WaterScarCity aims to help cities prepare against the challenges imposed by more frequent and intense heat and drought conditions. It builds upon our expertise in operational hydrological forecasting to provide our partner cities of Fribourg, Sion and Winterthur with sub seasonal forecasts of water supply and demand distributed spatially within the cities and between the different user groups. These forecasts up to a lead time of 46 days and neighborhood spatial resolution are essential for decision-support tools to proactively manage water scarcity. We use state-of-the-art machine learning models that benefit from data richness and detailed spatial information available for urban environments. We plan to combine graph-based spatial features (e.g. pipeline and road networks, building density, land use, urban greenness and heat indices) with a Temporal Fusion Transformer model to improve time series predictions of water supply and demand based on ensemble meteorological forecasts.
In summary our objectives are:
Develop models for water supply and demand that use detailed spatial information from the cities.
Provide actionable sub seasonal forecasts for water management.
Develop a methodology that is transferable to other cities.
Graphical abstract
Building densities of the three selected cities: Fribourg, Sion, Winterthur