Urban Rainstorm Waterlogging Risk Identification,Prediction,and Emergency Resource Dispatching

Urban rainstorm waterlogging disasters have characteristics of dramatic devastation,profound impact,and suddenness,and they may be caused as a result of extreme weather events such as typhoons and rainstorms.It may also lead to secondary and derivative disasters such as dangerous house collapse,landslides,water backflow,and post-disaster plague.The formation mechanism of urban rainstorm waterlogging is complex and affected by many factors.Besides climate and meteorological conditions,it is affected by topography,surface,drainage network,urban roads,buildings,structures,etc.Therefore,the questions of how to improve the accuracy of waterlogging risk identification and prediction,accurately extract the high-risk areas of waterlogging,detect the trend of waterlogging disaster in either a short term or a long term,and provide decision supports for emergency dispatching are challenging and crucial,which are the focus of this research.To address these issues,this thesis first summarizes the main factors affecting the spatial distribution of waterlogging risk by considering the development rules of urban rainstorm waterlogging disasters and proposes a waterlogging risk identification method(MDF-H)that combines the multi-source data fusion with hydrodynamics methods.By constructing a multi-source data layer and dividing the total study area,sub-catchment,and grid into three calculation levels,a total of 12 influencing factors(i.e.,rainfall intensity,rainfall duration,rainfall interval,historical average monthly rainfall statistics,soil erosion ratio,digital elevation model(DEM),building height,surface slope,land cover type,rainwater grate,drainage network,urban roads)are considered as model variables.Based on the selected 12 variables,we propose a two-layer probabilistic assessment algorithm of the drainage capacity with real-time integrated runoff coefficient for large-scale urban areas.The applicability of the method to the large-scale study area and the ability to identify the scope and depth of the waterlogging areas are verified by a case study in Shenzhen.To improve the accuracy of the short-term waterlogging prediction,this research proposes a multi-strategy-mode urban waterlogging prediction framework(MSMWP)to obtain the risk identification of waterlogging areas.Based on the water depth data from urban flooding monitoring stations and rainfall data from meteorological stations,the comprehensive steps for data preprocessing,model establishment,strategy consideration,and regression algorithm settings are investigated.The optimal parameter configuration is determined by Ground Truth Data.The accuracy of the model for short-term prediction of waterlogging depth in the next 15 minutes exceeded 93%,which can provide a certain time for authorities to make decisions.In response to the lack of real-time waterlogging depth data at waterlogging points or the prediction of medium-term waterlogging,this research proposes a method for the medium-term prediction of waterlogging depth based on feature extraction and transfer at waterlogging areas.A method is proposed to predict the depth of inundation based on the transfer of inundation point feature extraction.To be specific,five statistical features of rainfall data(i.e.,the mean,maximum value,standard deviation,kurtosis,and skewness using statistical analysis and domain knowledge)are extracted,and four features of rainfall data(i.e.,rainfall intensity,seasonality,rainfall interval,and soil infiltration capacity)are considered.Based on the feature similarity matching algorithm,the model can migrate its prediction capability and provide reference suggestions for identifying similar flood-prone sites for installing new waterlogging sensor sites.The model enables the prediction of waterlogging depth for long future prediction periods with an average absolute error of approximately 2 cm,which improves prediction accuracy by a factor of 2.5 to 3compared to the featureless time series prediction method.When a waterlogging disaster occurs in a mega city,emergency measures must be adopted according to the dynamic situation of the waterlogging disaster.Accurate waterlogging risk identification and prediction results could provide support through producing pre-designed emergency dispatching plans and emergency resource dispatching on-site.This research proposes a dynamic dispatching method for an emergency response to disaster situations.Based on the classical A-star algorithm,the Driving Preferences-A-star(DP-A-star)shortest path algorithm,which can select the exercise route preference,is proposed to improve the calculation efficiency of the optimal emergency dispatching path.A dynamic risk assessment is conducted to evaluate a disaster’s risk and emergency response level,according to the prediction results of waterlogging depth at the disaster sites and population-density distribution.A response configuration scheme is set for each response level,enabling the more accurate and efficient dynamic deployment of emergency vehicles and personnel in each dispatch sub-period.The computation time of emergency resource dispatching is over 30% less than classical A-star and Dijkstra algorithms.