Bridging the Gap Between Model Assumptions and Realities in Leak Localization for Water Networks

Localising leaks in pressurised water distribution networks (WDNs) is crucial for reducing water loss but remains challenging because of model uncertainties and limited sensor data. Nevertheless, many state-of-the-art methods rely on idealised assumptions that are perfectly known, like time-invariant demands, noise-free pressure sensors, a single, stationary leak, and a known leak-free baseline. These assumptions rarely hold in practice, creating a gap between expected performance and field reality. This article provides a comprehensive review of current leak localisation techniques based on sensor data and hydraulic or data-driven models. This study critically examines how recent studies have addressed these unrealistic assumptions. Advanced methods incorporate demand uncertainty and sensor noise into leak detection algorithms to improve robustness, estimate unknown demand variations using physics-informed machine learning, and employ Bayesian inference to locate multiple simultaneous leaks. The analysis indicates that accounting for such real-world complexities markedly improves localisation accuracy; for instance, even minor demand estimation errors or sensor noise can dramatically degrade performance if not addressed. Finally, bridging the gap between the models and reality is essential for the practical deployment of water utilities. Thus, this review recommends that future studies integrate uncertainty quantification, adaptive modelling, and enhanced sensing into leak localisation frameworks, thereby guiding the development of more resilient and field-ready leak management solutions.