale.
The general tendency of decreasing im-pacts with increasing basin scale does notapply to river training works. On the con-trary, river training impacts are likely toincrease with catchment size as there is atendency for larger settlements and hencelarge-scale flood protection works at largerstreams (Blöschl et al., 2007). The cumu-lative effects of river training works onfloods in large basins are difficult to assess.Large-scale hydraulic models are neces-sary that are able to consider the effects offlood protection, such as river dykes, onthe flood waves. Therefore, reliable quan-tifications of these effects are rare. To com-plicate matters, the effects are expectedto vary with flood magnitude. For exam-ple, Apel et al. (in press) investigated theimpact of dyke breaches along the lowerRhine. They showed that dyke overtoppingand successive dyke breaching lead to large
retention effects due to the inundation ofthe dyke hinterland. Since large retentionvolumes are activated as consequence ofdyke failures, flood peaks are significantlyreduced downstream of breach locations.These effects, however, occur only for rarefloods with return periods larger than ap-prox. 400 years (Apel et al., in press). Lam-mersen et al. (2002) analysed the effects ofriver training works and retention measureson the flood peaks along the river Rhine.The construction of weirs along the upperRhine in the years 1955–77 accelerated theflood wave, leading to a higher probabilitythat the flood peak of the Rhine coincideswith the peaks of its tributaries, such as theNeckar. After 1977, extensive retentionmeasures along the main stream have beenplanned and partially implemented. Aver-aged across many flood events, the rivertraining works have increased the floodpeaks at Cologne and the retention mea-sures have decreased the peaks, howeverto a smaller extent. Today’s flood peaksat Cologne are expected to be a few per-cent higher than before the extensive rivertraining works in the 1950s.
The detection of coherent flood trends atmany sites in a geographic region may al-low distinguishing climate-related changesfrom other anthropogenic changes. Al-though local effects and anthropogenic in-fluences, such as flood control measures,may markedly influence the at-site floodbehaviour, such changes are not expectedto cause coherent changes over a large ge-ographical area.
Table 2.1 lists nine recent studies whichanalyzed flood trends in a large number ofbasins. All studies used the annual maxi-mum streamflow (AMAXF) as flood indi-cator. Other indicators, that are less often
12
2.1 Introduction
used, are peak over threshold time series(POT) or streamflow percentiles. In twocases only, seasonal AMAXF have beenused, differentiating between summer andwinter floods. Table 2.1 shows that thereare no ubiquitous flood trend patterns andthat seasonally and regionally different pat-terns in flood trends have to be expected.This result calls for trend analyses that takeinto consideration seasonal and spatial dif-ferences. Most of the studies compiledin Table 2.1 have their regional focus onNorth-America. Only few studies on floodtrends in Europe, covering large regions orentire countries, could be found. These arethe studies of Robson et al. (1998) and Rob-son (2002) for UK, and of Lindström andBergström (2004) for Sweden. Besides,there are recent studies on flood trends forlarge areas or sub-catchments such as Lam-mersen et al. (2002), Pfister et al. (2004a),or Pinter et al. (2006), who studied issueson the flood hazard for parts of the Rhinecatchment. Also studies for parts of theElbe and Weser catchments were compiled,e.g. by Helms et al. (2002) or Mudelseeet al. (2006).
Main problems of flood trend analysisare data availability and data reliability.Many discharge time series are short andare not suited for analyzing extreme events.Kundzewicz et al. (2005) suggested a min-imum length of 50 years for flood trenddetection. For some gauges there are sys-tematic discharge observations in the rangeof 100 years or even more. Such time se-ries are very valuable; however, the qualityof these data has to be examined carefully.Lindström and Bergström (2004) empha-sised the need to balance availability andreliability: very long discharge time seriesmight not be reliable, but reliable series
might be too short.There exist some studies on flood
trends in Germany, which are however re-stricted to specific regions or catchments.Mudelsee et al. (2006) analyzed floodtrends during the last 500 years in theWerra catchment (sub-catchment of theWeser). Winter flood hazard showed anincrease during the last decades, whereasthe summer flood hazard showed a long-term decrease from 1760 on. Mudelseeet al. (2004) analyzed winter and summerflood trends at six gauges at the middleElbe and middle Odra and found signifi-cant downward trends in the occurrencerates of winter floods and no significanttrends for summer floods during the 20th
century. Moreover, they found significantvariations of occurrence rates for heavyfloods during the past centuries and notabledifferences between Elbe and Odra. Simi-larly, Grünewald (2006) illustrated that theseasonal distribution of floods at the gaugeDresden/Elbe varied significantly duringthe last 1000 years. Caspary and Bárdossy(1995) analyzed AMAXF of gauges alongthe river Enz in south-western Germany(sub-catchment of the Rhine) for the pe-riod of 1930–1994. They identified signif-icant upward trends in AMAXF. Bendix(1997) found significant trends in magni-tude, whereas Pinter et al. (2006) found sig-nificant upward flood trends in magnitudeas well as frequency in the Rhine catch-ment at the gauges Cologne (1900–2002)and Bonn. An increased flooding proba-bility was also suggested for the middleand lower Rhine by Pfister et al. (2004a).In the Danube and Rhine catchments (for5 gauges with varying time periods) up-ward trends in AMAXF were detectedby Caspary (1995) and Caspary and Bár-
13
2 Flood trends in Germany
dossy (1995). KLIWA (2007) analyzedflood trends of 158 gauges in southernGermany. Long time series of 70–150years mostly revealed no trends. How-ever, the study of the last 30 years showedat many gauges significant upward trendsin AMAXF. Moreover, the frequency ofwinter floods increased since the 1970s inmany basins.
This compilation of the trend analysesfor German rivers shows that there is no un-ambiguous pattern of flood trends acrossGermany. Further, the studies availableare limited to selected regions or singlebasins. There is no comprehensive studyon flood trends in Germany which coversthe entire country. This gap is filled by thispaper for the period 1951–2002. This isa period with (1) a good coverage of dis-charge sites with reliable observations, and(2) significant increases of concentrationsof atmospheric greenhouse gases. Section2 and 3 introduces the data and the meth-ods, respectively. In section 4 the resultsare presented for eight flood indicators. Fi-nally, the findings are discussed againstthe background of studies on recent tem-poral changes in atmospheric circulationpatterns (section 5). In particular, it is dis-cussed if the identified changes are causedby climate variability or by other drivers.
2.2 Data
Discharge time series were obtained fromthe water authorities of different federalstates in Germany. Since the data are partof the hydrometric observation network ofthe water authorities in Germany, the obser-vations are regularly checked and can beassumed to be of good reliability, although
it is acknowledged that flood peak measure-ments are frequently associated with con-siderable errors. Sites were selected witha catchment size of at least 500 km2. Inthat way, small catchments were excludedfrom the analysis but still a large numberof gauging stations and a satisfying spa-tial coverage of Germany were obtained.Although there is considerable uncertaintyabout the scale where changes in land useand land management in a specific basincannot be seen anymore in the basin floodhydrograph (Blöschl et al., 2007), 500 km2
seems a reasonable choice for the lowerlimit. Beyond that scale, most of the effectsof land use and land management are ex-pected to have been faded out (e.g. Ihringer,1996; Michaud et al., 2001; Bronstert et al.,2002). A common time period between1.11.1951 and 30.10.2002 was used (hy-drological year in Germany: 1 Novemberto 31 October). Small gaps in the data ofup to one year were marked as "missingvalues". This was necessary at only fivegauges. Time series with larger succes-sive gaps were excluded from the analysis.Finally, time series of mean daily stream-flow from 145 gauges in Germany wereincluded in the analysis. They are rela-tively homogeneously distributed acrossGermany (Fig. 2.1). 43 stations are locatedin the Danube catchment, 37 in the Rhinecatchment, 32 in the Elbe catchment and 27in the Weser catchment. The catchments ofEms and the small German part of the Odraare represented by four and two gauge sta-tions, respectively. Only the Maas and theBaltic Sea catchments are not representedin the study.
Germany is located in the thermal-hydrologic transition zone between the At-lantic Western Europe and the continen-
14
2.2 Data
tal climate in eastern Germany. The en-tire country is influenced throughout theyear by westerly atmospheric circulationtypes. However, the influence decreasesfrom west to east. The Rhine and Wesercatchments are dominated by westerly,north-westerly and south-westerly circula-tion types with associated mid-latitude cy-clone rainfall (Beurton and Thieken, 2009).High pressure systems occur rarely exceptfor spring, and Vb-weather regimes areinfrequent in north-eastern part. Floodsoccur predominantly during the mild andwet winter. The Weser as well as partsof the upper Rhine catchments are foundin the transition zone from Atlantic tocontinentally influenced climates. There,floods also occur mainly during the win-ter time however the share of summerflood events increases from west to east(Beurton and Thieken, 2009). The Elbe,Danube and Odra catchments are charac-terized by a smaller influence of westerly,north-westerly and south-westerly circu-lation types, a larger share of high pres-sure systems, and the occurrence of Vb-weather regimes. The Vb-weather regimeis a trough over Europe, which can bringlong-lasting heavy rainfalls causing de-structive floods in these catchments. Al-though winter floods dominate in the Elbe,Odra and northern Danube catchments,summer floods can reach remarkable dis-charges as experienced in 1997, 2002,2005 (DKKV, 2003). The southern part ofthe Danube catchment is dominated highpressure systems, especially during fall andwinter. Westerly, north-westerly and south-westerly circulation types are less frequent.In this region, summer floods dominate.
Selected results are shown exemplarilyfor the gauges Cologne (catchment size
!
!!
!
!
!
!
!
!!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!!
!
!!
!
! !
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!!
!
!
!
!
!
!
!!
!
!
!
!
!
!
!
!
!!
!
!
!
!
!
!
!
!
!!
!
!!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!!
!
!
!
!
!!
!
!!
!
!
!
!
!
! !!
!
!!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
Elbe
Rhine
Weser
Danube
Baltic Sea
Em
s
Od
ra
Maas
0 50 100 150 20025Kilometers
–
< 0 m
0 - 100 m
100.1 - 200 m
200.1 - 450 m
450.1 - 700 m
700.1 - 1,000 m
1,000.1 - 1,200 m
> 1,200 m
! Gauge
Cologne
Donauwoerth
Figure 2.1: Location of the analyzed gauges,
main rivers, large river basins and elevation
above sea level (in m)
144323 km2) in the Rhine catchment andDonauwoerth (catchment size 15037 km2)in the Danube catchment (Fig. 2.1). Thesegauges were selected because their be-haviour can be seen to be representativefor most gauges in Germany. The gaugeCologne is dominated by winter floodsand slowly rising water levels, which istypical for most gauges in the Rhine,Weser, as well as parts of the Elbe catch-ments. The discharge behaviour at Donau-woerth (Danube) is dominated by summerfloods and represents gauges in the moun-tain ranges with faster runoff regimes, es-pecially in the catchments of Elbe andDanube.
Eight flood indicators, which are listedin Table 2.2, were included in our study.
15
2 Flood trends in Germany
Flood indicator Abbreviation RemarksAnnual maximum daily AMAXF Maximum discharge formean streamflow [m3/s] each hydrological year
(1 Nov - 31 Oct)Annual winter maximum AWMAXF Maximum discharge fordaily mean streamflow [m3/s] each hydrological winter
(1 Nov - 31 Mar)Annual summer maximum ASMAXF Maximum discharge fordaily mean streamflow [m3/s] each hydrological summer
(1 Apr - 31 Oct)Peak-over-threshold POTXM Discharge peaks abovemagnitude [m3/s] threshold; on average
X events per yearPeak-over-threshold POT3F Annual number of dischargefrequency peaks above threshold;
on average 3 events per yearSummer peak-over-threshold SPOT3F Annual number of summerfrequency discharge peaks above thresholdWinter peak-over-threshold WPOT3F Annual number of winterfrequency discharge peaks above threshold
Table 2.2: Flood indicators studied for all gauges
These comprise annual maximum stream-flow series (AMAX) as well as peak overthreshold series (POT). Annual maximumdaily mean streamflow, i.e. the largest dailymean streamflow that occurs in each hydro-logical year, is the most common indicatorin flood trend studies. In some studies,POT series are used since they are consid-ered to include more information and thusallowing to reveal better the temporal pat-tern of flood occurrence (Svensson et al.,2006). Besides the detection of trends inflood magnitude, they offer the possibilityto analyze the flood frequency, i.e. changesin the number of floods occurring eachyear.
We selected the 52 largest independentflood events (POT1) and another serieswith on average three events per year forthe POT time series (POT3). For our time
frame of 52 years (1951–2002) the POT3samples include the largest 156 indepen-dent discharge peaks. In order to ensureindependence of the different flood events,we tested different time spans of 10, 20and 30 days. Svensson et al. (2005) usedthresholds which depended on catchmentsize: 5 days for catchments < 45000 km2;10 days for catchments between 45000 and100000 km2; 20 days for catchments >100000 km2. In our study, 85% of thecatchments are smaller than 45000 km2.Following Svensson et al. (2005) a 10 daytime span would be sufficient for mostgauges. Visual inspection of the hydro-graphs of some of the larger catchmentsas well as the spatial distribution on themap of the trend results of the differentflood indicators with different time spanssupported the time frame of 10 days to be
16
2.3 Methodology
sufficiently long to ensure independence ofthe extracted flood peaks. POT1 and POT3variables were selected by the magnitudeof the flood events (POT1M, POT3M) andthe frequency per year (POT3F). For this,the number of POT3M events for everyyear was counted.
In addition to the annual flood time se-ries, seasonal time series were derived, dis-tinguishing between winter (1 November -31 March) and summer (1 April - 31 Octo-ber). For example, the annual winter maxi-mum streamflow time series (AWMAXF)consists of the largest daily mean dischargeof the winter period of each year. In thecase of the POT time series, POT3F wasseparated in summer and winter events.For example, the winter POT3F time se-ries (WPOT3F) indicates the number offloods within the winter period, given that,on average, three events were selected peryear.
2.3 Methodology
There are different possibilities for test-ing for change in hydrological time series(e.g. Kundzewicz and Robson, 2004). Inthis study we used the Mann-Kendall test(Kendall, 1975), a robust non-parametrictest. The Mann-Kendall test is particularlyuseful for the analysis of extreme, not nec-essarily normally-distributed data (Kunkelet al., 1999). It has been used by manystudies on trends in hydrological time se-ries (e.g. Chen et al., 2007). We appliedthe 2-sided option with 10% significancelevel.
The Mann-Kendall test requires the datato be serially independent. von Storch andNavarra (1995) found that, if the data are
positively serially correlated, the Mann-Kendall test tends to overestimate the sig-nificance of a trend. To correct the data forserial correlation, the procedure of trendfree pre-whitening (TFPW) was applied,which is described in detail in Yue et al.(2002a, 2003). Firstly, the trend of a timeseries is estimated by the non-parametrictrend slope estimator developed by Sen(1968). This estimation of the trend slopeβ is more robust than a normal linear re-gression (Yue et al., 2003). β is the medianof all pair wise slopes in the time series:
β =x j − xi
j− i(2.1)
for all i < j; xi,x j = discharge values inyears i, j. Secondly, the calculated trend isremoved from the original series:
Yt = Xt −β ∗ t (2.2)
with Xt being the original time series andt the time. Third, the lag1-autocorrelation(acf) is calculated. If no significant auto-correlation is found, the Mann-Kendall testis directly applied to the original time se-ries. Otherwise, the lag1-autocorrelation isremoved from the time series:
Y′
t = Yt −ac f ∗Yt−1 (2.3)
The Y′
t time series should now be free ofa trend and serial correlation. Finally, thefirstly removed trend is included back intothe time series.
Y′′
t = Y′
t +β ∗ t (2.4)
The resulting time series Y′′
t is a blendedtime series including the original trend butwithout autocorrelation.
In trend detection studies, that analyzemany sites within a region, it is interest-ing to assess the field significance, i.e. the
17
2 Flood trends in Germany
significance of trends across the region(Douglas et al., 2000; Burn and Hag El-nur, 2002; Svensson et al., 2006). Thisis done by comparing the number of ob-served significant trends with the numberexpected within the region. Douglas et al.(2000) found that the existence of spatialcorrelation between sites may inflate theresults of change detection, if the spatialcorrelation is not accounted for. They pro-posed a bootstrapping test for assessing thefield significance of trends with preservingthe cross-correlation among sites. How-ever, this approach might be suitable onlyfor the case that the majority of trends in
a region are uniform, i.e. either upward ordownward (Yue et al., 2003). Therefore,we applied a slightly refined approach, pro-posed by Yue et al. (2003), which assessesthe field significance of upward and down-ward trends separately. In short, the testworks as follows (for details see Yue et al.(2003)):
1. The selected range of years is resam-pled randomly with replacement. Anew set is obtained with differentyear order but with the same length.
2. The observation values of each siteare rearranged according to the new
500
1000
500
1000
500
1000
500
1000
500
1000
0
4
8
0
4
8
1950 1960 1970 1980 1990 20000
4
8
5000
10000
5000
10000
5000
10000
5000
10000
5000
10000
0
4
8
0
4
8
1950 1960 1970 1980 1990 20000
4
8
Figure 2.2: Observations and linear regression trends in the flood indicators given in Table 2.2. Solid
lines indicate significant trend (10% significance level), dotted lines indicate no trend. Left column:
gauge Cologne; right column: gauge: Donauwoerth. From top to bottom: AMAXF, AWMAXF,
ASMAXF, POT1M, POT3M, POT3F, WPOT3F, SPOT3F
18
2.4 Results
year set obtained in step 1. In thisway, the spatial correlation of obser-vation values is preserved, whereasthe temporal order is destroyed.
3. The Mann-Kendall test is appliedto the synthetic time series of eachsite. At the given significance level,the number of sites with signifi-cant upward trends (N∗
up) and down-ward trends (N∗
down), respectively, iscounted.
4. By repeating steps 1–3 1000 times, 2samples with a sample size of 1000each are obtained.
5. The probability of the number of sig-nificant upward (downward) trendsNobs
up (Nobsdown) for the observed time
series is assessed by comparing Nobsup
(Nobsdown) with the empirical cumula-
tive distribution of N∗up (N∗
down). Ifthis probability is smaller than thesignificance level, then the trend isjudged to be field-significant.
2.4 Results
2.4.1 Results for the gauges
Cologne/Rhine and
Donauwoerth/Danube
We are mainly interested in the question,if there are coherent spatial patterns offlood trends across Germany during thelast five decades. However, to facilitatethe understanding of the spatial results, thetrend analyses are exemplarily discussedfor the gauges Cologne/Rhine and Donau-woerth/Danube (locations see Fig. 2.1).
Fig. 2.2 shows the observations and thelinear regression trends in the eight floodindicators given in Table 2.2. For bothsites, significant upward trends in AMAXFwere found. The comparison of the annualmaxima with the seasonal maxima showsthat, in the case of Cologne, AMAXF isdetermined by floods in the winter season.Summer floods are significantly smallerthan winter floods. Both seasonal time se-ries show upward trends, however, theyare not significant at the 10% significancelevel. In the case of Donauwoerth, summerfloods are only slightly smaller than win-ter floods. Increasing trends were detectedin both seasons; however, the trend in thewinter season is not significant. Althoughthe linear regression trends in AMAXF andAWMAXF have equal gradients, the trendin AWMAXF is not significant due to thelarger standard deviation of AWMAXF.
Contrary to AMAXF, no trends inPOT1M and POT3M were identified forboth gauges. Actually, both trend lines ofPOT3M show small decreases. That meansthat a significant increase in the numberof discharge peaks above the thresholddoes not necessarily comply with a signif-icant increase in the magnitude of thesepeaks - a result that was also found bySvensson et al. (2005). To understandthis discrepancy, the POT3M time serieswere further separated in the upper, mid-dle and lower third. Fig. 2.3 comparesAMAXF with these three samples. ThePOT thirds have a much smaller range com-pared to AMAXF. For both gauges, thePOT time series show no or only mild in-creases, whereas AMAXF grows signifi-cantly. This marked increase is mainly aresult of several very small annual floodsthat were lower than the POT3M threshold.
19
2 Flood trends in Germany
AMAXF POT upper third POT middle third POT lower third
100
200
300
400
500
600
700
800
900
1000
Dis
charg
e in m
³/s
Donauwoerth
AMAXF POT upper third POT middle third POT lower third
2000
3000
4000
5000
6000
7000
8000
9000
10000
Dis
charg
e in m
³/s
Cologne
1951 1956 1961 1966 1971 1976 1981 1986 1991 1996 2002100
200
300
400
500
600
700
800
900
1000
Dis
ch
arg
e in
m³/
s
Hydrological Year
Donauwoerth
POT lower third
POT middle third
POT upper third
AMAXF
1951 1956 1961 1966 1971 1976 1981 1986 1991 1996 20022000
3000
4000
5000
6000
7000
8000
9000
10000
11000
Dis
ch
arg
e in
m³/
s
Hydrological Year
ColognePOT lower third
POT middle third
POT upper third
AMAXF
Figure 2.3: Observations and linear regression trends in AMAXF and in the three time series that
resulted from the stratification of POT3M in the upper, middle and lower third (upper panel). Lower
panel: Boxplots of the four samples (Red line: median; box: interquartile range; whiskers: minimum
and maximum value; crosses: outliers)
Interestingly, these small discharge valuesoccurred exclusively during the first halfof the time period. The larger dischargepeaks, represented by POT upper third, in-creased only slightly.
The POT3F time series of both gaugesshow a very similar behaviour (Fig. 2.2).Trends in POT3F are upward and signif-icant. For both sites, the frequency ofdischarge peaks above the POT3M thresh-old increased, although the magnitude ofthese events (POT3M) did not experiencea significant change. The seasonal separa-tion of POT3F yielded significant increas-
ing trends for winter (WPOT3F), i.e. thenumber of high discharge events duringthe winter season grew. For both cases,the number of discharge peaks in summer(SPOT3F) above the POT3M threshold in-creased as well; however, this increasedoes not suffice to be significant.
2.4.2 Spatial distribution of
significant trends
In this section the spatial distribution ofsignificant upward and downward trends isshown and the field significance is calcu-
20
2.4 Results
lated for the different flood indicators. Allmaps showing the magnitude and directionof significant trends use the same mark-ers: Upward arrows indicate significantincreasing trends, and downward arrowsshow significant decreasing trends. Thesize of the arrows corresponds in all mapsto the relative increase ∆XR within 52 years(1951–2002):
∆XR =X∗
2002 −X∗1951
X∗100% (2.5)
X∗2002 and X∗
1951 is the value of the esti-mated trend line at the end and at the startof the analyzed time period, respectively.X is the mean value of the time series ofthe period 1951–2002.
The majority of the 145 gauges showedat least one significant result when ana-lyzed for trend in the eight flood indi-cators. In the Elbe catchment, no trendin any of the flood indicators was foundfor nearly 60% of the gauges, whereasin all other catchments 50% - 75% ofthe gauges showed at least one signifi-cant trend. 42% of the Danube gauges,46% of the Rhine gauges and 30% of theWeser gauges showed at least two signifi-cant trends. These numbers already hint toregional differences: The sites in the Elbebasin showed less change in flood indica-tors compared to sites in the Rhine, Weserand Danube catchments.
Figure 2.4 shows the spatial distribu-tion of significant trends in AMAXF. At41 gauges (28% of all sites) significantincreasing trends were detected, whereasonly two gauges showed significant de-creasing trends. An interesting spatial pat-tern emerges: All sites with significanttrends are located in the southern, west-ern and central parts of Germany. A rela-
tively sharp line from northwest to south-east can be drawn, which separates the re-gion with trends from the region withouttrends. Along the middle and lower Rhinemain river as well as along the Danubemain river most of the gauges show signifi-cant trends. In the Weser and Elbe catch-ments there are only some gauges withincreasing trends in the upper reaches ofsome sub-catchments.
The trend analyses for the winter max-ima gave similar results as the analyses forthe annual maxima. Significant upwardtrends in winter maxima were identifiedat 23% of all sites. No significant down-ward trends were detected. The spatial pat-tern is only slightly different: The gaugeswith significant upward trends for annualwinter maximum are found in a diagonalband stretching from northwest to south-east of Germany (Fig. 2.5; left). North andsouth of this band were no or only non-significant trends detected. In the Rhineand Danube catchments, the lower num-ber of trends in AWMAXF, compared toAMAXF, is mainly due to a smaller num-ber of significant trends along the mainrivers (Rhine, Danube).
