Europe is drying out in silence

Why a wet year no longer fixes it

Drought, as most people picture it, is a dry summer: cracked soil, a thin harvest, low rivers. Then rain, and recovery. The drying now spreading across parts of Europe doesn't follow that pattern: it's a more dangerous kind of dry, because it doesn't end when the rain comes back.

Most of the continent's fresh water actually sits not in its rivers but beneath them, in aquifers: the slow, deep stores that a well draws on and that a dry month never touches. Think of it as a savings account. For years across central and southern Europe, more has come out each season than has seeped back in, and the balance has started to fall. A wet winter refills the soil, lifts the rivers, and the news moves on. The account below keeps emptying regardless.

And that's what makes the loss so easy to miss. The surface can look entirely healthy, green fields, flowing streams, while underneath, the reserve is draining away. By the time a well runs short, or a river fails in high summer, the decline has usually been under way for years already, invisible to anyone standing on the ground.

This isn't a projection, and it isn't an artist's impression. It's the satellite record. Weighed from orbit, Europe's mean stored-water balance has fallen from around zero in 2003–06 to roughly −10 cm of water-equivalent in 2022, −9 cm in 2023 and −9 cm in 2025 (annual means; the wet 2024 lifted it only to −5 cm): about half a centimetre a year on average, with the sharpest fall between 2013 and 2018. Across the continent's statistical regions, 96% show a falling trend in total stored water over 2002–2026, and the rest are flat or rising. Some of the steepest losses sit over hard-worked aquifers, the Po Valley, and the Konya and Amik basins to the south-east, places where supply and irrigation lean hardest on the reserve below.

How underground water is read from space

You cannot photograph groundwater. It sits under soil and rock, hidden from any ordinary satellite image. But water has weight, and weight can be felt from orbit, as the amount held under a region rises or falls, it leaves a faint trace that instruments in space can track, month after month, across an area the size of a province.

The instrument doing the weighing is a pair of satellites chasing each other 220 km apart in the same orbit, nicknamed Tom and Jerry: first launched in 2002, then replaced by the near-identical GRACE-FO pair, launched on 22 May 2018 after a battery failure on one of the original satellites had ended the first mission in October 2017. Neither carries a camera. Each tracks the other's position to within roughly a micron using a microwave, and now laser, ranging link; when the lead satellite passes over a mass concentration, a filling reservoir, a draining aquifer, Earth's gravity tugs it fractionally ahead, stretching the gap between the two by a distance thinner than a red blood cell. That stretch, accumulated and inverted over a month, is the entire measurement.

Strip away the other things that move with the seasons, ice, snow, the swelling and shrinking of rivers, and what's left is a running tally of the total water stored in the ground, deep aquifers included. It isn't a snapshot of a single day, the way a rain gauge or a photograph is. It's a balance, summed over years, read independently of cloud, rainfall and the colour of the crop. That independence is really the whole point. It can't be talked out of by one rainy week.

From a continental blur to a single district

There's a catch, and it's resolution. Read straight from orbit, the signal is coarse, a single measurement smears across an area roughly a third the size of Poland. A map at that scale can tell you a continent is drying. It can't tell a mayor whether her town's wells are at risk, or a lender whether one particular farm sits over a falling water table.

Closing that gap is the harder, less glamorous half of the work: bringing a blur the size of a region down to the level of a single district. It means teaching the coarse signal the local detail it lacks: the lie of the land, the soils, the rainfall, the way water actually moves in that place. Done carefully, and only as far as the evidence allows, it turns a headline into a figure a decision can rest on. Not "Europe is drying" so much as "this district is losing water at this rate, and here is how sure the estimate is." The published literature on this problem splits into two families: dynamic downscaling, where the coarse signal is assimilated into a physical land-surface or hydrology model that already runs at fine resolution, and statistical downscaling, where the relationship between the coarse trend and finer-scale predictors, soil, precipitation, land cover, is learned directly, commonly with random forests, boosted regression trees or partial least squares regression. Each family fails differently, which is one more reason the checks below matter as much as the model itself.

