Réf. Petrow & Merz 2009 - A

Référence bibliographique complète

PETROW, T., MERZ, B. 2009. Trends in flood magnitude, frequency and seasonality in Germany in the period 1951–2002. Journal of Hydrology, 371, 129-141.

Abstract: During the last decades several destructive floods in Germany led to the impression that the frequency and/or magnitude of flooding has been increasing. In this study, flood time series are derived and analyzed for trends for 145 discharge gauges in Germany. A common time period of 52 years (1951– 2002) is used. In order to obtain a country-wide picture, the gauges are rather homogeneously distributed across Germany. Eight flood indicators are studied, which are drawn from annual maximum series and peak over threshold series. Significant flood trends (at the 10% significance level) [were detected] for a considerable fraction of basins. In most cases, these trends are upward; decreasing flood trends are rarely found and are not field-significant. Marked differences emerge when looking at the spatial and seasonal patterns. Basins with significant trends are spatially clustered. Changes in flood behavior in northeast Germany are small. Most changes are detected for sites in the west, south and center of Germany. Further, the seasonal analysis reveals larger changes for winter compared to summer. Both, the spatial and seasonal coherence of the results and the missing relation between significant changes and basin area, suggest that the observed changes in flood behavior are climate-driven.

Floods, Trend, Non-stationarity, Climate change, Land use change, Germany

Organismes / Contact

GeoForschungsZentrum Potsdam, Section Engineering Hydrology, Telegrafenberg, 14473 Potsdam, Germany: thpetrow@gfz-potsdam.de (T. Petrow), bmerz@gfz-potsdam.de (B. Merz)

(1) - Paramètre(s) atmosphérique(s) modifié(s)
(2) - Elément(s) du milieu impacté(s)
(3) - Type(s) d'aléa impacté(s)
(3) - Sous-type(s) d'aléa

Pays / Zone
Massif / Secteur
Site(s) d'étude
Période(s) d'observation
Germany   Only the results concerning the whole country or the Danube and Rhine catchments (alpine regions) are quoted in this page     1951– 2002

(1) - Modifications des paramètres atmosphériques

Informations complémentaires (données utilisées, méthode, scénarios, etc.)

(2) - Effets du changement climatique sur le milieu naturel

Sensibilité du milieu à des paramètres climatiques
Informations complémentaires (données utilisées, méthode, scénarios, etc.)

The southern part of the Danube catchment is dominated high pressure systems, especially during fall and winter. Westerly, north-westerly and south-westerly circulation types are less frequent. In this region, summer floods dominate.


(3) - Effets du changement climatique sur l'aléa

There exist some studies on flood trends in Germany, which are however restricted to specific regions or catchments [see synthesis and references in the study]. In the Danube and Rhine catchments (for five gauges with varying time periods) upward trends in AMAXF were detected by Caspary (1995) and Caspary and Bárdossy (1995). KLIWA (2007) analyzed flood trends of 158 gauges in southern Germany. Long time series of 70–150 years mostly revealed no trends. However, the study of the last 30 years showed at many gauges significant upward trends in AMAXF. Moreover, the frequency of winter floods increased since the 1970s in many basins.

[The] compilation of the trend analyses for German rivers shows that there is no unambiguous pattern of flood trends across Germany. Further, the studies available are limited to selected regions or single basins. There is no comprehensive study on flood trends in Germany which covers the entire country. This gap is filled by this paper for the period 1951–2002.

