Maintaining water quality in the face of population growth and global climate change is no small order. For Waterschap Aa en Maas, machine learning held the answer. Working with Royal HaskoningDHV Digital, the water company implemented bespoke algorithms designed to predict the most effective measures for providing clean water.
- ChallengeEnsure a clean and consistent supply of water to the Dutch public
- ClientWaterschap Aa en Maas
- LocationThe Netherlands
- Period2019 / 2020
- SolutionA new predictive model based on machine learning algorithms
Insight into the ecological tipping points
Detailed information into the influence of control variables
Better insights in how toimprove the quality and supply of water
Clean water. Bright futures.
The water authority Aa en Maas works to provide safe, clean and sufficient water to the Dutch public, maintaining dykes, purifying waste water, and monitoring the effects of climate change.
Education and innovation are central to the water authorities' activities. That’s why, when faced with a growing list of challenges, Aa en Maas implemented machine learning models to determine which measures would have the greatest effect on the quality of its water.
“As a water board, we are looking for effective measures to improve the ecological quality of our water systems,” says Frank van Herpen, Advisor Water Quality at Aa en Maas. “This requires the right level of insight–and for that we need the right technology.”
Machine learning and predictive analytics
Aa en Maas worked with Royal HaskoningDHV Digital to develop a bespoke predictive model that could help generate the required level of insight.
The machine learning algorithms are trained with historical data from past ecological quality ration (EKR) scores, and implemented in Deltares’ WFD Explorer –a software program designed to help companies keep in line with the Water Framework Directive.
This insight enables the company to predict EKR scores based on a number of variables, including meandering, congestion, shading and phosphorous concentrations of the water. Aa en Maas can then take the appropriate steps to ensure the requisite quality and quantity of water is available to the public.
In many cases, using machine learning models can be a black box experience, with limited insight into the intricate workings of the algorithms.
However, Aa en Maas wanted to know which control variables have the most influence on the EKR score, where tipping points were, and which combinations of measures would be most effective at producing clean water. All of this meant a deeper level of understanding was required.
“Having insight into the ecological tipping points is really important to us,” says Van Herpen. "We know that these tipping points are present in lakes, but much less is known about this for our type of water system consisting of ditches, canals and small rivers.”
Aa en Maas chose to use SHAP (SHapley Additive exPlanations) values to gain insight into its analytics models. This is a game theory concept used to deduce the contribution of each player to the outcome of a game.
In this case, the control variables are the players and the outcome is the predicted EKR score. With visibility of these contributions, Aa en Maas was able to see exactly how the algorithm arrived at a certain prediction.
This helps the company to understand the effectiveness of every measure it implements and, importantly, identify those all-important ecological tipping points.
An insight into the future
Working with Royal HaskoningDHVDigital, Aa en Maas integrated its forecasting models with WFD Explorer, while also making them available for use in separate programs. The program then visualizes the results of the SHAP values and makes them available to the relevant parties in an easy to digest report.
“A lot of time has been put into the visualisation of the data, which makes it easy for us to share the results,” says Van Herpen. “This is just one of the ways the collaboration with the content specialists and data specialists from Royal HaskoningDHV Digital has been very important in this project.”
Going forward, the insights generated by the predictive modelling will be available in an interactive tool, designed to help water authorities support policy design.
For now though, Aa en Maas has the insights it needs to improve the quality and supply of its water. And, with detailed information into the influence of control variables, the most effective combination of measures can be identified, helping to maximise the return on the company’s future investments and innovations.
...The collaboration with the content specialists and data specialists from Royal HaskoningDHV Digital has been very important in this project.
Frank van Herpen
Advisor Water Quality, Water board Aa en Maas