A water table is one of the hardest things in the earth sciences to read honestly. It answers to rain that fell years ago, to a neighbour’s pump and to abstraction a hundred kilometres away, to the slow breathing of wet and dry decades. And yet it produces exactly the kind of evidence that travels fastest in public argument: a single line on a chart, sloping down. The distance between how confidently that line is read and how little it can actually prove is where a great deal of mischief lives.
The instruments have improved dramatically. Since 2002 the GRACE satellites — a joint mission of NASA and the German Aerospace Center, continued since 2018 by its successor GRACE-FO — have let hydrologists weigh changes in total water storage from orbit, by detecting minute variations in the Earth’s gravity field. For the first time we can watch aquifers rise and fall at continental scale. What that record shows, over and over, is that measuring the decline is the easy part; attributing it is the hard one.
What the satellites actually show
Depletion is real and widespread: across many of the world’s great aquifers, extraction has outrun natural recharge. But when researchers try to separate the human signal from the natural one, the natural one keeps proving larger than expected. A 2019 analysis in Scientific Reports found that changes in groundwater storage were driven more by climate variability — recharge and precipitation — than by pumping, in part because a heavily pumped aquifer settles into a quasi-equilibrium that masks the human hand. Work on the major aquifers of the United States points the same way: interannual climate swings dominate the long-term record in most of them.
The models do not agree with each other
Attribution also depends on the model you choose, and the models disagree. One widely cited comparison found that global hydrological models — the ones that explicitly include pumping — overestimated depletion in heavily exploited U.S. aquifers by roughly two-and-a-half times relative to what the GRACE satellites measured. Read that slowly: the very tools built to quantify human impact can overstate it severalfold. Attribution within a single basin can even flip with the period studied — one steppe region moved from a climate-dominated decline to a human-dominated one only when the window shifted. None of this makes pumping innocent. It means that “the level fell, therefore this actor drained it” is not a finding. It is a hypothesis awaiting a model, controls, and independent replication.
A downward line is a question, not a verdict. Turning it into a culprit takes modelling, controls, and someone with no stake in the answer.
The standard of proof
What, then, does responsible attribution require? At a minimum: measurement by someone other than the interested party; a hydrogeological model that can reproduce the observed decline and rule out rainfall, recharge and competing users; and enough transparency that a third party can repeat the work. Where those conditions are met, a causal claim earns its weight. Where a decline is announced by the same body that campaigns on it, from data no one else has audited, the honest posture is neither belief nor dismissal but suspended judgement pending scrutiny. A recent Institut Kandrave note on evidence in basin disputes put it neatly: the confidence of a claim and the independence of its source ought to move together, and too often they move apart.
This matters well beyond the seminar room, because water attribution now carries real political and financial weight — shaping investment, regulation, and the reputation of named projects and organisations. Getting it wrong in either direction has a cost. That is exactly why the discipline should hold the line on what a chart can, and cannot, show.