The case study aims to improve estimation of 100-year return values of river discharge at the target location, exploring how this will be affected by the climate change for the Danube River in Bratislava, Slovakia.
Using the existing long-time observations of river discharge, the 100-year return values are estimated using the non-stationary frequency analysis. The same approach is used for the future period (2015‒2100), making use of simulated discharge series as an impact indicator from the Pan-European data sets. The ratio of the estimates of the 100-year flood return level based on future simulations vs. past observations (CCQ100) is the indicator of the climate change.
The largest increase in the 100-year flood is +36%, while the largest decrease is estimated as to -33%. The results highly depend on the particular hydrological model, and such a stratification of the outcomes is important from the perspective of the client.
Case Study Description
Flood frequency analysis is a part of the operational services of the Slovak Hydrometeorological Institute. So far, estimation of T-year floods has generally been based on local data (at-site approach), using the principle of the stationarity of the environment, and analyzing annual or seasonal maxima. While in the last couple of years, regional approaches to a flood frequency analysis have been successfully adopted, no methods for accounting for the non-stationarity of the environment were developed and implemented in the practice. Similarly, little efforts have been done towards implementing the peaks-over-threshold methodology (where all independent events exceeding a pre-defined threshold are taken into account), which is generally considered as a more rigorous statistical approach when compared to the traditional block maxima method.
The case study, therefore, aims at developing methods for flood frequency estimation methods where the non-stationarity of the environment is accounted for, and the data are analyzed using the peaks-over-threshold methodology.
The case study is expected to reveal information on how the frequency of floods will increase/decrease, and how the number of warning situations will increase/decrease. As a long-term goal, improvements to warning strategies are expected. The outputs should serve as input for decision making regarding residence policy and investment into flood control structures.
The working hypothesis is that the improved statistical methods of frequency analysis with the combination of the SWICCA climate impact indicator will reveal more insight into the expected behavior floods with low probability of occurrence, and this knowledge might be transformed into flood management and adaptation plans.
The boundary banks of the Danube River in Bratislava are designed on the basis of the discharge corresponding to return period of 100 years (Q100), and are constructed with a sufficient reserve (reliability) to resist against even larger floods. The prevention of the city against Q100 is ensured generally by the firm boundary banks of the river, and at some places, mostly in the city center, by mobile barrages. In the year 2013, Bratislava experienced an extraordinary event corresponding to a magnitude of about a 100-year flood, and this pointed at the potential weaknesses of the flood prevention system: at some locations there was a very tiny reserve at the banks and the barrages. It is thought that at the most vulnerable locations, the height of the mobile barrages should be increased by 20-30 cm, while at the weakest points of the boundary banks, it would be enough to build higher walls of sandbags in the case of need. The outcomes of our local project will serve in re-evaluation of these hypotheses – they may be used in assessing the degree of the reliability of the current system and possible strengthening/increasing the dam heights.
In the project we only make use of a single impact indicator, namely time series of river discharges for the target location. We have at hand a plenty of different time series to derive the desired flood statistics Q100. The information on Q100 on the basis of 33 scenarios together with the knowledge on Q100 from the past decades are useful at least from two aspects: (i) from the qualitative point of view, i.e., one can see what percentage of the scenarios yield considerable increase/decrease/no change, and (ii) from the quantitative point of view, i.e., one can assess the recent status of the flood prevention system, both in the light of the worst and the best scenarios. The largest values of CCQ100 (i.e., the ratio of Q100,future / Q100,past) may directly and indirectly indicate the amount of necessary investments (financial, material, logistical, political etc.) into the flood prevention system. On the other hand, even the best scenarios (cases with CCQ100 < 0.95) may convey valuable information. In this case the buildings that have been constructed on the basis of ‘old’ estimates of Q100 would not need to be rebuilt, they may be declared as flood safe, eventually as protected against Q1000. Furthermore, some of the new constructions (dams, bridges) will have lower costs of realization and running expenses.