A smaller number of significant trends(29 gauges corresponding to 20% of allgauges) were found for the summer max-ima (ASMAXF). In contrast to AWMAXF,where all detected trends are upward, thetrend analysis of ASMAXF resulted in thesame number of upward and downwardtrends. Moreover, there is a clear spatialdistinction between the regions with up-ward and downward trends, respectively(Fig. 2.5; right). Only gauges in centraland northern Germany in the catchmentsof Weser, Odra and Elbe show downwardtrends, whereas the upward trends are ex-
21
2 Flood trends in Germany
0 100 20050Kilometers
-62
-20
1 - 15
16 - 30
31 - 45
46 - 57
no Trend
AMAXF POT1M
-25 - -10
-6
1 - 10
11 - 20
21 - 30
31 - 36
no Trend0 100 20050
Kilometers
Figure 2.4: Spatial distribution of trends in annual maximum daily mean streamflow - AMAXF (left)
and in peak-over-threshold magnitude with on average one event per year - POT1M (right); (Upward
arrows: significant increasing trend; downward arrows: significant decreasing trend; circles: no
significant trend; size of arrows: relative change within 52 years; Mann-Kendall test, 2-sided option;
10% significance level)
clusively found at gauges in southern andwestern Germany in the Rhine and Danubecatchments.
At 18% of the gauges significant trendsin the POT1M time series could be de-tected. Due to the spatial concentrationin central Germany field significance wasobserved. As could be expected, manygauges show significant trends in AMAXFas well as in POT1M. A similar spatialpattern was detected for the POT2M vari-able (not shown), with however less sig-nificant trends (16%). The POT3M timeseries show almost no significant trends
across Germany. Only 7% of the gaugeshave significant trends. These are not spa-tially clustered, but are rather randomlydistributed all over Germany (not shown).The gauges Cologne/Rhine and Donau-woerth/Danube are two examples for thisbehaviour where significant changes inAMAXF are not matched with significantchanges in POT3M.
In contrast to POT3M, significanttrends in the peak-over-threshold fre-quency POT3F were identified at manygauges: 25% of all gauges show an increas-ing trend, 1% a decreasing trend. With the
22
2.4 Results
0 100 20050Kilometers
no Trend
1 - 15
16 - 30
31 - 45
46 - 54
AWMAXF ASMAXF
-66 - -40
-39.9 - -20
-19.9 - 0
0.1 - 20
20.1 - 40
40.1 - 59.4
no Trend 0 100 20050Kilometers
Figure 2.5: Spatial distribution of trends in seasonal maximum series - AWMAXF (winter, left map),
ASMAXF (summer, right map); (Upward arrows: significant increasing trend; downward arrows:
significant decreasing trend; circles: no significant trend; size of arrows: relative change within 52
years; Mann-Kendall test, 2-sided option; 10% significance level)
exception of two gauges in the Elbe catch-ment, only gauges in the Rhine and Danubecatchments show a significant change inflood frequency (Fig. 2.6). The relativechange of the POT3 frequency is ratherlarge with values up to 140%. This up-per value means that the number of dis-charge peaks above the threshold has in-creased approximately fivefold, from oneevent per year in the 1950s to five eventsper year at the end of the study period. Thespatial distribution of gauges with signif-icant trends is very similar to the resultof AMAXF. Again, a relatively sharp line
from northwest to southeast Germany canbe observed which separates the region ofno trend from the one with positive trends.The seasonal separation of the POT3F vari-able illustrates very well that the majorityof the positive trends is caused by signif-icant upward trends in the frequency ofthe winter floods, whereas POT3F summerevents only increase at three gauges in theDanube catchment (Fig. 2.7). Again, theRhine and Danube catchments are mainlyaffected by the changes in the flood dis-charge behaviour.
23
2 Flood trends in Germany
Table 2.3 summarizes the results forthe eight flood indicators. Field signifi-cance at the 10% significance level wasdetected for AMAXF, AWMAXF, POT3Fand WPOT3F. In all four cases, upwardtrends are the cause for the changes in floodbehaviour. No field significance could befound for decreasing trends for all flood in-dicators. The changes in the summer floodbehaviour (ASMAXF, SPOT3F) are toosmall to be counted as field significant.
0 50 100 150 20025Kilometers
POT3F
-95
0 - 50
51 - 100
101 - 139
no Trend
Figure 2.6: Spatial distribution of trends in
the peak-over-threshold frequency - POT3F
(Upward arrows: significant increasing trend;
downward arrows: significant decreasing trend;
circles: no significant trend; size of arrows: rel-
ative change within 52 years; Mann-Kendall
test, 2-sided option; 10% significance level)
2.4.3 Scale-dependency
Finally, it is assessed if a scale-dependencycan be found in the trend analyses, i.e. itis assessed if large changes are related tosmall or large basins, respectively. To thisend, the relative changes in each floodindicator were plotted against the basinarea, and significant changes were marked(Fig. 2.8). No scale-dependency can be ob-served. There are no spatial scales wheresignificant changes are concentrated. Onthe contrary, significant changes and nochanges, respectively, are found at all spa-tial scales.
2.5 Discussion
The analysis of trends in eight flood indica-tors for 145 gauges across Germany yieldsa number of interesting results. Overall,it can be summarized that the flood haz-ard in Germany increased during the lastfive decades, particularly due to an in-creased flood frequency. Marked differ-ences emerge when looking at the spatialand seasonal patterns and at different floodindicators. An important observation isthat sites with upward and downward floodtrends are spatially clustered. Changes inthe flood behaviour in northeast Germanyare small. Most changes were detectedfor sites in the west, south and center ofGermany. Further, the seasonal analysis re-vealed larger changes for winter comparedto summer.
The results are summarized in Fig. 2.9which highlights the fraction of gaugeswith significant changes, stratified accord-ing to flood indicators and according tothe large river basins Danube (D), Rhine
24
2.5 Discussion
0 100 20050Kilometers
no Trend
1 - 30
31 - 60
61 - 90
91 - 107
WPOT3F SPOT3F
-58
0.1 - 30
30.1 - 60
60.1 - 80
80.1 - 112
no Trend
0 100 20050Kilometers
Figure 2.7: Spatial distribution of trends in seasonal peak-over-threshold frequency - WPOT3F
(winter, left map), SPOT3F (summer, right map); (Upward arrows: significant increasing trend;
downward arrows: significant decreasing trend; circles: no significant trend; size of arrows: relative
change within 52 years; Mann-Kendall test, 2-sided option; 10% significance level)
(R), Elbe (E) and Weser (W). Mostly in-creasing trends were detected, with largeshares of significant trends in AMAXFand POT3F. Approximately 1/3 of the sitesin the western and southern parts of Ger-many (Danube, Rhine, Weser) have signif-icant upward trends in AMAXF, whereasthere are almost no upward trends in east-ern Germany (Elbe). Upward trends inAMAXF in the Rhine and Weser basinscan be attributed to trends in the winter sea-son, since the flood regime is dominatedby winter floods, i.e. the largest share of
annual maxima in the Weser basin and inthe middle and lower Rhine basin occursin the winter season. This is also the rea-son, why the relatively large number ofgauges with downward trends in maximumsummer floods in the Weser basin is not re-flected in the AMAXF.
Compared to Rhine and Weser, the sitesin the Danube catchment are much moreinfluenced by summer floods. Accord-ingly, the upward trends of AMAXF inthe Danube basin are mainly dominated byupward trends in summer floods. However,
25
2 Flood trends in Germany
% of gauges withincreasing trend decreasing trend no trend
AMAXF 28 1 71AWMAXF 23 0 77ASMAXF 10 10 80POT1M 17 2 81POT3M 5 2 93POT3F 25 1 74SPOT3F 2 1 97WPOT3F 17 0 83
Table 2.3: Percentages of gauges showing significant trends; bold numbers indicate field significance
also the frequency of floods (POT3F) in-creased significantly at many gauges, espe-cially along the main river Danube, whichis visible in both seasonal POT3F. An in-creasing frequency in the winter is sup-posed to be caused by higher winter tem-peratures, and hence, earlier snow meltingin the mountain ranges.
The annual maxima for the Elbe gaugesshowed a small number of significantchanges with a similar share of upwardtrends in winter (AWMAXF) and down-ward trends in summer (ASMAXF). In-creasing trends in the winter maximawere mostly found in the Saale catchment,which is the most western sub-catchmentof the Elbe river basin and which shows asimilar trend pattern as the neighbouringWeser catchment. The sites with decreas-ing trends in summer flood magnitude arerather randomly distributed in space.
The spatial and seasonal coherence ofthe results suggests that the observedchanges in flood behaviour are climate-driven. This conclusion is further sup-ported by the missing relation betweensignificant changes in the discharge se-
ries and basin area. Impact of land-coverchanges or of river training works wouldbe expected to show scale-dependency.However, from our analysis we concludethat there are no preferred spatial scaleswhere significant changes could be de-tected. Therefore, it is interesting to eval-uate, whether or not our results are in linewith studies on changes in climate. To thisend, our results are qualitatively comparedto those of recent investigations that an-alyze changes in atmospheric circulationpatterns. It has been shown that there is aclose link between the occurrence and per-sistence of atmospheric circulation patternsand floods in Germany (e.g. Bárdossy andCaspary, 1990; Pfister et al., 2004a; Petrowet al., 2007).
Gerstengarbe and Werner (2005) com-pared daily data of two time periods (1881–1910 and 1975–2004) and found for thesummer large upward trends in the fre-quency of circulation patterns from thesouth (tripled frequency with a step changein the 1940s). During the same time periodthe northwesterly patterns decreased atthe same magnitude (Mittelgebirge Weser,
26
2.5 Discussion
0
50
−30
0
40
−40
−50
0
50
0
40
−20
−20
0
30
0
100
0
100
103
104
105
0
100
Catchment size [km²]
Figure 2.8: Relative change [%] as function of basin area. Triangles indicate significant changes at
the 10% significance level. From top to bottom: AMAXF, AWMAXF, ASMAXF, POT1M, POT3M,
POT3F, WPOT3F, SPOT3F
Elbe). Gerstengarbe and Werner (2005)found small decreases for the summer inthe westerly, northern and easterly circula-tion patterns.
For the winter, Gerstengarbe and Werner(2005) found increasing trends of westerlyatmospheric circulation types. Addition-ally, a longer duration period of the per-sistence of the circulation patterns was ob-served. This yields a larger flood hazardthrough circulation patterns which are gen-erally not very prone to cause flood eventsbut may be increasingly hazardous due to alonger persistence time. Long-lasting west-erly atmospheric circulation types cause
eventually a large-scale saturation, lead-ing to rapid runoff processes. This is thenfinally observed in upward trends of theAWMAXF in the northern Rhine, Weserand Elbe catchments as well as in the up-ward trends of WPOT3F in the Rhine catch-ment. For the middle and lower stretchesof the Rhine, increased flooding probabil-ities for the winter season have been sug-gested by Pfister et al. (2004a). During thesecond half of the 20th century increasedwinter rainfall totals and intensities havebeen observed. At the same time, stronglinks between changes in atmospheric cir-culation patterns and flood occurrence have
27
2 Flood trends in Germany
been identified. Increasing westerly atmo-spheric circulation types correlate with in-creasing winter precipitation and are sup-posed to be responsible for the increase inflood probabilities.
Moreover, Gerstengarbe and Werner(2005) found a decreasing percentage ofeasterly circulation patterns during the win-ter, which cause cold and dry winters es-pecially in the catchments of Odra, Elbeand the Weser. UBA (2006) found signifi-cant upward trends in winter temperaturesduring the last 100 years. These findingsalso fit our results of upward trends in thewinter maximum discharges in the Elbeand Weser catchments, which are causedby more rain induced flood events due tomilder winters and an intensified zonal cir-
culation (Gerstengarbe and Werner, 2005;UBA, 2006).
2.6 Conclusion
Our study of flood trends at 145 runoffgauges, distributed all over Germany,shows that there is no ubiquitous increaseof flood magnitude and/or frequency in thesecond half of the 20th century, as it is of-ten asserted in the media. However, sig-nificant flood trends were detected for aconsiderable fraction of basins. In mostcases, these trends are upward; decreasingflood trends were rarely found and werenot field-significant. The joint analysis ofmany sites within one region allowed as-
AMAXF AWMAXF ASMAXF POT1M POT3M POT3F SPOT3F WPOT3F0
10
20
30
40
50
Pe
rce
nta
ge
of
ga
ug
es w
ith
sig
nific
an
t tr
en
ds
R
R
R
R
R
R
R
DW
E
E
E
E
E
E
E
D
D
D
D
D
D
D
W
W
W
W
Figure 2.9: Percentages of gauges with significant trends per catchment and flood indicator. Dark
grey bars show percentage of upward trends, light grey bars show percentages of downward trends;
abbr. for the catchments are D - Danube, R - Rhine, W - Weser, E - Elbe
28
2.6 Conclusion
sessing the spatial and seasonal coherenceof flood trends: Basins with significanttrends were spatially clustered. Changesin flood behaviour in northeast Germanyare small. Most changes were detectedfor sites in the west, south and center ofGermany, i.e. in the catchments of Rhine,Weser and Danube. The seasonal analysisrevealed larger changes for winter com-pared to summer. From the results weconcluded that the observed changes inflood behaviour are climate-driven. It waspossible to qualitatively link our resultsto trends in frequency and persistence ofatmospheric circulation patterns above Eu-rope. As already shown by Pfister et al.(2004) for a smaller area, orographic obsta-cles heavily influence the spatial distribu-tion of the rainfall and runoff processes. Achanging behaviour of circulation patternsis likely to cause changes in rainfall totals,
which in turn heavily affects discharge andwater levels in the rivers. The relationshipbetween circulation patterns, flood mag-nitude and/or frequency and the influenceof the topography will be further investi-gated. Our findings underline the need tothoroughly analyze the flood behaviour forchanges when estimates for flood designor flood risk management are needed.
Acknowledgements
We thank the GeoForschungsZentrum Pots-dam (GFZ) and the Helmholtz Associationof National Research Centres for their fi-nancial support. The study was part of theHelmholtz Young Scientists Group "Infor-mation and modelling systems for largescale flood situations" at GFZ. We dedi-cate our special thanks to the authoritiesthat provided data.
29
3 Changes in the flood hazard through
changing frequency and persistence of
circulation patterns
Abstract
The link between trends in circulation patterns and trends in the flood magnitude is studiedfor 122 meso-scale catchments in Germany for a period of 52 years (1951–2002). Floodtrends, significant at the 10% level, are detected for a large number of catchments. Thecatchments are pooled into three regions, based on flood seasonality and flood trends.Field-significant increasing trends are found for winter in Regions West and East. Forsummer, increasing and decreasing flood trends are detected for Regions South and East,respectively. The temporal behaviour of three flood indicators of each region is compared toatmospheric indicators derived from circulation patterns. Significantly increasing frequencyand persistence of flood-prone circulation patterns intensify the flood hazard during thewinter season throughout Germany. Moreover, a trend towards a reduced diversity ofcirculation patterns is found causing fewer patterns with longer persistence to dominate theweather over Europe. This indicates changes in the dynamics of atmospheric circulationswhich directly influence the flood hazard. Longer persistence of circulation patterns whichin general do not favour large precipitation amounts may lead to large runoff coefficientsdue to soil-moistening and hence cause floods.
Petrow, Th., J. Zimmer, and B. Merz. 2009. Changes in the flood hazard through
changing frequency and persistence of circulation patterns. Natural Hazards and
Earth System Sciences, 9, 1409–1423.
31
3 Changes in the flood hazard through changing circulation patterns
3.1 Introduction
Changes in the atmospheric dynamicsand their links to hydrological processesare an important aspect in the discus-sion about climate change. During thelast decades, many devastating floods oc-curred in Europe giving rise to the dis-cussion whether or not flood-triggering at-mospheric patterns may have significantlychanged. Many studies evaluated trendsin climatic variables such as the North-Atlantic Oscillation (NAO), changes inENSO phenomenon (El Nino/Southern Os-cillation) or in circulation patterns (CP)and linked these with precipitation or tem-perature. CPs are either derived by 500 hPageopotential height fields from reanalysesdata or from classification schemes, forinstance by Hess and Brezowsky (1952).Bárdossy and Caspary (1990) used the CPcatalogue of Hess and Brezowsky for theperiod 1881–1989 and found significantchanges in the frequency of daily, seasonaland annual data of several CPs leadingto milder and wetter winters in Europe.Steinbrich et al. (2005) showed for south-western Germany (Baden Wuerttemberg)that the link between CPs and large precip-itation events varies strongly depending onthe seasonal and regional conditions. Theyfound most of the analyzed heavy precipita-tion events to be triggered by only few CPs.Werner et al. (2008) detected CP and pre-cipitation trends in the Elbe catchment inthe period 1951–2003. During the winterseason, the number of days with precipita-tion tripled. Also, increases in frequencyand duration of west and north-west cir-culation patterns were observed (Werneret al., 2008). Pauling and Paeth (2007)identified an increase in extreme winter
precipitation during the last 300 years overEurope. Casty et al. (2005) found a closecorrelation between the NAO index andtemperature and precipitation indices in theEuropean Alps during the last 500 years.A clear relation between precipitation andNAO was detected by Feidas et al. (2007)for Greece for the period 1955–2001. Theyobserved downward trends in winter andannual precipitation which were correlatedwith rising trends in the hemispheric cir-culation modes of the NAO. Santos et al.(2007) emphasized the strong link betweenNAO and heavy precipitation events overEurope.
Only few studies investigated the linkbetween atmospheric and flood indicators.Kingston et al. (2006) reviewed on stud-ies about the connection between climate,streamflow and atmospheric circulations(esp. NAO and Arctic Oscillation (AO)).Svensson et al. (2006) reported correla-tions between trends in the NAO indexand floods for Europe. McKerchar andHenderson (2003) found changes in sev-eral hydrological variables in New Zealandwhich were consistent with changes inthe Interdecadal Pacific Oscillation. Ja-cobeit et al. (2006) determined large-scaleCP for historical flood events with thehelp of Reanalysis data. They identifiedCPs that are relevant for the flood hazardin Europe. The most important CPs fortriggering prominent discharge peaks are(1) for summer the Vb-pattern, westerlyflows with southerly components as well astroughs and (2) for winter westerly windswith changing north/south components. Astudy of meso-scale snow-free catchmentsin France and Spain by Bárdossy andFiliz (2005) identified flood-producing cir-culation patterns with the help of large-
32
3.1 Introduction
scale sea-level pressure fields. Bouweret al. (2006) identified a closer relation-ship between mean winter discharge andWZ circulation patterns (after Hess andBrezowsky, 1952) than between dischargeand NAO. Bouwer et al. (2008) calculatedcorrelations of four atmospheric variables(NAO, AO, westerly-cyclonic circulationpattern, sea-level pressure differences) andwinter precipitation to annual mean andmaximum winter discharges at 11 gaugesin Central Europe. They found that annualmaximum discharges are more sensitive tochanges in atmospheric circulations thanmean discharges. So far, the relationshipbetween peak discharges and atmosphericvariables has been investigated for Ger-many only for selected regions. Belz et al.(2007) analyzed flood discharges, CPs ac-cording to Hess and Brezowsky, and arealprecipitation for the Rhine catchment. CPswere classified into wet and dry patterns.The pattern WZ was studied in more de-tail, since it is the most frequent patternand comprises the days with the largestprecipitation amounts. Belz et al. (2007)found increasing trends in wet CPs (whichalso contain WZ), areal precipitation anddischarge for the period 1951–2000. Aneven higher significance level of increas-ing trends was found for winter maximumdischarges compared to increasing trendsin annual maximum discharges. Casparyand Bárdossy (1995) found an increase inthe pattern WZ for winter, leading to a dra-matic increase in the flood hazard for south-western Germany. Mudelsee et al. (2004)studied flood trends in the Elbe and Odracatchments and found downward trends inwinter and no significant changes duringsummer.
Since these studies are limited to se-
lected regions in Germany, a countrywidepicture of trends in floods and atmosphericpatterns cannot be drawn. Our study closesthis gap by presenting results of floodtrends and trends in circulation patternsfor 122 meso-scale catchments across Ger-many for the period 1951–2002. The trendbehaviour of eight flood indicators at 145gauges (500–159300 km2) in Germanywas already analyzed for the same periodby Petrow and Merz (2009). Their findingsform the basis for the here presented study.Petrow and Merz (2009) detected trendsin peak discharge, which were spatiallyand seasonally clustered. A missing rela-tion between discharge changes and basinarea suggests that the observed changesin flood behaviour are climate-driven. Incontrast to the study by Petrow and Merz(2009), we here present results of time-varying multiple trend tests both for floodand CP indicators. McCabe and Wolock(2002) also used this approach and foundpatterns of significant changes in differentdischarge variables in the United States forthe period 1941–1999. For the evaluationof a possible link between flood and atmo-spheric patterns, correlations between peakdischarges and CPs were computed similarto other studies (e.g. Feidas et al., 2007;Bouwer et al., 2008).
The gauges were pooled into regions (cf.Douglas et al., 2000; Merz and Blöschl,2009). The pooling into three regions re-flects different flood regimes across Ger-many. A catchment-independent poolingwas favoured over a catchment-based ap-proach to account for the characteristic sea-sonality of peak discharges in each region.Moreover, the flood trend results observedby Petrow and Merz (2009) showed thatregions of similar flood trends do not nec-
33
3 Changes in the flood hazard through changing circulation patterns
essarily coincide with catchment borders.Thus, the catchment-independent approachenables us to draw a more precise pictureof the observed changes. For all regionsconclusions are finally presented to whatextent parallels between trends in floodmagnitudes and trends in circulation pat-terns can be found.
3.2 Data
3.2.1 Discharge Data
Discharge data of meso-scale catchmentswere used for this study. We includedcatchments of at least 500 km2 in orderto minimize the influence of land man-agement measures on the flood behaviour(Bronstert et al., 2002). The largest ana-lyzed catchment is the river Mosel at thegauge Cochem (27088 km2). A commontime period between 1 November 1951 and30 October 2002 was used (hydrologicalyear in Germany: 1 November to 31 Oc-tober). Svensson et al. (2006) suggest aminimum length of 50 years for the analy-sis of flood trends. Shorter time series maynot capture a possible trend, whereas verylong series of up to 100 years may haveother shortcomings as for instance changesof the measuring procedure over time. Thechosen time period was seen as compro-mise between a minimum length and therequirements for reliability and availabilityof data.
Each of the 122 gauges was assigned toone of three regions (Fig. 3.1), which arecharacterized by homogeneous seasonalflood histograms and flood trends. The re-gions were extracted through a GIS-basedmulti-criteria analysis. Histograms of the
Region West
Region East
Region South
meso-scale Catchments0 50 100 150 20025
Kilometers±
Location of meso-scale catchments and regions
Figure 3.1: Discharge gauges of 122 meso-
scale catchments and their assignment to one of
the three regions. Catchment borders illustrate
the spatial coverage of the dataset.
seasonal flooding frequencies as presentedby Beurton and Thieken (2009) as well astrend results of eight flood indicators byPetrow and Merz (2009) were comparedin a spatially-explicit manner in order toidentify homogeneous regions. A changein the assignment of gauges to one or theother region is visible along the main riversof Danube and Weser. This is caused bydifferences in seasonal histograms of theflood indicators. For instance, the assign-ment of the gauges along the Danube toRegion South can be explained by the dom-inance of summer maximum dischargeswhich is characteristic for the southern trib-utaries rather than for the northern ones.
34
3.2 Data
Figure 3.1 gives an overview of the loca-tion of the gauges and regions.
Region West (yellow dots) comprises49 discharge gauges which are located inthe Rhine, Weser, Ems and Danube catch-ments. This region has a winter dominatedflood regime. Frequently, westerly windscause flooding especially during winter.Trends in winter maximum discharges in-crease both in magnitude and frequency(Petrow and Merz, 2009). Summer floodsplay a minor role.
Region East (blue dots) consists of 41gauges in the Weser, Ems and Elbe catch-ments. Maximum discharges occur pre-dominantly during winter in this region,however, summer peak discharges have alarger share than in Region West and canreach remarkable extents as experienced in2002 and 2005 (DKKV, 2004; Beurton andThieken, 2009). Winter floods increase inRegion East, whereas summer floods de-crease. This is the only region in Germany,in which summer floods significantly de-crease (Petrow and Merz, 2009).
