Identifying Urban Water Adaptation Measures Through Detailed Process-based Modeling

The sponge city planning and implementation projects are usually highly invested projects with long-term return benefits.Success of sponge cities depends on numerous unpredictable factors such as master plans,population growth and climate changes.Thus,there is an urgent need to develop a planning method that enhances the robustness of urban water infrastructure plans and the adaptability of its implementation pathways against these uncertainties.In this study,a green-grey urban drainage planning and implementation pathway identification model was developed based on the DAnCE4Water cloud platform.Several mathematical and computational methods such as Fuzzy decision making,Rough set theory and Artificial neural network were adopted to reduce the cognitive uncertainty from decision makers.Current work also focused on optimizing the robustness of green-grey drainage co-planning,increasing the adaptability of long-term water infrastructure planning,accelerating the simulation process,and enhancing the stability of the acceleration process.Current work could be concluded as follows:1)A GIS-fuzzification process was proposed and integrated with Hierarchical Fuzzy Decision Making method for green drainage system planning.Compared with the commonly used Multi-Criteria Decision Making method,the proposed method could effectively provide equivalent decision support with 58%reduction in the amount of user-defined data,thus reducing the uncertainties from data availability and users.2)A green-grey urban drainage planning model was developed.The model was applied to Elster Creek,Melbourne to re-plan the existing drainage network under same green infrastructure planning.Evaluated with Graph Theory,the average degrees of the modelled and existing network are 1.009 and 1.029,while the modularity being 0.945 and 0.958,indicating similar topological structure;Evaluated with SWMM under a 5-year designed storm,the existing network had a total flooding volume of 558.30m³,while the modelled network being410.57m³?310.62m³ and 47.93m³,respectively designed for 1-year,2-year and 5-year return period.The flooding risk was significantly reduced,and the robustness of the cooperated system was increased.The total cost of these three modelled networks was 0.72,0.80 and 1.09 times of the existing network.3)On the basis of DAnCE4Water cloud platform,a green-gray drainage system implementation pathway generation model was proposed.The modularized model has been proved for its capacity on the exploration,evaluation and adaptability optimization of multi-strategy and multi-objective urban water infrastructure implementation pathways.By applying the proposed method,the drainage system implementation pathways of Scotchman's Creek,Melbourne for the period of 2015-2035 was designed and optimized with 2.93 million simulations.4)An exploration acceleration module based on artificial neural network is designed and applied in the above case study,which saved about 80%of the total simulation time.An error-distribution based method was proposed to improve the stability of the acceleration process.With same amount of data for training and testing,the prediction error?Root mean square error,RMSE?was 9.80.With octupling the testing data volume,the error?RMSE?only increased by7.9%.The proposed method has an excellent prediction accuracy for scenarios with only green infrastructure updates,with the error ranging from-1.26%to-0.22%,and a minor inaccuracy for grey infrastructure updates,with the error ranging from 5.56%to 14.95%.5)A dynamic acceleration exploration module based on the rough set theory was also designed and tested in the current study.A parameter named‘Significance?Sig?'was introduced in this module to dynamically control the accuracy of the machine learning process and offer probability output of predicted results.The total time-saving capacity increased from 16.45%to 82.47%,the global RMSE increased from 6.63 to 17.52,with the learning accuracy decreased(Sig_min decreases from 0.5 to 0.25).The relative standard deviations of the time-saving capacity and the global RMSE were 0.73%and 1.82%respectively,when tested with a randomly unstable time sequence of the input data,which indicated a high stability of the acceleration process.The prediction accuracy for the scenarios with only grey infrastructures was excellent,with the error ranging from 0.00%to 0.05%.For the prediction of green infrastructure performance,there is a high but stable background noise,?RMSE?ranging from 16.29 to 19.18.Current research results provide an effective technical means and decision-making platform for improving the long-term adaptability of sponge city planning and infrastructure implementation pathways.