Results for the gauge [...] Donauwoerth/Danube:
[...] Significant upward trends in AMAXF were found. [...] In the case of Donauwoerth, summer floods are only slightly smaller than winter floods. Increasing trends were detected in both seasons; however, the trend in the winter season is not significant. Although the linear regression trends in AMAXF and AWMAXF have equal gradients, the trend in AWMAXF is not significant due to the larger standard deviation of AWMAXF. Contrary to AMAXF, no trends in POT1M and POT3M were identified [...]. Actually, both trend lines of POT3M show small decreases. That means that a significant increase in the number of discharge peaks above the threshold does not necessarily comply with a significant increase in the magnitude of these peaks – a result that was also found by Svensson et al. (2005). To understand this discrepancy, the POT3M time series were further separated in the upper, middle and lower third. [...] The POT thirds have a much smaller range compared to AMAXF. [...] The POT time series show no or only mild increases, whereas AMAXF grows significantly. This marked increase is mainly a result of several very small annual floods that were lower than the POT3M threshold. Interestingly, these small discharge values occurred exclusively during the first half of the time period. The larger discharge peaks, represented by POT upper third, increased only slightly. The POT3F time series show a very similar behavior. Trends in POT3F are upward and significant. For both sites, the frequency of discharge peaks above the POT3M threshold increased, although the magnitude of these events (POT3M) did not experience a significant change. The seasonal separation of POT3F yielded significant increasing trends for winter (WPOT3F), i.e. the number of high discharge events during the winter season grew. For both cases, the number of discharge peaks in summer (SPOT3F) above the POT3M threshold increased as well; however, this increase does not suffice to be significant.

Spatial distribution of significant trends:
The majority of the 145 gauges showed at least one significant result when analyzed for trend in the eight flood indicators. [...] Forty-two percent of the Danube gauges and 46% of the Rhine gauges [...] showed at least two significant trends. The sites in the Elbe basin showed less change in flood indicators compared to sites in the Rhine, Weser and Danube catchments.

[...] . At 41 gauges (28% of all sites) significant increasing trends were detected [for AMAXF], whereas only two gauges showed significant decreasing trends. An interesting spatial pattern emerges: all sites with significant trends are located in the southern, western and central parts of Germany. A relatively sharp line from northwest to southeast can be drawn, which separates the region with trends from the region without trends. Along the middle and lower Rhine main river as well as along the Danube main river most of the gauges show significant trends. [...]

The trend analyses for the winter maxima gave similar results as the analyses for the annual maxima. Significant upward trends in winter maxima were identified at 23% of all sites. No significant downward trends were detected. The spatial pattern is only slightly different: The gauges with significant upward trends for annual winter maximum are found in a diagonal band stretching from northwest to southeast of Germany. North and south of this band were no or only non-significant trends detected. In the Rhine and Danube catchments, the lower number of trends in AWMAXF, compared to AMAXF, is mainly due to a smaller number of significant trends along the main rivers (Rhine, Danube).

A smaller number of significant trends (29 gauges corresponding to 20% of all gauges) were found for the summer maxima (ASMAXF). In contrast to AWMAXF, where all detected trends are upward, the trend analysis of ASMAXF resulted in the same number of upward and downward trends. Moreover, there is a clear spatial distinction between the regions with upward and downward trends, respectively. Only gauges in central and northern Germany in the catchments of Weser, Odra and Elbe show downward trends, whereas the upward trends are exclusively found at gauges in southern and western Germany in the Rhine and Danube catchments.

At 18% of the gauges significant trends in the POT1M time series could be detected. Due to the spatial concentration in central Germany field significance was observed. As could be expected, many gauges show significant trends in AMAXF as well as in POT1M. A similar spatial pattern was detected for the POT2M variable, with however less significant trends (16%). The POT3M time series show almost no significant trends across Germany. Only 7% of the gauges have significant trends. These are not spatially clustered, but are rather randomly distributed all over Germany. The gauges Cologne/Rhine and Donauwoerth/Danube are two examples for this behavior where significant changes in AMAXF are not matched with significant changes in POT3M.

In contrast to POT3M, significant trends in the peak-over-threshold frequency POT3F were identified at many gauges: 25% of all gauges show an increasing trend, 1% a decreasing trend. With the exception of two gauges in the Elbe catchment, only gauges in the Rhine and Danube catchments show a significant change in flood frequency. The relative change of the POT3 frequency is rather large with values up to 140%. This upper value means that the number of discharge peaks above the threshold has increased approximately fivefold, from one event per year in the 1950s to five events per year at the end of the study period. The spatial distribution of gauges with significant trends is very similar to the result of AMAXF. Again, a relatively sharp line from northwest to southeast Germany can be observed which separates the region of no trend from the one with positive trends. The seasonal separation of the POT3F variable illustrates very well that the majority of the positive trends is caused by significant upward trends in the frequency of the winter floods, whereas POT3F summer events only increase at three gauges in the Danube catchment. Again, the Rhine and Danube catchments are mainly affected by the changes in the flood discharge behavior.

