Simulating urban drainage hydraulics is computationally demanding, limiting its application in tasks that require real-time or repeated simulations. Graph Neural Networks (GNNs) are promising metamodels, but the effect of their internal components and transferability potential remain underexplored. This study addresses these gaps through two main contributions: (1) a systematic evaluation of key architectural components, including graph layer type, processor depth, and prediction window with links to physical transport dynamics; and (2) transferability experiments across domains (across two distinct drainage networks) and tasks (from head to flow prediction). As case studies, we selected two combined sewer networks in The Netherlands that differ in their hydraulic dynamics. We find that metamodels with moderate depth and a ten-step prediction window achieve high accuracy (RMSE of 2–5 cm for hydraulic heads and 0.02 m³/s for flowrates). They also reach speed-ups of up to four orders of magnitude higher compared to the physics-based model, SWMM, when executing parallel simulations in GPU. Based on our two case studies, we find that pre-trained metamodels with full fine-tuning effectively adapt to a new task within the same domain, whereas cross-domain transfer requires appropriate normalization and fine-tuning. Furthermore, joint training on both case studies enables the metamodel to capture representations of both systems, suggesting potential for more general applicability. These findings demonstrate that metamodel architecture can reflect physical system behavior and offer practical guidance for building fast, accurate, and generalizable GNN-based metamodels—establishing a foundation for their use in applications such as uncertainty analysis, design optimization, and nowcasting.

Wastewater surveillance (WWS) of viruses can aid public health officials in monitoring community infection dynamics and act as an early warning system for the introduction of viral infectious diseases. In recent years, agile, low-cost devices called passive samplers have proven to be indispensable for targeted wastewater surveillance. However, the viral uptake kinetics are unexplored for most viruses, limiting the understanding of optimal deployment times and the representativeness of this sampling method for assessing community viral shedding. This study investigates the uptake kinetics of CrAssphage, Pepper Mild Mottle Virus, Human Adenovirus 40/41, Human Norovirus genogroup II, Enterovirus, and SARS-CoV-2 on electronegative membrane passive samplers. Viral uptake was modeled by linear and pseudo-first-order uptake models for up to 48 h (adjusted R²: 0.89–0.99), with minimal saturation for 48 h. Bench-scale experiments revealed enrichment of Human Adenoviruses 40/41 on membranes compared to all other viral targets for 24–48 h deployment (p < 0.05), while differences were less pronounced with shorter deployment durations. This work highlights virus-specific interactions with passive samplers and how deployment times can affect the relative concentrations of viruses detected. Understanding these kinetics is critical for selecting appropriate sampling strategies and normalization methodologies for WWS of viral infectious diseases.

Automated sewer defect detection has advanced through deep learning, particularly supervised methods using CCTV images, but based on large annotated datasets. This paper proposes a semi-supervised learning (SSL) approach to reduce labeling demands. The method comprises self-supervised pre-training on unlabeled images using SwAV (Swapping Assignments between multiple Views) followed by fine-tuning for multi-label classification. Experiments on the Sewer-ML dataset demonstrate that the SSL approach, trained on only 35k labeled images, achieves an F1-score of 69.11%, and F2_CIW of 54.22%, surpassing the fully supervised baseline trained from scratch on 1.04 million images. Increasing the unlabeled pre-training data further enhances performance, while ImageNet initialization consistently outperforms training from scratch. Self-supervised learning also helps mitigate the effects of mislabeled data, which is observed to be present even in the Sewer-ML ground truth. Overall, self-supervised learning provides an accurate, scalable, and cost-effective alternative to fully supervised approaches, particularly in data-scarce or imperfectly labeled scenarios.

