Even if tourism represents only a fraction of water consumption at the global scale, in some regions such demand may represent a critical share. The Mediterranean coast is an emblematic case, since it receives many millions of visitors in summer, when rainfall is scarce and there is competition with water demand for irrigation. Many important tourism locations face permanent limitations or temporary restrictions on urban demand due the occurrence of drought events, including the last one in 2022. As the climate changes, water scarcity events have become more frequent and the need for addressing tourism water demand, starting from its understanding, has become a priority.

The study presents analyses carried out in two important beach tourism cities (Rimini, in Italy, and Benidorm, in Spain), where water demand models were developed and validated over historical data and then applied for future climate scenarios.

Two modelling experiments were carried out:

- Modelling of monthly water demand at municipal scale, through implementation of non-linear (stepwise linear regression) and non-linear (based on parsimonious architectures of artificial neural networks) models for the Rimini urban area;

- Modelling of hotels water demand: the monthly consumption series of the hotels in Benidorm were averaged to obtain the time series of a "typical hotel”, modelling was carried out and its performance assessed also in relation to the anomalous pandemic years 2020 and 2021.

In both types of model implementation (municipal and hotel scale), the input variables included climatic factors (monthly precipitation, number of rainy days, maximum and minimum temperatures) and socio-economic factors (in particular indicators on tourist attendance, resident population and tariff, where available).

Finally, a set of 10 EC-JRC raw and bias-corrected Euro Cordex climate simulations were processed and validated against local ground observations on the control period; the corresponding rainfall and temperature series simulated for the future decades (up to 2100) under the RCP8.5 scenario were then used in input to the water demand models in order to analyse the impact of expected climate change on the water demand.

How to cite: Toth, E. and Neri, M.: Tourism water demand modelling in Mediterranean cities under current and future climate, EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-18575, https://doi.org/10.5194/egusphere-egu24-18575, 2024.

Water Distribution Networks (WDNs) are crucial for meeting current and future urban water demands. Knowledge of future water demand at different time scales is fundamental for optimal WDN design and operations. Various predictive models of water demand have been proposed in the literature, ranging from traditional time series analysis to cutting-edge machine learning and deep learning techniques. However, the task of forecasting urban water demand remains mostly decoupled from the design of the optimal control of the related WDN. Current performance assessment of demand predictive models focuses predominantly on forecast accuracy, overlooking their practical implications on WDN operations. Meanwhile, the existing research on WDN management often assumes either perfect knowledge of future water demands or employs empirical approximations and aggregate statistics to estimate future water needs. This study bridges this gap by scrutinizing the actual operational value of water demand forecasts in designing optimal operations of WDNs.

Here, we develop a forecast-informed optimal WDN control framework to evaluate the operational value of water demand forecasts for WDN operations. Our framework comprises two main modules. The first module computes water demand forecasts. We comparatively evaluate a suite of different forecasting models —including Seasonal Auto-Regressive Integrated Moving Average with eXogenous factors, Multilayer Perceptron, Convolutional Neural Network, and Long Short-Term Memory (a network augmented with an attention mechanism) —across various forecast horizons (1/6/12 hours and 1/3/7/14 days ahead). The most accurate forecasts are integrated into the second module, an economic nonlinear Model Predictive Control (MPC) algorithm designed to optimize WDN operations. Within this module, an Artificial Neural Network (ANN) – based surrogate model encapsulates the hydraulics of the physical WDN, enhancing the optimization process while reducing complexity.

We demonstrate our forecast-informed optimal WDN control framework on a real WDN of a rural town with approximately 10,000 inhabitants. Water demand data collected in the Netherlands for a period of 10 years (2007-2017) at 5-minute resolution and corresponding meteorological data are used to train the water demand forecasting models. The MPC module computes the optimal control sequence for 7 pumps and 1 valve in the WDN to minimize pump energy costs while meeting water demands and ensuring safety storage in 5 tanks. Initial findings reveal that the ANN-based surrogate model can accurately incorporate the WDN characteristics (R² > 0.85), facilitating its integration into the MPC for an efficient and simplified representation of a real-world WDN. Further, MPC fed by 24-hour ahead water demand forecasts achieves potential energy savings of approximately 18% compared to a benchmark rule-based control strategy. Our framework yields a versatile simulation-based optimization tool for evaluating the impact of demand forecasts on WDN management. Future research efforts will aim at refining and comparing deterministic and stochastic water demand forecasts within the MPC framework, under diverse operational objectives.

