Groundwater is depleting rapidly because of easy and affordable technologies to access and exploit the common resource. So there is an urgent need to identify potential recharge zones.
However, many factors seem to affect any location’s recharge capacity: the underlying structure’s percolation capability, rainfall dissemination, topographic characteristics, drainage density, and land use. It is not easy for researchers to wade through all the data in search of insights. In such cases, machine learning algorithms are useful.
Nitheshnirmal Sadhasivam from the Bharathidasan University, Tiruchirappalli teamed up with researchers from the US, Australia and Iran to test three algorithms that had given good results for similar problems: support vector machine, random forest, and multivariate adaptive regression spline. Which one is best for identifying potential groundwater recharge zones?
To check, they selected Firuzkuh County, to the north-east of Tehran, since a lot of data had already been generated from the area. With altitudes ranging from about 1100 to more than 4000 metres above sea level, the eastern part of the area had more than 100 quanats, underground channels that transport water to settlements. And the western and northern parts had more than 1700 springs. Underground channels and caves in carbonate rocks played a role in the storage and movement of groundwater.
For computation, the team chose sixteen factors which earlier literature claimed influenced groundwater recharge. Data from the eight meteorological stations, data from a digital elevation model of the area at 30 metre resolution, lithological factors from a geological map, water flow, drainage density from topographic maps etc. filled up the requirements for most factors.
There was, however, one important factor missing: permeability. So the researchers undertook a survey, using the double ring infiltrometer. Seven three-member teams went around the area, setting up the system five centimetres below the surface, pouring water into the inner ring and measuring the rate of change in water level for two hours. They thus collected permeability data from 2000 locations.
Using 70% of the data, they trained all three models to classify areas with low, moderate, high, and very high potential for groundwater recharge. The models were then used to predict groundwater recharge potential in the remaining data.
“All three learning algorithms gave excellent results, as expected. But models based on random forests showed better predictive accuracy than those based on support vector machines and multivariate adaptive regression spline”, says Sadhasivam, Bharathidasan University, Tiruchirappalli.
Are all sixteen factors equally useful in predicting areas with potential for groundwater recharge? The team used the LASSO algorithm, a regressive technique, to identify and distinguish between most and less effective factors. Minimum and maximum temperatures and annual rainfall were most effective factors, as expected. Drainage density, elevation, slope angle, aspect etc. influenced recharge to a lesser extent. But, surprisingly, the topographic wetness index, the multiresolution index of valley bottom flatness, fault density, lithology, distance from faults, distance from rivers and land use turned out to be the least effective factors.
The random forest machine learning algorithm helps us find patterns in huge amounts of geological and meteorological data that we cannot handle otherwise. Identifying areas with potential for groundwater recharge has immediate applications, especially in India given the geological and meteorological diversity.
“Constructing structures for groundwater recharge in areas with higher potential for groundwater recharge and restricting groundwater withdrawal from areas with least potential can help manage groundwater resources in a sustainable manner”, says Sadhasivam.
J, Environ. Manag., 265: 110525 (2020)
P Ramesh, Nehru memorial college, Puthanampatti
D C Jharia, National Institute of Technology, Raipur
and Ravi Mishra, National Centre for Polar and Ocean Research, Goa
STEAMindiaReports, a free resource for Indian media.
A resource that can never be over-exploited.