For the Slovak Hydrometeorological Institute, the temporal and strategic aspects of the problem are very important. Flood prevention arrangements may be implemented relatively quickly, within the next 5 years. Plans for long-term investments and for flood protection, with regard to the results of the evaluation of the case study, will be analyzed and implemented gradually in the horizon of 20 to 50 years.
The analyses are only carried out for the target site of Bratislava, thus this is a single-site frequency analysis with no spatial aspects.
During the duration of the SWICCA project (from November 2015 till the date of the current report, August, 2017), there were 3-4 face to face meetings with the client at their offices. These consultations were not scheduled on a regular basis, but they were organized as soon as any new partial results appeared or it was necessary to discuss some specific problems (e.g., identification of independent flood events) in detail. Otherwise we kept informing each other in the form of e-mails, and discussed the problems/questions immediately as they emerged.
The communication with the client went smoothly, without any problems, and the client was very enthusiastic during the whole period. They pointed out that the most important benefits of this co-operation were as follows:
‘It is very positive that we kept organizing meetings. New ideas came out how one can make use of the results of the case study in practice and what consequences they can bring.’
‘We really appreciate the way of communication “purveyor vs. client” that was adopted within the SWICCA project. Such a form of communication is a kind of a novelty for us since it is perhaps the first time that a company or a research institute (namely the MicroStep-MIS) comes to solve certain problems hand in hand with us, people from the practice. We consider it progressive. Generally, researchers from the Academy or the University of Technology deal with a number of hydrology related projects; however, only a minority of their outcomes is made use in the practice. The concept of SWICCA indeed bridges this gap.’
The results obtained so far are fascinating and there is a good potential to extend the study to further sites of Slovakia, for instance within a framework of a domestic research project. The main focus could be then given to smaller catchments in different parts of the country that might be more vulnerable to the climate change like the extremely robust Danube River.
River discharge values are used from the pan-European datasets. It is a point analysis, focused at the target site of Bratislava, at the Danube River (approximately 48.14 lat. and 17.10 long.). The desired time resolution of the data is 1-day, particularly in order to derive peaks-over-threshold data. The simulated data covers the period until the end of the 21st century.
In the case study, long time observations of the river discharge at the Bratislava station are used. These data are readily available at the client’s database.
Step 1a: Downloading impact indicator of river discharge – Future simulations of river discharge (in daily time resolution) are downloaded from the Pan-European databases for the target site, which is Bratislava, the capital city of Slovakia. Only a point estimation is carried out. The overall goal is to estimate return values corresponding to 100 years.
Step 2a: Extracting annual maxima and peaks-over-threshold data – Long time series of daily discharge are necessary to derive reliable data sets of annual maxima and peaks-over threshold for the flood frequency analysis in Step 3a.
Step 3a: Non-stationary flood frequency estimation – Flood frequency estimation is carried out based on different data sets (block maxima vs. peaks-over threshold) and on the basis of different approaches (stationary vs. non-stationary assumption).
Step 1b: Getting the observed time series of river discharge – The task in this step is similar to that in Step 1a. The only difference is that return values corresponding to 100 years are estimated on the basis of the observed time series of river discharges. The local data cover several decades in the past, and are readily available at the client’s database.
Step 2b: Extracting annual maxima and peaks-over-threshold data – Step 2b is identical with Step 2a above.
Step 3b: Non-stationary flood frequency estimation – Step 3b is identical with Step 3a above.
Step 4: Climate change indicator – The climate change indicator is simply the ratio of the estimates of the 100-year flood return level based on the future simulations vs. on the past observations.
Step 5: Plan of action – The analysis is expected to reveal information on how the frequency of floods will increase/decrease, and how the number of warning situations will increase/decrease. In a long-term goal, improvement of warning strategy is expected. The outputs should serve as input for decision making in residence policy and investment into flood control structures.
Lessons learnt during the workflow process:
Step 1b: One of the limiting factors of the analysis stems in the shortness of the observed data series (1984‒2014). This fact is directly represented by considerably wide confidence intervals of the return level estimates.