How to cite: Hao, W., Cominola, A., Vertommen, I., and Castelletti, A.: A Water Demand Forecast-informed Framework for Optimal Control of Urban Water Distribution Networks, EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-12778, https://doi.org/10.5194/egusphere-egu24-12778, 2024.

In the past decades, the potential of numerical optimization for the automated design of drinking water distribution networks has been extensively studied. In particular, evolutionary algorithms have been shown to be a powerful and versatile tool for several design tasks. In the past few years in the Netherlands, drinking water utilities have started to embrace this approach more and more to explore new design philosophies as well as to address immediate asset management decision challenges. Key to meaningful application has been the possibility to iteratively and flexibly develop the optimization problem throughout the design process. The traditional 'benchmark problems' from academia provide a strong starting point for a design process, giving utility experts a taste of the possibilities. Subsequently, however, the problem definition has to be adapted and fine-tuned in order to keep up with the evolving perspective of the utility experts on the design problem. During this type of practical implementation, it frequently occurs that questions emerge which greatly increase the complexity of the optimization task without an approach being readily available from scientific literature, requiring workarounds to be created on the spot. Here, we present recent examples of such questions and their workarounds, which we ran into while tackling different practical design challanges, namely: how to incorporate deal with the prohibitively large complexity of a pipe dimension optimization for the city of Amsterdam, and how to incorporate topological properties regarding flushability as a performance constraint into a sectorization problem for the city of Rotterdam.

How to cite: van Laarhoven, K., Mitrović, D., Vertommen, I., and Hillebrand, B.: Numerical optimization for drinking water distribution network design: ideas and questions provided by practice, EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-15003, https://doi.org/10.5194/egusphere-egu24-15003, 2024.

The long-term design of Water Distribution Systems is a difficult task, with complex non-linear relationships, multiple objectives, and a decision space constrained by a discrete set of feasible actions. Numerous and heterogeneous sources of uncertainty also influence the complex landscape of solutions the decision-makers must explore. In response to this complexity, much of the current research is devoted to developing innovative methodologies to design systems that cope with these uncertainties, aiming at robust or flexible solutions.

In this study, we investigate a source of uncertainty whose role in the long-term design and planning of the infrastructure is often overlooked: operational uncertainty, i.e., the uncertainty stemming from the missing knowledge on the future values of the operational variables (e.g., pumping speeds and valve positions). From the design perspective, this represents an additional source of uncertainty for two reasons: first, the implemented control strategy is unknown (e.g., pump scheduling vs Model Predictive Control), and finally, in the case of any feedback control strategy, the optimal control actions depend on the uncertainties’ realisation, unknown during the design phase. Unlike other types of uncertainty, which stem from external factors beyond our control, operational uncertainty comes from the control decision variables, which can be subjected to cost-effective adjustments in the future.

The "Anytown" (Walski et al. 1987) case study is used as a benchmark to optimise reliability and cost, accounting for design and operational aspects. This classical optimisation problem combines irreversible design decision variables (e.g., pipe duplication) and adjustable controls (e.g., pump speed). Conventional and widely accepted optimisation techniques (e.g., Evolutionary Algorithms and Linear Programming) are used to solve the coupled operation and design problem, with the focus being the interplay between the control and design decision variables.

Gaining insight into the relationship between these variables will help us develop a metric to assess operational flexibility, a measure of a system's ability to adapt to changing conditions over time, adjusting its operations without requiring expensive changes in design. Developing such a metric would be particularly beneficial to create adaptive WDS, where the systems are built in phases, and the adaptation of the control decision variables allows for a delay of costly capital expenditure associated with design actions.

We show preliminary results on the influence of different problem formulations on the Pareto set of ideal designs and their operational uncertainty.

Walski, Thomas M., E. Downey Brill, Johannes Gessler, Ian C. Goulter, Roland M. Jeppson, Kevin Lansey, Han-Lin Lee, et al. 1987. “Battle of the Network Models: Epilogue.” Journal of Water Resources Planning and Management 113 (2): 191–203. https://doi.org/10.1061/(ASCE)0733-9496(1987)113:2(191).

How to cite: Zanutto, D., Castelletti, A., and Savic, D.: On the impact of operational uncertainties on water distribution system design, EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-5953, https://doi.org/10.5194/egusphere-egu24-5953, 2024.

