Optimization of water work operations during critical flood events using neural networks and data visualization

In this work we aim to understand how current questions in hydrogeology can be answered using new methods of data analysis developed in recent years. Water works supplying drinking water to large cities represent a hydrogeological challenge of great global interest. Ensuring optimal water quality not only under normal conditions but also during flood events is a public health issue. The pollution processes associated with flood events are the result of a combination of numerous factors at the basin scale, which complicates their prediction. Numerical models are powerful tools to simulate and manage groundwater flow around water works, but due to their high computational costs, simplifications and assumptions must be made, which reduces modelling precision. 

We selected a water work located in a subalpine fluvio-glacial aquifer, providing water to a large city. The water work is compound of several drains that extract the water from the aquifer by gravity. Hydrochemistry is stable under normal conditions but changes drastically during flood events, with a decrease in water quality. Due to the vast amount of data, on hydraulic heads, river levels and hydrochemistry, that is available from over 40 locations across the relevant area, modern data analysis tools perfectly complement the numerical model.  

The goal of our work is to review how to predict critical flood events and optimize water work operations accordingly by complementing numerical models with new methods of data analysis. To reach the ultimate goal of building a decision support system for water work operations state-of-the art data visualization tools must be combined with machine learning methods such as deep neural networks. These methods have a lower computational cost than numerical models, which makes them suitable for real-time predictions. They can also answer questions that are too complex for the numerical model.  

We provide an overview on the current literature on data visualization tools and neural networks for ground water modelling and suggest approaches relevant for the selected site. Customized data visualization tools are used to allow both researchers and water work operators gain information directly from the data, without further computations. A neural network trained with parameters describing rainfall in the area as well as groundwater and river levels is able to predict the correlation between rain events and water levels. A second neural network links river and groundwater levels to water quality at the water work. In a next step, water quality at the water work under different conditions is correlated with different modes of operation.