Region South is represented by 32gauges which are located in the Rhine andDanube catchments. Two gauges are lo-cated in the Rhine catchment, all othergauges are located either along the mainriver of the Danube or along its southerntributaries. The two gauges in the Rhinecatchment have larger shares of winter dis-charges compared to the other gauges inthe Danube catchment. The Danube re-gion is dominated by summer flood events.However, winter peak discharges signif-icantly increase in magnitude at manygauges in the region (Petrow and Merz,2009).
In this study, three flood indicators wereanalyzed for each gauge. These com-
prise annual maximum streamflow series(AMAXF) as well as seasonal maximumseries (AWMAXF and ASMAXF). Annualmaximum daily mean streamflow, i.e. thelargest daily mean streamflow that occursin each hydrological year, is the most com-mon indicator in flood trend studies. Theanalysis of seasonal maximum series en-ables a more differentiated picture of floodtrends. Annual winter maximum dischargeseries (AWMAXF) were derived from databetween 1 November and 31 March of thefollowing year and consist of the largestdaily mean discharge of each winter sea-son. Summer maximum discharge series(ASMAXF) were derived for the period of1 April - 31 October.
3.2.2 Circulation patterns
For the analysis of trends in circulation pat-terns different classification systems areavailable, which are either manual (basedon subjective knowledge) or automated nu-merical schemes. The widely used man-ual classification scheme by Hess and Bre-zowsky (1952) is currently the only oneavailable, which captures the large-scaleEuropean pattern, while still focusing onlocal details (James, 2007). Buishand andBrandsma (1997) compared three classi-fication schemes of CPs and found thatthe subjective Großwetterlagen classifica-tion by Hess and Brezowsky (1952) yieldsequally good results as the two other ob-jective schemes. Therefore, the schemeby Hess and Brezowsky (1952) was usedin this study. Moreover, the use of thisclassification facilitates the comparison ofour results to other studied conducted forGerman catchments.
Daily data of the "Catalogue of Großwet-
35
3 Changes in the flood hazard through changing circulation patterns
terlagen in Europe 1881–2004" after Hessand Brezowsky (1952) was available forthis study (Gerstengarbe and Werner,2005). The catalogue provides a subjec-tive classification for every day about thedominant circulation pattern (CP) over Eu-rope which is derived based on the spatialdistribution of pressure systems over Eu-rope as well as the location of frontal zones.The catalogue distinguishes 30 differentCPs (one is classified to be a "transitionclass"). The CPs comprise the zonal cir-culation form, the mixed circulation formas well as the meridional circulation form(Table 3.1).
The most important CPs with respectto the flood hazard in Germany are WZ,WS, NWZ, and TRM. The first three pat-terns are frequent and comprise 25% of theoverall distribution for Germany. Theseare westerly winds of varying direction(from north to south). The pattern TRMis better known to be the "Vb-weather pat-tern" and is represented by a trough overCentral Europe (van Bebber, 1891). Lowpressure systems move from the Gulf ofGenoa northwards to Poland. Large pre-cipitation amounts can be accumulated andmay be enhanced along the northern slopesof the Alps and the mountain ranges inCentral and Eastern Europe. Several de-structive floods were triggered by TRM, asexperienced for instance in the Elbe andDanube catchments in 2002 and 2005 (Ul-brich et al., 2003).
The influence of a CP on the floodregime varies from region to region. Peakdischarges in Region West are often causedby westerly, south-westerly and north-westerly circulation types (Beurton andThieken, 2009). High pressure systemsare rarely responsible for floods in Region
West. Floods occur predominantly duringmild and wet episodes in winter. RegionEast is also largely influenced by westerlywinds, however, the north-westerly patternplays a more important role than in Re-gion West. A larger share of high pressuresystems and the occurrence of Vb-weatherregimes distinguish the region from RegionWest. Region South is dominated by highpressure systems, especially during fall andwinter. Westerly, north-westerly and south-westerly circulation types are less frequentcompared with the other regions. Peak dis-charges occur predominantly during sum-mer.
Daily data of the CPs were analyzed fortrends in four variables:
1. the number of days of each CP peryear,
2. the number of events of each CP peryear (independently of its length),and
3. the mean duration of each CP peryear, and
4. the maximum duration of each CPper year.
These variables were analyzed on an an-nual basis and for winter and summer sea-sons. The number of days per year gives anindication of the frequency. The numberof events was counted, independently ofthe CP length, in order to gain informationabout the variability of CP. The persistenceof CP is particularly important for the floodhazard. There are numerous examples oflong-lasting CPs that are accompanied bya sequence of weaker precipitation events,which finally cause large floods.
36
3.2 Data
Form of Circulation No. Circulation pattern name Abbr.Zonal Circulation 1 West wind, anti-cyclonic WA
2 West wind, cyclonic WZ3 Southern west wind WS4 Angular west wind WW
Mixed circulation 5 South-west wind, anti-cyclonic SWA6 South-west wind, cyclonic SWZ7 North-west wind, anti-cyclonic NWA8 North-west wind, cyclonic NWZ9 High pressure system, Central Europe HM
10 High pressure bridge over Central Europe BM11 Low pressure system, Central Europe TM
Meridional circulation 12 North wind, anti-cyclonic NA13 North wind, cyclonic NZ14 High pressure Iceland-Norwegian Sea, anti-cyclonic HNA15 High pressure Iceland-Norwegian Sea, cyclonic HNZ16 High pressure, British Isles HB17 Trough Middle Europe TRM18 North-east wind, anti-cyclonic NEA19 North-east wind, cyclonic NEZ20 High pressure Fennoscandia, anti-cyclonic HFA21 High pressure Fennoscandia, cyclonic HFZ22 High pressure Norwegian Sea- HNFA
Fennoscandia, anti-cyclonic23 High pressure Norwegian Sea- HNFZ
Fennoscandia, cyclonic24 South-east wind, anti-cyclonic SEA25 South-east wind, cyclonic SEZ26 South wind, anti-cyclonic SA27 South wind, cyclonic SZ28 Low Pressure, British Isles TB29 Trough, Western Europe TRW30 Transition, no classification U
Table 3.1: Classification of the circulation form and its specific pattern after Hess and Brezowsky
(1952)
To each value of the discharge AMAXFseries of each gauge, the flood triggeringCP was assigned, in order to evaluate thelink between changes in CP over time andflood trends. Depending on the catchmentsize, a time lag of one to three days wasapplied (Duckstein et al., 1993; Frei et al.,
2000; Bárdossy and Filiz, 2005). For smallcatchments with 500–5000 km2, a time lagof one day was assumed, catchments with5001–20000 km2 had a time lag of twodays and larger catchments of three days.For example, a catchment of 600 km2 hadan AMAXF entry on 19 March 1951. Then,
37
3 Changes in the flood hazard through changing circulation patterns
it was assumed that the flood triggering CPoccurred on 18 March 1951. Althoughthere is some uncertainty when assigninga "mean" concentration time to all catch-ments of a certain size range, the time lagis regarded to be sufficiently precise, es-pecially when considering that the medianpersistence of CPs is from three to fivedays. Except for the transition class "U"(CP 30), all CPs persist for at least threedays and often up to 10 days.
3.3 Trend detection
The robust non-parametric Mann-Kendall(MK) test and a resampling approach wereused for detecting trends both in peak dis-charge and circulation patterns (Kendall,1975). The MK test is particularly usefulfor the analysis of extremes and requires nospecific distribution. It has been used by avariety of studies on hydro-meteorologicaltrends (e.g. Chen et al., 2007; Feidas et al.,2007). We applied the two-sided optionwith 10% significance level. The MK testrequires the data to be serially independent.von Storch and Navarra (1995) found that,if the data are positively serially correlated,the test tends to overestimate the signifi-cance of a trend. To correct the data forserial correlation, the procedure of trendfree pre-whitening (TFPW) was applied,which is described in detail in Yue et al.(2002a, 2003) and Petrow and Merz (2009).In short, a trend of a time series is esti-mated by the non-parametric trend slopeestimator (Sen, 1968). A possible trendis then removed from the original series.Thereafter, the lag1-autocorrelation is cal-culated. If no significant autocorrelation isfound, the MK test is directly applied to the
original time series. Otherwise, the lag1-autocorrelation is removed from the timeseries. The data series is now regarded tobe free of trend and serial correlation. Fi-nally, the firstly removed trend is includedback into the time series resulting in a se-ries that includes the original trend withoutautocorrelation.
Discharge data were analyzed for eachgauge separately and aggregated for eachregion as a regional composite. These com-posite series were derived as follows: se-ries of AMAXF (analogue ASMAXF, AW-MAXF) from every gauge were drawn. Forall time series the TFPW-methodology wasapplied. After that, all series were normal-ized (by subtracting the mean and dividingthe result by the standard deviation of thetime series). Then, the regional compos-ite was calculated by averaging all normal-ized discharge time series of a given re-gion. The TFPW procedure causes in someinstances small deviations, which lead tolarger magnitudes of seasonal MAXF com-pared to AMAXF (cf. Fig. 3.2).
Maximum discharge and CP time se-ries were analyzed for trends not onlyfor the entire period (1951–2002), butalso with the help of moving windows ofvarying time lengths (multiple trend tests).Through this methodology it is possible todetect changes in trends over time and todistinguish recent trends from trends thatare stable over longer time periods (Mc-Cabe and Wolock, 2002). All possible pe-riods of 20 to 52 years within the investi-gated time series were analyzed for trends.The trend matrix shows for each variableand changing time periods the resulting sig-nificance level. These levels were derivedby means of a resampling approach.
38
3.3 Trend detection
1955 1960 1965 1970 1975 1980 1985 1990 1995 2000−2
−1
0
1
2
3
Region West
1955 1960 1965 1970 1975 1980 1985 1990 1995 2000−2
−1
0
1
2
3
Region East
1955 1960 1965 1970 1975 1980 1985 1990 1995 2000−2
−1
0
1
2
3
Region South
AMAXF
AWMAXF
ASMAXF
Figure 3.2: Composite maximum discharge time series with trends per region and flood indicator
(note: in Region West the trend lines of AMAXF and AWMAXF are almost the same and therefore
not easily distinguishable)
1. The time series is resampled ran-domly without replacement. A newtime series is obtained with the samevalues but different year order.
2. A linear trend line is fitted to thenew time series and its slope is cal-culated.
3. By repeating steps 1 and 2 1000times, a sample of slope values ofsize 1000 is obtained.
4. The significance level of the ob-served time series is determined bycomparing its slope with the empir-ical cumulative distribution of the
slope values of the resampled timeseries.
In regional trend detection studies it isinteresting to assess the field significance,i.e. the significance of trends across the re-gion (Douglas et al., 2000; Burn and HagElnur, 2002; Svensson et al., 2006). Dou-glas et al. (2000) determined the field sig-nificance by calculating a regional criticalvalue for the Mann-Kendall test, which isderived through a bootstrapping approach.The number of stations to show a trendby chance for the specific region is deter-mined. Thereafter, the number of observedtrends is compared with the number of ex-
39
3 Changes in the flood hazard through changing circulation patterns
pected trends for the region. Douglas et al.(2000) found that the existence of spatialcorrelation between sites may inflate the re-sults of change detection, if the spatial cor-relation is not accounted for and proposeda bootstrapping test for assessing the fieldsignificance of trends while preserving thecross-correlation among sites. However,this approach might only be suitable for thecase that the majority of trends in a regionare uniform, i.e. either upward or down-ward (Yue et al., 2003). Therefore, we ap-plied a slightly refined approach, proposedby Yue et al. (2003), which assesses thefield significance of upward and downwardtrends separately. A detailed description ofthe methodology can be found in Yue et al.(2003) or Petrow and Merz (2009).
3.4 Results
3.4.1 Trends in flood data
Composite discharge time series of thethree maximum series are shown for allregions in Fig. 3.2. Upward trends (MKtest, 10% SL) were found for all three com-posite series for Regions West and Southand for winter and annual maximum se-ries for Region East. A downward trendwas detected for the composite summermaximum series (ASMAXF) of RegionEast. Many trend lines in Fig. 3.2 havealmost the same slope and are thereforenot easily distinguishable. Discharge dataof each gauge were tested with the MKtest for upward and downward trend (SL10%). Table 3.2 shows the results for theflood indicators and regions. Large differ-ences are evident in the number of trendsper variable and region.
Many upward trends were found in theRegions West and South. The three gaugeswith upward AMAXF trends in RegionEast are clustered in the southern part ofthe region and are located in the vicinity ofgauges with trends of Region West (Petrowand Merz, 2009). Upward trends in AW-MAXF are spatially clustered in CentralGermany affecting the northern part of Re-gion West and the southern part of RegionEast. A different pattern evolves for thesummer series: downward trends are, ex-cept for one gauge in Region South, ex-clusively found in Region East, whereasupward trends are concentrated in RegionSouth. In the following, trend results of dif-ferent time lengths of the composite seriesare presented, which offer the possibilityto study the temporal variability of floodtrends.
Figure 3.3 shows multiple trend tests(MK test, 10% SL) for varying time pe-riods of the composite series of AMAXF,AWMAXF and ASMAXF for the three re-gions. Upward trends are reflected in Fig-ures 3.3, 3.8 and 3.9 by numbers from95 to 100, and downward trends by num-bers from 5 to 0, respectively. The x-coordinate shows the starting year and they-coordinate the ending year of the ana-lyzed period, leading to time series lengthsof 20 to 52 years. The result of the trendtest for the time period 1951–1970 is givenin the lower left corner of the trend matrix.In the upper left corner the result of the en-tire series (1951–2002) can be found. Onthe diagonal, trend results of 20-year timeperiod are shown, beginning in the lowerleft corner with the period 1951–1970, pro-gressing with a step of one year and endingin the upper right corner with the period1983–2002. In the first row the results for
40
3.4 ResultsE
nd
ing
year
Region West − AMAXF
1960 1970 19801970
1980
1990
2000
Region West − AWMAXF
1960 1970 19801970
1980
1990
2000
Region West − ASMAXF
1960 1970 19801970
1980
1990
2000
En
din
g y
ear
Region East − AMAXF
1960 1970 19801970
1980
1990
2000
Region East − AWMAXF
1960 1970 19801970
1980
1990
2000
Region East − ASMAXF
1960 1970 19801970
1980
1990
2000
Beginning year
En
din
g y
ear
Region South − AMAXF
1960 1970 19801970
1980
1990
2000
Beginning year
Region South − AWMAXF
1960 1970 19801970
1980
1990
2000
Beginning year
Region South − ASMAXF
1960 1970 19801970
1980
1990
2000
0
20
40
60
80
100
0
20
40
60
80
100
0
20
40
60
80
100
Figure 3.3: Significance level of trends of different flood indicators for varying time periods for
Region West (first row), Region East (second row), and Region South (third row). Results between
95 and 100 indicate upward trends, results between 0 and 5 downward trends.
Region West are shown, in the second rowfor Region East and finally in the third rowfor Region South. In the following, theresults are discussed for each region sepa-rately.
3.4.1.1 Region West
Many upward trends are evident for RegionWest for AMAXF and AWMAXF. Bothmatrices have similar patterns (Fig. 3.3).Time series of different lengths ending lat-est in 1982 show a slightly downward ten-dency. Also, series of the last 20 to 25years show small downward tendencies.All other time series which cover different
time periods ending in 1982 or later showincreases (mostly significant). Time serieswith at least 30 years show almost alwaysupward trends. Interestingly, a relativelyfast change of increases and decreases canbe seen. The analysis of the diagonal withtime periods of 20 years shows at first aperiod of decreasing annual and wintermaximum discharges which ends in thebeginning year 1962. This is followed bya period until 1977 where upward trendsare detectable. From the beginning year1978 onward, the time series show againno or only minor downward changes inthe discharge data. This general pattern ismore or less clearly seen for all regions for
41
3 Changes in the flood hazard through changing circulation patterns
Number of gauges Number of gaugesRegion Flood indicator with upward trend with downward trendWest AMAXF 21 1(49 gauges) AWMAXF 20 0
ASMAXF 3 0East AMAXF 3 0(41 gauges) AWMAXF 8 0
ASMAXF 0 12
South AMAXF 9 1(32 gauges) AWMAXF 4 0
ASMAXF 8 1
Table 3.2: Result of MK test (10% SL) of different flood indicators for each region (bold numbers
indicate field significance)
AMAXF and AWMAXF.Multiple trend test were also performed
for each gauge and flood variable. A rel-atively heterogeneous spatial pattern re-sulted (not shown). As it can be expected,the summation per variable of the individ-ual matrices of all gauges revealed a goodcorrelation with the regional composite foreach variable.
Although the regional composite of thewinter maximum discharge shows a verysimilar pattern compared to the annualmaximum discharge, the downward ten-dencies of the last 20 to 25 years are morepronounced during winter. In contrast toAMAXF, a spatial pattern of the temporaltrend behaviour is visible for AWMAXF.The gauges in the northern part of RegionWest show all similar patterns and domi-nate the composite. All other gauges havechanging patterns over time. These resultsare not shown here due to the limited read-ability when plotting a large number ofmatrices onto a regional map.
Summer maximum discharges play a mi-nor role in Region West. The trend patternof ASMAXF (Fig. 3.3) differs greatly for
the last two decades. Decreases of differ-ent magnitudes are detectable for all timeseries since 1990. During the first yearsof the analyzed time span, the downwardchanges are similar to those in AMAXFand AWMAXF. Over the entire period,there are however nearly no significanttrends detectable.
3.4.1.2 Region East
Although the seasonal distribution of floodevents is similar to Region West with amajority of large discharges occurring dur-ing winter, the trend pattern for RegionEast shows many differences compared toRegion West (Fig. 3.3, second row). Avery heterogeneous pattern is visible forAMAXF with almost no trends. Periods ofincreasing and decreasing discharges alter-nate. Although the pattern of AWMAXF isdominated by increasing discharges, theseare usually not significant at the 10% SL.The overall picture has some similarity toAWMAXF of Region West with clusteredperiods of upward and downward cycles.
In contrast, the summer series show
42
3.4 Results
many downward trends (significant at the10% SL), especially for time series endingin 1989 or later. There is a relatively sharpchange from no trend towards a downwardtrend. Nearly all time series, which cover(parts of) the last two decades, show down-ward trends. The overall pattern is morepronounced compared to Region West.
3.4.1.3 Region South
Upward trends were detected for the threeflood indicators in Region South (Fig. 3.3,third row). The AMAXF results showupward trends for a relatively short timeframe beginning around 1960. Addition-ally, there are a few downward trends inshort time series in the early 1950s, whichcan be found in all three flood indicators.Although the winter maximum dischargeshows many upward trends, this pattern isnot visible in AMAXF. The reason lies inthe large percentage of summer events inAMAXF. Thus, the absence of significanttrends in ASMAXF causes the small num-ber of trends in AMAXF. Downward andupward cycles are more pronounced in AW-MAXF, whereas the changes in AMAXFand ASMAXF are less abrupt. Changes ofthe summer series reveal a more variablepicture. Here, upward and downward pe-riods dominate the pattern. Most of themare not significant.
3.4.2 Identification of flood
triggering circulation
patterns
In order to identify flood triggering CPs forthe three regions, the frequency of CPs thatare associated with annual maximum dis-
charges (AMAXF) was derived. DifferentCPs are relevant for triggering peak dis-charges in each region. Histograms of theflood triggering CPs were calculated foreach gauge. Thereafter, mean frequenciesfor every region were calculated (Fig. 3.4).The differentiation into winter and sum-mer reflects the discharge behaviour of therespective regions.
Regions West and East both show a win-ter dominated flood regime that is mainlyinfluenced by only few CPs. 62% of themaximum discharges in Region West aretriggered by the circulation patterns: WZ,WS, SWZ and NWZ. In Region West thedominance of WZ is more pronounced thanin Region East, where the other remain-ing CPs play a more important role in trig-gering large discharges. The circulationpatterns TM, TRM and TRW are impor-tant during the summer in the Regions Eastand South, when they are associated withlarge precipitation amounts that may causefloods. Note, that the importance of theCPs TM, TRM, TRW is not directly visi-ble in Fig. 3.4, as they occur seldom.
Region South is characterized by a dif-ferent seasonal flood behaviour. Sum-mer maximum discharges dominate theAMAXF series. Winter peak dischargesare triggered by the same CPs as in Re-gion West, namely by WZ, WS and NWZ.Although summer floods have also largeshares of WZ and NWZ, the circulation pat-terns BM, HB, TRM, NEZ and TRW playan important role for the summer flood haz-ard.
3.4.3 Trends in daily CP data
Daily data of CPs were analyzed for trendwith the Mann-Kendall test (10% signifi-
43
3 Changes in the flood hazard through changing circulation patterns
Figure 3.4: Mean frequencies of flood causing CPs in AMAXF series for the Region West, East and
South
44
3.4 Results
cance level). Four variables that capturethe behaviour of the CPs over time wereselected. These are the number of dayswith a certain CP per year, the number ofCP events (independent of its length) andthe mean and maximum durations per year.These variables were derived for the com-
plete hydrological year and for the winterseason and summer season, respectively(Table. 3.3). Since the trend results of themean and maximum duration are very sim-ilar, Table 3.3 only shows the findings ofthe number of days, number of events and
Number of days Number of Events Mean persistenceName of per year per year in days per year
CP CP All Wi Su All Wi Su All Wi Su1 WA 19 93 -17 0 0 0 42 79 182 WZ 43 90 18 0 0 0 49 69 44
3 WS -31 0 0 0 0 0 0 0 04 WW -75 -22 -60 -75 0 -61 0 0 05 SWA 50 218 0 0 0 0 34 112 06 SWZ -13 -45 61 0 0 0 0 -36 78
7 NWA 72 0 0 0 0 0 73 0 08 NWZ -16 68 -74 -33 0 -61 38 118 -36
9 HM -29 0 -43 -44 0 -51 35 22 31
10 BM 84 59 113 21 0 0 51 48 46
11 TM 0 0 0 0 0 0 32 0 012 NA 0 0 0 0 0 0 0 0 013 NZ -24 0 0 0 0 0 0 0 014 HNA -19 0 0 0 0 0 0 0 015 HNZ -36 0 0 0 0 0 0 0 016 HB -11 -55 57 0 0 0 14 0 163
17 TRM 33 0 48 0 0 0 29 50 3318 NEA -57 0 0 -75 0 0 0 0 019 NEZ -61 0 0 -65 0 0 -39 0 020 HFA -28 0 0 -39 0 0 21 0 021 HFZ 0 0 0 0 0 0 0 0 022 HNFA 57 0 260 0 0 0 82 0 384
23 HNFZ -50 0 0 0 0 0 0 0 024 SEA 0 0 0 0 0 0 0 0 025 SEZ 0 0 0 0 0 0 0 0 026 SA 0 0 0 0 0 0 0 0 027 SZ 0 0 0 0 0 0 0 0 028 TB 0 0 0 0 0 0 33 0 1829 TRW 58 52 40 0 0 0 41 68 2730 U -45 0 -44 -52 0 -50 0 0 0
Table 3.3: Relative change within 52 years in % (bold numbers indicate trends (MK test, 10% SL);
grey rows show flood relevant CPs)
45
3 Changes in the flood hazard through changing circulation patterns
the mean duration per year. Six CPs showno trend at all for all variables and the dif-ferent datasets. Four CPs revealed slightchanges in only one variable, which werehowever not significant.
3.4.3.1 Number of days
The majority of CPs shows changes in thenumber of days (MK test, 10% SL). How-ever, for only eight CPs the trend is alsosignificant for the annual dataset (column"All"). Upward trends were found for theannual dataset for the three CPs WZ, BMand TRW. It is important to note that allthree CPs hold a considerable potential forfloods. WZ is important throughout Ger-many, whereas the patterns BM and TRWonly play an important role for RegionSouth. Downward trends were detected forthe patterns WW, NEA, NEZ, HNFZ, andU. Among these, only NEZ is importantfor the flood hazard (again only in RegionSouth). All other changes were not signifi-cant. Figure 3.5 shows two histograms ofthe number of days per CP for the annualdataset (top), winter (middle) and summer(bottom), respectively. On the left side thehistograms for the decade 1951–1960 arefound and on the right side for the decade1991–2000. In all three datasets, a statisti-cally significant shift is visible towards asmaller number of CPs, which dominatesthe weather (Chi2 test; 10% SL). Thus,CPs, as for instance WZ, which had al-ready a large share of days per year evenincreased in the frequency, whereas lessfrequent CPs in general decreased. Thereare some exceptions that are also impor-tant for the flood hazard. For instance, thepattern TRM increased in frequency, al-though it rarely occurs. This is especially
important for the summer flood hazard, al-though the changes are also detectable forthe winter and the annual datasets.