The effective environmental management of combined sewer systems requires reliable estimation of discharge and pollutant loads conveyed at the outlet during rainstorms. This study investigates how, with a lumped modelling approach, it is possible to reproduce the quality characteristics of discharged water, provided that high temporal resolution experimental data of pollutant concentrations are available. The methodology is applied to the combined sewer of a real urban drainage network where a continuous high resolution monitoring campaign of water quality and quantity has been carried out at an overflow structure location near the outlet of the drainage system. The lumped modelling approach has been implemented in the Storm Water Management Model (SWMM) with hydrological parameters estimated from cartographic information, based on recently proposed methodology that allows reliably simulating the storm hydrographs without model calibration. A semi-distributed model has been also developed using the SWMM with hydrologic parameters randomly sampled to fit the measured hydrographs of different training and validation data. The results obtained show that the uncalibrated lumped model simulates the observed hydrographs with similar performance as with the semi-distributed model (i.e., the normalized Nash-Sutcliff efficiency index of the validation set is 0.753 for the uncalibrated lumped model and 0.765 for the best-performing sampled parameter set of the semi-distributed model). The water quality parameters describing the build-up and wash-off of total dissolved solids (TDS) in a lumped model have been calibrated too, as well as those describing the mixing and consumption of dissolved oxygen (DO). The results show that a lumped modelling approach can reproduce the water quality dynamics in a combined sewer system, representing a promising tool for effective environmental management. However, event-specific calibrated parameter values have been obtained in some cases, which require further investigation and still limit the general applicability of the obtained results, thus confirming that setting up a reliable model requires water quality measurements.

Enhancing sediment accumulation monitoring techniques in sewers will enable a better understanding of the build-up processes to develop improved cleaning strategies. Thermal sensors provide a solution to sediment depth estimation by passively monitoring temperature fluctuations in the wastewater and sediment beds, which allows evaluation of the heat-transfer processes in sewer pipes. This study analyses the influence of the flow conditions on heat-transfer processes at the water-sediment interface during dry weather flow conditions. For this purpose, an experimental campaign was performed by establishing different flow, temperature patterns, and sediment depth conditions in an annular flume, which ensured steady flow and room-temperature conditions. Numerical simulations were also performed to assess the impact of flow conditions on the relationships between sediment depth and harmonic parameters derived from wastewater and sediment-bed temperature patterns. Results show that heat transfer between water and sediment occurred instantaneously for velocities greater than 0.1 m/s, and that sediment depth estimations using temperature-based systems were barely sensitive to velocities between 0.1 and 0.4 m/s. A depth estimation accuracy of ±7 mm was achieved. This confirms the ability of using temperature sensors to monitor sediment build-up in sewers under dry weather conditions, without the need for flow monitoring.

Computational models for water resources often experience slow execution times, limiting their application. Metamodels, especially those based on machine learning, offer a promising alternative. Our research extends a prior Graph Neural Network (GNN) metamodel for the Storm Water Management Model (SWMM), which efficiently learns with less data and generalizes to new UDS sections via transfer learning. We extend the metamodel’s functioning by adding flowrate prediction, crucial for assessing water quality and flooding risks. Using an Encoder–Processor–Decoder architecture, the metamodel displays high accuracy on the simulated time series. Future work is aimed at incorporating more physical principles and testing further transferability.

Sediments in urban drainage systems (UDS) significantly impact their operation, so effective strategies are required to reduce their negative effects. Monitoring sediment accumulation provides valuable insights into sediment characteristics, sediment transport dynamics, and system performance. However, the effectiveness of monitoring systems is limited due to cost constraints and installation challenges. This study describes the development and application of a new system based on temperature dynamics to measure sediment depths in sewer systems. The methodology involves the analysis of temperature time series under dry weather flow conditions to identify harmonic patterns between wastewater and sediment-bed temperatures. These patterns are increasingly attenuated by increasing sediment depth. This study combines a system called MONitoring Temperatures in SEdiments (MONTSE), which integrates a dual-probe heat-pulse (DPHP) method to characterize sediment thermal properties, and a surrogate model, which includes temperature pattern analysis, to estimate sediment depths. Likewise, laboratory-scale experiments were performed to validate the temperature monitoring system and the surrogate model performance. The maximum absolute errors in measured sediment depths were less than 22 mm, and the uncertainty of the system was estimated at ±7.3 mm. Groundbreaking measurements of thermal properties of UDS sediments were also reported. Reliable information on sediment depths and properties was provided, so the system could significantly optimize sewer system operation and cleaning strategies.

