Pollutants in groundwater: uncovering blind spots with machine learning

Eawag researchers Joel Podgorski and Michael Berg have developed a model that can be used to determine the risk of groundwater contamination – for example, with arsenic, fluoride or nitrate – on a worldwide basis despite extensive gaps in the measurement data.

Over 500 million people worldwide have no access to safe drinking water and instead obtain their water from rivers and lakes or groundwater wells, for example, thereby putting their health at risk. In many regions of the world, groundwater naturally contains substances that are hazardous to health – primarily arsenic or fluoride. And in many affected areas, the groundwater is not tested for these naturally occurring, harmful “geogenic” substances. So, how can the contaminant load and its effects on the population be determined if barely any measurement data is available?

Michael Berg, head of Eawag’s Water Resources and Drinking Water department, and his colleague Joel Podgorski have found a way to solve this problem with machine learning. They have developed a wide-area modelling technique that can be used to calculate whether a geogenic contaminant in the groundwater exceeds the guideline limit value defined by the World Health Organization (WHO), based on prevailing factors such as climate and geology. The researchers train the model with the limited available measurement data, and the computer uses statistical algorithms to learn which combinations of factors cause levels to exceed the limit values. This allows the computer to draw conclusions about areas where no measurement data is available, resulting in an extensive hazard map showing areas where high levels of groundwater contamination are likely.

Population density and water use are key factors

The next step after that involves feeding population figures and data on water usage behaviour into the calculation. “The strength of our model is that we can identify not only the areas of risk but also how many people in those areas are affected – in other words, the areas where measures are most urgently needed and will have the greatest impact,” says Michael Berg. These predictions are therefore extremely valuable for authorities operating in the affected regions. “For example, China has used them as the basis for targeting its nationwide groundwater monitoring campaign at areas with the highest levels of arsenic contamination according to our modelling,” says Berg, “and it turns out that our predictions have been almost universally correct.”

Arsenic map: Once a popular way of murdering people, arsenic is today the cause of an insidious case of mass poisoning – with the Eawag model showing that some 220 million people worldwide consume groundwater contaminated with arsenic on a daily basis. Found at widespread locations around the world, this harmful geogenic substance becomes a hazard in areas where many people consume untreated groundwater – especially in parts of India, Pakistan, Southeast Asia and China, but also at certain locations in African countries, the USA, Mexico or Argentina. (Map: Joel Podgorski, Michael Berg (2020). Global threat of arsenic in groundwater. Science, 368, 845–850)

Using the same methodology, Podgorski and Berg can not only predict naturally occurring groundwater contamination but also that which is caused by humans. At present, they’re looking into nitrate pollution in Swiss groundwater. Nitrate concentrations are measured regularly at 500 locations in Switzerland – but there are lots of blind spots in between. The researchers are therefore working to develop a prediction model based on soil types, landscape forms, farming culture and population density with a view to uncovering nitrate hotspots, thereby providing the basis for targeted measures. Berg says, “Maps of this kind also serve as a basis for decisions on where it makes sense to drill new wells for monitoring purposes – or whether the cultivation of certain crops that need especially high levels of fertiliser should ideally be dispensed with in a given area.”

Fluoride map: As an additive in toothpaste, fluoride protects our teeth against decay. In groundwater, however, too high concentrations are a cause for concern: fluoride is one of the most common geogenic pollutants worldwide. Eawag’s model shows that large parts of Africa are more likely to have harmful fluoride levels in their groundwater. Other badly affected regions include the Middle East, Central Asia, China and eastern Brazil. There are also hotspots in the south-western USA and in Australia – although almost no one is at risk in those areas, as they don’t drink untreated groundwater. That is not the case in Africa and Asia, however, where many people have no choice but to consume groundwater contaminated with fluoride. According to the model calculations, 180 million people worldwide are likely to be affected. (Map: Joel Podgorski, Michael Berg (2022). Global analysis and prediction of fluoride in groundwater. Nature Communications, 13, 4232)

Groundwater Assessment Platform (GAP)

Eawag makes the global and regional hazard maps of groundwater contamination with arsenic, fluoride and other harmful geogenic substances freely available online at gapmaps.org. The Groundwater Assessment Platform (GAP), developed by Eawag with financial support from the Swiss Agency for Development and Cooperation (SDC), also allows experts and authorities from around the world to visualise their own data using the Eawag model and to produce their own risk maps of arsenic and fluoride contamination.

A contribution to the SDGs:

Created by Isabel Plana for the Info Day Magazine 2023