For the winter and summer seasons, lesstrends were detected. Figure 3.6 shows re-sults of the seasonally differentiated trendsfor selected CPs. These six CPs are im-portant for at least one of the three re-gions. WS, a pattern that also frequentlytriggers floods, is not shown in the di-agram because for both seasons and allthree variables non-significant decreaseswere found. The patterns WZ, BM andTRW show for all datasets upward changeswhich are however not always significant.Interesting results were found for the pat-tern NWZ which is important through-out Germany. When testing the entiredataset for trend, no trend was detected forNWZ. A look at the seasonal time seriesshowed, however, that during the winter anon-significant increase of 68% was found,whereas the summer time series revealed asignificant downward trend of -74%. Fig-ure 3.4 shows that the NWZ pattern is espe-cially important for triggering winter peakdischarges throughout Germany and alsosummer peak discharges in Region South.The SWZ pattern has the opposite develop-ment: a downward trend during the winterand a non-significant increase during thesummer months.
3.4.3.2 Number of events
The analysis of the number of events witha particular CP (independent of its length)reveals less change than for the numberof days (Table 3.3). Only eight CPs showchanges, seven of these are trends (signif-icant at the 10% SL). During the winterthere are no changes at all, and for the
46
3.4 Results
Figure 3.5: Comparison of CP frequencies for the decades 1951–1960 and 1991–2000
47
3 Changes in the flood hazard through changing circulation patterns
1960 1970 1980 1990 20000
20
40
60
WZ
1960 1970 1980 1990 20000
10
20
30
NWZ
1960 1970 1980 1990 20000
20
40
60
BM
1960 1970 1980 1990 20000
10
20
30
TRM
1960 1970 1980 1990 20000
10
20
30
TRW
1960 1970 1980 1990 20000
10
20
30
SWZ
Winter Trend
Summer Trend
Figure 3.6: Number of days per year in selected summer and winter daily CP series
summer season there are only four trends,which are significant at the 10% SL. Withthe exception of one increase (for BM),which is however not significant, all otherchanges were downward.
3.4.3.3 Mean persistence
Conditions in the catchment prior to a floodplay an important role for the flood hazard.Consecutive precipitation events due to in-creasing persistence of specific CPs con-tribute to soil saturation, leading to higherrunoff, possibly even for weak precipita-tion events. Our analysis reveals many up-ward trends in CP persistence. In the entiredataset, 12 out of 30 CPs show significantchanges in the mean persistence: Upward
trends were detected for 11 CPs, and onlyone downward trend was found for NEZ(cf. Table 3.3). Eight of the 11 CPs alsorevealed upward trends in the maximumpersistence. The only downward trend inthe maximum duration was also found forNEZ. Figure 3.7 shows trends in the meanduration of selected CPs for summer andwinter separately. The patterns WZ, NWZand BM all have trends in the mean du-ration. WZ and BM show for all threedatasets upward trends. The patterns NWZand SWZ show again opposite trends forwinter and summer. However, the trenddirections are similar to those found forthe frequency of days: for SWZ significantupward summer trends and for NWZ sig-nificant upward winter trends. The pattern
48
3.4 Results
1960 1970 1980 1990 20000
5
10
15
WZ
1960 1970 1980 1990 20000
5
10
15
SWZ
1960 1970 1980 1990 20000
5
10
15
NWZ
1960 1970 1980 1990 20000
5
10
15
BM
1960 1970 1980 1990 20000
5
10
15
TRM
1960 1970 1980 1990 20000
5
10
15
TRW
Winter Trend
Summer Trend
Figure 3.7: Mean duration of CPs per year in selected summer and winter daily CP series
TRM also shows increases which are how-ever only significant for the annual datasetand the winter season.
3.4.4 Trend development of
selected circulation
patterns
3.4.4.1 Winter
All three flood regions are dominated dur-ing the winter season by the circulationpatterns WZ, WS and NWZ (cf. Fig. 3.6).Figure 3.8 shows the significance level oftrends for the number of days per winterand the mean duration matrices. Upwardtrends are reflected by the levels of 95 to100, and downward trends by 5 to 0, re-
spectively. Upward trends were detectedin the number of days per winter as wellas in the mean duration for WZ. NWZalso shows upward trends in frequency, es-pecially when including the most recentyears. Large changes are also visible inthe mean duration of NWZ during win-ter, where a significant upward trend is de-tectable, when including the last years inthe time series, no matter how long theseries progresses into the past. WS doesnot change very much over time. Thenumber of events of WZ and NWZ doesnot show a change during winter, for WSa slight decline is visible (not shown inFig. 3.8). These results show therefore anincreased flood hazard during the winterfor all regions of Germany. This is espe-
49
3 Changes in the flood hazard through changing circulation patterns
En
din
g y
ea
r
Number of days WZ
1960 1970 19801970
1975
1980
1985
1990
1995
2000
Number of days of WS
1960 1970 19801970
1975
1980
1985
1990
1995
2000
Beginning year
En
din
g y
ea
r
Mean duration of WZ
1960 1970 19801970
1975
1980
1985
1990
1995
2000
1960 1970 19801970
1975
1980
1985
1990
1995
2000
Beginning year
Mean duration of NWZ
Number of days NWZ
1960 1970 19801970
1975
1980
1985
1990
1995
2000
Beginning year
Mean duration of WS
1960 1970 19801970
1975
1980
1985
1990
1995
2000
0
20
40
60
80
100
0
20
40
60
80
100
Figure 3.8: Significance level of trends in the number of days and mean duration during winter of
the circulation patterns WZ, WS and NWZ. Results between 95 and 100 indicate upward trends,
results between 0 and 5 downward trends.
cially important for the Regions West andEast, where WZ and NWZ trigger largeshares of winter peak discharges.
3.4.4.2 Summer
The trend picture for the summer sea-son is more differentiated for all regions(Fig. 3.9). Although the percentages ofWZ-triggered summer peak discharges aremuch lower than for winter, WZ playsan important role for the flood hazard (cf.Fig. 3.4). In Region South an increasedflood hazard is visible, especially for thegauges in the Rhine catchment. Significantincreases (results between 95 and 100) in
the duration of WZ are evident when in-cluding the last 10 years into the dataset.No changes in number of days and num-ber of events were detected. In contrastto the winter season, NWZ is significantlydecreasing (results between 5 and 0) dur-ing summer in all three CP variables (Ta-ble 3.3). Downward changes are howeveronly significant, when including at least 35years into the trend test (Fig. 3.9). Thisdecrease in the flood hazard caused by theNWZ pattern is again only relevant for Re-gion South, since in the other two regionsthere are rarely NWZ-triggered summerflood events. The Vb-pattern TRM is in-creasing in all three variables. When in-
50
3.4 ResultsE
nd
ing
ye
ar
Number of days WZ
1960 1970 19801970
1975
1980
1985
1990
1995
2000
Number of days of TRM
1960 1970 19801970
1975
1980
1985
1990
1995
2000
Beginning year
En
din
g y
ea
r
Mean duration of WZ
1960 1970 19801970
1975
1980
1985
1990
1995
2000
Beginning year
Mean duration of TRM
1960 1970 19801970
1975
1980
1985
1990
1995
2000
Number of days NWZ
1960 1970 19801970
1975
1980
1985
1990
1995
2000
Beginning year
Mean duration of NWZ
1960 1970 19801970
1975
1980
1985
1990
1995
2000
0
20
40
60
80
100
0
20
40
60
80
100
Figure 3.9: Significance level of trends in the frequency and mean duration during summer of the
circulation patterns WZ, NWZ and TRM. Results between 95 and 100 indicate upward trends, results
between 0 and 5 downward trends.
cluding the last decade in the trend test,the number of days and the mean durationshow upward trends.
3.4.4.3 Correlation between seasonal
MAXF and combinations of CP
Although there is only poor visual agree-ment between multiple trend test matri-ces of the different flood variables and in-dividual CPs (cf. Fig. 3.3, Fig. 3.8, andFig. 3.9), it is interesting to investigatethe correlation of the time series betweenseasonal MAXF and combinations of CPs.The mean frequencies in Fig. 3.4 highlightthe fact that a number of different CPs in-
fluences the peak discharge behaviour inall regions. We conducted therefore cor-relation analyses of seasonal compositeMAXF and combinations of the most im-portant and/or frequent CPs (number ofdays) based on the histograms in Fig. 3.4.All three regions show significant correla-tions (at the 10% SL) for the combinationsof composite winter maximum series andthe sum of days with WZ-NWZ and WZ-WS-NWZ, respectively.
Figure 3.10 exemplarily shows for Re-gions West (first row) and East (secondrow) results of four combinations of CPand seasonal discharge. These combina-tions were chosen based on (1) the domi-
51
3 Changes in the flood hazard through changing circulation patterns
nance of winter peak discharges in RegionsWest and East, and (2) the large numberof significant correlations for summer inRegion East (see below). An overall goodagreement of the fluctuations of seasonalMAXF and the different combinations ofCPs (number of days) can be seen. Theresults show better agreements for both re-gions for the winter season (first row bothdiagrams and lower left diagram) than forsummer. Although the coefficients of cor-relation are quite low (0.43–0.51) they arestatistically significant. These low corre-lation coefficients are caused by (1) thesmall potential of a frequent CP to causea flood event, and (2) the relatively pooragreement of the series during the first twodecades.
For Region South, which is predomi-nantly affected by summer floods, onlynon-significant correlations were foundduring summer, which are therefore notshown. The ASMAXF series of that re-gion comprises many CPs, which can onlybe poorly represented by a small numberof CPs as in Fig. 3.10. The combina-tion with summer maxima (lower right dia-gram) in Region East exemplifies the com-plex relation during summer. Region Eastshows many significant correlations of thecomposite summer maximum series anddifferent combinations of number of CP-days. These are WZ-TRM, TM-TRM, WZ-NWZ-TRM, and WZ-NWZ-TRM-TRW.Even the correlation between the compos-ite ASMAXF of Region East with the num-ber of days of TRM is statistically signif-icant. This is an important finding sinceTRM only comprises about 4% of the en-tire CP, but regularly triggers peak dis-charges during summer in Regions Eastand South. Figure 3.10 (lower right sub-
plot) presents therefore also a combinationfor summer in Region East, the one withthe highest correlation. The overall agree-ment of fluctuations is, however, worsecompared to the three subplots for win-ter combinations. The combination of CPdays exhibits much more variability thanthe composite summer discharge series.
3.5 Discussion
A number of interesting trend results wasfound for each of the three flood regions(West, South and East Region). A linkbetween trends in winter maximum dis-charges and the frequency and persistenceof the CPs WZ and NWZ could be foundfor the Regions West and East. For sum-mer, the link is not that obvious, however,also detectable for selected CPs and theRegions East and South. In the following,the findings of the study are discussed sep-arately for each region.
3.5.1 Region West
Region West is dominated by winter peakdischarges which are significantly increas-ing in magnitude. Several other studiesfound similar results for at least parts of theregion (Caspary, 1995; Caspary and Bár-dossy, 1995; Belz et al., 2007; Petrow andMerz, 2009). The study of Hennegriff et al.(2006) is in good agreement with our find-ings regarding the temporal dynamics ofsignificant flood trends (cf. Fig. 3.3). Theyalso detected at many gauges significantupward trends in AMAXF and AWMAXFfor time series beginning in the 1970s.
The overwhelming part of peak dis-charges in Region West is triggered by
52
3.5 Discussion
1950 1960 1970 1980 1990 2000−3
−2
−1
0
1
2
3
Number of winter days with WZ, WS or NWZand AWMAXF in Region West
1950 1960 1970 1980 1990 2000−3
−2
−1
0
1
2
3
Number of winter days with WZ, WS or NWZand AWMAXF in Region East
1950 1960 1970 1980 1990 2000−3
−2
−1
0
1
2
3
Number of winter days with WZ or NWZand AWMAXF in Region West
1950 1960 1970 1980 1990 2000−3
−2
−1
0
1
2
3
Number of summer days with WZ, NWZ or TRM and ASMAXF in Region East
AWMAXF
WZ−WS−NWZ
AWMAXF
WZ−WS−NWZ
AWMAXF
WZ−NWZ
ASMAXF
WZ−NWZ−TRM
R = 0.51
R = 0.43 R = 0.38
R = 0.44
Figure 3.10: Comparison of seasonal MAXF data and combinations of CPs (number of days) for
Region West (first row) and Region East (second row)
westerly winds of varying direction: WZ,WS, SWZ and NWZ. Although no majortrends were detected for WS, the other CPsrevealed significant changes, which affectthe flood hazard in that region. With onlyslight changes in the number of events, theincreasing number of days and persistenceof WZ and NWZ cause the flood hazardto rise. Belz et al. (2007) also detected in-creasing trends of discharge and WZ forthe winter season in the Rhine catchment.Gerstengarbe and Werner (2005) foundfor the winter season increasing trends ofwesterly atmospheric circulation types, too.Additionally, a longer duration period ofthe persistence of the circulation patternswas observed. This yields a larger flood
hazard through circulation patterns whichare generally not very prone to causingflood events but may be increasingly haz-ardous due to a longer duration. Long-lasting westerly atmospheric circulationtypes cause eventually large-scale soil satu-ration, leading to higher runoff coefficients.For the middle and lower stretches of theRhine, increased flooding probabilities forthe winter season have been suggested byPfister et al. (2004a). During the secondhalf of the 20th century increased winterrainfall totals and intensities have been ob-served.
53
3 Changes in the flood hazard through changing circulation patterns
3.5.2 Region East
Region East is characterized by a similarwinter discharge and CP regime as in Re-gion West. The dominating patterns WZand NWZ increase in frequency and persis-tence and will therefore intensify the floodhazard during the winter season. Mudelseeet al. (2006) also found an increase in thewinter flood hazard during the last decadesfor parts of the region. Gerstengarbe andWerner (2005) found decreasing percent-ages of easterly circulation patterns duringthe winter, which cause cold and dry win-ters especially in Region East. At the sametime, the number of days with precipitationtripled during winter in combination withincreases in the frequency and duration ofthe patterns WZ and NWZ (Werner et al.,2008). These findings fit to our results ofupward trends in the winter maximum dis-charges in Region East which are causedby more rain induced flood events due tomilder winters and an intensified zonal cir-culation (Gerstengarbe and Werner, 2005).
Summer floods play a more importantrole than in Region West, especially floodstriggered by TM and TRM (Petrow et al.,2007). We found decreasing trends in sum-mer maximum discharges in Region East(cf. Table 3.2). A decrease in the flood haz-ard would be expected to be also visible ina decrease in flood-prone CPs during sum-mer. The most frequent pattern WZ showsan upward trend in the duration. However,this pattern usually does not cause largefloods in the region. In contrast the pat-tern TRM is better known to trigger largefloods in the area. Although increases inthe number of days and in the persistenceof TRM were detected, these are not sig-nificant. Thus, a decreasing flood hazard
based on the trends found in the dischargedata of Region East is possible, despitethese non-significant increases in the dura-tion of flood-prone CPs.
3.5.3 Region South
Region South is dominated by summermaximum discharges. Also for this region,an increasing flood hazard can be founddue to increasing trends in the patterns WZ,SWZ, TRM and TRW which play an im-portant role for AMAXF discharges in theregion. Our results show an upward trendin the persistence of SWZ during summer.This pattern is prone to triggering heavyconvective rainfall during summer, whichregularly causes local flood events. Also,Gerstengarbe and Werner (2005) found forsummer large upward trends in the fre-quency of SWZ (tripled frequency witha step change in the 1940s).
3.6 Conclusions
Analyses of trends in flood hazard and inflood-triggering CPs show a regionally andseasonally differentiated picture for Ger-many. We investigated flood time seriesof 122 meso-scale catchments in Germanyand their triggering circulation patterns.Our analysis detected discharge trends (atthe 10% SL) for a large number of catch-ments, as well as in the frequency and per-sistence of flood-favouring circulation pat-terns. Of particular interest is a significantincrease in (1) the flood relevant CPs, aswell as in (2) the very frequent CPs. Signif-icant correlation (at the 10% SL) betweenthe frequency of CPs and seasonal floodtime series was detected.
54
3.6 Conclusions
We found a trend (significant at 10% SL)towards a reduced diversity of CPs, caus-ing fewer patterns with longer persistenceto dominate the weather over Europe (cf.Fig. 3.7). This indicates changes in the dy-namics of atmospheric circulations whichare of direct relevance to the flood haz-ard. Longer persistence of CPs may lead toconsecutive precipitation events. Althoughthe single events may have rather low pre-cipitation amounts, the succession of sev-eral events may lead to saturated catchmentconditions. This is particularly importantfor winter peak discharge, which are inmany cases triggered by WZ or NWZ pat-terns. Both are weather patterns that donot favour extreme precipitation. How-ever, very wet preconditions may causelarge runoff coefficients paving the wayfor flooding. Rapp and Schönwiese (1996)found upward trends in winter precipita-tion in the period 1891–1990 for largeparts of south-western and western Ger-many. These results are in agreementwith our findings and other studies for theRhine catchment, which show significantincreases in precipitation extremes and inthe flood hazard during winter (e.g. Cas-pary and Bárdossy, 1995; Hundecha andBárdossy, 2005; Belz et al., 2007; Petrowand Merz, 2009).
The investigated time span of 52 yearsin our study is relatively short compared tolow-frequency climate variability. Thereare well organized modes of climate vari-ability at different time scales and this vari-ability may have a significant impact onthe occurrence and magnitude of floodsby changed atmospheric moisture trans-port (Hirschboeck, 1988). For example,Llasat et al. (2005) compiled a catalogue offloods for three basins in north–east Spain
since the 14th century and found episodesof 20 to 40 years with markedly increasedoccurrence of catastrophic floods. Sturmet al. (2001) compiled a catalogue withfloods in Central Europe from 1500 untiltoday. Basins in Central Europe show clus-tering of floods. Given low–frequency cli-mate variations, a much longer time periodwould have been favourable. However, acompromise between data availability andspatial coverage had to be found, whenconducting a countrywide study. Althoughlong series would capture a broader pic-ture of the variability, a good spatial cover-age with shorter time series was favouredover a long period of more than 100 yearsat only few stations. Further, long floodtime series covering 100 or more years,are often associated with considerable un-certainty. For example, Glaser and Stangl(2004) stress that the direct comparison isproblematic between reconstructed histor-ical floods and measured data due to thedifferent derivation of the datasets. ForFinally, the considered time period is par-ticularly interesting, since global warmingis supposed to be of minor effect before thesecond half of the 20th century.
The presented results have implicationsfor the flood risk management, especiallyfor flood design measures. Petrow et al.(2008) compared a stationary and an insta-tionary flood frequency analysis approach.They showed for the period of 1951–2002that the stationary estimation (which as-sumes no trend in the data) may underes-timate discharges of extreme events. Ow-ing to the many detected flood trends inour study, a revised estimation of extremeevents, which incorporates the instationar-ity inherent in the data, seems appropriatefor the affected catchments.
55
3 Changes in the flood hazard through changing circulation patterns
Acknowledgements
We thank the Helmholtz Centre Potsdam –German Research Centre for Geosciences(GFZ) and the Helmholtz Association ofNational Research Centres for their finan-cial support. We dedicate our specialthanks to the authorities that provided data.
56
4 Aspects of seasonality and flood generating
circulation patterns in a mountainous
catchment in south-eastern Germany
Abstract
Analyses of discharge series, precipitation fields and flood producing atmospheric circu-lation patterns reveal that two governing flood regimes exist in the Mulde catchment insouth-eastern Germany: frequent floods during the winter and less frequent but sometimesextreme floods during the summer. Differences in the statistical parameters of the dischargedata can be found within the catchment from west to east. The discharges are comparedto a number of landscape parameters that influence the discharge in the subcatchments.Triggering circulation patterns were assigned to all events of the annual maximum dischargeseries in order to evaluate which circulation patterns are likely to produce large floods. Itcan be shown that the cyclone Vb-weather regime (TM, TRM) generates the most extremeflood events in the Mulde catchment, whereas westerly winds produce frequently smallfloods. The Vb-weather pattern is a very slowly moving low pressure field over the Gulf ofGenoa, which can bring large amounts of rainfall to the study area. It could also be shownthat even with the two flood regimes estimates with the annual maximum series provide asafer flood protection with a larger safety margin than using summer maximum dischargeseries for extreme summer floods only. In view of climate change it is necessary to integrateknowledge about catchment characteristics, the prevailing flood regime or the trends ofweather patterns in the estimation of extreme events.
Published as: Petrow, Th., B. Merz, K.-E. Lindenschmidt, and A.H. Thieken. 2007.
Aspects of seasonality and flood generating circulation patterns in a mountainous
catchment in south-eastern Germany. Hydrology and Earth System Sciences, 11:
1455–1468.
57
4 Aspects of seasonality and flood generating circulation patterns
4.1 Introduction
Limited data on extreme and thus rareflood events complicate the accurate es-timation of design discharges (e.g. Francéset al., 2001; Benito et al., 2004; Merz andThieken, 2005). Numerous approacheshave been developed for flood estimation,which include statistical approaches suchas flood frequency analysis (FFA), theuse of envelope curves as well as rainfall-runoff modelling with hydrological models.The focus in this study is set on the FFA.
The most common methods for FFA useannual maximum series (AMS) and peakover threshold series (POT) (Institute ofHydrology, 1999). The AMS and POT se-ries can also be extracted for summer orwinter seasons, when, for instance, oneflood process type (e.g. floods triggered bysnow melting) is of special interest. Sev-eral distribution functions such as the Gum-bel, Weibull, Generalized Extreme Value,or the Pearson type III can be fitted tothe data (Hosking and Wallis, 1997; In-stitute of Hydrology, 1999). Althoughthese functions and possibilities exist asto which data to integrate, large uncertain-ties still remain when estimating extremeevents. There is much debate about thelength of the data series. Short series maynot capture the entire flood variability andvery long series may not reflect station-ary conditions (e.g. Bárdossy and Pakosch,2005; Khaliq et al., 2006)). Moreover, itis questionable whether or not an AMSis stationary when the discharges reflectdifferent flood producing processes. In-dependence, homogeneity and stationarityare required characteristics of the data tolegitimate flood frequency analysis (Ste-dinger, 2000; Kundzewicz and Robson,
2004). However, often these criteria arenot satisfied due to climatic change and/oranthropogenic influence (Webb and Be-tancourt, 1992; Klemes, 1993; Jain andLall, 2000; Sivapalan et al., 2005; Svens-son et al., 2005; Khaliq et al., 2006). Inde-pendence is almost always given, when an-alyzing annual maximum series, whereaspartial series have to be carefully examinedin order to avoid miscounting one floodevent as two. Usually, a threshold of sev-eral days is included in the extraction of thedata, which defines the minimal time be-tween two floods to ensure independenceof the events. This threshold can compriseup to 30 days depending on the catchmentarea and discharge conditions. Stationaryconditions seldom exist due to changes inclimate, land-use or in the vulnerability ofthe study area, although these are oftenassumed (Merz, 2006). Moreover, the dy-namics of atmospheric processes and floodgeneration have to be taken into account inthe study of stationarity and independenceand further in the FFA (Merz and Blöschl,2003; Sivapalan et al., 2005).
The relationship between climate andflood generation has been of growing inter-est and study (Webb and Betancourt, 1992;Kästner, 1997; Jain and Lall, 2000; Bár-dossy and Filiz, 2005; Steinbrich et al.,2005; St. George, 2007). Steinbrich et al.(2005) analyze the correlation between cir-culation patterns (CP) and heavy rain forthe south-western part of Germany (Baden-Wuerttemberg). Kästner (1997) found thatonly five out of thirty different weatherpatterns are susceptible to produce floodevents in Bavaria. Three catchments insouthern Germany (Bavaria), which havedifferent discharge characteristics and aredifferently influenced by snow melting,
58
4.2 Study area and data
were studied. Kästner (1997) found theVb-weather regime to be most susceptiblefor the generation of large floods. Thisweather system is a low pressure systemthat moves very slowly from the Gulf ofGenoa northwards. It can accumulate largeamounts of moist and warm air over theMediterranean Sea, which is transformedinto large precipitation amounts that fallalong the northern slopes of the Alps andmountain ranges in Central and Eastern Eu-rope. It is therefore interesting to analyzethe relationship of circulation patterns andflood generation in the study area.