Similar to other infrastructure sectors, the water distribution network is also undergoing increasing digitalisation, including real-time measurement of current system statuses. While high-resolution data is frequently available at the main points of the networks, the main challenge lies in remotely obtaining high-resolution water consumption data. Water meters are usually installed at remote and underground locations without a connection to the power grid, requiring battery-powered devices and reliable and energy-efficient wireless communication technologies. For an efficient large-scale implementation, documented practical experiences in real world applications are rare.

In this work, the experiences gained of a large-scale implementation of digital water meters in a demonstration project are presented. The case study includes 163 customer sites with the majority of single-family houses, aiming to measure water consumption data at a temporal resolution of 15 min in near real-time. In contrast to the electricity sector, there is no EU-wide legal regulation for the installation of digital meters. Instead, the requirements are depending on the specifications of the digital water meter type and are subject to the European General Data Protection Regulation (GDPR) at the intended spatial and temporal resolution, as detailed information about the user behaviour can be revealed. Subsequently, active costumer agreement was obtained in form of a signed declaration of consent and the approval rate was significantly increased through a pro-active approach of the network operator (e.g., detailed and personal information about the project). In total, around 70% of households were equipped with a commercially available digital water meter, using mioty® for the remote read-out. Initially facing data gap problems, after comprehensive software updates and improved antenna positions, the quality of service (as the ratio between received data packages and theoretically expected measurement data) still varies between 10 and 100% depending on the installation site, but during the last month of operation, 84% of the meters transmitted at least 75% of the expected data.

In combination with inflow and pressure measurement data, the measured water consumption data is afterwards used for an early warning system designed for detecting new leakages, serving as an exemplary application for digital water meters. The leakage detection and leakage localisation are implemented as data-based and model-based approach, respectively, and the system was tested on ten engineered leakage events with leakage sizes between 0.1 and 2.0 l/s. Despite a temporal failure in the data communication, strong fluctuations in the pressure data, and changing operating conditions, even small leakages could be timely detected, and the possible leakage area could be successfully narrowed down to 10 to 40% of the network length for the subsequent on-site fine search.

How to cite: Oberascher, M., Maussner, C., Hinteregger, P., Knapp, J., Halm, A., Kaiser, M., Gruber, W., Truppe, D., Eggeling, E., and Sitzenfrei, R.: Experiences from a large-scale implementation of digital water meters used for improved leakage management, EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-8150, https://doi.org/10.5194/egusphere-egu24-8150, 2024.

With 120 Mio. m³ per year of lost water globally, leakages in drinking water distribution networks (WDN) still pose a major challenge to water utilities, furthermore, resulting in a multitude of cascading effects such as operational disruptions, environmental hazard, property damage, and sanitary issues. In the last decades there has been a growing focus on leakage detection within the scientific community leading to the development of numerous computer-based solutions for leakage detection. Despite these developments, practical approaches employed by water utilities in their leak management routines still primarily rely on in-situ acoustic devices in combination with periodic water audits, altogether falling short of ensuring continuous system monitoring and leaving much further potential for leakage reduction. Conclusively, further dissemination and widespread implementation of automatic leakage detection technology in the near future will be paramount to contain water losses and foster robust and climate-resilient water supply systems.

Currently available computer-based technologies for leakage detection can be categorized either as data-driven or model-based, primarily depending on their requirement of a hydraulic model. Algorithms for leakage detection based on hydraulic models may accurately detect the occurrence and location of leakages, yet they are highly sensitive to model inputs and, thus, are required to be well calibrated. On the other hand, data-driven models operating on the premise of anomaly detection merely require data without any anomaly, i.e., leakage, for their calibration. However, these data-driven models cannot compete with the localisation accuracy of model-based leakage detection, as they do not incorporate geophysical information about the underlying WDN. Altogether, while yielding great improvement over in-situ technology, the requirements of automatic leakage detection technology still hamper its practical implementation. While both model-based and data-driven approaches have different requirements, their combination may ultimately enable mitigation of high technical requirements and, thus, enhance its practical applicability, thereby potentially facilitating a more efficacious, robust, and widespread implementation of leakage detection technology in water distribution networks.

In this work, we explore the trade-off between model-based and data-driven leakage detection on the basis of two award-winning state-of-the-art leakage detection algorithms developed by our consortium in previous research, i.e., the data-driven LILA and the model-based Dual Model. Through the integration of both algorithms into a unified application, we aim to mitigate technical barriers and bolster detection robustness. To validate our approach, we quantitatively evaluate its performance regarding false alarms, time-to-detection, and localisation accuracy against the individual algorithms while considering different levels of confidence and availability regarding the input data, i.e., hydraulic model, water demand estimation, and pressure data.