For the eight flood indicators, field significance at the 10% significance level was detected for AMAXF, AWMAXF, POT3F and WPOT3F. In all four cases, upward trends are the cause for the changes in flood behavior. No field significance could be found for decreasing trends for all flood indicators. The changes in the summer flood behavior (ASMAXF, SPOT3F) are too small to be counted as field significant.

Finally, it is assessed if a scale-dependency can be found in the trend analyses, i.e. it is assessed if large changes are related to small or large basins, respectively. To this end, the relative changes in each flood indicator were plotted against the basin area, and significant changes were marked. No scale-dependency can be observed. There are no spatial scales where significant changes are concentrated. On the contrary, significant changes and no changes, respectively, are found at all spatial scales.

The analysis of trends in eight flood indicators for 145 gauges across Germany yields a number of interesting results. Overall, it can be summarized that the flood hazard in Germany increased during the last five decades, particularly due to an increased flood frequency. Marked differences emerge when looking at the spatial and seasonal patterns and at different flood indicators. An important observation is that sites with upward and downward flood trends are spatially clustered. Changes in the flood behavior in northeast Germany are small. Most changes were detected for sites in the west, south and center of Germany. Further, the seasonal analysis revealed larger changes for winter compared to summer.

[...] Mostly increasing trends were detected, with large shares of significant trends in AMAXF and POT3F. Approximately 1/3 of the sites in the western and southern parts of Germany (Danube, Rhine, Weser) have significant upward trends in AMAXF, whereas there are almost no upward trends in eastern Germany (Elbe). [...]

Compared to Rhine and Weser, the sites in the Danube catchment are much more influenced by summer floods. Accordingly, the upward trends of AMAXF in the Danube basin are mainly dominated by upward trends in summer floods. However, also the frequency of floods (POT3F) increased significantly at many gauges, especially along the main river Danube, which is visible in both seasonal POT3F. An increasing frequency in the winter is supposed to be caused by higher winter temperatures, and hence, earlier snow melting in the mountain ranges.

The spatial and seasonal coherence of the results suggests that the observed changes in flood behavior are climate-driven. This conclusion is further supported by the missing relation between significant changes in the discharge series and basin area. Impact of land-cover changes or of river training works would be expected to show scale-dependency. However, from [their] analysis [the authors] conclude that there are no preferred spatial scales where significant changes could be detected.

Therefore, it is interesting to evaluate, whether or not [the present] results are in line with studies on changes in climate. To this end, [the] results are qualitatively compared to those of recent investigations that analyze changes in atmospheric circulation patterns [see review in the study]. It has been shown that there is a close link between the occurrence and persistence of atmospheric circulation patterns and floods in Germany (e.g., Bárdossy and Caspary, 1990; Pfister et al., 2004a; Petrow et al., 2007).


Paramètre de l'aléa
Sensibilité des paramètres de l'aléa à des paramètres climatiques
Informations complémentaires (données utilisées, méthode, scénarios, etc.)

Eight flood indicators were included in the study. These comprise annual maximum streamflow series (AMAX) as well as peak over threshold series (POT) [see details in the study].

Flood indicators:
•AMAXF: Annual maximum daily mean streamflow (m3/s)
AWMAXF: Annual winter maximum daily mean streamflow (m3/s)
ASMAXF: Annual summer maximum daily mean streamflow (m3/s)
POTXM: Peak-over-threshold magnitude (m3/s)
POT3F: Peak-over-threshold frequency (number of peaks)
SPOT3F: Summer peak-over-threshold frequency
WPOT3F: Winter peak-over-threshold frequency

Annual maximum daily mean streamflow, i.e. the largest daily mean streamflow that occurs in each hydrological year, is the most common indicator in flood trend studies. In some studies, POT series are used since they are considered to include more information and thus allowing to reveal better the temporal pattern of flood occurrence (Svensson et al., 2006). Besides the detection of trends in flood magnitude, they offer the possibility to analyze the flood frequency, i.e. changes in the number of floods occurring each year.