More information about flood gener-ating processes can be gained when ex-tending the study from one gauge sta-tion to the hydrological behaviour of sub-catchments and neighbouring regions (Har-lin and Kung, 1992; Merz et al., 2006;Ouarda et al., 2006). Harlin and Kung(1992) extract for each sub-catchment themost extreme measured events and sim-ulate the simultaneous occurrence of thefloods which has not been observed yet. Ofspecial interest for the flood hazard estima-tion of ungauged areas is also the regionalFFA which incorporates flood process in-formation from neighbouring catchments(e.g. Stedinger, 1983; Hosking and Wal-lis, 1997; Institute of Hydrology, 1999)).Regionally valid distribution functions arefitted to data of preferably independentgauges within a region, which exhibit, ingeneral, better fits (Merz, 2006).
In this paper the flood discharge charac-teristics of the Mulde catchment in south-eastern Germany are analyzed accordingto stationarity, their spatial distribution ofthe statistical moments and the relation-ship between landscape characteristics andflood peaks. Additionally, the relation-
ship between the dominating weather pat-tern in Europe and the flood generationin this catchment is discussed. The fol-lowing questions will be answered basedon this analysis: Which landscape com-ponents (geology, soil, groundwater flow,land-use, precipitation) contribute to theflood discharge regime? Can seasonal orspatial differences be distinguished? Dospecific circulation patterns exist whichtrigger large events? And finally, are the re-quirements for the flood frequency analysiswith AMS for this catchment fulfilled?
4.2 Study area and data
4.2.1 Study area
The Mulde catchment is a sub-catchmentof the Elbe River basin in south-easternGermany. The southern boundary ismarked by the mountain ranges of theErzgebirge, which coincides with theCzech-German border. The catchment hasa total area of 6171 km2 (at the gauge BadDüben) and has three large sub-catchments(Zwickauer Mulde, Zschopau, FreibergerMulde), which drain the upper, mountain-ous part of the catchment (Fig. 4.1). Withinonly 20 km, the tributaries Zschopau andFreiberger Mulde disembogue near thegauge Erlln (gauge 13, Fig. 4.1) into theZwickauer Mulde and form the VereinigteMulde ("Joined Mulde"), which disem-bogues near the city of Dessau into theElbe River.
The elevation ranges from 52 m to1213 m a.s.l. with approx. 2/3 of the areabeing lowlands and 1/3 mountains (500–1213 m a.s.l.) (Fig. 4.1). The mountainranges in the south cause fast runoff re-
59
4 Aspects of seasonality and flood generating circulation patterns
sponses to rainfall events in the tributaries,whereas in the major part of the catchmentslower runoff responses dominate. The an-nual precipitation ranges from 500 mm inthe lowlands to 1100 mm in the mountainranges.
The landscape characteristics of thecatchment such as geology, soil, hydro-geology and land-use parameters wereevaluated to gain information about thevariability of the discharge behaviour.Therefore, the catchment was split intothree zones, which correspond to the threelarge subcatchments (Fig. 4.1).
")
")
")
")
")
")
")
")
")
")
")
")
")
")
")
3
4
5
7
2
9
6
8
1
1210
14
13
11
15
Legend
") Discharge Gauge
River
Elevation in m a.s.l.
Value
52 - 250
260 - 500
510 - 750
760 - 1,200
Sub-catchment
$
$
$
$
$
Dresden
Berlin
Cologne
Frankfurt
Munich
Magdeburg
0 10 20 30 405Kilometers
Vere
inig
te M
uld
e
Fre
iberg
er M
uld
e
Zschopau
Zwic
kauer
Mul
de
Figure 4.1: Study area Mulde catchment: left:
discharge gauge locations (numbered accord-
ing to Tab.4.1) and the digital elevation model;
right: geographical location in Germany
The region has a long history of largeflood events. First written documentsabout floods, the corresponding water lev-
els and damage can be found from the 9thcentury onward and more detailed docu-ments starting from the 14th century (Pohl,2004). It is noteworthy that large winterfloods with ice blockage as well as summerfloods from torrential storms or long last-ing frontal rains caused high damages oninfrastructure and agriculture, often withfatalities.
During the last 100 years, three extremeflood events occurred in the study area,namely in July 1954, July 1958 and Au-gust 2002. These events will be analyzedin more detail in this paper. All of themwere caused by large torrential storms. Thefloods in 1954 and 2002 were triggered byVb-weather systems. Both flood events inthe fifties caused high damage in differentparts of the catchment, whereas in 2002the entire catchment was affected. Thisflood caused a damage of 11.6 Billion ein Germany alone (DKKV, 2004; Thiekenet al., 2006). As a consequence of the floodhistory, flood defence measures play an im-portant role and have been extended untilthe present day (DKKV, 2004). Numer-ous flood retention basins and dams wereconstructed, which are mainly located inthe upper part of the catchment, and signifi-cantly influence the discharge downstream.
4.2.2 Data
4.2.2.1 Discharge data
Over 60 discharge and water level gaugesexist in the Mulde catchment. The earliestmeasurements at regular intervals beganin 1910 at two gauges. In order to evalu-ate the influence of a dam before includingdata from the downstream discharge gaugeinto the dataset, daily differences of inflow
60
4.2 Study area and data
Number Gauge Basin Elevation Period of Mean max. Highestarea [m.a.s.l.] Measure- annual flood value of
[km2] ments discharge observation[m3/s] period
1 Aue 1 362 349 1928–2002 66 3152 Niederschlema* 759 314 1928–2002 111 5853 Zwickau-Poelbitz* 1030 255 1928–2002 128 6834 Wechselburg 1 2107 160 1910–2002 213 10005 Streckewalde 206 410 1921–2002 30 1456 Hopfgarten* 529 357 1911–2002 81 4207 Pockau 1 385 397 1921–2002 69 4498 Borstendorf 644 356 1929–2002 91 5409 Lichtenwalde 1575 253 1910–2002 218 1250
10 Kriebstein UP 1757 183 1933–2002 231 135011 Berthelsdorf 244 377 1936–2002 35 36012 Nossen 1 585 204 1926–2002 69 69013 Erlln 2983 133 1961–2002 329 155014 Golzern 1* 5442 118 1911–2002 517 260015 Bad Düben 1 6171 82 1961–2002 474 1760
Table 4.1: Analyzed discharge gauges in the study area (* stations with one year of missing values)
versus outflow of five large dams for theperiod 1991–2002 were compared. Moreinformation from the dam authorities wasnot available. Inflow and outflow floodpeaks were compared and the downstreamstations were excluded from the datasetif the flood peak differences were greaterthan 10%, and if there were at least fiveaffected flood events during this 10 yearperiod. Additionally, daily time series ofdischarge gauges that are in the immedi-ate vicinity of a dam were compared todaily discharge data from neighbouringgauges at other tributaries. Time seriesof discharge gauges that did not reflect thehydrograph at the compared gauge stationwere excluded from the dataset. AMS (hy-drological year from November to Octo-ber) were extracted from daily maximumdischarges.
A subset of discharge gauges was se-lected for this analysis which met the fol-lowing criteria:
• the time series must have a length ofat least 40 years,
• the sub-catchment area is larger than100 km2,
• the flood AMS exhibits no trend,
• the discharge gauges are distributedacross the catchment and have a dis-tance of at least 3 km between eachother.
15 discharge gauges meet these criteria;they are listed in Fig. 4.1 and Table 4.1.For better readability, the gauge stationsare listed in all tables in the same orderbeginning with those located in the south-west (Zwickauer Mulde), then progressing
61
4 Aspects of seasonality and flood generating circulation patterns
north and east (Zschopau, Freiberg Mulde)and ending with gauges located in the Vere-inigte Mulde (cf. Fig. 4.1).
4.2.2.2 Precipitation Data
Precipitation data were available from theGerman Weather Service (DWD) at 49 sta-tions in and around the Mulde catchment(see Fig. 4.2). The data cover the timeperiod between 1952 and 2002 on a dailybasis. Daily areal precipitation was calcu-lated based on cubic interpolation for eachof the 15 sub-catchments (corresponding tothe discharge stations) for the comparisonof precipitation and discharge.
4.2.2.3 Atmospheric circulation
patterns
Information about the predominant Euro-pean circulation pattern for each day wasavailable from the "Catalogue of Großwet-terlagen in Europe 1881–2004" (Gersten-garbe and Werner, 2005). The cata-logue distinguishes three large circulations,which are divided into 30 different circula-tion patterns (one is classified to be a "tran-sition class") (Tab. 4.2). The Vb-weathersystem is represented by the patterns TM(low Middle Europe) and TRM (TroughMiddle Europe).
The circulation patterns comprise thezonal circulation form, the mixed circu-lation form as well as the meridional cir-culation form. For every day a circulationpattern is assigned to be the dominant onefor Europe. Through the specific distri-bution of lows and highs over Europe, itmay therefore be possible that the domi-nant circulation pattern of a particular dayis not necessarily representative for the
Mulde catchment. This is for instance thecase, if the Mulde catchment is still underthe influence of a weakened low, whichis however already situated above EasternEurope, whereas the dominating Europeancirculation pattern is above Western Eu-rope. However, other than this catalogue,more detailed meteorological data for thestudy area were not available.
")
")
")
")
")
")
")
")") ")
")
")
")
")
")
")
")
")
")
")
")
")
")
")
")")
")
")
")
")
")
")
")
")
")
")
")
")
")
")
")
")
")
")
")
")
")
")
")
0 10 20 30 405Kilometers
Legend
") Precipitation station
Rivers
Catchment area
Figure 4.2: Locations of the 49 precipitation
stations in and around the study area
4.3 Methodology
4.3.1 Flood frequency analysis
The distribution-free and non-parametricMann-Kendall test for Trend (one-sidedtest; significance level: α = 0.05) wasused for the detection of trends in the data.
62
4.3 Methodology
Form of Circulation No. Circulation pattern name Abbr.Zonal Circulation 1 West wind, anti-cyclone WA*
2 West wind, cyclone WZ*3 Southern west wind WS*4 Angular west wind WW*
Mixed circulation 5 South-west wind, anti-cyclone SWA*6 South-west wind, cyclone SWZ*7 North-west wind, anti-cyclone NWA*8 North-west wind, cyclone NWZ*9 High pressure system, middle Europe HM*
10 High pressure circuit over middle Europe BM*11 Low pressure system, middle Europe TM*
Meridional circulation 12 North wind, anti-cyclone NA13 North wind, cyclone NZ14 High pressure Iceland, anti-cyclone HNA15 High pressure Iceland, cyclone HNZ*16 High pressure, British Isles HB*17 Trough Middle Europe TRM*18 North-east wind, anti-cyclone NEA19 North-east wind, cyclone NEZ*20 High pressure Fennoscandia, anti-cyclone HFA*21 High pressure Fennoscandia, cyclone HFZ22 High pressure Norwegian Sea- HNFA
Fennoscandia, anti-cyclone23 High pressure Norwegian Sea- HNFZ
Fennoscandia, cyclone24 South-east wind, anti-cyclone SEA25 South-east wind, cyclone SEZ*26 South wind, anti-cyclone SA27 South wind, cyclone SZ28 Low Pressure, British Isles TB*29 Trough, Western Europe TRW*30 Transition, no classification U
Table 4.2: Classification of the form of circulation and its specific pattern (* indicates circulation
patterns which are relevant for AMS discharges in the Mulde catchment)
Since small trends in the data may notbe detectable, for instance by the Mann-Kendall test (Bárdossy and Pakosch, 2005),a regional test of stationarity was con-ducted with all 15 data sets (Lindström andBergström, 2004). To this end, several dataseries from the same region, that cover the
same period of measurements, are testedjointly (also with the Mann-Kendall test).For comparison, the discharge data weredivided by the MAF (mean maximum an-nual flood discharge) of the respective se-ries. AMS of 13 gauge stations with datafrom 1936 to 2002 and of two gauges with
63
4 Aspects of seasonality and flood generating circulation patterns
data from 1961 to 2002 were included.Independence of the data was ensured by
using AMS data, which were also checkedfor possible dependent values around theturn of a hydrological year. For this, athreshold time of 7 days between two AMSfloods was included, which guarantees theindependence of two close-by flood events,since the time of concentration for thisbasin is smaller than 7 days.
Flood frequency analyses were per-formed with seven different distribu-tion functions (Gumbel, Weibull, 2-parametric LogNormal, Generalized Ex-treme Value (GEV), General Logistics(GL), 3-parametric LogNormal, and Pear-son type III) with both the Method of Mo-ments and with the L-Moments (Hosk-ing and Wallis, 1997; Institute of Hydrol-ogy, 1999). The GEV and GL distributionfunctions (both with L-Moments) revealedthe best fits based on the Kolmogorov-Smirnov test and visual examination rel-ative to the empirical probabilities (Testhypothesis: F(x) = CDF for all x withα = 0.05). Emerging consensus can befound in many studies worldwide that theGEV distribution reveals the best fits (Pear-son, 1991; Onoz and Bayazit, 1995; Vo-gel and Wilson, 1996; Douglas and Vogel,2006). The Institute of Hydrology (1999)also describes the "theoretical and histori-cal importance" of the GEV. Hence, subse-quent analyses were performed using theGEV.
4.3.2 Spatial distribution of
flood characteristics
The spatial distributions of the statisticalmoments of the AMS, such as skewness
and coefficient of variation, were analyzedto detect possible differences among sub-catchments. The spatial extent and distribu-tion of the three most extreme flood events(July 1954, July 1958, August 2002) wereanalyzed in more detail. For every eventand gauge station, return periods (GEV,L-Moments) were calculated. These es-timates were then assigned to each riversegment upstream of the 15 gauge stationsin order to analyze the flood characteristicsin a spatially explicit manner.
Moreover, the AMS of 11 gauge stationswith data from 1929 to 2002 (74 years)were studied with respect to the spatial dis-tribution and magnitude of flood events.To this end, the number of different floodevents per year in the catchment was ana-lyzed. If all 11 gauges have their highestdischarge of a certain year on the same day(+/- 1 day), the number of flood events forthat year will be one. The other extreme isthat all gauges have their highest peak atanother time of the year. In that case, thenumber of flood events for that year is 11.
4.3.3 Relationship between
precipitation maxima and
discharge maxima
The relationship between precipitationmaxima and discharge maxima was studiedin more detail. Areal precipitation was cal-culated for the three large sub-catchments(Zwickauer Mulde: gauge Wechselburg;Zschopau: gauge Lichtenwalde; FreibergerMulde: gauge Nossen) and the VereinigteMulde at the gauge Golzern. Precipita-tion sums of 24 h, 48 h and 72 h of theflood events were compared to dischargemaxima. The four discharge stations are
64
4.3 Methodology
distributed over the entire catchment andrepresent the large sub-catchments. Rain-fall AMS were extracted from the precipita-tion data and then compared on a seasonalbasis to the discharge AMS to determine,how many large precipitation events arereflected in the discharge AMS.
4.3.4 Circulation pattern and
flood generation
Daily data of circulation patterns between1911 and 2002 were analyzed in orderto obtain an overview about the seasonaldistribution and frequency of the circu-lation patterns in Europe. Additionally,the circulation patterns, which are trigger-ing the AMS discharges, were assignedto the AMS flood data of the gauge Golz-
ern. The gauge at Golzern is representativefor the entire catchment, because it com-prises 88% of the catchment area. As thefirst gauge at the Vereinigte Mulde it rep-resents the influence of the two large sub-catchments. Moreover it has a long timeseries (1911–2002) compared to nearbygauges such as Bad Düben or Erlln (both43 years).
From the AMS data, empirical probabil-ities were assigned to the flood events andthen combined with the circulation patterndata. With this information, it is possible toestimate the potential of a circulation pat-tern to generate a flood of a certain returnperiod.
Normalized AMS from 15 stations (1936-2002)
y = 0.005x - 8.8063
0
1
2
3
4
5
6
7
8
9
10
11
1933 1938 1943 1948 1953 1958 1963 1968 1973 1978 1983 1988 1993 1998 2003
HQ
/M
AF
Figure 4.3: Regional trend test based on discharge data of 15 stations.
65
4 Aspects of seasonality and flood generating circulation patterns
4.4 Results
4.4.1 Testing for trends in the
flood AMS
The one-sided Mann-Kendall test for in-creasing trend (significance level α = 0.05)revealed no trends for all 15 gauge stations.The trend test for regional stationarity wasperformed with the normalized AMS ofthe 15 gauge stations. As Fig. 4.3 shows,the data exhibit a very small positive trendin the regional trend analysis. When theflood event from August 2002 was ex-cluded from the data, the slightly positivetrend became slightly negative. The Mann-Kendall test showed no trend (significancelevel α = 0.05). Therefore the data wereused for flood frequency analysis.
4.4.2 Seasonal occurrence and
magnitude of floods
Two dominant flood process types in theMulde catchment can be extracted from
the data. During March and April, a firstpeak in the discharge AMS occurs duringsnow melt and "rain on snow" flood events.The second peak occurs in July and Au-gust, when large torrential storms traversethe area (Table 4.3). At all 15 discharge sta-tions winter floods (November-April) com-prise a larger part of the AMS than summerfloods. In the upper western part of theErzgebirge (corresponding to the gaugesat Aue, Niederschlema, Zwickau), the per-centage of summer and winter floods in theAMS is almost equal (e.g. Aue: 46% sum-mer floods; 54% winter floods), whereasin the eastern part of the catchment winterfloods have larger percentage (59%–69%).
The winter floods are usually smallevents with a low return period. They con-stitute at all 15 gauges only 8–21% of the20% of the largest floods. Summer floodevents, on the other hand, are less frequent,but cover a larger proportion of extremeevents (26–39%). In Fig. 4.4 the data ofTable 4.3 are summed up for all 15 gauges.Additionally, the monthly distribution of
Gauge Jan Feb March April May June July Aug Sept Oct Nov DecAue 8 4 9 24 8 5 15 8 7 4 3 5Niederschlema 5 5 12 23 8 8 15 7 4 4 1 7Zwickau 4 5 11 20 8 9 16 7 4 4 3 8Wechselburg 12 8 13 9 5 8 17 9 1 2 5 12Streckewalde 11 9 16 17 5 7 17 9 0 4 1 5Hopfgarten 13 10 14 11 7 8 12 7 1 5 1 11Pockau 11 11 17 10 10 6 12 7 2 4 2 7Borstendorf 8 9 20 14 9 5 11 7 1 4 3 8Lichtenwalde 13 14 19 10 6 5 9 9 1 2 1 11Kriebstein 9 11 19 14 7 7 10 7 1 3 1 10Berthelsdorf 7 13 24 7 9 3 10 7 1 1 1 13Nossen 10 16 23 5 6 4 9 6 3 3 3 12Erlln 10 12 26 10 7 2 7 12 2 2 0 10Golzern 14 12 16 9 5 7 11 9 2 3 3 8Bad Düben 10 10 26 12 7 2 10 10 2 2 0 10
Table 4.3: Monthly relative frequency of discharge AMS (in percent)
66
4.4 Results
0
25
50
75
100
125
150
175
200
225
January February March April May June July August September October November December
His
tog
ram
of
dis
ch
arg
e A
MS
per
mo
nth
Number of discharge AMS (sum over 15 gauges)
Number of 20% largest AMS floods per gauge (sum over 15 gauges)
Figure 4.4: Monthly distribution of the number of discharge AMS, summed up over the 15 gauges
for all AMS floods and for the 20% largest events
the 20% largest flood events is shown.Again, it is visible that winter floods have alarge percentage of the AMS, but the mostextreme events occur during the summer.From these analyses we could concludethat summer flood events play a more im-portant role for the flood hazard estimationof extreme events, which would necessi-tate the usage of Summer Maximum Se-ries (SMS) instead of AMS. A comparisonof return periods estimated with AMS andSMS for the three extreme flood eventsshowed however that estimated return pe-riods up to 270 years are at all 15 gaugesmuch lower with AMS. As an example re-turn periods (GEV) for the gauge Aue areshown for the three floods 1954: 48 (AMS),65 (SMS); 1958: 7 (AMS), 10 (SMS);
2002: 115 (AMS), 143 (SMS). Thus, alarger discharge would be needed to esti-mate the same return period, e.g. a designdischarge of 100 years when using AMScompared to SMS. Estimates for return pe-riods larger than 270 years show, howeverlower values with SMS. Therefore, floodprotection measures designed on the basisof AMS estimated return periods providesafety margins, even for extreme summerevents up to 250 years.
4.4.3 Spatial distribution of
flood characteristics
The AMS of 11 gauge stations with datafrom 1929 to 2002 (74 years) were stud-ied with respect to the spatial distribution
67
4 Aspects of seasonality and flood generating circulation patterns
January 1938
0
1
2
3
4
5
6
7
8
9
10A
ue
1
Nie
ders
ch
lem
a
Zw
ickau
-
Po
elb
itz
Wech
selb
urg
Str
eckew
ald
e
Ho
pfg
art
en
Po
ckau
Bo
rste
nd
orf
Lic
hte
nw
ald
e
No
ssen
Go
lzern
Retu
rnp
eri
od
July 1954
0
20
40
60
80
100
120
140
160
Au
e1
Nie
ders
ch
lem
a
Zw
ickau
-
Po
elb
itz
Wech
selb
urg
Str
eckew
ald
e
Ho
pfg
art
en
Po
ckau
Bo
rste
nd
orf
Lic
hte
nw
ald
e
No
ssen
Go
lzern
Retu
rnp
eri
od
July 1958
0
10
20
30
40
50
Au
e1
Nie
ders
ch
lem
a
Zw
ickau
-
Po
elb
itz
Wech
selb
urg
Str
eckew
ald
e
Ho
pfg
art
en
Po
ckau
Bo
rste
nd
orf
Lic
hte
nw
ald
e
No
ssen
Go
lzern
Retu
rnp
eri
od
October 1960
0
1
2
3
4
5
6
7
8
9
10
Au
e1
Nie
ders
ch
lem
a
Zw
ickau
-
Po
elb
itz
Wech
selb
urg
Str
eckew
ald
e
Ho
pfg
art
en
Po
ckau
Bo
rste
nd
orf
Lic
hte
nw
ald
e
No
ssen
Go
lzern
Retu
rnp
eri
od
August 1984
0
1
2
3
4
5
6
7
8
9
10
Au
e1
Nie
ders
ch
lem
a
Zw
ickau
-
Po
elb
itz
Wech
selb
urg
Str
eckew
ald
e
Ho
pfg
art
en
Po
ckau
Bo
rste
nd
orf
Lic
hte
nw
ald
e
No
ssen
Go
lzern
Retu
rnp
eri
od
August 2002
0
100
200
300
400
500
600
700
800
Au
e1
Nie
ders
ch
lem
a
Zw
ickau
-
Po
elb
itz
Wech
selb
urg
Str
eckew
ald
e
Ho
pfg
art
en
Po
ckau
Bo
rste
nd
orf
Lic
hte
nw
ald
e
No
ssen
Go
lzern
Retu
rnp
eri
od
SD = 1
SD = 1
SD = 1 SD = 263
SD = 13
SD = 47
Figure 4.5: Variation of return periods for six different floods (SD = standard deviation)
and magnitude of flood events. To this end,the number of different flood events peryear in the catchment was analyzed. In 13years of the 74-year period, one flood eventoccurred that affected all 11 sub-basins,whereas in 18 years no dominant floodevent (i.e. four to seven flood events peryear) could be identified. These are sum-mer and winter events. In most years (27)three different flood events are related toAMS discharges.