Step 1a: First, shortness of the observed data series also influenced the definition of the common period to derive the statistical characteristics of the observed and the modelled data for bias correction. We had to restrict ourselves to a period of a length of 17 years (1984‒2000). Second, the selected method of bias correction might have influenced the outcomes. Since we are focusing on extremes, it may be more rigorous (and more time demanding, though) to apply a bias correction method based on the similarity of the empirical distribution functions.
Step 2: Subjective decisions cannot be completely eliminated in determination of the threshold value in the peaks-over methodology. Even though we attempted to select the threshold on the basis of objective criteria (graphical tools), there were still a number of plausible and statistically correct alternatives to the threshold selection. The topic need further investigation in the future.
Step 3: Estimates of the 100-year flood Q100 are associated and displayed together with their 90% confidence intervals (CIs). It is straightforward to interpret CIs in the case of the observed data since the uncertainty is only related to the size data sample and the adopted statistical approach. The sample represents real data with no further uncertainty. Nevertheless, CIs are really hard to interpret in the case of the future data sets since the estimates are not based on REAL data: there is a great portion of uncertainty related to the corresponding emission scenarios, climate and hydrological models, and all of these are manifested somehow in the resulting time series.
Step 4: The most important findings when analyzing the final results of the case study are:
- There are no major differences in the overall results from the aspect of the data sample selection (i.e., block maxima vs. peaks-over-threshold methods).
- Increases dominate in case of two hydrological models (HYPE and Lisflood).
- The VIC model only yields decrease in the estimate of the 100-year flood
- The overall performance of the data sets are rather balanced:
- increase / no change / decrease appears in 14 / 7 / 12 cases for the AMS/GEV method, and
- increase / no change / decrease appears in 14 / 6 / 13 cases for the POT/GPA method.
- Overall, one can expect that the change in 100-year flood can reach ~1/3 of its value in both directions.
It is, however, important that the above presented (objective) findings should be supplemented with some subjective comments from the side of the client. Their representative (Dr. D. Lešková) raised some concerns related to the performance of the hydrological model VIC. It is suspicious the estimates of the 100-year flood based on the VIC outputs are the lowest one overall. This induces questions whether the VIC model is appropriate for modelling river discharges in the Central European settings, particularly for the Danube River, which, despite its large size, still behaves as a mountain type river in Bratislava. Deeper and more detailed study of the VIC model structure is necessary in the near future to decide on the questions raised during our consultations, and before pushing forward the final results of the case study to the local policy makers. As Dr. Lešková further stressed, they have no similar objections against the other two hydrological models since they are familiar with them. The HYPE model, or more precisely, the HBV model is used in the operational practice at the Slovak Hydrometeorological Institute, while the Lisflood model is also known to them since it is used within the international project EFAS (www.efas.eu).
Improved methodology of data selection. Traditional methods frequency analysis have predominantly been based on the selection of the largest flood (the least favorable flood situation) during each year/season. Using such a data sample, flood quantiles corresponding to return periods of 100, 1000 etc. can be estimated, using relatively simple statistical methods. This approach, however, has some drawbacks. First, the sample of annual flood maxima is affected by the years, during which no considerable floods occurred, and secondly, in the case of years with multitude of floods, one has to ignore the second, third etc. largest flood. These drawbacks of a traditional flood frequency analysis are overcome by the peaks-over-threshold methodology where the data samples are built on the basis of all floods exceeding a pre-defined threshold.