How to cite: Daniel, I., Steffelbauer, D., Steins, E., and Cominola, A.: Enhancing technology transfer by combining data-driven and model-based leakage detection in drinking water distribution networks, EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-16713, https://doi.org/10.5194/egusphere-egu24-16713, 2024.

Drinking water distribution systems operate to transport high-quality treated water across large distances to entire populations, yet water quality and contamination events occur between treatment and tap. Deployment of water quality instrumentation within drinking water distribution systems enables such events to be better understood. Specifically, in-network turbidity sensors offer a unique opportunity to measure network discolouration events, which are difficult to predict and may pose health risks to end users. However, extracting actionable information from the increasing volume of water quality data represents a major challenge to realising the true benefits of digitalisation. Typically this involves manual interpretation of time series plots, which is time-consuming and impractical for larger sensor networks. There is therefore a need to develop automated algorithmic approaches to process and integrate the turbidity signals. However, the information that is of interest and such algorithms should detect is uncertain. This study employed crowd-sourcing exercises with groups of domain experts to identify significant features within turbidity time series data from real-world distribution systems. The labelled data derived from these exercises delivers valuable insights and a critical benchmark for evaluating algorithmic methods designed to replicate human interpretation. Reflecting on the outcomes of the labelling tasks led to the development of a turbidity event scale that differentiates between advisory (< 2 NTU), alert (2 < NTU < 4), and alarm (> 4 NTU) level events. This event scale provides network operators with tools required to manage discolouration events both reactively and, crucially, proactively. A time-based averaging method, centred on data from the same time each day, proved most effective in identifying the advisory events, when compared to popular time series forecasting approaches. The event scale is demonstrated on a real-world example not included in the labelling exercises, showcasing the practical benefits and scalability of this data-driven approach. The automation of event detection and categorisation developed here offers the potential to obtain actionable insights to protect the quality of drinking water as it passes through ageing network infrastructure.

How to cite: Gleeson, K., Husband, S., Gaffney, J., and Boxall, J.: Crowd-sourced Turbidity Event Scale for Proactive Management of Drinking Water Quality in Distribution Systems, EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-17202, https://doi.org/10.5194/egusphere-egu24-17202, 2024.

A water distribution network (WDN) is a complex system that supplies drinking water from water treatment plants to consumers. WDN consists of various elements that ensure an efficient and reliable water supply. Among all these elements, valves are critical components that control the flow and pressure within the network and can isolate segments (the smallest parts of the WDN that can be isolated without interrupting service in the entire WDN) for maintenance or repair purposes. There are several studies in the literature on the optimal positioning of valves, in general the proposed methods are treated as optimization problems with one or more objectives aimed at reducing installation costs while ensuring high system reliability of the WDN (Creaco et al., 2010).

In recent years, water distribution networks have become increasingly vulnerable to contamination risks (WHO, 2014). Various factors contribute to this vulnerability, such as malfunctions in chlorination equipment, low pressure, contaminants entering water tanks and inadvertent connections between drinking and non-drinking water sources. When contamination is detected, the quickest remedial action that a water utility can take is to isolate the water section by closing gate valves.

The objective of this study is to find the optimal placement of gate valves in the water distribution network (WDN) to address the vulnerability of water quality, effectively isolate the contamination and minimize the residual concentration of contaminants. The proposed methodology is based on community detection algorithms used by sociologists to detect community structures in social networks (Traag, 2014). A community C can be described as a group of nodes with a high density of links between them and low density of links between different groups (or communities).

In this work, the community detection algorithm proposed by Girvan and Newman (2002) is used to identify groups of densely connected nodes in the WDN, and then isolation valves are placed on the boundary pipes between the different groups of nodes without performing hydraulic simulations. Different edge weights are tested to improve the placement of the isolation valves and reduce the risk of water pollution. The proposed methodology will be tested on a real water distribution network in southern Italy.

References

World Health Organization. (2014). Water safety in distribution systems. World Health Organization. https://apps.who.int/iris/handle/10665/204422

Creaco, E., Franchini, M. & Alvisi, S. Optimal Placement of Isolation Valves in Water Distribution Systems Based on Valve Cost and Weighted Average Demand Shortfall.  Water Resour Manage 24, 4317–4338 (2010)

Traag, Vincent. 2014. Algorithms and Dynamical Models for Communities and Reputation in Social Networks. Springer International Publishing.

