The sub-project 4 “Flood typology – controls in a changing world” will focus on developing process-based flood typology, which will account for the atmospheric, catchment and river system processes of floods, as all three are important in characterising the essential flood properties. Depending on the generating processes flood event types varies in their resulting characteristics such as the shape of the flood hydrograph, their spatial coverage, time of occurrence within the year, dependence on antecedent soil moisture. Hence a classification enables to break down the plethora of different flood events in typical representatives by a clear description of similarities and differences in flood producing processes and resulting event characteristics. The classification allows us to compare only those events, which stems from similar processes and will give deeper insight in the spatio-temporal changes of differing flood producing factors.
The objective of the first phase of the project is to develop a hierarchical typology of flood events in Germany and Austria. The classification will be built on indicators of flood processes and resulting event characteristics such as the intensity, duration and spatial coverage of flood producing precipitation event, weather patterns, antecedent soil moisture states and snow. Using the classification we will analyse how flood types changed regionally over the study period and how flood types change from small to more extreme floods. In the second phase we will analyse flood events in very small and very large river catchments in Germany. Very small catchments are defined in this proposal, by catchments where the typical travel time of flood events can be smaller than one day. Hence such flood events cannot be analysed on the basis of daily data. Very large catchments are defined in this proposal as catchment where flood event characteristics are mainly affected by routing effects and the coincidence of flood peaks from sub-catchment. The research will be based on the typology approach developed in the first period, but will extend the analysis by focusing on sub-daily processes in selected small catchments and routing effects in larger catchments, which requires a deeper analysing of process of the flood waves along the river network. Additionally we will analyse the drivers of changing flood types and we will tackle the question of how flood type occurrence will change in future if e.g. climatic conditions, such as the frequency of rainfall event with high intensities will change. The analysis of the future evolution of flood types will be based on a scenario approach.
To enable classification of single flood events, first rainfall-runoff events will be separated from continuous time series similar to approach of Merz et al. (2006). The algorithm is designed to automatically separate events according the characteristics of hydrographs. The algorithm will separate base flow, identify runoff events, attribute rainfall events and refine multiple-peak events based on stochastic properties of distributions of event runoff coefficients. In contrast to classical flood events selection approaches by taking maximum annual peak flow or peak over threshold, the approach will not only provide peak flow, but delivers other important event characteristics, such as start/end point of the event, volume, shape of the event flood hydrograph. After separating single flood events, indicators describing important event characteristics, such as peak flow, event duration, flood time scale (Gaal et al., 2012) will be calculated for the single flood events and their spatio-temporal variability will be analysed in respect to climate and landscape characteristics. The analysis will give first insight in the spatio-temporal patterns of flood event characteristics (independent of flood generating processes).
In the next step potential flood types and indicators, describing event characteristics, will be developed. First, potential flood types will be hypothesized based on existing flood typologies (e.g. Merz and Blöschl, 2003) and process studies of single flood events (e.g. Blöschl et al., 2013). Second, a set of indicators (“flood event signatures”) will be derived, which represent the catchment functioning and are used to associate single events in flood types. Hence, these indicators have to grasp the different factors of the dominating flood generating processes, such as climatic forcing, catchment response and routing effects. The indicators can be event specific and contain information on intensity, duration and spatial coverage of flood producing precipitation events, soil moisture states and snow cover, the spatial coverage of the flood events, the reaction time of flood runoff and the shape of the flood hydrograph. Indicators can also represent more static catchment characteristics such as information on topography, land use or amount of impermeable soils.
The indicators can be derived from meteorological data (e.g. intensity and duration of precipitation event), hydrological data such as runoff observations (e.g. shape of hydrograph, base flow ratio) or hydrological modelling results (e.g. soil moisture state, snow cover). Daily soil moisture dynamics and runoff processes of the past years are simulated in 4 km resolution for the target region using the hydrological model mHM (Samaniego et al., 2013). Indicators describing precipitation and climate processes will be derived using information about duration, intensity, spatial extent, circulation type and storm path of precipitation events. Indicators describing catchment response and routing effects will be based on the catchment’s soil moisture state, snow cover, shape of the flood hydrograph, the centre of mass of flood hydrographs, flood wave celerity among others. Tab. 1 shows examples of potential flood types and indicators describing event characteristics. Fig. 1, taken from Merz and Blöschl (2003), shows for three different flood types a set of indicators, which allow grasping the essence of the flood processes. For each flood event, a set of indicators will be derived, which will be the base for further classification.
