Integrating empirical data into the quantification of human-flood interactions

Abstract

In flood risk management the aim is to reduce the risk, either through reducing the frequency of flooding (e.g. by constructing levees), through reducing exposure (e.g. through flood zoning) or by reducing the vulnerability of the objects and people at risk (e.g. by taking precautionary measures at the property level). Besides their clear positive effects, these actions may have unintended consequences that may often result in unexpected phenomena due to feedbacks in the system. Socio-hydrology tries to understand and explain the phenomena that result from long-term interactions between humans and water by studying and modelling the human-flood system as a whole. While research so far has shown the potential of these models in exploring the behaviour of human-flood systems, there is a lack of integration of empirical data in socio-hydrological modelling of human-flood interactions.The aim of this thesis is to build upon, combine and extend existing methods to integrate empirical data in the quantification of human-flood interactions. In Chapter 2, the state of the art is presented through a literature review on observed phenomena in human-flood systems and the models that have been used to investigate these phenomena. The existing models and their applications show great potential for flood risk management from adopting an approach that considers the feedback loops in the system. A dynamic approach with coupled human-flood models can help explore the different dynamics of risk at varying points in time and space that may arise because of the interplay between the system components.In Chapter 3, a case study of the seven regions of England (as defined by the Environment Agency) is presented. Survey data collected between 1996 and 2004 are used to investigate the relationship between prior experience of flooding and public preparedness to flooding. It is found that in some regions, there are greater increases in preparedness with increasing experience of flooding than in other regions. By analysis of additional data sources, an explanation is provided that may account for the differences. The severity of the impact of flooding, when it occurs, appears to positively influence public preparedness for future flooding. The presence of structural defences seems to negatively influence the preparedness, because this leads to people feeling protected and feeling less need to further prepare for future flooding. In addition, the fact that residents do not feel responsible for flood risk mitigation, but rather feel that the government should take more measures, appears to negatively influence the preparedness as well.In Chapter 4, the adaptation effect (i.e. when frequent flooding results in increased preparedness and reduced vulnerability) is investigated in Dresden, Germany, using a model that captures the feedbacks in this human-flood system. The parameters of this model are estimated with empirical data using Bayesian Inference. This approach allows for obtaining reasonably accurate estimates, even though the data are rather uncertain. A sensitivity analysis shows that, in general, if data is available for the end of the investigated and modelled time period, it is most valuable and results in less bias in the estimated parameters. Furthermore, data on flood awareness are shown to be the most important data for correctly estimating the parameters of this model and accurately capturing the system dynamics. Importantly, it is shown that using additional data for the other variables cannot compensate for the absence of awareness data. The approach in this chapter demonstrates that combining socio-hydrological modelling and empirical data gives additional insights into the human-flood system, such as quantifying the forgetfulness of the society, which would otherwise not be easily achieved by socio-hydrological models without data or by standard statistical analysis of empirical data.In Chapter 5, a framework is developed for the comparison of long term flood risk in human-flood systems. Using an adapted version of the model developed for Dresden and the Elbe, hypothetical human-flood systems are generated and placed in the framework. The application of the framework suggests that the long-term behaviour of human-flood systems is not influenced by individual societal characteristics, like the forgetfulness or risk taking behaviour of a system. The long term flood risk is more influenced by the shape of the flood frequency curve, i.e. whether it is quite skewed and floods occur unexpectedly to the inhabitants or whether it is less skewed and therefore society regularly experiences flooding of similar magnitudes. The behaviour of two real world systems, Dresden (on the River Elbe) and Cologne (on the river Rhine) is compared. This demonstrates that the system of Dresden is more prone to experiencing shocks and therefore the long term flood loss is driven by hydrology, whereas long term flood risk in Cologne is more influenced by the amount of exposure and preparedness. In this way, the framework allows for contrasting and comparing these empirical systems and can contribute to the understanding of why these systems behave in a certain way.The findings of this thesis result in a number of recommendations for socio-hydrological model building, data collection and flood risk management. Bayesian Inference is recommended for use in socio-hydrological modelling, since it enables the inclusion of the different types of data and information that is required for understanding human-flood systems (i.e. both quantitative and qualitative data with varying uncertainties). Hierarchical models are useful for the analysis of data for multiple regions or systems at the same time in a way that reduces the uncertainty of estimations, which partly solves the problem of data scarcity. Regarding data collection, the analyses have shown that information on awareness and preparedness may often be more important than damage data. The analyses in this thesis have demonstrated that it would be very useful if this data would be available more consistently in both time and space and strategies to collect such data regularly and consistently should be explored (e.g. through census data collection). Socio-hydrological models have great potential to contribute to flood risk management. These type of models are not predictive, but they can assist in exploring the possibility space. Human-flood models supported with empirical data enables particular systems to be located in this possibility space and placed in a global context. Differences in the hydrological and societal setting can be identified and relevant adjustments can be made. This makes it possible to explore the potential trajectories that are specific for a system and infer possible future dynamics of flood risk based on those observed elsewhere in the past.