Groundwater studies may be tainted by "survivor bias"

Bad sinks tend to be excluded from studies of groundwater levels, a problem that could skew the results wherever monitoring is used to decide government policies and expenditures.

Researchers at the University of Waterloo have discovered the problem by examining a gap between scientific data and anecdotal evidence in southern India.

Reports of thousands of wells and satellite images taken between 1996 and 2016 suggest that groundwater levels were rising, which is good news in an area where it is of vital importance for the region. ;Agriculture.

At the same time, however, field workers were hearing more and more stories from farmers about dewatering wells, suggesting that levels were falling.

The researchers solved the apparent paradox by first obtaining census data that supported the anecdotal evidence. For example, it has shown that more farmers are digging deep, expensive wells in the hard rock aquifer.

"If indeed groundwater levels rise, why would farmers choose to pay more and dig deeper wells?" asked Nandita Basu, professor of civil and environmental engineering. "It did not make sense."

The researchers then examined the well data and found that those with missing water level data were often excluded from the analysis because they were considered unreliable.

When the excluded wells were reintroduced into the mix, the results confirmed the evidence provided by farmers that groundwater levels were decreasing and not increasing.

"They were systematically collecting wells with a lot of data and potentially ignoring wells drying up because they had incomplete data," said Tejasvi Hora, a PhD in engineering. student who conducted the research.

The culprit has been identified as something called "survival bias," a statistical phenomenon that results in the exclusion of negative data.

When the wells dried up, there was no water level to report. This created gaps in reporting on these wells, and their incomplete data was later rejected as being less than complete data from good wells that had not dried up.

Basu, also a professor of Earth and Environmental Sciences and a member of the Waterloo Water Institute, said that the lesson learned from southern India was applicable everywhere in the world to monitoring and analyzing the level of groundwater.

"Our main argument is that bad data is good data," she said. "When you have wells with a lot of missing data points, it tells you something important." Take note of it.

"Whenever you focus solely on complete data, you should step back and ask if there is a reason for incomplete data, a systematic bias in your data source," Hora said.

Basu and Hora collaborated with Veena Srinivasan, a researcher in a think tank about the environment in India.

An article about their work, The Paradox of Groundwater Recovery in South India, is published in the newspaper Geophysical Research Letters.