In Fig. 4.5 six different flood events atthe 11 analyzed gauges and their respec-tive return periods are shown. The re-turn periods were estimated with the GEV(L-Moments). The six flood events com-prise the three largest events in the catch-ment (1954, 1958, 2002) and three smallcatchmentwide events. Events with dis-charges that correspond up to a 10-yearpeak discharge are mostly homogenouslydistributed across the catchment. They
68
4.4 Results
Figure 4.6: Estimated return periods (GEV, L-Moments) for the floods in 1954, 1958, 2002 (period
1929–2002 (above)) and the corresponding precipitation fields (below). Note that for a better illus-
tration of the spatial distribution the classes of discharge return periods and precipitation amounts
differ
have similar return periods at all gaugesand exhibit a standard deviation of 1. Thisis shown for the floods in January 1938, Oc-tober 1960 and August 1984. Events withdischarges larger than a 10-year peak ex-hibit increasing spatial distinctions as wellas increasing standard deviations. This isillustrated by the floods in 1954, 1958 and2002. Depending on the location of theprecipitation field, one or the other sub-catchment is more affected during a largeflood event. Figure 4.6 shows the spatialdistribution of the return periods that werecalculated for the observed discharges ofthe three most extreme flood events (1954,
1958, 2002) in the Mulde catchment (upperpart) and the corresponding areal precipita-tion events (lower part). The return periodcalculated for a certain gauge was assignedto the river segment upstream of the gauge.A marked spatial distribution can be seen.For the flood event in 1954, high return pe-riods were calculated for the western partof the catchment. This is explained by therainfall event that had its centre in the west-ern part. The floods in 1958 and 2002 werecaused by precipitation events with theircentres east of, or in the eastern part ofthe study area. Figure 4.6 illustrates thedirect relationship between the location of
69
4 Aspects of seasonality and flood generating circulation patterns
")
")
")
")
")
")
")
")
")
")
")
")
")
")
")
")
")
")
")
")
")
")
")
")
")
")
")
")
")
")
Legend
Skewness
") 2.1 - 2.5
") 2.6 - 3.0
") 3.1 - 3.5
") 3.6 - 4.5
") 4.6 - 6.7
0 10 20 30 405Kilometers
A) B) Legend
Coefficient of Variation
") 0.7
") 0.8
") 0.9 - 1.1
") 1.2 - 1.3
Figure 4.7: Skewness (A) and coefficient of variation (B) of the discharge AMS for the 15 gauges
Zwickauer Zschopau FreibergerMulde Mulde
Landuse Urban Areas 12% 7% 7%Agricultural Land 52 % 60% 70%Forest 32% 33% 18%
Soil Cambisols and Planosols 88% 94% 90%Hydrogeology no or small local 94% 99% 99%
groundwater reservoirsGeology Metamorphic or plutonic rocks 66% 91% 85%
Table 4.4: Percentages of the dominating landscape characteristics
the precipitation field and the flood returnperiod for the three events.
More similar statistical moments werefound along the tributary rivers rather thanaccording to the elevation of the gauge lo-cations. In the beginning the assumptionwas made that the gauges in the moun-
tains of the Erzgebirge can be groupedtogether to exhibit similar statistical mo-ments as well as the gauges in the low-lands. However, increasing values of thestatistical moments occur from west to eastthat corresponds to the division of the sub-catchments. Figure 4.7 shows the spa-
70
4.4 Results
24 h 48 h 72 hGauge Summer Winter Summer Winter Summer WinterWechselburg 65% 7% 61% 7% 70% 10%Lichtenwalde 88% 20% 88% 7% 71% 14%Nossen 78% 15% 89% 26% 83% 26%Golzern 59% 20% 68% 17% 68% 20%
Table 4.5: Percentages of agreement between precipitation AMS (precipitation sums of 24h, 48h
and 72h) and discharge AMS
tial distribution of the skewness (A) andthe coefficient of variation (B) for the 15gauges. The subcatchment of the Zwick-auer Mulde and the western part of theZschopau (gauges 1–6 in Table 4.1) aremore homogeneous and differ significantly(CI 95%) in its statistical moments fromthe eastern part of the catchment. These re-sults suggest a different distribution of theprecipitation in the subcatchments whichin turn leads to differences in the dischargebehaviour. Another possibility is that thelandscape characteristics are largely re-sponsible for these differences, which isdiscussed in the following section.
4.4.4 Landscape characteristics
The land-use is dominated forest cov-ered mountains and intensively used agri-cultural lowland. The proportion ofagriculturally-used areas increases fromwest to east and south to north, whereasthe percentage of forest decreases. Urbanareas only play a role in the sub-catchmentZwickauer Mulde with two larger cities(Zwickau, Chemnitz). Meadows and pas-tures are homogenously distributed acrossthe area with a slightly larger area in theupper middle Erzgebirge.
Table 4.4 shows the main percentagesof the analyzed landscape characteristics.
It can be seen that no major differences insoil (type of soil with information on soildepth, texture, conductivity, etc.), bedrock,groundwater flow and land-use can bedistinguished among the three large sub-catchments. As we can see landscape char-acteristics, such as soil and hydro-geology,do not vary much between the subcatch-ments. Although there are slight differ-ences in the land-use, there is much evi-dence in the literature that during extremeevents the land-use only plays a minor role(e.g. DKKV, 2004). Thus, the dominantinfluence seems to be exerted by precipita-tion and weather characteristics, which isdiscussed in the following two sections.
4.4.5 Relationship between
precipitation AMS and
discharge AMS
AMS of precipitation and discharge weretherefore compared to determine how wellprecipitation and discharge AMS coin-cide. Different precipitation AMS wereextracted from sums of one, two and threedays. A time lag of two days between theprecipitation event and the discharge peakwas allowed. Table 4.5 shows exemplarilyfor four discharge stations the percentagesof agreement for summer and winter sepa-
71
4 Aspects of seasonality and flood generating circulation patterns
Monthly distribution of AMS discharges at Golzern and the assigned circulation pattern
0
500
1000
1500
2000
2500
3000
0 1 2 3 4 5 6 7 8 9 10 11 12 13
Month
Dis
ch
arg
ein
m³/
s
Low British Isles (TB) North-west winds (NWZ) other CP Vb (TM, TRM) Westerly winds (WA-WW)
Figure 4.8: Monthly distribution of AMS discharges at Golzern and the assigned circulation pattern
rately. During the winter, the precipitationevents are not so clearly and directly re-flected in the discharge data (agreement7–26%). One reason for this can be foundin the topography of the catchment.
During the winter time, large amountsof the precipitation can fall as snow in theErzgebirge and the water is stored in thesnowpack. The discharge generation is de-layed until melting starts. Therefore, thetriggering circulation pattern,which mayhave brought a major snow cover, cannotbe directly related to the corresponding dis-charge peak. On the contrary, a direct con-nection between a large summer rain eventand a large discharge can be found in thesummer throughout the catchment (agree-ment 59–89%). Based on these findingsthe question was posed if large summer
flood events can also be related to a spe-cific circulation pattern. This question willbe answered in the following section.
4.4.6 Circulation pattern and
flood generation
First of all, daily information about thedominating European circulation patternbetween 1911 and 2002 were analyzed.For the entire period, westerly winds (WA-WW) cover about 25% of the total cir-culation patterns; high pressure weatherregimes (all circulation patterns beginningwith the letter "H") cover about 27%. Theproportion of the Vb-weather regime (TMand TRM) is relatively low with 6.5%.
The analysis of the discharge AMS atthe gauge Golzern shows that approx. 60%
72
4.4 Results
0
2
4
6
8
10
12
14
16
18
20
22
24
26
WA WZ WS WW SWA SWZ NWA NWZ HM BM TM HNZ HB TRM NEZ HFA SEZ TB TRW
Fre
qu
en
cy
of
cir
cu
lati
on
patt
ern
s(A
MS
1911
-2002)
Winter Summer
Figure 4.9: Histogram of the circulation patterns at the gauge Golzern that generated AMS discharges
between 1911 and 2002 (abbr. see Table 4.2)
occur during the winter time and 40% dur-ing the summer time. Only 19 out ofthe 30 circulation patterns (cf. Table 4.2)play a role in creating AMS dischargesin the Mulde catchment. Thus, 11 outof 30 CPs have not created an AMS dis-charge within the 92 years. In the win-ter (November-April), the cyclonal west-ern and northwestern patterns (WA-WW;NWZ) play the dominant role in floodgeneration, because they account for 84%of the AMS winter discharges and 100%for the floods from November until Febru-ary (see Fig. 4.8). The summer AMSdischarges are generated by several dif-ferent CPs, though mainly by westerlycyclones (WA-WW), north-east cyclones(NEZ) and the troughs over central Europe
(TM, TRM). Figure 4.9 illustrates the dis-tribution separately for summer and winter.
To answer the question, which circu-lation pattern is likely to generate largefloods in the Mulde catchment, the floodpotential was calculated as the probabilityfor a flood quantile HQT , given a certainCP: P(HQT |CPX) = nHQT
nCPXwhere nHQT
is the number of flood events larger thanHQT (e.g. the 10-year flood) that havebeen triggered by a certain circulation pat-tern CPX, whereas nCPX is the numberof days with the corresponding circulationpattern. It is important to note that alreadyfor small return periods (5 years) the Vb-weather regime (TM, TRM) has the high-est flood potential (Fig. 4.10). These cir-culation patterns occur seldom, however
73
4 Aspects of seasonality and flood generating circulation patterns
they are associated with high dischargepeaks. Their flood potential is even morepronounced for floods of larger return pe-riods. Weather patterns, such as the west-erly and north-western cyclones, which areresponsible for most of the winter AMSdischarges, play only an important role forreturn periods of max. 10 years.
There exist also Vb-weather regimesthat generated floods with low return peri-ods at the gauge Golzern. However, theyoften caused high damage in other catch-ments in Europe and had their precipitationcentre outside the Mulde catchment. Thisis for example the case for the flood inApril 1930 in Bavaria, the August 1984flood in Switzerland, and the flood in July1997 in the Odra catchment, when the
Czech Republic and Poland were heavilyaffected (Grünewald et al., 1998; Wasser-wirtschaftsamt Bayreuth, 2006).
Analyses of the other gauge stations aswell as historical records of large floods inthe Mulde catchment show similar resultswith the highest floods being generated byVb-weather regimes. From this analysiswe can conclude that although Vb-weatherpattern do not occur often in the Europeanweather regime they carry a large flood riskin the Mulde catchment.
4.5 Conclusions
Analyses of discharge series, precipitationfields and flood producing atmospheric
0.00
0.02
0.04
0.06
0.08
0.10
0.12
2-year flood 5-year flood 10-year flood 25-year flood 50-year flood 90-year flood 100-year flood
nu
mb
er
of
AM
Sd
isc
ha
rge
trig
ge
red
by
the
CP
/
nu
mb
er
of
da
ys
wit
hth
eC
P
Westerly winds (WA - WW) North-west winds (NWZ) Vb-weather system (TM; TRM) Low British Isles (TB) other CP
Figure 4.10: Flood potential of different circulation patterns to cause a flood of a certain return
period
74
4.5 Conclusions
circulation patterns revealed two govern-ing flood regimes in the Mulde catchmentin south-eastern Germany: (1) frequentfloods during the winter with generally lowreturn periods and (2) less frequent floodsduring the summer, which can reach re-markable flood peaks. Differences in thestatistical parameters of the discharge dataare found in the catchment from west toeast, which are however not reflected inthe landscape characteristics such as soil,elevation or land-use. It is suspected thatthe location and the duration of the precipi-tation field are the most influencing factorsfor the discharge.
The usage of SMS could seem appropri-ate for extreme events in this catchment.However, return periods based on SMSrevealed underestimations of extreme dis-charges up to a return period of 270 years.Estimates for even larger events showed un-derestimations with the AMS. Thus, floodprotection measures for design floods up to250 years based on estimations from AMSare still recommended. From these anal-yses we can conclude that for catchmentswith two or more flood regimes it is not al-ways necessary to separate these from the
AMS given that the extreme events are wellrepresented by the AMS and thus flood pro-tection measures are designed with safetymargins. However, a thorough analysis ofthe flood characteristics of a catchment aswell as flood producing weather regimesis of great importance for reliable floodestimates. In view of the climate changeit is necessary to gain information aboutweather regimes that trigger large floodevents in the region of interest and pos-sible trends of these. With the combinedinformation of catchment characteristics,flood behaviour and weather patterns, theuncertainty in the estimation of extremeevents can be reduced.
Acknowledgements
We thank the GeoForschungsZentrum Pots-dam (GFZ) and the Helmholtz Associationof National Research Centres for their fi-nancial support. The study was part of theHelmholtz Young Scientists Group "Infor-mation and modelling systems for largescale flood situations" at GFZ. We dedi-cate our special thanks to the authoritiesthat provided data.
75
5 Summary and Conclusions
For the period 1951–2002, 145 dis-charge series across Germany were ana-lyzed for flood trends using eight differentflood indicators. These comprised annualand seasonal maximum series (AMAXF,AWMAXF, ASMAXF) as well as peak-over-threshold series of flood magnitude(POTxM) and frequency (POTxF). Season-ally differentiated series were studied tobetter distinguish changes in the flood haz-ard for both seasons (winter and summer).
Data from the gauge Golzern in theMulde catchment were investigated inmore detail for aspects of flood seasonalityand the relationship between CPs and floodpeaks. The analysis of CP data was con-ducted based on daily data from the "Cata-logue of Großwetterlagen in Europe 1881–2004" by Hess and Brezowsky (1952).Four indicators were derived from theseCP data, which capture different aspectsof the frequency and persistence. Theseare the number of days and events a CPwas detected in each year, and the meanand maximum durations of each CP peryear. Trend tests were performed for vary-ing time periods. Finally, CP and flooddata were correlated.
In the introduction, research questionswere posed concerning the existence offlood trends in Germany as well as the in-fluence of atmospheric circulation patternson discharge series. In this Chapter, thesequestions are discussed based on the results
in Chapters 2 to 4. Consequences for floodhazard estimation are outlined by using theexample of flood frequency analysis. Fur-thermore, data constraints, which becameapparent during the thesis, are discussed inthe context of data availability and uncer-tainty. Finally, an outlook indicates furtherresearch directions.
5.1 Main findings
In the following, the research questionswill be answered briefly, highlighting themain findings of the work.
Can significant flood trends be
detected in Germany?
Significant trends were detected at 64% ofthe gauges in at least one of the eight floodindicators. Regional differences emerge:in the Elbe catchment only 50% of the sitesshow trends in at least one flood indicator.In contrast, 60% - 76% of the gauges ofRhine, Weser and Danube show trends in atleast one flood indicator. Trends in at leasttwo indicators were detected at 47% of theDanube sites, 49% of the Rhine gauges and33% of the Weser. In summary, sites in theRhine, Weser and Danube catchments re-vealed more trends than sites in the Elbebasin. Spatial clusters with significant up-ward or downward trends were identified.Partly, they appear to be field significant,
77
5 Summary and Conclusions
i.e. they are representative for the entireregion. Thus, no ubiquitous increase in theflood magnitude and frequency emergesfor the entire country. In the following, thetrend results are separately summarized foreach of the four large river basins.
The Rhine catchment is characterized bymany upward trends. More than 40% ofthe gauges show significant increases inannual maximum series as well as in thefrequency of floods (POT3F, cf. Fig. 2.9).All sites located along the main river ofthe Rhine between Mainz and Düsseldorfshow significant trends in AMAXF. Themajority of increases in the frequency offloods (POT3) and in the annual maximaare attributed to upward trends in winterfloods.
Besides the Rhine basin, the Danuberiver basin is most affected by changes inflood discharge behaviour. In contrast tothe Rhine, the sites in the Danube catch-ment are much more dominated by sum-mer floods. Accordingly, upward trends inannual maxima are mainly attributed to up-ward trends in summer floods. More than1/3 of the series show significant increasesin the annual maxima, and more than 40%of the sites upward trends in the frequencyof floods (POT3F). Gauges along the mainriver of the Danube are especially affectedby these trends: 89% of these gauges hadsignificant increases in POT3F.
The Elbe river basin shows a trend pat-tern that significantly differs from the onesof Rhine and Danube. At only 6% of thesites were significant trends detected inthe annual maximum series. On the con-trary, similar shares (21–25%) of upwardtrends in winter (AWMAXF) and down-ward trends in summer (ASMAXF) weredetected. Increasing trends in the winter
maxima were mostly found in the Saalecatchment, which is the most western sub-catchment of the Elbe river basin. Thissub-catchment shows a trend pattern thatis similar to the one in the neighbouringWeser catchment. Downward trends weredetected at 1/4 of the gauges. Althoughthese trends are field significant, the par-ticular sites where these occur are ratherrandomly distributed in space.
The Weser represents a transitional zonebetween the western and eastern floodregimes. This is visible in two points: (1)upward trends in the annual and the wintermaximum series were detected at 1/3 of thegauges, which are all located in the south-western and southern parts of the catch-ment, and (2) significant downward trendsin summer maximum series were detectedat 1/5 of the sites, which are all located inthe south-eastern and north-eastern partsof the catchment. This clear spatial distinc-tion of trends was therefore also consideredin the determination of the three regions inChapter 3.
In summary, a surprisingly large numberof sites show significant trends in one ormore of the eight flood indicators. Mostchanges were detected for sites in western,southern and central Germany, with partlyspatial clusters. Trends in winter floodsare larger compared to the ones in summerfloods.
Is there a link between trends in
CPs and trends in flood
discharges?
After summarizing the main results onflood trends in the previous section, the re-sults on CP trends are now briefly outlined
78
5.1 Main findings
and the possible link between both aspectsis discussed. In the following, findings forthe three regions, which were defined inChapter 3, are summarized.
Floods in Region West (parts of theRhine, Weser, Ems and Danube catch-ments) are predominantly triggered by thepatterns WZ and NWZ (for abbreviationssee Tab. 3.1). Figure 3.4 illustrates well thedominance of winter floods in the annualmaximum discharges in this region. Oftenthese are triggered by the two patterns WZand NWZ. An increase in the flood hazardin Region West is attributed to significantupward trends in the frequency and persis-tence of both CPs.
Region East (parts of the Elbe and Wesercatchments) is mostly affected by the samedevelopment of these two flood relevantCPs during winter as Region West. There-fore, also Region East is affected by an in-creasing winter flood hazard. Gerstengarbeand Werner (2005) found that the numberof days with precipitation tripled duringwinter and the frequencies and durationof the patterns WZ and NWZ increased(Werner et al., 2008), while easterly CPs,which cause cold and dry winters espe-cially in Region East, decrease at the sametime. Both studies fit the results of thisthesis well. An increase in rain-inducedflood events can therefore be expected, dueto milder winters and an intensified zonalcirculation.
In Region East, summer floods play amore important role than in Region West,through rare but extreme events. However,decreasing trends were detected in summermaximum discharges (cf. Table 3.2). Thesedecreases would be expected to result in adecrease in flood-prone CPs during sum-mer. Interestingly, the most frequent pat-
tern WZ shows an upward trend in the du-ration. This pattern usually does not causelarge floods in the region. In contrast, thepattern TRM, better known to trigger largefloods in the area, does not allow for defi-nite conclusions. For the Mulde catchment,it was possible to show that TRM is ofgreat importance for the summer flood haz-ard. The flood potential was estimated fordifferent return periods at Golzern gauge inorder to capture the entire range from fre-quent to rare flood events. Even for smallreturn periods, the pattern TRM appearedto dominate. Although increases in thenumber of days per year and the mean per-sistence of TRM were visually observedin Chapter 3, these, however, are not sig-nificant at the 10% SL. A clear picture forthe future flood hazard cannot be drawnbased on these findings of field-significantdecreases in discharges on the one hand,and a slight increase in the CP that triggersextreme floods in the area on the other.
Region South is dominated by maxi-mum discharges during summer. An in-creasing flood hazard was found due toincreasing trends in the patterns WZ, SWZ,TRM and TRW, which play an importantrole for AMAXF series in the region (cf.Fig. 3.4). An upward trend in the persis-tence of SWZ during summer may lead toan increase in local thunderstorm-inducedflash floods. This is in agreement withGerstengarbe and Werner (2005), who de-tected a tripled frequency of SWZ duringsummer.
Moreover, a significantly smaller varietyof CPs was detected for the 1990s com-pared to the 1950s (cf. Chapter 3, Fig. 3.5).This is of direct relevance for the floodhazard as fewer patterns with longer per-sistence now dominate the weather across
79
5 Summary and Conclusions
Europe. Longer persistence of CPs maylead to consecutive precipitation events,whereby single events may have ratherlow precipitation amounts. This succes-sion of precipitation may lead to saturatedcatchment conditions. This finding is par-ticularly important for winter floods. Inmany cases these are triggered by WZ orNWZ patterns. Both patterns usually donot favour large storms. However, very wetpre-conditions may cause large runoff co-efficients, which in turn may lead to floodevents.
In summary, a link between CPs and theflood hazard was identified for large partsof Germany. A distinct number of CPsdominate the weather pattern in Germany.A subset of these CP could be shown totrigger large floods. Statistically signifi-cant correlations between composite max-imum discharge series and combinationsof frequent CPs were found for many com-binations and regions, which supports thehypothesis of a supposed link between cir-culation patterns and flood peaks.
5.2 Consequences for the
flood hazard
estimation
For the study in the Mulde catchment, onlydischarge series without a significant trendwere included in the analysis. A compari-son of annual and seasonal maximum se-ries revealed large deviations in the estima-tion of return periods. The use of annualdischarge data provided larger safety mar-gins for flood protection measures of upto 250 years return period than summer se-ries. Around return periods of 250 years a
change occurred leading to a larger safetymargin for more extreme events when us-ing summer maximum series. The use ofboth seasonal and annual maximum seriesis recommended for flood hazard assess-ment and consequently, for flood designmeasures against the background of a pos-sible poor reflection of the seasonality inannual maxima. This was shown for in-stance for the Weser catchment (cf. Chap-ter 2).
The results of Chapter 2 show that theassumption of stationarity for the purposeof flood estimation cannot be held anymore for 64% of the gauges, due to signif-icant trends in at least one flood indicator.An instationary flood frequency analysisthat considers trends is therefore recom-mended for flood estimation purposes forthese catchments (see also Zhang et al.,2004; Coles, 2001).
In a complementary study, trend resultsbased on the Mann-Kendall test (cf. Chap-ter 2) were compared with trend resultsestimated with an instationary GEV model(Petrow et al., 2008). The study revealed asimilar trend pattern for both approaches(MK and GEV) at the large scale. How-ever, flood protection measures are usu-ally performed on the local scale, whichnecessitates the examination of trends forthe area of interest. The Rockenau gauge,which is situated along the Neckar river(Rhine basin), serves as an example. Fig-ure 5.1 shows the annual maximum seriesof this Rockenau gauge (1951–2002) withthe fitted trend line of the instationary GEVmodel (top diagram). The trend is signif-icant at the 10% significance level. Thelower part of Fig. 5.1 presents differentflood frequency functions for the stationaryand instationary GEV models. Contrary to
80
5.3 Data constraints
the stationary model, the flood frequencychanges with time when using the instation-ary model. Both curves from the instation-ary model show larger discharge amountsfor a flood with 100 years return periodcompared to the stationary one. Whencomparing both curves of the instationarymodel, the increase in the flood hazard be-comes even more evident due to the largerdischarge estimate for 2020 compared to2002.
Starting from this gauge-specific analy-sis, stationary and instationary approacheswere also applied to all 145 discharge se-ries (annual and seasonal maximum series)studied in Chapter 2. The stationary GEVmodel reveals lower discharge estimatesfor a flood of 100 years return period forat least 75% of the gauges compared to theinstationary model. The results for winterand annual series show significant trendsat many sites. Summer series do not showclear tendencies.
These findings underline again the im-portance of incorporating different ap-proaches into the flood frequency estima-tion. These could comprise the use of in-stationary models for the estimation of rareevents, but also the use of model chains,which enables the consideration of the sev-eral aspects of the flood risk. However, thelatter approach goes beyond the scope ofthe thesis and has to be accomplished inanother study.
5.3 Data constraints
In this section, some of the main contraintsthat arose during data analysis, are dis-cussed. The first part covers aspects ofthe uncertainty caused by different lengths
of the time series and the grade of accuracyof the investigated discharge data. The sec-ond part focuses on aspects related to theflood triggering CPs, namely the assign-ment of the flood triggering CP as wellas the uncertainty in the results due to theinfluence of snow.
5.3.1 Flood data
Inherently, data on extreme flood eventscontain a large amount of uncertainty.When regular flood measurements began,estimates of the water level were obtainedduring a flood event and later convertedinto discharge estimates. Nowadays, sev-eral methods for discharge recordings areavailable, which yield more accurate re-sults for mean discharges. Nevertheless,the accurate estimation of flood peaks hasnot yet been achieved during large floodevents such as one in 2002 in Saxony or theones in 2005 in Switzerland and Austria.
For the estimation of flood design mea-sures, preferably long time series areneeded. A careful analysis of the data pro-vided is needed, if flood indicators suchas annual maximum series are to be in-vestigated. For this study, discharge datawith varying information were obtainedfrom the water authorities: daily mean dis-charges from all stations were provided.Some water authorities provided additionalinformation such as daily maximum dis-charges or monthly maximum discharges.It could be shown for the Mulde catchmentthat deviations of up to 28% occur betweenthe daily mean and maximum discharges.Table 5.1 presents such deviations for se-lected flood events measured at the Golz-ern 1 gauge (Mulde catchment).
