Urban wastewater systems are critical infrastructures whose monitoring is complicated by sparse sensing, stochastic inflows, and complex hydraulic behavior. Physics-based hydraulic models provide accurate simulations but are computationally demanding and unsuitable for real-time operation. We adapt a graph neural network surrogate model GATRes to wastewater networks and train it on synthetic data generated from calibrated hydraulic simulations. The approach is evaluated on two wastewater systems: a municipal network in the Netherlands and the open Shunqing stormwater dataset from China. In the Dutch network, the model achieves estimation errors within 20-50 cm across multiple rainfall scenarios, which aligns with the precision expected by water domain experts. Results demonstrate strong overall accuracy under sparse sensing and across unseen rainfall scenarios, while highlighting challenges near hydraulic structures such as pumps and outfalls, where errors are higher but remain within acceptable ranges. These findings show that GNN surrogates can provide an accurate and scalable approach for state estimation in wastewater networks, contributing to monitoring and operational decision-making in smart cities.
Ghabi et al. (Thu,) studied this question.