A more rigorous way to consider the climate change. The Slovak legislation prescribes that the effects of the climate change have also be considered in estimation of the T-year flood return levels. Hydrological studies in Slovakia try to follow these recommendations, but it can be generalized that they stick to a simple adoption of a trend analysis of the observed (past) data and extending the trends towards the future decades. Such an approach may be highly questionable when trends in extremes are examined. SWICCA offers a new platform for dealing with similar problems. A number of simulated data from different climate model runs (which a priori reflect the climate change through different greenhouse gas emission scenarios) and hydrological models can be accessed via the SWICCA Demonstrator. The researcher only has to adopt the same statistical tools to these data as in the case of the observed ones. A comparison of the results based on the past vs. future data sets may indicate changes or trends which are not perceived recently but should be considered in hydrological / climatological studies and engineering applications.
Costs of flood prevention. Increased frequency of floods does not necessarily imply increased volumes of individual flood events. Nevertheless, it may indicate enhanced cost of flood prevention and work supply during floods (monitoring, dispatching, flood forecasting services etc.), and at the same time, considerably lower costs to cover the flood damages. Flood prevention cost are estimated as to 2% of the total cost related to the floods (damages on the properties of state, general public, agriculture etc.).
Cross-border aspects. Protection of Bratislava the capital city of Slovakia that lies along the stream of the Danube River has an international, cross-border aspect. It is recently governed by the Border Waters Commission. The Slovak-Austrian Commission of Border Waters is responsible for the changes in the water regime, for the possible threats of the countries due to unfavorable impacts of floods, for water management during dry periods etc. A similar commission has been operating between Slovakia and Hungary.
Ignoring the preparations for the consequences of the anticipated climate change means not being prepared for possible disastrous floods and the related losses of lives and extraordinary damages to the properties of state, municipalities, businessmen and general public. On the other hand, preparedness for the expected climate change yields optimal (with a given accepted risk) measures of flood protection: mobile barrages, firm dams/dikes, ensuring protection or relocation of objects/people in threat, protection of historical objects etc. Cost savings and the economic benefits as a result of the adaptation measures may be quantified on the basis of flood risk and vulnerability of the target site(s) and plans for management of flood risk.
The results of the current case study may indicate whether the constructions for flood prevention were well designed, overdesigned or underdesigned. On the other hand, it is hard to judge, whether they were designed in a correct way.
‘No doubts, we have to deal with the climate change adaptations since we need to protect property of the state and general public, people’s lives, cultural heritage etc. On the other hand, there are also some points against the adaptations... It seems to me that these are only figures or technical solutions which are purely adopted without any regard to the hydrological processes in the nature. In my opinion, it would be much wiser (and perhaps cheaper) to move away a bit from the streams and not to build artificial barriers and toboggans for the rivers ... better to live in harmony with the nature instead of endless fighting against the natural phenomena such as floods. Nevertheless, these are rather philosophical and predominantly political questions, and not hydrological ones.’ Dr. D. Lešková, representative of the client.
The results obtained within the framework of our local project will have consequences in decision making at different levels:
First, the results will be reported to the Slovak Water Management Enterprise. This body is the caretaker of the Danube River, and it is responsible for any technical arrangements in flood prevention. The estimated discharge values with the return periods of 100 (and possibly even 1000) years will be transformed to water levels on the basis of the stage discharge relation curve. The new, critical water levels of Danube will reveal whether the current degree of flood prevention of the capital city of Bratislava is sufficient or it is necessary to re-define them.
Second, the Department of Water at the Ministry of Environment of the Slovak Republic will be informed about the results of the analysis. This department as a governing unit is responsible for the co-ordination of the activities during any flooding event in Slovakia.
Third, since Danube is a boundary stream between Slovakia and Austria, a consultation with the representatives of this neighboring country is expected. On the annual, bi-lateral meeting of The Slovak-Austrian Commission of Border Waters, the finding of the case study will be presented and the willingness of the Austrian colleagues and institutions to re-evaluate their flood prevention adaptations will be discussed.
Ladislav Gaál, Ph.D.
Relevant EU Policy
Purveyor: Ladislav Gaál, Ph.D.
Client: Slovak Hydrometeorological Institute
Enhanced level of the Danube River in Bratislava during the last serious flooding situation in Bratislava (Photos by Ladislav Gaál, June 5th, 2013)