The spatial distribution of significant upward and downward trends is shown and the field significance is calculated for the different flood indicators.


Climate change is not the only possible driver of change in flood time series. [...] Most of the basins in Germany have undergone widespread land use changes, significant volumes of flood retention have been implemented in the last decades, and many rivers have experienced river training works. In particular, the active floodplains of many rivers in Germany have been reduced through the construction of dykes. [...] Different studies [...] found little or no influence of land use on flood discharge. Blöschl et al. (2007) argued that the impact of land use changes on floods is a matter of spatial scale. In small basins land use changes can significantly alter the runoff processes, effecting flood magnitude and frequency. However, these effects are expected to fade with increasing basin scale. The general tendency of decreasing impacts with increasing basin scale does not apply to river training works. On the contrary, river training impacts are likely to increase with catchment size as there is a tendency for larger settlements and hence large-scale flood protection works at larger streams (Blöschl et al., 2007). The cumulative effects of river training works on floods in large basins are difficult to assess. [...] To complicate matters, the effects are expected to vary with flood magnitude. [...] [see references in the study]

The detection of coherent flood trends at many sites in a geographic region may allow distinguishing climate-related changes from other anthropogenic changes. Although local effects and anthropogenic influences, such as flood control measures, may markedly influence the at-site flood behavior, such changes are not expected to cause coherent changes over a large geographical area.

Discharge time series were obtained from the water authorities of different federal states in Germany. Since the data are part of the hydrometric observation network of the water authorities in Germany, the observations are regularly checked and can be assumed to be of good reliability, although it is acknowledged that flood peak measurements are frequently associated with considerable errors. Sites were selected with a catchment size of at least 500 km². In that way, small catchments were excluded from the analysis but still a large number of gauging stations and a satisfying spatial coverage of Germany were obtained. Although there is considerable uncertainty about the scale where changes in land use and land management in a specific basin cannot be seen anymore in the basin flood hydrograph (Blöschl et al., 2007), 500 km² seems a reasonable choice for the lower limit. Beyond that scale, most of the effects of land use and land management are expected to have been faded out. A common time period between 1.11.1951 and 30.10.2002 was used (hydrological year in Germany: 1 November–31 October). Small gaps in the data of up to one year were marked as ‘‘missing values”. This was necessary at only five gauges. Time series with larger successive gaps were excluded from the analysis. Finally, time series of mean daily stream- flow from 145 gauges in Germany were included in the analysis. They are relatively homogeneously distributed across Germany. Forty-three stations are located in the Danube catchment, 37 in the Rhine catchment, 32 in the Elbe catchment and 27 in the Weser catchment.

Selected results are shown exemplarily for the gauges Cologne in the Rhine catchment and Donauwoerth (catchment size 15,037 km²) in the Danube catchment. These gauges were selected because their behavior can be seen to be representative for most gauges in Germany. [...] The discharge behavior at Donauwoerth (Danube) is dominated by summer floods and represents gauges in the mountain ranges with faster runoff regimes, especially in the catchments of Elbe and Danube.

[See Methodology in the study]

(4) - Remarques générales

Only few studies on flood trends in Europe, covering large regions or entire countries, could be found. [...] Main problems of flood trend analysis are data availability and data reliability. Many discharge time series are short and are not suited for analyzing extreme events. Kundzewicz et al. (2005) suggested a minimum length of 50 years for flood trend detection. For some gauges there are systematic discharge observations in the range of 100 years or even more. Such time series are very valuable; however, the quality of these data has to be examined carefully. Lindström and Bergström (2004) emphasised the need to balance availability and reliability: very long discharge time series might not be reliable, but reliable series might be too short.