These deviations have an influence on
81
5 Summary and Conclusions
1950 1955 1960 1965 1970 1975 1980 1985 1990 1995 20000
500
1000
1500
2000
2500
Year
Dis
ch
arg
e (
m3/s
)
Trend of location parameter of instationary GEV
Annual maximum discharge series
0 500 1000 1500 2000 2500 3000
101
102
Discharge [m3/s]
Retu
rn p
eri
od
[Y
ear]
instationary model, 2002
instationary model, 2020
stationary model
Figure 5.1: Annual maximum discharge series of gauge Rockenau/Neckar catchment (top) and flood
frequency functions for the stationary and instationary GEV model (bottom)
82
5.3 Data constraints
date of daily mean daily maximum deviationflood discharge [m3/s] discharge [m3/s] in %11 July 1954 1310 1740 2531 July 1955 523 577 106 July 1958 875 1090 2018 October 1960 544 615 129 December 1974 970 1020 513 August 2002 1880 2600 28
Table 5.1: Deviation of mean and maximum daily discharges for selected flood events at gauge
Golzern (Mulde catchment)
the results of flood data analysis. There-fore, the most accurate data, representingthe flood peaks, were used. In Chapter 4,flood data based on daily maximum dis-charges were studied. This was particularlyimportant for the flood frequency analysissince the magnitude of floods determinesthe discharge estimated for the consideredreturn period. In Chapters 2 and 3 analyseswere performed with daily mean dischargedata, since not all water authorities pro-vided daily maximum data. This may havea slight influence on the trend results offlood magnitude indicators such as maxi-mum series or POTM series.
5.3.2 Data on circulation
patterns
Data on the dominant atmospheric cir-culation pattern over Europe were avail-able for every day of the studied period.The coarse scale of information causessome uncertainty associated with the data.For each day, only one CP over Europeis determined. Atmospheric features ofsmaller scale, such as shortwave troughsand ridges, are not included. Therefore, insome occasions, a pattern different fromthe dominant one may cause a local flood
event. For example, a local flash floodin southern Germany might be caused bya south-westerly flow component, whilethe dominant pattern over Europe is de-termined to be WZ. Since more detailedinformation was not available, it is difficultto quantify the associated error.
The assignment of the triggering CPswas an important procedure of the study,which however bears much uncertainty.For the assessment of a possible link be-tween CPs and flood peaks, the flood trig-gering CP was automatically assigned tothe flood data depending on the size of therespective catchment. The concentrationon meso-scale catchments was attributedto the complex assignment of the flood trig-gering circulation patterns in large catch-ments with different sub-catchments. Forexample, flood peaks at the Cologne gaugeat the Rhine River can have its source indifferent upstream regions. These are, forinstance, the Upper Rhine in Switzerland,or one or more sub-catchments like Mosel,Neckar or Main. Depending on the loca-tion of the flood triggering precipitationfield and the distance to the Cologne gauge,a time lag of 2 to 6 days would have to beassigned. Since the meso-scale catchmentsanalyzed here cover most of the country, a
83
5 Summary and Conclusions
model for large nested catchments was notdeveloped.
Another issue in the assignment of a trig-gering CP is seasonality. Winter precipi-tation is often accumulated as snow forseveral days or weeks. This applies toa large number of German catchments inthe middle mountain ranges. During thesnow melting season it is difficult to es-timate which of the past circulation pat-terns can be designated as the one thattriggers a spring flood event. It is ques-tionable whether these events should beinvestigated separatly. Some studies likeDuckstein et al. (1993) or Bárdossy andFiliz (2005) have chosen catchments thatare not affected by snow. However, mostparts of Germany are dominated by winterflood events, which are partly influencedby snow melting. Therefore, it was notreasonable to exclude winter data from theanalysis. The assessment of the uncertaintyassociated with the assignment of CPs dur-ing winter would require more informationof the affect of the snow on each catch-ment. A possible solution is to performa flood typology according to Merz andBlöschl (2003) to separate catchments withand without snow influence. Thereafter, arevised assignment of the flood triggeringCP to annual and winter discharge max-ima can be performed, which enables thequantification of the uncertainty. Unfortu-nately, the time frame of the project did notallow conducting these analyses during theproject.
Although there is a large amount of un-certainty associated with the CP data, theresults presented in Chapters 3 and 4 arein lines with those presented in Kästner(1997); Pfister et al. (2004a); Grünewald(2006); KLIWA (2007).
5.4 Influence of the
selected time period
The time period used in this study is animportant factor influencing the results oftrend detection studies. Chapters 2 and 3examine a time period (1951–2002) thatcovers significant changes at many gauges.Several studies have already documentedchanges of hydrological variables duringthe 1970s (e.g. Franks, 2002; McCabe andWolock, 2002). Moreover, it is a period inwhich there was a significant increase ingreenhouse gases (Solomon et al., 2007).
It was a declared goal of the study tohave a good temporal and spatial cover-age with nearly no gaps in the dischargedata. Consequently, a compromise be-tween time length and data availability hadto be found. Currently, discharge data rep-resenting more than 1000 sites are avail-able in Germany. However, A large num-ber of gauge data could not be includedin the study due to gaps of more thanone year or short time lengths. Especiallyeastern Germany is poorly represented.For instance, 29 meso-scale catchmentsin the Elbe basin did not fit the criteria ofour study due to short time series lengths.Thus, only 32 out of 61 meso-scale catch-ments in the area could be investigated. Inthe end, 145 gauge stations with data be-tween 1951 and 2002 provided us with agood spatial and temporal coverage withthe exception of eastern Germany. Withthe help of field significance tests, conclu-sions for entire regions could be drawn (fordetails see Chapter 2). In the study pre-sented in Chapter 4 (Mulde catchment), allavailable data were used without defininga common period for all gauges. Conse-
84
5.5 Future research prospects
quently, discharge series with lengths of 43to 91 years were investigated.
When focusing only on the Mulde catch-ment (Chapter 4), the comparison of the re-sults gained in Chapters 3 and 4 reveals aninteresting point. Although seven gaugeswere analyzed in both studies, the preva-lent patterns vary for the different time pe-riods. Especially results of very long seriesof up to 91 years are not directly compa-rable to the study covering 52 years. Thisis the case, for instance, with the prevail-ing weather pattern at the Golzern gauge.For the period of 1951–2002 the most fre-quent CP associated with the annual max-imum series was found to be NWZ (cy-clonic north-western pattern), whereas dur-ing the period 1911–2002 it was the WZpattern (cyclonic western pattern). How-ever, the frequency of both patterns wasalmost equal (12 WZ vs. 11 NWZ). De-spite these deviations, it is important tonote that the main findings of both studiesin Chapters 3 and 4 for this region coincidewell.
5.5 Future research
prospects
Despite the findings summarized above, anumber of research directions remain forthe future. One research focus should beset on the analysis of catchments smallerthan 500 km2. These catchments are of-ten heavily influenced by flood protectionmeasures and land-use changes. Thus, itwould be of particular interest to performanalogue flood trend analyses with dis-charge series of these catchments. Withsuch results, a validation of the concep-
tional model by Blöschl et al. (2007) wouldthen be possible. According to Blöschlet al. (2007) the impact of land-use is afunction of the catchment size. In contrast,the impact of climate variability on floodsdoes not change with catchment size.
Another aspect of interest regardingsmall catchments would be the analysisof flood-triggering CPs, since small moun-tainous catchments are mainly affectedby flash floods, which often occur duringspring and summer. A shift in the im-portance and frequency of flood relevantCPs is to be expected, especially in pre-alpine catchments. For instance, the south-westerly pattern SWZ is known for caus-ing intense thunderstorms during summerin southern and central Germany, regularlytriggering local floods. These may not bedetectable on the meso-scale.
The coupling of flood data with NAO orother atmospheric pressure indices shouldalso be applied in future research. TheNAO index represents the difference innormalized sea-level pressure anomaliesbetween the Azores and Iceland and is esti-mated through automated procedures. Sim-ilar to Bouwer et al. (2008), a compari-son of different atmospheric indicators andtheir reflection in the flood data could beperformed.
The findings of this PhD thesis empha-size again the importance of incorporatinginstationary approaches into flood hazardestimation. The use of these models for theestimation of rare events has to be imple-mented on a large scale, but complementedby the use of model chains. In that way,a more holistic picture of the flood hazardis gained, which enables more appropriateand precise flood management measures.
85
Bibliography
Adamowski, K. and Bocci, C. (2001). Geostatistical regional trend detection in river flowdata. Hydrological Processes, 15, 3331–3341.
Apel, H., Merz, B., and Thieken, A. (in press). Influence of dike breaches on flood frequencyestimation. Computers & Geosciences.
Bárdossy, A. and Caspary, H. (1990). Detection of Climate Change in Europe by AnalyzingEuropean Atmospheric Circulation Patterns from 1881 to 1989. Theoretical and AppliedClimatology, 42, 155–167.
Bárdossy, A. and Filiz, F. (2005). Identification of flood producing atmospheric circulationpatterns. Journal of Hydrology, 313, 48–57.
Bárdossy, A. and Pakosch, P. (2005). Wahrscheinlichkeiten extremer Hochwasser untersich ändernden Klimaverhältnissen. Wasserwirtschaft, 7-8, 58–62.
Belz, J., Braehmer, G., Buitefeld, H., Engel, H., Grabher, R., Hodel, H., Krahe, P., Lam-mersen, R., Larina, M., Mendel, H.G., Meuser, A., Müller, G., Plonka, B., Pfister, L., andvan Vuuren, W. (2007). Das Abflussregime des Rheins und seiner Nebenflüsse im 20.Jahrhundert - Analyse, Veränderungen, Trends. Technical report, Kommission für dieHydrologie des Rheingebietes (KHR), Koblenz and Lelystad.
Belz, J., Engel, H., and Weber, H. (2004). Die Donau und ihr Einzugsgebiet. Hydrologieund Wasserbewirtschaftung, 48 (2), 80–84.
Bendix, J. (1997). Natürliche und anthropogene Einflüsse auf den Hochwasserabfluss desRheins. Erdkunde, 51, 292–308.
Benito, G., Lang, M., Barriendos, M., Llasat, C., Francés, F., Ouarda, T., Thorndycraft, V.,Enzel, Y., Bárdossy, A., Coeur, D., and Bobée, B. (2004). Use of systematic, palaeofloodand historical data for the improvement of flood risk estimation - Review of Scientificmethods. Natural Hazards, 31, 623–643.
Beurton, S. and Thieken, A. (2009). Seasonality of floods in Germany. HydrologicalSciences Journal, 54 (1), 62–76.
Blöschl, G., Ardoin-Bardin, S., Bonell, M., Dorninger, M., Goodrich, D., Gutknecht, D.,Matamoros, D., Merz, B., Shand, P., and Szolgay, J. (2007). At what scales do climatevariability and land cover change impact on flooding and low flows? HydrologicalProcesses, 21 (9), 1241–1247. Invited Commentary.
87
Bibliography
Bouwer, L., Vermaat, J., and Aerts, J. (2006). Winter atmospheric circulation and riverdischarge in northwest Europe. Geophysical Research Letters, 33, L06403.
Bouwer, L., Vermaat, J., and Aerts, J. (2008). Regional sensitivities of mean and peakriver discharge to climate variability in Europe. Journal of Geophysical Research, 113,D19103.
Bronstert, A., Niehoff, D., and Bürger, G. (2002). Effects of climate and land-use changeon storm runoff generation: present knowledge and modelling capabilities. Journal ofHydrology, 16 (2), 509–529.
Buishand, T. and Brandsma, T. (1997). Comparison of circulation classification schemesfor predicting temperature and precipitation in th e Netherlands. International Journal ofClimatology, 17, 875–889.
Burn, D. and Hag Elnur, M. (2002). Detection of hydrologic trends and variability. Journalof Hydrology, 255, 107–122.
Caspary, H. (1995). Recent winter floods in Germany caused by changes in the atmosphericcirculation across Europe. Physics and Chemistry of the Earth, 20, 459–462.
Caspary, H. and Bárdossy, A. (1995). Markieren die Winterhochwasser 1990 und 1993das Ende der Stationarität in der Hochwasserhydrologie infolge von Klimaänderungen?Wasser und Boden, 47 (3), 18–24.
Casty, C., Wanner, H., Luterbacher, J., Esper, J., and Böhm, R. (2005). Temperature andPrecipitation Variability in the European Alps since 1500. International Journal ofClimatology, 25, 1855–1880.
Chen, H., Guo, S., Xu, C., and Singh, V. (2007). Historical temporal trends of hydro-climatic variables and runoff response to climate variability and their relevance in waterresources management in the Hanjiang basin. Journal of Hydrology, 344, 171–184.
Coles, S. (2001). An introduction to Statistical Modeling of Extreme Values. Springer.
Disse, M. and Engel, H. (2001). Flood Events in the Rhine Basin: Genesis, Influences andMitigation. Natural Hazards, 23 (2–3), 271–290.
DKKV (2003). Hochwasservorsorge in Deutschland. Lernen aus der Katastrophe 2002 imElbegebiet, Schriftenreihe des DKKV, volume 29. Deutsches Komitee für Katastrophen-vorsorge e.V., Bonn. 152 pp.
DKKV (2004). Flood Risk Reduction in Germany - Lessons Learned from the 2002Disaster in the Elbe Region, DKKV Publications, volume 29e. Deutsches Komitee fürKatastrophenvorsorge e.V., Bonn. 40 pp.
88
Bibliography
Douglas, E. and Vogel, R. (2006). Probabilistic behavior of floods of record in the UnitedStates. Journal of Hydrologic Engineering, 11 (5), 482–488.
Douglas, E., Vogel, R., and Kroll, C. (2000). Trends in floods and low flows in the UnitedStates: impact of spatial correlation. Journal of Hydrology, 240, 90–105.
Duckstein, L., Bárdossy, A., and Bogárdi, I. (1993). Linkage between the occurrence ofdaily atmospheric circulation patterns and floods: an Arizona case study. Journal ofHydrology, 143, 413–428.
Feidas, H., Noulopoulou, C., Makrogiannis, T., and Bora-Senta, E. (2007). Trend analysisof precipitation time series in Greece and their relationship with circulation using surfaceand satellite data: 1955-2001. Theoretical and Applied Climatology, 87, 155–177.
Francés, F., Glade, T., and P., A. (editors) (2001). The Use of Historical Data in Natu-ral Hazards Assessments, chapter Incorporating non-systematic information to floodfrequency analysis using the maximum likelihood method. Kluwer Academic Publishers.
Franks, S. (2002). Identification of a change in climate state using regional flood data.Hydrology and Earth System Sciences, 6 (1), 11–16.
Frei, C., Davies, H., Gurtz, J., and Schär, C. (2000). Climate dynamics and extremeprecipitation and flood events in Central Europe. Integrated Assessment, 1, 281–299.
Gerstengarbe, F.W. and Werner, P. (2005). Katalog der Großwetterlagen Europas (1881-2004) nach Paul Hess und Helmut Brezowsky. PIK-Report 100, Potsdam Institute forClimate Impact Research. 6th revised edition.
Glaser, R. and Stangl, H. (2004). Climate and Floods in Central Europe since AD 1000:Data, methods, results and consequences. Surveys in Geophysics, 25, 485–510.
Grünewald, U. (2006). Extreme hydro(meteoro-)logical Events in the Elbe Basin. Österre-ichische Wasser- und Abfallwirtschaft, 58 (3-4), 27–34.
Grünewald, U., Kaltofen, M., Rolland, W., Schümberg, S., Chmielewski, R., Ahlheim,M., Sauer, T., Wagner, R., Schluchter, W., Birkner, H., Petzold, R., Radczuk, L.,Eliasiewicz, R., Paus, L., and Zahn, G. (1998). Ursachen, Verlauf und Folgen desSommer-Hochwassers 1997 an der Oder sowie Aussagen zu bestehenden Risikopotentia-len. Technical report, DKKV.
Harlin, J. and Kung, S. (1992). Parameter uncertainty and simulation of design floods inSweden. Journal of Hydrology, 137, 209–230.
Helms, M., Büchele, B., Merkel, U., and Ihringer, J. (2002). Statistical analysis of the floodsituation and assessment of the impact of diking measures along the Elbe (Labe) river.Journal of Hydrology, 267, 94–114.
89
Bibliography
Hennegriff, W., Kolokotronis, V., Weber, H., and Bartels, H. (2006). Klimawandel undHochwasser - Erkenntnisse und Anpassungsstrategien beim Hochwasserschutz. Korre-spondenz Abwasser und Abfall, 53 (8), 770–779.
Hess, P. and Brezowsky, H. (1952). Katalog der Großwetterlagen Europas. DeutscherWetterdienst in der US-Zone 33.
Hirschboeck, K. (1988). Flood Geomorphology, chapter Flood hydroclimatology, pp.27–49. Wiley New York.
Hosking, J. and Wallis, J. (1997). Regional frequency analysis. Cambridge University Press,New York.
Hundecha, Y. and Bárdossy, A. (2005). Trends in daily precipitation and temperatureextremes across Western Germany in the second half of the 20th century. InternationalJournal of Climatology, 25, 1189–1202.
Ihringer, J. (1996). Hochwasser aus ländlichen und städtischen Gebieten. Geowisschen-schaften, 14 (12), 223–530.
Institute of Hydrology (1999). Flood Estimation Handbook, volume 1-5. CrowmarshGifford, Wallingford.
Jacobeit, J., Philipp, A., and Nonnenmacher, M. (2006). Atmospheric circulation dynamicslinked with prominent discharge events in Central Europe. Hydrological Sciences Journal,51 (5), 946–965.
Jain, S. and Lall, U. (2000). Magnitude and timing of annual maximum floods: Trends andlarge-scale climatic associations for the Blacksmith Fork River, Utah. Water ResourcesResearch, 36 (12), 3641–3651.
James, P. (2007). An objective classification method for Hess and Brezowsky Grosswetter-lagen over Europe. Theoretical Applied Climatology, 88, 17–42.
Johnson, S. and Stefan, H. (2006). Indicators of climate warming in Minnesota: Lake icecovers and snowmelt runoff. Climatic Change, 75, 421–453.
Jones, P., Hulme, M., and Briffa, K. (1993). A comparison of Lamb circulation types withan objective classification scheme. International Journal of Climatology, 13, 655–663.
Kästner, W. (1997). How do Großwetterlagen become flood-producing weather situationsin Bavaria? DGM, 41 (3), 107–112.
Kendall, M. (1975). Rank Correlation Measures. Charles Griffin, London.
90
Bibliography
Khaliq, M., Ouarda, T., Ondo, J.C., Gachon, P., and Bobée, B. (2006). Frequency analysisof a sequence of dependent and/or non-stationary hydro-meteorological observations: Areview. Journal of Hydrology, 329, 534–552.
Kingston, D., Lawler, D., and McGregor, G. (2006). Linkage between atmospheric circula-tion, climate and streamflow in the northern North Atlantic: research prospects. Progressin Physical Geography, 30 (2), 143–174.
Klemes, V. (1993). Probability of extreme hydrometeorological events - a different approach.In Z.W. Kundzewicz, D. Rosbjerg, S.P. Simonovic, and K. Takeuchi (editors), Extremehydrological events: precipitation, floods and droughts, IAHS Publications, volume 213,pp. 167–176.
KLIWA (2007). 3. KLIWA-Symposium. In Klimaveränderung und Konsequenzen für dieWasserwirtschaft, KLIWA Reports, volume 10. Arbeitskreis KLIWA, Stuttgart.
Kundzewicz, Z. (2008). River floods in the changing climate - Observations and projections.Proceedings of the 4th International Symposium on Flood Defence, 6–8 May 2008,Toronto, Canada.
Kundzewicz, Z., Graczyk, D., Maurer, T., Pinskwar, I., Radziejewski, M., Svensson, C.,and Szwed, M. (2005). Trend detection in river flow series: 1. Annual maximum flow.Hydrological Sciences Journal, 50 (5), 797–810.
Kundzewicz, Z. and Robson, A. (2004). Change detection in hydrological records - a reviewof the methodology. Hydrological Sciences Journal, 49 (1), 7–19.
Kunkel, K., Andsager, K., and Easterling, D. (1999). Long-Term Trends in ExtremePrecipitation Events over the Conterminous United States and Canada. Journal ofClimate, 12, 2515–2527.
Lamb, H. (1972). British Isles weather types and a register of the daily sequence ofcirculation patterns 1861-1971. Geophysical Memoir 116, HMSO, London.
Lammersen, R., Engel, H., van de Langemheen, W., and Buiteveld, H. (2002). Impactof river training and retention measures on flood peaks along the Rhine. Journal ofHydrology, 267, 115–124.
Lindström, G. and Bergström, S. (2004). Runoff trends in Sweden 1807-2002. HydrologicalSciences Journal, 49 (1), 69–83.
Llasat, C., Barriendos, M., Barrera, A., and Rigo, T. (2005). Floods in Catalonia (NESpain) since 14th century. Climatological and meteorologiscal aspects from historicaldocumentary sources and old instrumental records. Journal of Hydrology, 313 (1-2),32–47.
91
Bibliography
McCabe, G. and Wolock, D. (2002). A step increase in streamflow in the conterminousUnited States. Geophysical Research Letters, 29 (24), 2185.
McKerchar, A. and Henderson, R. (2003). Shifts in flood and low-flow regimes in NewZealand due to interdecadal climate variations. Hydrological Sciences Journal, 48 (4),637–654.
McNeil, V. and Cox, M. (2007). Defining the climatic signal in stream salinity trends usingthe Interdecadal Pacific Oscillation and its rate of change. Hydrology and Earth SystemSciences, 11, 1295–1307.
Merz, B. (2006). Hochwasserrisiken: Grenzen und Möglichkeiten der Risikoabschätzung. E.Schweizerbart’sche Verlagsbuchhandlung (Nägele und Obermiller), Stuttgart. Habilita-tionsschrift.
Merz, B. and Thieken, A. (2005). Separating natural and epistemic uncertainty in floodfrequency analysis. Journal of Hydrology, 309, 114–132.
Merz, R. and Blöschl, G. (2003). A process typology of regional floods. Water ResourcesResearch, 39 (12), 1340.
Merz, R. and Blöschl, G. (2009). Process controls on the statistical flood moments - a databased analysis. Hydrological Processes, 23, 675–696.
Merz, R., Blöschl, G., and Parajka, J. (2006). Raum-zeitliche Variabilität von Ereignisab-flussbeiwerten in Österreich. Hydrologie und Wasserbewirtschaftung, 50 (1), 2–11.
Michaud, J., Hirschboeck, K., and Winchell, M. (2001). Regional variations in small-basinfloods in the United States. Water Resources Research, 37 (5), 1405–1416.
Mudelsee, M., Börngen, M., Tetzlaff, G., and Grünewald, U. (2004). Extreme floods incentral Europe over the past 500 years: Role of cyclone pathway Zugstrasse Vb. Journalof Geophysical Research, 109, 1–21.
Mudelsee, M., Deutsch, M., Börngen, M., and Tetzlaff, G. (2006). Trends in flood riskof the River Werra (Germany) over the past 500 years. Hydrological Sciences Journal,51 (5), 818–833.
Onoz, B. and Bayazit, M. (1995). Best-fit distributions of largest available flood samples.Journal of Hydrology, 167, 195–208.
Ouarda, T., Cunderlik, J., St-Hilaire, A., Barbet, M., Bruneau, P., and Bobée, B. (2006).Data-based comparison of seasonality-based regional flood frequency methods. Journalof Hydrology, 330, 329–339.
92
Bibliography
Pauling, A. and Paeth, H. (2007). On the variability of return periods of European winterprecipitation extremes over the last three centuries. Climate of the Past, 3, 65–76.
Pearson, C. (1991). New Zealand Regional flood frequency analysis using L-Moments.Journal of Hydrology, 30, 53–64.
Petrow, T., Delgado, J., and Merz, B. (2008). Trends der Hochwassergefährdung inDeutschland (1951–2002) und Konsequenzen für die Bemessung. Wasserwirtschaft, 11,24–28.
Petrow, T. and Merz, B. (2009). Trends in flood magnitude, frequency and seasonality inGermany in the period 1951–2002. Journal of Hydrology, 371, 129–141.
Petrow, T., Merz, B., Lindenschmidt, K.E., and Thieken, A. (2007). Aspects of seasonalityand flood generating circulation patterns in a mountainous catchment in south-easternGermany. Hydrology and Earth System Sciences, 11 (4), 1455–1468.
Pfister, C. (1999). Wetternachhersage - 500 Jahre Klimavariationen und Naturkatastrophen.Paul Haupt Verlag Bern.
Pfister, L., Drogue, G., El Idrissi, A., Iffly, J.F., Poirier, C., and Hoffmann, L. (2004). Spatialvariability of trends in the rainfall-runoff relationship: A mesoscale study in the MoselBasin. Climatic Change, 66, 67–87.