Girvan , M. E. J. Newman, Community structure in social and biological networks, Proc. Natl. Acad. Sci. 99(12) (2002) 7821–7826.

How to cite: Santonastaso, G. F., Di Nardo, A., and Greco, R.: Valve isolation placement to mitigate contaminant spreading in water distribution network, EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-18948, https://doi.org/10.5194/egusphere-egu24-18948, 2024.

Detailed hydrodynamic models like MIKE+ provide accurate representations of the complex behavior of urban drainage systems. Surrogate models simplify this high-resolution representation to minimize the computational cost at the expense of reduced accuracy. As such, they are particularly useful when timely and repetitive simulations are needed.
Physics-based models of urban drainage systems typically include both hydrological (surface runoff) and hydraulic (network collection) components, with the computational cost mostly associated with the latter. The constructed surrogate model retains the hydrological representation of the original MIKE+ model but lumps the hydraulic network to a single collector downstream. This approach caters use cases where the full catchment description is needed.
We apply Muskingum routing to the catchment runoff to emulate the hydraulic routing between the catchment and the collector. The parameters of the routing function (delay and smoothing factor) are defined for each catchment as a function of the distance from the collector. We calibrate two global proxy parameters to optimize the performance of the surrogate: average velocity as a proxy of the individual delay, and smoothing range as a proxy of the individual smoothing factor.
The calibration of the proxy parameters is fully automated, given upper and lower bounds and the number of trials, and utilizes a Bayesian optimization algorithm. The objective of the autocalibration is minimizing the RMSE of the collector discharge for a synthetic rainfall event with 1-year return period.
To validate the calibrated surrogate, we simulated synthetic rainfalls with return periods both lower and higher than the calibration one and compared the modelled discharge with the results of the original model.
Our results for the 1-year rainfall show that the surrogate achieves a 25 times speedup in execution time compared to the original model, while introducing and 0.1% error in the accumulated volume of the event, a 3 minute error in the peak time, and a 5,8 % error in the peak discharge. A similar or better performance was obtained with lower return periods, but the performance of the surrogate quickly degrades with higher return periods.
Further research could focus on testing additional calibration objectives, e.g. timing and magnitude of the peak, as well as investigate methods to extend the validity of the surrogate beyond the calibration return period.

How to cite: Palmitessa, R., Murray, R. W., Andersson, H., and Mariegaard, J. S.: Autocalibration of hydraulically simplified model for urban drainage systems, EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-18429, https://doi.org/10.5194/egusphere-egu24-18429, 2024.

Urban drainage system (UDS) plays an import role in city urbanization. Defective pipes in UDS can lead to unanticipated damages such as blockage or seepage. Previous studies have identified the locations of defects in UDS using inverse optimization models. However, these studies overlook the uncertainty introduced by errors in monitoring and simulation. In addition, the multi-point defect problem is computationally slow for Markov chain Monte Carlo methods due to high dimensional parameters space. To address these issues, the paper employs a hybrid approach on USEPA Stormwater Management Model, leveraging the efficiency of a genetic algorithm (GA) to identify an optimal solution space and the precision of an adaptive Metropolis (AM) algorithm to yield a dependable estimation of the posterior probability distribution (PPD). Firstly, a modified multi-population GA is applied to maximize the exploration of the model space, generating an initial PPD. Then, AM algorithm is used to explore the final PPD of each pipe health status variable.

Two UDS cases are used to validate the method. The first case generates 400 sets of randomized multi-point seepage scenarios with monitoring flow sequences. The metrics accuracy and Matthews correlation coefficient are used to evaluate the binary diagnosis performance. The statistical results of metrics suggest that the method is effective in diagnosing the location and seepage extent of defective pipes, including in complex scenarios of multi-point seepage. The effects of seepage location, monitoring error, and data richness on the results are also analysed. In addition, comparison reveals that appropriate GA can maximize the exploration of the model space and attenuate the “genetic drift” effect. The second case considers a UDS with multi-point blockage. The application results suggest that the proposed method offers a comprehensive representation of the PPD on each pipe blockage status. The proposed method is easy to implement and can present uncertainty in the form of probability, which will help to narrow down the scope of defect monitoring and reduce the cost of detection.