Two different numbers describe this coarseness, and it's worth being precise about which is which. The grid on which modern mascon products are distributed runs roughly 0.25°–0.5°, about 27–55 km on a side, though the mascons themselves are larger, about 1°–3° or 110–330 km. But gravimetric error is spatially correlated across neighbouring cells, so the resolution at which two grid cells actually carry statistically independent information is coarser still, closer to the "third of Poland" figure above, once that correlation and the leakage-correction filtering are accounted for. That gap between the grid and the real independent resolution is exactly why downscaling isn't a refinement: it's unavoidable. It's also why the choice of solution matters before any downscaling begins: the older spherical-harmonic representation expands the gravity field as a global sum of wave-like functions at roughly 1° sampling, while a mascon solution instead fits equal-area surface patches directly, which acts as a built-in smoother and handles land-ocean leakage differently: the two approaches can and do disagree on the same basin, which is why a figure should always state which family of solution it comes from. Worth naming one confound rather than hiding it: in Alpine-adjacent basins, the same gravimetric signal that tracks groundwater also picks up mass loss from melting glaciers, and the two have to be separated before a groundwater trend can be trusted, an effect documented in GRACE-based studies of German water storage. It's exactly the kind of failure mode a district gets flagged for. Not quietly averaged past.Further reading · ScienceDirect: GRACE remote sensing for groundwater monitoring · ESA: Enhancing GRACE data resolution

How the maps are checked

A map is only as good as the checking behind it. Each district's trend gets set against independent measurements on the ground, wells and boreholes that sample the water directly, and against the surface water balance of rain, evaporation and soil. Where the data is thin, or the checks disagree, the district gets flagged rather than quietly filled in. The right scorecard here is trend correlation and RMSE against the borehole record. Not the Nash-Sutcliffe or Kling-Gupta efficiency scores used for streamflow: those apply to a discharge hydrograph, and a storage trend is a different kind of series entirely.

The discipline is plain and unfashionable: publish the uncertainty, and say "we don't know" when that's the honest answer. Skill figures belong after a layer has cleared independent benchmarks, not before.

This kind of fusion already has an operational precedent worth naming: NASA's own GRACE Data Assimilation product feeds the satellite water-mass signal into the Catchment Land Surface Model at a global 0.25° grid, refreshed weekly, and its groundwater and soil-moisture percentiles are folded directly into the US Drought Monitor. Applied to European districts rather than US counties, and checked against a different ground network, the idea, a coarse orbital measurement disciplined by a physical model and validated against wells, is the same one that already underwrites a national drought product, not a novel bet.

The memory of deep water

Deep aquifers don't run on the weather's clock. Rain refills the soil in weeks and the rivers in a season, but water bound for the deep stores moves slowly, filtering down through rock over years. A reserve drawn down across two decades carries that history with it: hydrologists call it memory. Each season's deficit doesn't get erased by the next one. It gets added to the ledger.

Which is why a single wet year, however welcome, doesn't turn the map blue. One generous winter sets a season's worth of recharge against years of accumulated shortfall, a single deposit against a long overdraft. The record behind this briefing shows exactly that: a wet year appears as an uptick, 2024 was one, the news declares the drought over, and the multi-year line turns down again.

There's a harder edge to it, too. In some ground, when the water table falls far enough, the emptied layers compact and the space the water occupied closes up behind it. Part of the storage then isn't merely empty: it's gone for good. A structural deficit isn't a dry spell waiting for rain. It's a reserve being spent, and the longer the drawdown runs, the more of the recovery stops being possible at all.

The record does not stand alone

A single map, however careful, invites a single question: is it real, or an artefact of one method? The decline mapped here isn't read in isolation. It sits inside a body of independent evidence that points the same way.