(5) - Syntèses et préconisations

[This] study of flood trends at 145 runoff gauges, distributed all over Germany, shows that there is no ubiquitous increase of flood magnitude and/or frequency in the second half of the 20th century, as it is often asserted in the media. However, significant flood trends were detected for a considerable fraction of basins. In most cases, these trends are upward; decreasing flood trends were rarely found and were not field-significant. The joint analysis of many sites within one region allowed assessing the spatial and seasonal coherence of flood trends: Basins with significant trends were spatially clustered. Changes in flood behavior in northeast Germany are small. Most changes were detected for sites in the west, south and center of Germany, i.e. in the catchments of Rhine, Weser and Danube. The seasonal analysis revealed larger changes for winter compared to summer. From the results [the authors] concluded that the observed changes in flood behavior are climate-driven. It was possible to qualitatively link [the] results to trends in frequency and persistence of atmospheric circulation patterns above Europe. As already shown by Pfister et al. (2004b) for a smaller area, orographic obstacles heavily influence the spatial distribution of the rainfall and runoff processes. A changing behavior of circulation patterns is likely to cause changes in rainfall totals, which in turn heavily affects discharge and water levels in the rivers. The relationship between circulation patterns, flood magnitude and/or frequency and the influence of the topography will be further investigated. [The present] findings underline the need to thoroughly analyze the flood behavior for changes when estimates for flood design or flood risk management are needed.

Références citées :

Bárdossy, A., Caspary, H.J., 1990. Detection of climate change in Europe by analyzing European atmospheric circulation patterns from 1881 to 1989. Theoretical and Applied Climatology 42, 155–167.

Blöschl, G., Ardoin-Bardin, S., Bonell, M., Dorninger, M., Goodrich, D., Gutknecht, D., Matamoros, D., Merz, B., Shand, P., Szolgay, J., 2007. At what scales do climate variability and land cover change impact on flooding and low flows? Invited commentary. Hydrological Processes 21, 1241–1247. doi:10.1002/hyp.6669.

Caspary, H., 1995. Recent winter floods in Germany caused by changes in the atmospheric circulation across Europe. Physics and Chemistry of the Earth 20, 459–462.

Caspary, H., Bárdossy, A., 1995. Markieren die Winterhochwasser 1990 und 1993 das Ende der Stationarität in der Hochwasserhydrologie infolge von Klimaänderungen? Wasser & Boden 47 (3), 18–24.

KLIWA, 2007. Klimaveränderung und Konsequenzen für die Wasserwirtschaft. KLIWA Berichte Heft 10., Stuttgart, 258 pp.

Kundzewicz, Z.W., Graczyk, D., Maurer, Th., Pinskwar, I., Radziejewski, M., Svensson, C., Szwed, M., 2005. Trend detection in river flow series: 1. Annual maximum flow. Hydrological Sciences Journal 50 (5), 797–810.

Lindström, G., Bergström, S., 2004. Runoff trends in Sweden 1807–2002. Hydrological Sciences Journal 49 (1), 69–83.

Petrow, Th., Merz, B., Lindenschmidt, K.-E., Thieken, A.H., 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.

Pfister, L., Kwadijk, J., Musy, A., Bronstert, A., Hoffmann, L., 2004a. Climate change, land use change and runoff prediction in the Rhine–Meuse basins. River Research and Applications 20, 229–241. doi:10.1002/rra.775.

Pfister, L., Drogue, G., El Idrissi, A., Iffly, J.-F., Poirier, Ch., Hoffmann, L., 2004b. Spatial variability of trends in the rainfall–runoff relationship: a mesoscale study in the Mosel basin. Climatic Change 66, 67–87.

Svensson, C., Kundzewicz, Z.W., Maurer, Th., 2005. Trend detection in river flow series: 2 flood and low-flow index series. Hydrological Sciences Journal 50 (5), 811–824.

Svensson, C., Hannaford, J., Kundzewicz, Z., Marsh, T., 2006. Trends in river flows: why is there no clear signal in observations? In: Frontiers in Flood Research, IAHS Publ. 305, pp. 1–18.