The derived indicators will be used to associate single events to flood types. Floods are the interplay of meteorological forcing, catchment response and routing effects. Different combinations of typical precipitation processes, runoff generation and routing effects may result in flood runoff. To capture the large number of different process combination in an effective way, a stepwise classification is envisaged. In a first step, the meteorological condition is analysed to distinguish, e.g. between synoptic and convective rainfall events, based on the meteorological indicators, such as rainfall duration and volume or the spatio-temporal pattern of the rainfall fields. In a second step, the processes leading to runoff are analysed. This step aims to decipher between different runoff generation processes, such as surface runoff due to infiltration excess, due to exceeding the catchment storage capacity, or due to snowmelt effects. To do this the indicators representing effects of runoff generation are used, such as the shape of the hydrograph, base flow ratio or soil moisture state of the catchment, given by the hydrological model mHM. In a last step routing effects will be analysed. Routing effects or the coincidence of flood peaks from sub-catchment affect flood runoff mainly larger catchments. For the classification of flood events in smaller catchments routing effect may be negligible and the last step is not needed.
|Level||Type||Generating processes||Event characteristics||Potential indicators|
|Long rain||Rainfall over several days or possibly weeks,
including Low-intensitiy rainfall.
The rainfall events are synoptic or frontal type
|Rainfall events often cover a large area up to
several thousands of square kilometres
Duration from days to weeks
| - rainfall duration
- max. rainfall intensity
- rainfall volume
- spatial coverage of rainfall
- time of occurence within year
- circulation patterns and air
|Short rain||Rainfall of short duration and high intensity||Depending on the rainfall patterns, the events
can be of a local or regional scale
Duration up to few days
|Rainfall burst||Short, high intensity rainfalls, mainly of convective
origin, mainly occur in summer oder late summer when
enough energy is available in the atmosphere
|Local rainfall events, covering only few
hectares to square kilometres.
Short duration (mins-hrs)
|Large event rainfall volume exceed catchment
|Can occur the whole year|| - catchment soil moisture state
- pre-event soil moisture state
- shape of hydrograph
- temperature changes
- snow cover dynamics (i.e.
snow water equivalent)
- catchment topography, -soil
type and land-use cover
|Due to large pre-event rainfall volumes, catchment
reaches storage capacity. Additional event rainfall
causes flood runoff
|High pre-event rainfall volumes, Large pre-
event base flow ratios,
Often multiple-peak events
|Snow melt||Snowmelt during fair weather perios often
associated with a rapid increase in air temperature
continuously raises the flows. Rainfall may occur but
is of relatively minor importance.
|Occurence in snowmelt period, significant
temperature changes, Slowly increasing and
decreasing limps of hydrographs.
|Rain on snow||Rain falls on an existing snow cover, which
facilitates overland flow. Rainfall additionally
enhances snowmelt as compared to dry spells.
Antecedent snowmelt may saturate large parts of the
catchment facilitating overland flow once rain starts.
|Fast reacting flood hydrographs, high event
runoff coefficients, large flood runoff
contribution from tributaries with snow cover
|High intensity rainfalls can locally exceed infiltration
capacity and overland flow occurs.
|Very fast catchment response, Dominating
runoff process only in smaller catchments, can
be enhanced by impervious land cover
|3 Routing effects||Superposition of
|Superposition of flood wave peaks from tributaries
considerably increase downstream flood peak
| - Centre of mass of flood hydrographs
- flood wave celerity
- shape of flood hydrograph
Table 1. Examples of potential flood types and indicators for the level 1 (meteorological forcing), level 2 (catchment response) and level 3 (routing effects). Note that types and indicators are only example and maybe modified and extended during the project. Final event flood types will be composed by a combination of the types form the three different levels
The three levels of flood generating mechanisms (meteorological forcing, catchment response and routing effects) will then be combined to form classes of flood types, such as flood type „convective rainfall + infiltration excess runoff + no routing effects“. It is assumed that by such a hierarchical classification, the different process combinations can be better represented, than in existing classification schemes where meteorological forcing, catchment response and routing effects are treated in a lumped way. However, if typical combinations of the three levels occur, such combination may be summarised in one type.
