Application Of Bayesian Network In Flood Risk Assessment Under Non-stationary Condition

In recent years,the occurrence of flood disasters has become more frequent under non-stationary conditions(e.g.,climate change and urbanization).It is urgent to manage urban flooding given the large number of social and economic problems induced by urban flood disasters.To alleviate the negative impacts of urban flooding,it is necessary to improve the assessment and optimization of flood management and adaptation measures.In this paper,using a portion of Z city as case study,the hydrodynamic model of the drainage network was established by applying tools such as the Geographic Information System(Arc GIS)and Storm Water Management Model(SWMM).The Bayesian network(BN)analysis was conducted with the support from software Bayes Server to establish a model for BN-based flood risk assessment.The details are as follows:(1)First,current research status of flood risk assessment and BN analysis were reviewed to explore the shortcomings of current flood simulation and assessment frameworks and tools,the advances of BN in urban flood risk assessment,built upon which the research content of this paper was established.(2)Flood adaptation measures were studied and evaluated by integrating different regional hydrology,water resources characteristic and related socio-economic and technical analysis.The design and selection of appropriate measures are based on national and local regulations and conditions,using the cost-effective principles.Studies on related SWMM model,flood risk assessment tools and BN applciation tehories and tools were carried out to provide theoretical support for the construction and use of BN-based flood risk assessment analysis.(3)Regarding flood risk assessment,the effect of pipe network enlargement was better than that of LID,in which the effect of green roof was better than that of biological retention.At the same time,it is shown that the peak of flood risk curve has shifted towards larger return periods after implementing the pipe enlargement measures.In terms of costs,the Bio-retention cell was more expensive than that of the Green roofbecause of the complexity in constructing the Bio retention facilities.At the same time,it could be found that the cost of the pipe network was the highest among all the investigated measures,but the correspoding cost to meet the 10-year design return period was however the smallest.Considering both the risk(i.e.,annual damage)and the cost of flood,the results show that the Green roof was better than the Bio-retention cell.For the10-year design standard,pipe enlargement was the best.(4)Evaluation model based on BN could visually reflect the degree of preference of different stakeholders and the related uncertainties and occurrence of each parameter.The Bayesian network-based decision-making platform turned out to be more transparent and conducive to facilitate the dialogue between decision-makers in flood risk assessment.Taken this study for example,the cost-effective rate μ(Mean)is 3.02 and theσ2(Variance)value is 5.3 when taking into account the prior probabilities of the nodes.At the same time,the Bayes Server tool can automatically update that the probability of occurrence of flood risk(EAD)above 1.5 reaches up to a maximum of 30%.(5)BN can realize the simulation and comparative analysis of a variety of irreversible flood risk assessment strategies;it could help the decision makers to reduce the biased investments in drainage design and construction by comparing and optimizing different adaptation scenarios considering different preferences to save economic investments and resources.For example,when the adaptation investments are constrained within the range of 0.2-0.3 billion,it is found that that the most sensitive parameters to the final cost-effective rate are the comprehensive quota,5-year and10-year pipe enlargement measures...