Veranstaltungsprogramm

The existing urban drainage infrastructure, which mainly consists of central solutions, has been extended in the last decades by decentralised system applications such as nature-based solutions (NBS) and is currently an area of greatest interest of the research community.

To further expand the knowledge on this topic, the "Combat of retrofitting Urban Drainage Networks with NBS" is organised at the UDM 2025. The aim of the combat is to optimise the retrofitting of an existing urban drainage network with seven different types of nature-based solutions. Therefore, a calibrated combined sewer network of an Alpine municipality in Austria is provided as an SWMM5 input file, together with implementation rules (e.g., type of nature-based solution is depending on land classification).

The submitted solutions will be evaluated based on seven performance indicators, including costs, biodiversity and metrics related to system performance improvements. The best teams are honoured during the closing ceremony, and the submitted solutions and the task will be published in a joint publication of all participants of the combat to support the dissemination of the case study as a benchmark network in future.

The UNIPV team proposes a multi-step methodology to optimize locations of nature-based solutions (NBSs) in the case study network of the Combat. Since the running time of the one-year long simulation of the case study network has order of magnitude of some hours, the first step of the methodology concerns simplification of the problem for the reduction of computational times. The second step of the methodology consists of the multi-objective genetic optimization of NBS locations, considering a pair of objective functions at a time, i.e. i) cost-flooding, ii) cost-biodiversity of NBS solutions, iii) cost-evaporation, iv) cost-water volumes addressed to the treatment plant, v) cost-water volumes discharged into the environment through the combined sewer overflow (CSO) device and vi) cost-pollutant mass discharged into the environment through the CSO device. This approach represents the simplification of an optimization in which all the seven objective functions are dealt with at the same time. The third and last step of the methodology makes use of optimization algorithms and manual refinements to obtain a single solution that compromises the solutions obtained in the second step of the methodology. After being tested against the constraints enforced by the Combat Organizers, this solution will be finally submitted.

Team name: Sewer or Later Prabhat Joshi, Matthew McGauley, Ricardo Reyes Sotomayor, Fabrizia Fappiano, Giovan Battista Cavadini The proposed approach for retrofitting urban drainage networks with nature-based solutions (NBS) consists of two key steps. In the first step, potential NBS implementation solutions are filtered and constrained using the objective functions for cost and biodiversity, as these objectives are independent of SWMM simulation outputs. This pre-selection ensures that only feasible combinations, which adhere to cost constraints and incorporate diverse NBS elements that enhance biodiversity, are considered. In the second step, optimization is performed by integrating the Python packages PySWMM (McDonnell et al., 2020) and Pymoo (Blank and Deb, 2020). This integration allows for optimizing the placement and extent of NBS elements based on SWMM simulation outputs and the defined objective functions. PySWMM enables dynamic simulation of SWMM input files within Python, while Pymoo supports the use of multi-objective evolutionary algorithms, for instance the NSGA-II algorithm (Deb et al., 2002). This method has been effectively utilized for SWMM model calibration (Rodriguez et al., 2024) and for cost optimization of NbS in flood prevention (Ur Rehman et al., 2024); however, it has yet to be applied for optimizing multiple objectives of NbS simultaneously. This method would offer a framework for balancing stormwater management goals with cost efficiency and biodiversity enhancement in urban environments, by selecting the solution that maximizes the objective functions.

We propose a multi-phase method to effectively optimise the location of nature-based solutions (NBS). During Phase 1, the largely branched drainage network will be partitioned into sections using Louvain algorithms for graph partitioning. The sensitivity of the system to NBS implementation in defined sectors is evaluated through automatic manipulation of the rainfall data for those sections, mimicking the hydrodynamic effects of the NBS on the overall system. This is done for the individual sectors and for all possible combinations of sectors, using the most impactful rainfall events on the objective function. The sensitivities of each sector will be ranked and used to steer the next optimisation in Phase 2. Here, a computationally efficient metamodel, which can mimic the flooding behaviour and uses the urban hydrology as an input will be deployed using transfer learning. The metamodel will be combined with a custom genetic algorithm: the optimisation-parameters are skewed towards the most impactful sectors and interventions, using a biased-random sampling based on the ranking of the partitioned sectors. The optimisation function used will include normalisation and weighting of the multi-objectives of the combat. Periodically, the highest scored solutions are evaluated with the SWMM-model to validate the proposed solutions.

Our approach begins by exploring various LID implementations and combinations at the individual subcatchment (SC) level. Performance is evaluated using peak runoff and evapotranspiration (ET). Using the gained knowledge, cost-efficient and inefficient configurations for each SC are identified, by varying the implementation area for green and road LIDs. Efficiency is assessed based on achieved peak runoff reduction or ET increase per euro of investment for the investigated configurations. This produces a catalogue of viable configurations and reduces the range of possible solutions significantly. To identify good global arrangements, suitable LID types and strategies for their placements are identified for each objective. The budget is distributed to optimize the biodiversity objective. For flood reduction, flooding nodes are identified in the reference scenario. LIDs are then placed in the vicinity of the identified flooding nodes. To reduce CSO volume and loading, LIDs are placed based on the estimated flow duration of each SC to the CSO, targeting SCs, that contribute to peak flow. Finally, leveraging this knowledge, we conduct event-scale simulations with random variations of implementations, using Pareto optimization to evaluate multi-criteria performance indicators. The variant with the best overall performance is then selected.

Our intended approach for the “Combat of Retrofitting Urban Drainage Networks with Nature-Based Solutions” leverages our AI-powered web application, Ziggurat (www.ziggurat.ai). Ziggurat automatically designs and optimizes sustainable urban drainage systems by integrating advanced algorithms from sewer system hydraulics, graph theory, mathematical optimization and artificial intelligence. It considers a comprehensive set of optimization variables, including network layout, pipe sizing, pumping, storage facilities, and the incorporation of various Low Impact Development (LID) measures. To handle the complexity of continuous rainfall series and the large-scale problem setting, we apply several preprocessing strategies. First, we extract a reduced set of critical rainfall events by analyzing the full time series to identify only the most relevant extreme events. This significantly decreases runtime while preserving the accuracy of the optimization objectives. Next, we employ subcatchment clustering to reduce the dimensionality associated with distinct drainage areas, streamlining the search space for solution candidates. Additionally, we replaces time-consuming rainfall-runoff simulations with data-driven emulators that offer rapid yet reliable performance estimates for different NBS scenarios. By combining these approaches, our team aims to efficiently identify an optimal configuration of nature-based solutions that meets the combat’s objectives of improving drainage performance, mitigating flooding, and enhancing multiple co-benefits.

This method seeks a local optimum solution for retrofitting the urban drainage network ‎under computational constraints. The simplifications used in this method are clustering sub-‎catchments into 2 major classes based on their distance to the WWTP and considering the ‎two largest rain events for the simulation to reduce the computational effort. Moreover, for ‎fulfilling the objective of the least manhole flooding, the junctions with the most surcharge ‎are considered. A weighting decision-making approach is applied within each cluster and ‎the seven NBSs are considered as the options to be weighted by the team members. The ‎characteristics of the NBSs plus the imperviousness and slope of the sub-catchments are ‎used in choosing the right NBS for each sub-catchment in each cluster. ‎ After selecting the possible NBSs for each cluster, multiple simulations are conducted and ‎the results act as the training dataset of a data-driven model, which takes the 7 NBS options ‎as inputs and predict the objectives (Cost, Biodiversity, etc.) as outputs of the model. The ‎trained model generates additional samples, and the results can be put in an evaluation ‎framework like a Pareto front or the TOPSIS analysis to find the optimal solution. ‎