Hydrant-Guided Water Distribution System Reconstruction from OpenStreetMap via Constraint Programming

Knowledge of Water Distribution System (WDS) pipe topology is a prerequisite for hydraulic simulation, interdependency analysis, and crisis-response planning, yet network records are withheld as security-sensitive critical-infrastructure information and rarely accessible to external researchers or emergency responders. Existing reconstruction methods require operator-provided inputs, and none exploits fire hydrant locations, despite hydrants being physical components of the pipe network. We formulate the placement of a tree-based WDS skeleton as a Constraint Programming (CP) optimisation over the OpenStreetMap (OSM) road network, with hard constraints enforcing connectivity and tree structure, and soft objectives for demand coverage, topographic-head plausibility, and hydrant-proximity guidance. The method requires only open data from OSM and a freely available digital elevation model (DEM). A spatially stratified k-fold cross-validation protocol using held-out hydrants enables quantitative evaluation without ground-truth pipe data. We evaluate the method on Darmstadt, Germany, comparing a literature Answer Set Programming (ASP) baseline, a CP formulation without hydrant guidance, and the hydrant-guided variant. On a city-wide instance with 554 hydrants, the hydrant-guided method reduces the mean distance from held-out hydrants to the nearest reconstructed pipe by 10.2% relative to the ASP baseline (a reduction of 2.76 m, 95% CI over folds 0.94–4.59 m; p < 0.001) and by 8.2–10.2% across three independent solver runs, while preserving full demand coverage and a comparable pipe count. Because utility pipe records are unavailable, this measures agreement with hydrant positions unseen by the optimiser, not with the actual pipe network. The results show that publicly mapped hydrant locations carry spatial information that generalises to hydrants withheld from the optimisation.