Non-residential water-consumption characterization and clustering: evidences from service-industry users in Northern Italy

Non-residential water consumption can represent a significant share of urban water demand, making the characterization of users belonging to the service industry (SI) pivotal for urban water management. However, the relationship between water consumption and SI user categories remains underexplored, limiting accurate modeling and the reliability of leakage assessments. This study aims to fill this gap by analysing hourly water-consumption data collected through smart meters over a one-year period for 44 labelled SI users located in Northern Italy and grouped into six categories: shops, catering services, offices, wellness services, recreational and training activities and police stations. The methodological framework relies on four phases: (i) data preprocessing, including the removal of periods affected by long-term closure or post-meter leakages; (ii) characterization of daily water consumption for each user; (iii) derivation of seasonal and annual average daily consumption profiles; and (iv) profile clustering based on the application of K-means and Gaussian Mixture Models (GMMs) and clustering performance evaluation by adopting the Adjusted Rand Index (ARI) and the Normalized Mutual Information (NMI). Results reveal substantial variability in both daily consumption and profiles, even among users belonging to the same SI category. Moreover, user categories do not closely correspond to the obtained clusters (ARI approximate to 0.10 - 0.20 and NMI approximate to 0.30 - 0.40), suggesting that user category alone may not be sufficient to describe actual water-consumption behaviours. The dataset is provided as an open-access library, supporting applications such as preliminary water-demand estimations, water-balance assessments, or modelling applications in contexts where detailed consumption data are unavailable.