An urban flood risk assessment method using the Bayesian Network approach

Abstract

Flooding is one of the most damaging natural hazards to human societies. Recent decades have shown that flooding constitutes major threats worldwide, and due to anticipated climate change the occurrence of damaging flood events is expected to increase. Urban areas are especially vulnerable to flooding, because these areas comprise large amounts of valuable assets. Flooding in urban areas can grow into significant disruptions and national threats unless appropriate flood risk management (FRM) plans are developed and timely adaptation options are implemented. FRM is a well-established process that aims to keep flood risk at, or reduce flood risk to, an acceptable level in flood prone areas. According to IPCC’s Summary for policy-makers (2014), risk management is an iterative process that is divided into 3 phases, which in this thesis are adapted to fit FRM terminology. Hence, FRM includes flood risk scoping, flood risk assessment (FRA), and adaptation implementation and involves an ongoing process of assessment, reassessment, and response. This thesis mainly focuses on the FRA phase of FRM. FRA includes hazard analysis and impact assessment (combined called a risk analysis), adaptation identification and adaptation assessment. The main task of FRA is to combine these assessments in a robust and systematic manner to provide valuable information to decision-makers by identifying suitable adaptation options and developing feasible adaptation strategies. In this study, a FRA method using the Bayesian Network (BN) approach is developed, and the method is exemplified in an urban catchment. BNs have become an increasingly popular method for describing complex systems and aiding decision-making under uncertainty. In environmental management, BNs have mainly been utilized in ecological assessments and water resources management studies, whereas climate risk studies have not yet fully adapted the BN method. A BN is a graphical model that utilizes causal relationships to describe the overall system where risk occurs. A BN can be further extended into a Bayesian Influence diagram (ID) by including decision and utility nodes, which are beneficial in decision-making problems. This thesis aims at addressing four specific challenges identified in FRA and showing how these challenges may be addressed using an ID. Firstly, this thesis presents how an ID can be utilized to describe the temporal dimension of flood risk in a coherent and systematic manner. Herein, risk is assessed in so called time slices, where each time slice represents one specific year. For each time slice, separate hazard analyses are conducted to assess the occurrence probability of hazards in that specific year. Time slices are connected with each other by connecting the adaptation nodes in the time slices. Secondly, this thesis recognizes the need for including a spatial dimension in FRA. An urban catchment is rarely homogenous, and there are areas that have a higher risk than others. From a decision-making point of view, a spatial risk profile may provide valuable insight in where risk is higher than acceptable and where additional adaptation measures are needed to keep risk at an acceptable level. In an ID, the urban catchment can be divided into subregions, and risk is described for each sub-region separately. Thirdly, the objective is to improve FRA by including multiple hazards caused by concurrent events. Concurrent events refer to two or more flood hazards that occur simultaneously. In such circumstances the hazards may interact, and total damage from such a concurrent event may be larger than for the hazards separately. Currently, FRA is mainly based on single hazard events, but with expected climate change impacts there may be a need to include several hazards into FRA to assure that risk is described correctly for identification of important adaptation. This thesis shows that IDs may serve as a good approach for inclusion of multiple hazards in FRAs. Lastly, the inclusion of multiple hazards in FRA may be challenging, among others because concurrent events are rare. However, with climate change, the annual variation of hazards may change, and concurrent events may become more frequent. Large-scale atmospheric circulation influences local and regional climate and is considered an important factor when aiming at improving our understanding of local weather conditions and the occurrence of extreme events. Hence, this thesis presents a study that explores the relationship between flood generating hazards and large-scale atmospheric circulation. This thesis concludes that IDs can serve as a good approach for describing the complex system in which flood risk occurs. The final product is a spatiotemporal FRA approach that can include the impacts from multiple hazards.