There are two different approaches possible to classify flood events. First, flood events can be classified in a forward mode by applying „if-then“ rules to indicators (Fig. 1). If a certain set of indicators are within given ranges of the flood type, then this event will be associated to the flood type, e.g. (to classify meteorological forcing) „if (rainfall intensity > xxx mm/hr) and (rainfall duration
Fig. 1 Diagnostic maps for classifying maximum annual floods according to process type. Examples of three process types are shown: (top) long-rain flood, (middle) flash flood, and (bottom) rain-on-snow flood. All symbols have been plotted around the catchment centroids. (from Merz and Blöschl, 2003)
A second approach is to classify event by using statistical clustering algorithms. Cluster algorithms aim to classify object according to given indicators in a way that the differences between classes are large, while the variability within classes is low. Clustering of similar events is similar to the grouping of similar catchment. Grouping of catchments is well addressed in hydrological literature and different clustering algorithms are available, which can be used in this project. A promising approach is the use of self-organizing maps (SOM). SOM is an unsupervised learning neural network algorithm that performs a non-linear mapping of the dominant structures present in a high-dimensional data field onto a lower-dimensional grid. As event classification is based on multi-dimensional indicators, such as an approach may be advantageous.
Existing flood event classifications and event analyses show that flood events are often a mixture of different processes. For example, long duration rainfall event can coincide with snow melt periods. In such cases a strict classification into one or the other type is hardly possible and Sikorska et al. (2015) propose a fuzzy clustering as an alternative. With a fuzzy classification, an event is not associated with only one type, but can be attributed to several types with changing weights. The selection of a classification approach in this project will be based on a first a preliminary analysis of potential flood types and derived indicators.
Dividing objects into classes are always subjective and classification results depend on available data and which indicators are selected to describe event characteristics. In this step we test the robustness of the event classification. For a set of events, given by random sampling, it is analysed how the use of different sets of indicators, to describe event characteristics, will result in varying classifications. It will also be analysed, how the results of the classification depend on input data. For example, indicators describing soil moisture state can be derived from mHM simulations (Samaniego et al., 2013) or by using another hydrological model. For several catchments in the study region soil moisture simulations using a simpler conceptual GR6J and HBV models are available. In the last step the sensitivity of the classification towards classification rules (e.g. different threshold values in if-then rules) will be examined. This will shed light on the uncertainty of the flood typology.
The frequency of flood type occurrence, i.e. how often events are assigned to one flood type, will be calculated for each catchment. The spatial variability in flood type frequency (Fig. 2) will be analysed with respect to the spatial changes in climate and landscape characteristics. Fig. 2 (from Merz and Blöschl, 2003) shows the spatial patterns of the frequency of flood types in Austria. The spatial pattern indicates that long-rain floods are the main causative process type of annual maximum floods in most catchments in Austria, particular in catchments at the northern fringe of the high Alps, long-rain floods are particularly common. The high Alps tend to act as a topographic barrier to North-Westerly airflows, and orographic enhancement often produces persistent rainfall which can result in floods. Flash floods, however, occur significantly less frequently and are only important in eastern Austria, specifically in the hilly region of Styria in South-Eastern Austria and in the hilly region of in North-Eastern Austria. The hilly terrain appears to increase the instability of the boundary layer and hence the likelihood of convective storms. Throughout Austria, the spatial pattern of flash flood occurrence is rather patchy which reflects the random and local nature of flash floods causing maximum annual floods.
