It’s not easy to determine how much water there is across a landscape. A measly 1% of Earth’s freshwater is on the surface, where it can be seen and measured with relative ease. But beneath that, measurements vary massively depending on water table depth and ground porosity we can’t directly see.
Reed Maxwell, a hydrologist at Princeton University, likes to think of rainfall, snow, and surface water as a checking account used for short-term water management needs and groundwater as a savings account, where a larger sum should, ideally, be building up over time.
But a new groundwater map by Maxwell and colleagues offers the highest-resolution estimate so far of the amount of groundwater in the contiguous United States: about 306,500 cubic kilometers. That’s 13 times the volume of all the Great Lakes combined, almost 7 times the amount of water discharged by all rivers on Earth in a year. This estimate, made at 30-meter resolution, includes all groundwater to a depth of 392 meters, the deepest for which reliable porosity data exist. Previous estimates using similar constraints have ranged from 159,000 to 570,000 cubic kilometers.
“It’s definitely a move forward from some of the previous [mapping] efforts,” said Grant Ferguson, a hydrogeologist at the University of Saskatchewan who was not involved in the research. “They’re looking at much better resolution than we have in the past and using some interesting techniques.”
Well, Well, Well
Past estimations of groundwater quantity have been based largely on well observations.
“That’s the really crazy thing about groundwater in general,” said Laura Condon, a hydrologist at the University of Arizona and a coauthor of the paper. “We have these pinpricks into the subsurface where there’s a well, they take a measurement of how deep down the water table depth is, and that’s what we have to work with.”
But not all wells are measured regularly. For obvious reasons, there tend to be more wells in places where more groundwater is present, making data on areas with less groundwater scarcer. And a well represents just one point, whereas water table depth can vary greatly over short distances.
Researchers have used these data points, as well as knowledge of the physics of how water flows underground, to model water table depth at a resolution of about 1 kilometer. They’ve also used satellite data to capture large-scale trends in water movement. But those data are of lower resolution: Data from NASA’s GRACE (Gravity Recovery and Climate Experiment) Tellus mission, for instance, have a resolution of about 300 kilometers, about 10,000 times coarser than the new map.
To demonstrate the value of high-resolution data, the team showed what happened when they decreased the resolution of their entire map from 30 meters to 100 kilometers—the spatial resolution of many global hydrologic models. The resulting more pixelated map estimated just above 252,000 cubic kilometers of water, an underestimation of 18% compared to the new map.
In addition to identifying groundwater quantities at high resolution, the new map reveals more nuanced information about known groundwater sources.
For instance, it shows that about 40% of the land in the contiguous United States has a water table depth shallower than 10 meters. “That 10-meter range is that range where you can have groundwater— plant— land surface interactions,” Condon said. “And so that’s just really pointing to how connected those systems are.”
Bias for Good
The new work used direct well measurements as well as satellite data — about a million measurements, made between 1895 and 2023 — along with maps of precipitation, temperature, hydraulic conductivity, soil texture, elevation, and distance of streams. Then, the scientists used the data to train a machine learning model.
In addition to its being able to quickly sort through so many data points, Maxwell noted another benefit of the machine learning approach that might sound unexpected: its bias. Early groundwater estimates were relatively simplistic, not accounting for either hydrogeology or the fact that humans themselves pump water out of the ground. The team’s machine learning approach was able to incorporate that information because evidence of groundwater pumping was present in the data used to train it.
“When you hear about bias in machine learning all the time, it’s usually in a negative connotation, right?” Maxwell said. “As it turns out, when you can’t disentangle the signal of groundwater pumping and groundwater depletion from the almost 1 million observations that we used to train this machine learning approach, it implicitly learned that bias.… It’s learned the pumping signals, it’s learned the human depletion signal.”
Maxwell and the other researchers hope the map can be a resource for regional water management decisionmakers, as well as for farmers making decisions about irrigation. Condon added that she hopes it raises awareness of groundwater in general.
“Groundwater is literally everywhere all the time,” she said. The map is “filled in everywhere, wherever you are. Some places it’s 300 meters deep, some places it’s 1 meter deep. But wherever you’re standing, dig down, and there’s water down there somewhere.”
This article was originally published on Eos.org. Read the original article.