How to cite: Zhou, C., Liu, P., Luo, X., Liu, Y., Xu, H., and Liu, W.: An efficient hybrid method of uncertainty estimation on defective pipe diagnosis in urban drainage system, EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-6491, https://doi.org/10.5194/egusphere-egu24-6491, 2024.

Utilizing digital measurement devices enhances the efficiency and reliability of urban water systems. For instance, digital water meters (DWMs) measure high-resolution water consumption at customer sites, serving both personal awareness raising and advanced water loss management in water distribution networks (WDNs). However, the pros and cons of these devices in terms of environmental impacts have yet to be fully unrevealed. This research aims to bridge this gap by conducting a comprehensive environmental assessment focused on the implementation of DWMs. The assessment considers not only the direct environmental impact of DWMs (e.g., due to production, installation, etc.) but also their indirect effects on the entire urban water cycle due to their usage. As an example of an indirect effect, using DWMs can reduce household water demand by promoting awareness. This leads to less freshwater treatment and pumping, decreased hot water and energy consumption in households, and a lower volume of wastewater generation. Thus, the current study categorizes the indirect effects on the urban water cycle into three scales: freshwater scale (including freshwater treatment and pumping energy), water user scale (involving energy consumption for water heating), and wastewater scale (including wastewater treatment).

Life cycle assessment (LCA) is used as a holistic approach to quantify environmental impacts. Accordingly, the system boundary is defined to encompass the entire life cycle of DWMs (from production to end-of-life), as well as the three scales reflecting indirect effects. An Alpine city in Austria with 105,000 inhabitants is selected as a case study, where the impacts of deploying DWMs are evaluated by defining three scenarios according to the requirements of the study area. These scenarios include: (1) Reducing 5% of total leakage, (2) Reducing 15% of water demand, and (3) combining (1) and (2). For each scenario, comprehensive datasets on resources, processes, and energy consumption are compiled, and impacts are quantified using the LCA software SimaPro 9.0.

Evaluating the environmental impacts of the study area in the existing situation (i.e., without any DWMs) shows that the water user scale (including energy for water heating) contributes to 80% of the total impacts. Thus, applying the second and third scenarios results in substantial energy savings across all scales (particularly water users) compared to the existing situation, reducing the environmental impacts ranging from 3 to 4 million kilograms of CO2 equivalent per year. This fluctuation is tied to the lifespan of the DWMs, extending from 2 to 8 years.

The proposed framework can explore the extent to which DWM deployment is sustainable, providing a blueprint for decision-makers to assess the effectiveness of similar interventions in different urban settings.

How to cite: Hajibabaei, M., Oberascher, M., Shahangian, S. A., Gschösser, F., and Sitzenfrei, R.: Environmental impact assessment of digital water meters throughout urban water cycle, EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-8882, https://doi.org/10.5194/egusphere-egu24-8882, 2024.

The urban water sector is increasingly turning to AI and deep learning to address the complex challenges posed by growing demographics, climate change, and urbanization. Despite the pressing need, this sector has been relatively slow in adopting these technologies compared to others, primarily due to its conservative nature. However, the recent advancements in generative AI have opened new frontiers for innovation, presenting a crucial opportunity for the urban water sector to accelerate its technological evolution. Expected regulations, particularly from institutions like the European Union, should not be viewed as a hindrance but as a catalyst for enhanced collaboration between academia, industry, and public stakeholders. Such collaboration is essential to finally push the development and adoption of reliable and safe AI systems, ensuring alignment with regulatory frameworks.

In this work, we first provide an overview of the latest trends in generative AI, focusing on how Large Language Models and Large Multimodal Models can benefit the urban water sector. Particularly, Large Multimodal Models can offer an added layer of explainability to predictive models working on imagery or other sensor data, a highly desirable feature for applications related to critical infrastructure. By literally asking these models to explain their decision-making processes with respect to the processed data streams, we can partially demystify the 'black box' nature of AI systems.

This potential is highlighted for a case study on sewer defect detection, utilizing a Large Multimodal Model that processes both text and imagery. The predictive results on the publicly available SewerML dataset are encouraging with respect to existing deep learning methods. More importantly, we show that explanations provided by the Large Multimodal Model can enlighten the decision-making process, making it more transparent. This added layer of explanation offers valuable insights and may set a new trajectory for developing trustworthy AI methodologies in critical water infrastructure management.

How to cite: Taormina, R.: The Potential of Generative AI for the Urban Water Sector, EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-15392, https://doi.org/10.5194/egusphere-egu24-15392, 2024.