Fig. 2. Regional patterns of the frequency of flood process types in Austria. A frequency of unity indicates that in a catchment all the maximum annual floods are due to one particular process while a frequency of zero indicates that this process never leads to a maximum annual flood. (a) Long-rain floods, (b) short-rain floods, (c) flash floods, (d) rain-on-snow floods, and (e) snowmelt floods. For nested catchments the frequencies of the smaller catchments have been plotted on top of those of the larger catchments, and only catchments smaller than 5000 km2 are shown.(from Merz and Blöschl, 2003)
The spatial patterns of flood type occurrence will be analysed in respect to geographical regions with typical climate and landscape characteristics, such as high alpine catchments, pre-alpine catchments, North German lowlands, etc. Classification results will be used to detect linkages of flood types and circulation patterns. Regression trees or self-organizing maps (SOM) will be used to detect multi-variate linkages of flood type and catchment characteristics. Regression trees and SOMs have been shown to be able to detect non-linear relationships in large and high-dimensional data sets. They are non-parametric approaches and are able to derive relationships without prior specification of the type of relations or the variables included. By using the multivariate approaches we will scrutinize which climate and landscape characteristics can foster the occurrence of particular flood types.
In a second step, the temporal changes in flood types will be analysed. To do this, the flood time series will be divided in time periods and the occurrence of a flood types will be compared between the different time periods. This will provide insight into whether a particular flood type, e.g. flash flood due to short extreme rainfall, has become more frequent in recent times. The detection of flood type changes will shed light on the question if there is a change in flood type occurrence between flood rich and flood poor periods.
In a third step, the change in flood types from small to more extreme floods will be analysed. This analysis will identify the differences of meteorological event conditions, catchment response and routing effects of extreme floods as compared to smaller events. The analysis of the factors controlling flood type for smaller to more extreme floods will be organised similar to the analysis of the spatial dependence of catchment characteristics by multivariate approaches.
The main outcomes of the project are a consistent classification of flood events in Germany and Austria and spatio-temporal maps of flood type occurrence and changes. In general, only a few studies on event classification in respect to generating processes and event characteristics have been published yet. Even though the classification of extreme events is a topic, which is only rarely addressed in hydrology, the benefit of such process based event classification has been indicated in several hydrological publications. For example, the utilization of flood types is recommended in Germany’s guidelines for estimating flood probabilities. Experiences from the PUB Initiative show, that processes similarities of events have a massive potential (Patil and Stieglitz, 2011) for the determination of hydrological similarities as a condition for transferability of flood information across catchment boundaries.
The project results can be applied in practice to improve design flood values. Moreover, the spatio-temporal maps of flood type changes show flood risk managers which type of flood events, including typical event characteristics, such as duration, spatial coverage, flood hydrograph dynamic, can be expected in a given catchment in future. This allows new better adaption of flood risk management. On the other hand a classification of extreme hydrological events by means of formation mechanisms and event characteristics will improve understanding of hydrological processes on a catchment scale. Thus it features an important step towards the improvement of process conception or new up-scaling concepts on catchment scales in hydrology.
Merz, R., Blöschl, G. & Parajka, J. (2006) Spatio-temporal variability of event runoff coefficients. Journal of Hydrology, 331, 591-604, doi:10.1016/j.jhydrol.2006.06.008
Merz, R. & Blöschl, G. (2003) A process typology of regional floods. Water Resources Research, 39, 12; 1340 - 1359.
Gaal, L., Szolgay, J., Kohnova, S., Parajka, J., Merz, R. Viglione, A., Blöschl, G. (2012) Flood time scales – understanding the interplay of climate and catchment processes through comparative hydrology, Water Resour. Res., 48, W04511, doi:10.1029/2011WR011509
Samaniego, L., R. Kumar, and M. Zink (2013), Implications of Parameter Uncertainty on Soil Moisture Drought Analysis in Germany, J. Hydrometeorol., 14(1), 47–68, doi:10.1175/JHM-D-12-075.1.
Blöschl, G., T. Nester, J. Komma, J. Parajka, and R. A. P. Perdigão (2013), The June 2013 flood in the Upper Danube Basin, and comparisons with the 2002, 1954 and 1899 floods, Hydrol. Earth Syst. Sci., 17(12), 5197–5212, doi:10.5194/hess-17-5197-2013.
Patil, S., and M. Stieglitz (2011), Hydrologic similarity among catchments under variable flow conditions, Hydrol. Earth Syst. Sci., 15(3), 989–997, doi:10.5194/hess-15-989-2011.
Sikorska, A. E., D. Viviroli, and J. Seibert (2015), Flood-type classification in mountainous catchments using crisp and fuzzy decision trees, Water Resour. Res., doi:10.1002/2015WR017326.