The Dutch Water authority Waterschap Limburg asked us to develop and implement a machine learning model that could forecast flow rates on the Rur River at Stah and generate early warnings in case of extreme peaks and flood risk.
By using machine learning, we identified complex patterns and relationships between precipitation, groundwater levels and water flows from historic data. The model uses these parameters and more to forecast flow levels 16 hours into the future, and feeds this information directly into the water authority’s existing FEWS systems.