On the ground, national well and borehole networks have logged falling water levels for years: long-known over-abstraction in southern Spain and Greece, and more recently, measured declines across Germany, the Netherlands, Belgium, Denmark and Poland. At the European scale, the European Environment Agency puts about 9% of the area of groundwater bodies in poor quantitative status, a share that has not fallen since its previous assessment, on a continent where close to two-thirds of drinking water is drawn from underground. And the drought of 2022 left central and southern Europe with the most widespread soil-moisture deficit in roughly six decades. Different instruments, different institutions, no shared assumptions. Same conclusion: across much of the continent, the water below is being drawn down faster than it returns.

What it changes

A falling aquifer isn't only an environmental story. It quietly rewrites value. Farmland over a draining water table will, in time, yield less and cost more to irrigate, and that fact belongs in its price, and in the risk on any loan secured against it. A city long comfortable on its groundwater may have fewer years of margin than its budgets assume. A planner zoning new housing, an insurer pricing a dry season, a utility deciding where to sink the next intake: each of them is making a bet on water they have, until now, simply been unable to see.

None of this changes the weather. What changes is who can see the loss while there's still time to act on it. Measured early, a slow drought is a problem to be managed. Noticed late, it's an emergency. So the value of weighing the water comes down to that: it moves the moment of knowing forward, from the year the well runs dry to the decade before it does.

The slow repricing of farmland

Farmland is priced as if the water beneath it were permanent. A hectare's value quietly capitalises everything the buyer expects it to keep doing: the yield, the cost of watering it, the reliability of both. Where the water table is falling, each of those assumptions is being eroded from below. Wells have to reach deeper. Pumping costs more. The driest weeks of the season arrive with less of a buffer beneath them.

Over the coming decade this turns into a sorting problem. Two farms can look identical from the road and from a satellite photograph, same soil, same rainfall, same crop, while sitting over reserves moving in opposite directions. Nothing at the surface distinguishes them today, though the multi-year water-mass trend does, and in time so will their prices, their irrigation bills, and the terms on which anyone will lend against them.

That last point is the quiet one. A mortgage secured on farmland is, in part, a loan secured on water, and it runs long enough for a structural trend to actually matter. This isn't a forecast of collapse. It's a slow repricing, and it'll show up first where the decline is steepest: markets reprice what they can see, and the trend simply makes the hidden half of the asset visible while the adjustment can still be gradual.

From a map to a decision

A trend only matters if someone changes a decision because of it. The practical use of the water-mass signal isn't the map itself: it's the handful of choices that sit downstream of it, and each reader of the map needs a different thing from it.

A municipality granting water permits is betting that the district's reserve can carry one more user. Set each permit against the multi-year balance rather than last summer's rainfall, and the bet becomes explicit: a district whose account is falling year after year shouldn't price new abstraction as if the reserve were stable. An insurer pricing a dry season is separating a bad year from a district where every future dry year will land harder, since the same event costs more over an emptying reserve. An agricultural lender is matching the term of a loan to the direction of the water balance beneath the collateral, because two portfolios that look alike today can diverge over the life of a mortgage.

None of them needs the raw signal. What they need is the direction, the rate and an honest margin of error, refreshed at a known cadence and flagged where the evidence runs thin: the same discipline that governs every figure in this briefing. The water below will keep its own account either way. The only question left is whether the people deciding above it read the ledger in time.

Data and method

Groundwater storage is not observed directly from orbit: a satellite gravity mission measures changes in total water mass, which combines groundwater, soil moisture, surface water, snow and ice. Groundwater is inferred by subtracting the other components, which is why the uncertainty has to be stated and why basins with glaciers or snow need extra care.

The native gravity grid is coarse: mascon products are distributed on grids of roughly 0.25°–0.5° (about 27-55 km), the mascons themselves span about 1°–3°, and the resolution at which neighbouring cells carry independent information is coarser still. District-level figures are therefore inferred by downscaling and carry larger uncertainty than the continental signal.

District figures are estimates and should be read with their stated margin of error. This text is for information only.