Pfister, L., Kwadijk, J., Musy, A., Bronstert, A., and Hoffmann, L. (2004a). Climate change,land use change and runoff prediction in the Rhine-Meuse Basins. River Research andApplications, 20, 229–241.
Pinter, N., van der Ploeg, R., Schweigert, P., and Hoefer, G. (2006). Flood magnification onthe River Rhine. Hydrological Processes, 20 (1), 147–164.
Pohl, R. (2004). Historische Hochwasser aus dem Erzgebirge. Institut für Wasserbau undTechnische Hydromechanik TU Dresden, Wasserbauliche Mitteilungen 28.
Rapp, J. and Schönwiese, C.D. (1996). Atlas der Niederschlags- und Temperaturtrends inDeutschland 1891-1990. Universität Frankfurt.
Robinson, M., Cognard-Plancq, A.L., Cosandey, C., David, J., Durand, P., Führer, H.W.,Hall, R., Hendriques, M., Marc, V., McCarthy, R., McDonnell, M., Martin, C., Nisbet,T., O’Dea, P., Rodgers, M., and Zollner, A. (2003). Studies of the impact of forestry onextreme flows: a European perspective. Forest Ecology and Management, 186, 85–97.
Robson, A. (2002). Evidence for trends in UK flooding. Philosophical Transactions:Mathematical, Physical and Engineering Sciences, 360, 1327–1343.
93
Bibliography
Robson, A., Jones, T., Reed, D., and Bayliss, A. (1998). A study of national trend andvariation in UK floods. International Journal of Climatology, 18, 165–182.
Santos, J., Corte-Real, J., Ulbrich, U., and Palutikof, J. (2007). European winter precipitationextremes and large-scale circulations: a coupled model and its scenario. Theoretical andApplied Climatology, 87, 85–102.
Sen, P. (1968). Estimates of the regression coefficient based on Kendall’s tau. Journal ofAmerican Statistical Association, 63, 1379–1389.
Sivapalan, M., Blöschl, G., Merz, R., and Gutknecht, D. (2005). Linking flood frequency tolong-term water balance: Incorporating effects of seasonality. Water Resources Research,41, W06012.
Solomon, S., Qin, D., Manning, M., Chen, Z., Marquis, M., Averyt, K., Tignor, M., andMiller, H. (2007). IPCC, 2007: Climate Change 2007: The Physical Science Basis. Con-tribution of Working Group I to the Fourth Assessment Report of the IntergovernmentalPanel on Climate Change.
St. George, S. (2007). Streamflow in the Winnipeg River basin, Canada: Trends, extremesand climate linkages. Journal of Hydrology, 332, 396–411.
Staatliches Amt für Umwelt und Arbeitsschutz (2005). Hochwasser-Aktionsplan Weser.Technical report, Staatliches Amt für Umwelt und Arbeitsschutz Nordrhein-Westfalen.
Stedinger, J. (1983). Estimating a regional flood frequency distribution. Water ResourcesResearch, 19 (2), 503–510.
Stedinger, J. (2000). Inland Flood Hazards - Human, Riparian and aquatic communities,volume 12, chapter Flood frequency analysis and statistical estimation of flood risk.Cambridge University Press, Cambridge.
Steinbrich, A., Uhlenbrook, S., Reich, T., and Kolokotronis, V. (2005). Raum-zeitliche Zusammenhänge zwischen Großwetterlagen und Starkniederschlägen in Baden-Württemberg. Wasserwirtschaft, 11, 14–19.
Strupczewski, W., Singh, V., and Feluch, W. (2001a). Non-stationary approach to at-siteflood frequency modelling. I. Maximum likelihood estimation. Journal of Hydrology,248, 123–142.
Strupczewski, W., Singh, V., and Mitosek, H. (2001b). Non-stationary approach to at-siteflood frequency modelling. III. Flood analysis of Polish rivers. Journal of Hydrology,248 (1-4), 152–167.
94
Bibliography
Sturm, K., Glaser, R., Jacobeit, J., Deutsch, M., Brázdil, R., Pfister, C., Luterbacher, J.,and Wanner, H. (2001). Hochwasser in Mitteleuropa seit 1500 und ihre Beziehung zuratmosphärischen Zirkulation. Petermanns Geographische Mitteilungen, 145, 14–23.
Svensson, C., Hannaford, J., Kundzewicz, Z., and Marsh, T. (2006). Trends in river flows:why is there no clear signal in observations? IAHS Publications, 305, 1–18.
Svensson, C., Kundzewicz, Z., and Maurer, T. (2005). Trend detection in river flow series:2. Flood and low-flow index series. Hydrological Sciences Journal, 50 (5), 811–824.
Thieken, A., Petrow, T., Kreibich, H., and Merz, B. (2006). Insurability and mitigation offlood losses in private households in Germany. Risk Analysis, 26 (2), 383–395.
UBA (2006). Anpassung an Klimaänderungen in Deutschland - Regionale Szenarien undnationale Aufgaben.
Ulbrich, U., Brücher, T., Fink, A., Leckebusch, G., Krüger, A., and Pinto, J. (2003). Thecentral European floods of August 2002: Part 2 - Synoptic causes and considerationswith respect to climatic change. Weather, 58, 434–443.
van Bebber, W. (1891). Die Zugstraßen der barometrischen Minima. MeteorologischeZeitschrift, 8, 361–366.
Vogel, R. and Wilson, I. (1996). Probability distribution of annual maximum, mean, andminimum streamflows in the United States. Journal of Hydrologic Engineering, 1 (2),69–76.
von Storch, H. and Navarra, A. (editors) (1995). Analysis of Climate Variability - Applica-tions of Statistical Techniques. Springer, New York.
Wasserwirtschaftsamt Bayreuth (2006). http://www.wwa-bt.bayern.de/gewaesser/wasserbau/hochwasserschutz/hwku.htm.
Webb, R. and Betancourt, J. (1992). Climate variability and flood frequency of the SantaCruz River, Pima County, Arizona. U.S. Geological Survey Water Supply Paper, p. 2379.
Weikinn, C. (1958). Quellentexte zur Witterungsgeschichte Europas von der Zeitwende biszum Jahr 1850. Hydrographie Teil 1 (Zeitwende-1500). Akademie Verlag Berlin.
Werner, P., Gerstengarbe, F.W., and Wechsung, F. (2008). Großwetterlagen and precipitationtrends in the Elbe river catchment. Meteorologische Zeitschrift, 17 (1), 61–66.
Yue, S., Pilon, P., and Cavadias, G. (2002b). Power of the Mann-Kendall Spearman’s rhotests for detecting monotonic trends in hydrological series. Journal of Hydrology, 259,254–271.
95
Bibliography
Yue, S., Pilon, P., and Phinney, B. (2003). Canadian streamflow trend detection: impacts ofserial and cross-correlation. Hydrological Sciences Journal, 48 (1), 51–63.
Yue, S., Pilon, P., Phinney, B., and Cavadias, G. (2002a). The influence of autocorrelationon the ability to detect trend in hydrological time series. Hydrological Processes, 16,1807–1829.
Zhang, X., Zwiers, F., and Li, G. (2004). Monte Carlo Experiments on the Detection ofTrends in Extreme Values. Journal of Climate, 17, 1945–1952.
96
Appendix
Overview of used discharge data and the providing water authority.
Gauge River basin Basin area Period of Water authority[km2] measurements
Achleiten Donau 76653 1900–2002 WSA RegensburgAffoldern Weser 1452 1940–2004 BfG KoblenzAken Elbe 69849 1935–2003 BfG KoblenzAllendorf Weser 5166 1941–2004 BfG KoblenzAltena Rhein 1190 1950–2005 LUA EssenAndernach Rhein 139549 1930–2006 WSD Südwest (Mainz)Aue 1 Elbe 362 1928–2002 StuFa PlauenBad Düben Elbe 6171 1961–2002 LfUG DresdenBad Kissingen Rhein 1587 1929–2003 LFW MünchenBad Mergentheim Rhein 1018 1929–2004 LFU KarlsruheBarby Elbe 94060 1899–2003 BfG KoblenzBerthelsdorf Elbe 244 1936–2002 StuFa ChemnitzBeuerberg Donau 954 1950–2002 LFW MünchenBirnbach Donau 865 1930–2003 LFW MünchenBodenwerder Weser 15924 1940–2005 WSD MitteBorstendorf Elbe 644 1929–2003 StuFa ChemnitzBrenneckenbrück Weser 1638 1945–2002 NLÖ HildesheimBurghausen Donau 6649 1900–2003 LFW MünchenCalbe-Grizehne Elbe 23719 1931–2003 BfG KoblenzCalvörde Elbe 732 1951–2003 LHW Sachsen-AnhaltCamburg-Stöben Elbe 3977 1931–2003 TLUG JenaCelle Weser 4374 1901–2004 BfG KoblenzChamerau Donau 1356 1930–2003 LFW MünchenCochem Rhein 27088 1900–2006 WSD Südwest (Mainz)Dillingen Donau 11315 1923–2003 WWA KrumbachDonauwörth Donau 15037 1923–2003 LFW MünchenDörzbach Rhein 1029 1923–2003 RP StuttgartDresden Elbe 53096 1852–2003 BfG KoblenzDüsseldorf Rhein 147680 1930–2006 WSD WestEichstätt Donau 1400 1929–2003 LFW MünchenEisenhüttenstadt Oder 52033 1920–2004 WSA EberswaldeErfurt Elbe 843 1930–2003 TLUG JenaErlln Elbe 2983 1961–2002 LfUG DresdenEschelbach Donau 1335 1930–2002 LFW München
97
Appendix
Gauge River basin Basin area Period of Water authority[km2] measurements
Frankenroda Weser 4214 1935–2004 TLUG JenaFreising Donau 3088 1950–2003 LFW MünchenFürstenfeldbruck Donau 1235 1920–2003 LFW MünchenGera Elbe 2186 1950–2003 TLUG JenaGerstungen Weser 3039 1931–2004 TLUG JenaGolzern 1 Elbe 5442 1910–2003 LfUG DresdenGöritzhain Elbe 532 1910–2003 StuFa ChemnitzGrafenmühle Donau 1436 1939–2003 LFW MünchenGrebenau Weser 2975 1950–2004 BfG KoblenzGreene Weser 2916 1940–2002 NLÖ HildesheimGreiz Elbe 1255 1924–2003 TLUG JenaGreven Ems 2842 1940–2004 BfG KoblenzGroße Schwülper Weser 1734 1925–2002 NLÖ HildesheimGuntershausen Weser 6366 1920–2004 BfG KoblenzHadmersleben Elbe 2758 1931–2003 LHW Sachsen-AnhaltHaltern Rhein 4273 1950–2005 LUA EssenHann-Münden Weser 12442 1931–2005 WSD MitteHarburg Donau 1578 1939–2003 LFW MünchenHavelberg Elbe 24037 1945–2003 BfG KoblenzHeitzenhofen Donau 5426 1920–2003 WWA RegensburgHeldra Weser 4302 1950–2004 BfG KoblenzHerrenhausen Weser 5304 1941–2005 WSD MitteHerzlake Ems 2226 1937–2002 NLÖ HildesheimHofkirchen Donau 47496 1900–2004 WSA RegensburgHof Elbe 521 1920–2003 WWA HofHohensaaten Oder 109564 1920–2004 WSA EberswaldeHopfgarten Elbe 529 1911–2003 StuFa ChemnitzHorb/Neckar Rhein 1113 1931–2004 WWA HofHundersingen Donau 2639 1929–2004 LFU KarlsruheIngolstadt Donau 20001 1923–2003 LFW MünchenInkofen Donau 3043 1926–2006 LFW MünchenIntschede Weser 37720 1940–2005 WSD MitteKalkofen Rhein 5304 1935–2006 WSD Südwest (Mainz)Kalteneck Donau 762 1920–2003 LFW MünchenKarlshafen Weser 14794 1940–2005 WSD MitteKaub Rhein 103488 1930–2006 WSD Südwest (Mainz)Kelheim Donau 22950 1923–2003 WWA LandshutKempten Donau 955 1919–2003 LFW MünchenKöln Rhein 144232 1845–2004 BfG KoblenzKriebstein UP Elbe 1757 1910–2002 LfUG DresdenLandau Donau 8467 1925–2003 LFW MünchenLandsberg Donau 2287 1900–2003 WWA Weilheim
98
Gauge River basin Basin area Period of Water authority[km2] measurements
Laufermühle Rhein 954 1926–2003 LFW MünchenLauffen Rhein 7916 1948–2006 WSD Südwest (Mainz)Lechbruck Donau 1714 1950–2003 WWA WeilheimLetzter Heller Weser 5487 1940–2004 BfG KoblenzLeun Rhein 3571 1935–2006 WSD Südwest (Mainz)Lichtenwalde Elbe 1575 1910–2003 StuFa ChemnitzMagdeburg Elbe 94942 1930–2003 BfG KoblenzMainz Rhein 98206 1930–2006 WSD Südwest (Mainz)Marklendorf Weser 7209 1941–2005 WSD MitteMaxau Rhein 50196 1921–2006 WSD Südwest (Mainz)Meinigen Weser 1170 1918–2004 TLUG JenaMellingen Elbe 627 1922–2003 TLUG JenaMünchshofen Donau 4014 1929–2003 LFW MünchenNägelstedt Elbe 716 1936–2003 TLUG JenaNeu Darchau Elbe 131950 1900–2004 BfG KoblenzNeubrück Rhein 1595 1950–2006 StUA KrefeldNeuhausen Rhein 11887 1930–2004 BfG KoblenzNiederschlema Elbe 759 1928–2002 StuFa PlauenNiedertrebra Elbe 894 1922–2003 TLUG JenaNossen 1 Elbe 585 1925–2003 LfUG DresdenNürnberg Rhein 1192 1910–2003 LFW MünchenOberaudorf Donau 9712 1900–2003 WWA RosenheimOberndorf Donau 26448 1925–2003 LFW MünchenOberstein Rhein 558 1937–2003 LFW MainzOffingen Donau 951 1940–2003 LFW MünchenOhrum Weser 813 1925–2002 NLÖ HildesheimOldisleben Elbe 4174 1922–2003 TLUG JenaOpladen Rhein 606 1951–2006 LUA EssenPassau Donau 26084 1920–2003 LFW MünchenPettstadt Rhein 7005 1922–2003 LFW MünchenPforzheim/Enz Rhein 1479 1931–2004 LFU KarlsruhePlattling Donau 8839 1925–2003 WWA DeggendorfPlochingen Rhein 3995 1918–2006 WSD Südwest (Mainz)Pockau 1 Elbe 385 1921–2002 StuFa ChemnitzPorta Weser 19162 1935–2005 WSD MitteRees Rhein 159300 1931–2006 WSD WestRegenstauf Donau 2660 1900–2002 WWA RegensburgRekingen Rhein 14718 1920–2004 BfG KoblenzRethem Weser 14730 1940–2005 WSD MitteRheine Ems 3740 1930–2004 BfG KoblenzRockenau Rhein 12710 1950–2006 WSD Südwest (Mainz)Rönkhausen Rhein 884 1950–2005 StUA Siegen
99
Appendix
Gauge River basin Basin area Period of Water authority[km2] measurements
Rotenburg Weser 2523 1920–2004 BfG KoblenzRottersdorf Donau 720 1939–2003 LFW MünchenRudolstadt Elbe 2678 1942–2003 TLUG JenaSchmittlotheim Weser 1202 1930–2004 BfG KoblenzSchwabelweis Donau 35399 1930–2002 WSA RegensburgSchwarmstedt Weser 6443 1941–2005 WSD MitteSchweinfurt Rhein 12715 1844–2006 WSA SchweinfurtSchwürbitz Rhein 2424 1940–2003 LFW MünchenSeebruck Donau 1399 1930–2003 LFW MünchenSpeyer Rhein 53131 1950–2006 WSD Südwest (Mainz)Staudach Donau 952 1920–2003 LFW MünchenStegen Donau 993 1930–2003 LFW MünchenStreckewalde Elbe 206 1921–2002 StuFa ChemnitzSylvenstein Donau 1138 1949–2003 LFW MünchenTorgau Elbe 55211 1935–2003 WSA DresdenTreuchtlingen Donau 982 1940–2003 WWA AnsbachTrier Rhein 23857 1931–2006 WSD Südwest (Mainz)Türkheim Donau 671 1950–2003 WWA KrumbachUnterjettenberg Donau 940 1900–2003 LFW MünchenUnterköblitz Donau 2004 1940–2003 WWA AmbergUnterlangenstadt Rhein 713 1931–2002 WWA HofVacha Weser 2246 1921–2004 TLUG JenaVersen Ems 8369 1941–2004 BfG KoblenzVilligst Rhein 2009 1950–2005 LUA EssenWahmbeck Weser 12996 1941–2005 WSD MitteWarnbach Donau 821 1940–2003 LFW MünchenWechselburg Elbe 2107 1910–2003 StuFa ChemnitzWegeleben Elbe 1215 1894–2003 LHW Sachsen-AnhaltWildenau Donau 712 1940–2003 LFW MünchenWittenberg Elbe 61879 1950–2003 BfG KoblenzWittenberge Elbe 123532 1899–2003 BfG KoblenzWolfsmünster Rhein 2131 1930–2003 LFW MünchenWolmirstedt Elbe 1503 1951–2003 LHW Sachsen-AnhaltWorms Rhein 68827 1936–2006 WSD Südwest (Mainz)Zwickau-Pölbitz Elbe 1030 1928–2003 StuFa Plauen
100
Acknowledgements
First of all, I would like to express my gratitude to my supervisor Prof. Dr. B. Merzfrom GFZ German Research Centre for Geosciences for his continuous support, fruitfuldiscussions and critical suggestions on how to improve my studies. Under his supervision, Iwas involved in two very interesting projects: first, I got in contact with the challenging topicof flood hazard management with the focus on mitigation measures and land-use planning inthe project "Flood Risk Reduction in Germany – Lessons Learned from the 2002 Disaster inthe Elbe Region". Later, the second project was realized by a "Helmholtz Young ScientistsGroup", consisting of four scientists under the supervision of Dr. K.-E. Lindenschmidt. Itwas funded by GFZ and the Helmholtz Association. The aim of the project was to studyextreme flood events for the Elbe and Mulde catchments in south-eastern Germany andto develop tools for flood hazard assessment and management. I would like to thank bothinstitution for their financial support.
Prof. Dr. A. Bronstert from the University of Potsdam supported me during the entirework. I dedicate special thanks to the reviewers of this thesis: Prof. Dr. B. Merz and Prof.Dr. A. Bronstert from the Institute of Geoecology, Prof. Dr. M. Disse from the Universitätder Bundeswehr München and Prof. Dr. R. Jüpner from the University of Kaiserslautern.
At GFZ Potsdam in the Section of Engineering Hydrology I experienced a stimulatingworking atmosphere and interesting discussions with all group members. I am particularlyindepted to Prof. Dr. A.H. Thieken (now at AlpS Innsbruck) for her scientific support as wellas for always having an open ear during difficult times. Particularly, I would like to thankSergiy Vorogushyn, Saskia Förster and Steffi Uhlemann for endless fruitful discussions.
I would like to express my gratitude to my husband Stefan Vater and my family. Theysupported me during all these years of research and shared with me good and difficult times.Without their encouragement, this work would have probably never been finished.
Last but not least, I would like to thank numerous water authorities such as BfG Koblenz,WSA Dresden, WSA Magdeburg and many others for providing discharge data.
The scientific focus of this PhD work changed for several reasons in the course of theproject. I am grateful for insights I gained in flood history and the estimation of historicevents, even if I finally could not include them in this dissertation.
101
Curriculum vitae
Personal Data
born on 10th September 1975 in Berlin
married, 1 son
Working Experience
2004 – 2009 Scientific Staff (PhD student) at GFZ German Research Centre forGeosciences, Section Hydrology
2006 – 2007 Teaching at the University of Potsdam, Institute of Geoecology
2003 – 2004 Scientific Staff at GFZ German Research Centre for Geosciences,Section Engineering Hydrology
2001 Practical work at ETH Zurich, Chair of Forest Ecology
2000 Practical work at Credit Valley Conservation in cooperation with theUniversity of Guelph (Canada)
1995 – 1996 Voluntary Ecological Year (FÖJ) at the Foundation for NatureConservation Berlin
Education
1996 – 2003 Studies of Geoecology at the University of Potsdam, Graduationwith honors as Dipl.-Geoecologist
2002 Diploma thesis at ETH Zurich „Modelling snow cover in apre-alpine valley with remotely sensed data“, supported by DAAD
1999 – 2000 Studies of Environmental Sciences at the University of Guelph,Canada
1995 Diploma from Secondary School (Max-Planck-Gymnasium, Berlin)
103
List of Publications
Kreibich, H., Petrow, T., Thieken, A., Müller, M., and Merz, B. (2005a). Consequences ofthe extreme flood event of August 2002 in the city of Dresden (Germany). In M. SavicD. A.and Marino, H. Savenije, and J. Bertoni (editors), Sustainable Water ManagementSolutions for Large Cities, IAHS Publications, volume 293, pp. 164–173. InternationalAssociation of Hydrological Sciences.
Kreibich, H., Thieken, A., Petrow, T., Müller, M., and Merz, B. (2005b). Flood lossreduction of private households due to building precautionary measures - Lessons learnedfrom the Elbe flood in August 2002. Natural Hazards and Earth System Sciences, 5,117–126.
Lindenschmidt, K.E., Fleischbein, K., Petrow, T., Vorogushyn, S., Theobald, S., and Merz,B. (2005). Model system development for the provisionary management of extremefloods in large river basins, with emphasis on model system uncertainty vs. complexity atdifferent scales. Advances in Geosciences, 5, 99–104.
Müller, M., Vorogushyn, S., Maier, P., Thieken, A., Petrow, T., Kron, A., Büchele, B., andWächter, J. (2006). CEDIM Risk Explorer - a map server solution in the project “RiskMap Germany”. Natural Hazards and Earth System Sciences, 6, 711–720.
Petrow, T., Delgado, J., and Merz, B. (2008). Trends der Hochwassergefährdung inDeutschland (1951–2002) und Konsequenzen für die Bemessung. Wasserwirtschaft, 11,24–28.
Petrow, T. and Merz, B. (2009). Trends in flood magnitude, frequency and seasonality inGermany in the period 1951–2002. Journal of Hydrology, 371, 129–141.
Petrow, T., Merz, B., Lindenschmidt, K.E., and Thieken, A.H. (2007). Aspects of seasonalityand flood generating circulation patterns in a mountainous catchment in south-easternGermany. Hydrology and Earth System Sciences, 11 (4), 1455–1468.
Petrow, T., Thieken, A., Kreibich, H., Bahlburg, C., and Merz, B. (2006). Improvementson flood alleviation in Germany - Lessons Learned from the Elbe flood in August 2002.Environmental Management, 38, 717–732.
Petrow, T., Thieken, A., Kreibich, H., and Merz, B. (2003). Vorsorgende Maßnahmen zurSchadenminderung. In Hochwasservorsorge in Deutschland. Lernen aus der Katastrophe2002 im Elbegebiet, Schriftenreihe des DKKV, volume 29, pp. 34–73. Deutsches Komiteefür Katastrophenvorsorge e.V. 152 pp.
105
List of Publications
Petrow, T., Zimmer, J., and Merz, B. (2009). Changes in the flood hazard through changingfrequency and persistence of circulation patterns. Natural Hazards and Earth SystemSciences, 9, 1409–1423.
Thieken, A., Grünewald, U., Merz, B., Petrow, T., Schümberg, S., Kreibich, H., Streiz,W., and Kaltofen, M. (2005). Flood risk reduction in Germany after the Elbe 2002flood: aspects of hazard mapping and early warning systems. In Proceedings of theinternational conference “Cartographic Cutting-Edge Technology for Natural HazardManagement”, Kartographische Bausteine, volume 30, pp. 145–156. TU Dresden, Insti-tut für Kartografie.
Thieken, A., Petrow, T., Kreibich, H., and Merz, B. (2006). Insurability and mitigation offlood losses in private households in Germany. Risk Analysis, 26 (2), 383–395.
106
Author’s declaration
I have prepared this dissertation without illegal assistance. The work is original exceptwhere indicated by special reference in the text and no part of the dissertation has beensubmitted for any other degree. This dissertation has not been presented to any otherUniversity for examination.
Theresia PetrowPotsdam, June 2009
107