Beneath the earth’s surface lies a notorious time-consumer for many of Denmark’s 98 municipalities as well as several private companies and utility firms. It has gradually become a regular phenomenon that groundwater mapping and the appertaining model calculations in climate adaption projects, drinking water drillings, etc. both take time and a lot of money. This will end now.
A new Industrial PhD collaboration between NIRAS and Aarhus University (Department of Geoscience and Department of Computer Science) will reduce the time consumption of groundwater calculations. The combination between Machine Learning, geo statistics, and hydrology will transform conventional and time-consuming calculation methods to a new state-of-the-art tool. This is going to contribute to increased flexibility and reduce time consumption – beneficial to both clients and employees.
Pragmatic and progressive calculation model
The collaboration will contribute to highlight the gain opportunities and the potential of Machine Learning. The project will culminate in a new, digital platform where the consultant is able to quickly alter the parameters of the calculation in the field and with the client. Based on an amount of collected data, which contribute to develop and train the calculation model, this provides the ability to make new groundwater calculations on the spot.
Current calculation methods do not allow for accurate adaption of the calculations as fast, because the precise calculations often take multiple hours to complete. Consequently, the consultant will have to go back to the office and calculate new scenarios and data. The project will thereby contribute to transform the time aspect of the calculations.
Confidence level of 80-85%
The new calculation tool will work on an approximate statistical confidence of 80-85%. This means that Machine Learning currently cannot provide a complete result, as there will be an uncertainty of 15-20% in the calculations. The results from the model are however moving towards the level of accuracy of the conventional and precise methods, but there will always be uncertainty connected to the model’s calculations.
The calculation rate of the new tool is expected to be a fraction of the conventional calculation rate. The calculation model will additionally through Machine Learning, learn from its mistakes, which is why the uncertainty will continuously drop and create more reliant results.
International potential and increased collaboration
The tool has already sparked an interest with our Swedish neighbours. The project will thus be tested in Sweden under varying geological combinations, once the tool is closer to being finished. Expectations for the tool are high, and hopes are that more countries will show an interest in the long run.