Data-driven Modelling of Urban Water Demand in Major European Cities: the Case Study of Milan, Italy.

The combined effect of population growth, urbanization, economic development, and climate change is changing the spatial and temporal patterns of water demands and exacerbating the stress on water resources. The global gap between water supply and demand is projected to reach 40% by 2030. Concerns on water security are dominant in cities with dense populations and more socio-economic activities. In this context, urban water demand management (UWDM) has emerged as a key measure to complement supply-side interventions to address water scarcity and governance issues in big European cities. Yet, there is still a lack of research fusing and comparing water demand characteristics and UWDM strategies across heterogeneous European contexts. A comparative and comprehensive urban water demand analysis at high spatial and temporal resolutions in major European urban centers could contribute to a better understanding of how different environmental, socio-economic, and political factors influence water use patterns, facilitating the scaling up of fine-scale UWDM practices into integrated regional, national, and European models. Here, we develop a holistic modeling approach incorporating trend detection, input identification, pattern characterization, and demand forecasting to understand the long- and short-term water demand dynamics. The first city we investigate is Milan, Italy. We explore the historical monthly water demands at the district level in 2017-2020 and daily data from individual meters in 2019-2021. Two main water use sectors can be distinguished: multiple households residential and commercial, industrial, and institutional buildings. Our preliminary results show a declining trend of total water use in 2017-2020, with a breakpoint identified at the end of 2019. Our comparative modeling study also shows that a hybrid model combining wavelet transform technique and artificial neural networks can achieve the best performance on 1-day short-term water forecast based on historical water demand only. Adding either temperature or precipitation variable does not improve the forecast accuracy. In the next steps, we will cross-correlate socio-demographic variables with water demands and apply our method to evaluate the urban development impact on water use patterns and inform efficient UWDM strategies