Model Predictive Control (MPC) of Urban Drainage Systems (UDS) has been established as a cost-effective method to reduce pollution. However, the operation of large UDS (containing over 20 actuators) can only be optimized by oversimplifying the UDS dynamics, potentially leading to a decrease in performance and reduction in users' trust, thus inhibiting widespread implementation of MPC procedures. A Heuristic And Predictive Policy (HAPPy) was set up, relying on the dynamic selection of the actuators with the highest impact on the UDS functioning and optimizing those in real-time. The remaining actuators follow a pre-set heuristic procedure. The HAPPy procedure was applied to two separate UDS in Rotterdam with the control objective being the minimization of overflow volume in each of the two cases. Results obtained show that the level of impact of the actuators on the UDS functioning changes during an event and can be predicted using a Random Forest algorithm. These predictions can be used to provide near-global optimal actuator settings resulting in the performance of the HAPPy procedure that is comparable to a full-MPC control and outperforming heuristic control procedures. The number of actuators selected to obtain near-global optimal settings depends on the UDS and rainfall characteristics showing an asymptotic real-time control (RTC) performance as the number of actuators increases. The HAPPy procedure showed different RTC dynamics for medium and large rainfall events, with the former showing a higher level of controllability than the latter. For medium events, a relatively small number of actuators suffices to achieve the potential performance improvement.

Urban drainage systems are composed of subsystems. The ratio of the storage and discharge capacities of the subsystems determines the performance. The performance of the urban water system may deteriorate as a result of the change in the ratio of storage to discharge capacity due to aging, urbanisation and climate change. We developed the graph-based weakest link method (GBWLM) to analyse urban drainage systems. Flow path analysis from graph theory is applied instead of hydrodynamic model simulations to reduce the computational effort. This makes it practically feasible to analyse urban drainage systems with multi-decade rainfall series. We used the GBWLM to analyse the effect of urban water system aging and/or climate scenarios on flood extent and frequency. The case study shows that the results of the hydrodynamic models and the GBWLM are similar. The rainfall intensities of storm events are expected to increase by approximately 20% in the Netherlands due to climate change. For the case study, such an increase in load has little impact on the flood frequency and extent caused by gully pots and surface water. However, it could lead to a 50% increase in the storm sewer flood frequency and an increase in the extent of flooding.

Small utilities often lack the required amount of data to train machine learning-based models to predict pipe failures, and hence are unable to harness the possibilities and predictive power of machine learning. This study evaluates the generalizability and transferability of a machine learning model to see if small utilities can benefit from the data and models of other utilities. Using nine Norwegian utilities’ datasets, we trained nine global models (by merging multiple datasets) and nine local models (by utilizing each utility's dataset) using random survival forest. Several pre-processing techniques including addressing left-truncated break data and break data scarcity are also presented. The global models and three of the local models were tested to predict the pipe failure of the utilities which were not included in their training datasets. The results indicate that the global models can predict other utilities with sufficient accuracy while local models have some limitations. However, if a representative utility with a sufficiently large (and information rich) dataset is selected, its model can predict the other utility's pipe breaks as accurate as the global models. Furthermore, survival curves for defined cohorts as proxies for uncertainty, and variable importance show that pipes with and without previous breaks behave extremely different. With the understanding of models’ generalizability and transferability, small utilities can benefit from the data and models of other utilities.

The ecological state of receiving water bodies can be significantly influenced by organic micropollutants that are emitted via stormwater runoff. Reported efforts to quantify the emission of micropollutants mainly focus on sampling at combined sewer overflows and storm sewer outfalls, which can be challenging. An alternative method, called fingerprinting, was developed and tested in this study. The fingerprinting method utilizes wastewater treatment plant (WWTP) influent samples and derives the proportion of stormwater in a sample. This is achieved by comparing the wet weather vs dry weather concentrations of substances-tracers which are present only in wastewater. It is then possible to estimate the concentration of organic micropollutants in stormwater runoff from measurements in the influent of a WWTP based on a mass balance. In this research, the fingerprinting method was applied in influent samples obtained in five WWTPs in the Netherlands. In total, 28 DWF and 22 WWF samples were used. The chosen tracers were ibuprofen, 2-hydroxyibuprofen, naproxen and diclofenac. Subsequently, the concentration in stormwater runoff of 403 organic micropollutants was estimated via the WWF samples. The substances that were present and analyzed included glyphosate and AMPA, 24 out of 254 pesticides, 6 out of 28 organochlorine pesticides, 45 out of 63 pharmaceuticals, 15 out of 15 PAHs, 2 of the 7 PCBs, and 20 of 33 other substances (e.g. bisphenol-A). A comparison with findings from other studies suggested that the fingerprinting method yields trustworthy results. It was also noted that a representative and stable dry weather flow reference concentration is a strict requirement for the successful application of the proposed method.