The answer is China’s Loess Plateau, and the species is black locust. Since the Grain for Green programme began in 1999, Robinia pseudoacacia has been planted across more than 10 million hectares of the plateau, accounting for close to 90 percent of its artificial forest. It grew fast, it held the slopes, and it drank. Beneath tens of thousands of those stands the soil is now desiccated well below the depth that rain can reach, vegetation greening has contributed to roughly a 30 percent fall in average annual runoff between 2000 and 2020, and Chinese planners have begun putting grass and shrubs where they once put trees.

This is what an environmental success story looks like when it overshoots. The erosion problem the plateau is famous for has largely been solved. The water problem it created has not.

The scale is worth holding onto. The Loess Plateau is a 640,000-square-kilometre expanse of wind-blown silt in north-central China, home to around 100 million people, and for most of the twentieth century it was the most eroded large landscape on Earth. Terraces, check dams, grazing bans and planting cut the sediment reaching the Yellow River by something like 85 to 90 percent. That worked. The question that has occupied plateau hydrologists for the last decade is what it cost in water.

The tree that did the job too well

Black locust was an obvious choice in 1999. It is a nitrogen-fixing legume, it tolerates poor soil, it grows quickly on bare ground, and it puts down roots fast enough to bind a slope before the next summer storm arrives. Foresters across the semi-arid world have reached for it for the same reasons.

What it also does is transpire. A hillside of mature locust moves far more water from soil to atmosphere than the millet stubble or bare loess it replaced. On a plateau where annual rainfall runs between roughly 300 and 600 millimetres, and falls in a handful of violent summer downpours rather than steadily through the year, that difference is not a rounding error. It is the difference between a soil profile that recharges and one that does not.

Rain on the plateau typically penetrates one to three metres before it stops. Below that lies water that arrived years or decades ago and moves very little. A deep-rooted plantation can reach past the annual infiltration depth and start withdrawing from that older store. When it does, a dried soil layer forms beneath the reach of ordinary rainfall — a persistent band of desiccated loess that does not refill between wet years.

Chinese soil scientists have a phrase for what happens to the trees standing on top of it. They call them little old man trees: stands twenty or thirty years old that have simply stopped growing, stunted at two or three metres, holding the slope in a nominal sort of way while delivering almost none of the canopy or biomass the planting was meant to produce. Estimates of how much of the plateau carries some degree of soil desiccation run to about two-fifths of its area.

What dry means three metres down

The mechanics have been mapped in some detail. A study in Hydrology and Earth System Sciences compiled field sampling and published data on deep soil moisture deficit under black locust and apple plantations and found the deepest drawdown beneath stands aged roughly 24 to 28 years, with the strongest effect on deep moisture across the 20-to-30-year window.

That timing matters. Plantations put in during the first flush of Grain for Green in 1999 and 2000 are now passing through exactly that age band. The problem is not ahead of the programme. It is arriving on schedule.

The same work identified where root water uptake simply stops: at the permanent wilting point, the moisture level below which a root can no longer pull water out of the soil regardless of how hard it tries. Once a profile is drawn down to that point, the tree cannot go deeper for relief, and the layer beneath it stays dry.

A review published this year in WIREs Water frames the situation as a hydrological debt. Plantations that mine legacy water from metres below the surface incur a deficit that can only be repaid by the rare, heavy events that drive water deep into the profile — and once a dried layer establishes itself, the system becomes self-limiting, because the profile has to be rewetted from the top before meaningful recharge can resume at all. The authors are blunt about the timescale: under the plateau’s current climate, that deep debt can be slowed or partly serviced, but is unlikely to be paid off within any horizon a land manager works to.

This is the part that separates the water story from the erosion story. Sediment responds to engineering within a decade. Deep soil moisture keeps a much longer memory.

The argument the scientists are still having

The foundational paper is Feng Xiaoming and colleagues in Nature Climate Change in 2016. Studying the plateau after fifteen years of Grain for Green, they found that the new planting had raised both net primary productivity and evapotranspiration, and that the rise in evapotranspiration had driven a statistically strong decline in the ratio of river runoff to annual rainfall across catchments. Over 2000 to 2010 they measured losses of roughly 2.4 millimetres of soil moisture and half a millimetre of runoff per year.

Their headline finding was a ceiling. From the revegetated area and the region’s human water demand, they estimated a threshold of about 400 grams of carbon per square metre per year in net primary productivity, above which the plateau’s population would start experiencing water shortages — and reported that productivity was already close to it. Under a bad combination of climate drying and heavy withdrawals the threshold could fall by more than a third; under a favourable one it could rise by over 40 percent.

That paper set the terms of the debate. It did not settle it.

In 2022, a team writing in Geophysical Research Letters reached a directly opposing conclusion. Examining long-term observations, they reported that surface water yield has been rising, not falling, across 82.3 percent of the plateau since revegetation began, at about 2.76 millimetres per year squared. Their argument is that earlier studies left out vegetation-climate feedbacks: greener ground alters the local energy and moisture balance in ways that can increase precipitation, and on the plateau those gains have outrun the extra evapotranspiration.

Attribution is contested in the same way. Some analyses of the middle Yellow River conclude that soil and water conservation work accounts for around 62 percent of the streamflow decline against 38 percent for climate; others put climate change above half. The honest position is that revegetation, reservoir operation, irrigation withdrawal and a shifting rainfall regime are all pulling on the same river, and separating them cleanly is still an open problem.

Where the evidence converges is on margins rather than direction. A 2026 analysis in Ecohydrology integrating remote sensing with water-balance modelling found water resource conflict becoming acute across the plateau, with available water approaching sustainable limits by 2020 and the pressure concentrated in central and northern Shanxi and northern Shaanxi. Its projections to 2100, interestingly, suggest a wetter future climate may ease the squeeze rather than deepen it.

So the plateau is not uniformly drying. It is running close to a limit that varies enormously from one valley to the next, and the places already past it are identifiable.

Redrawing the map by rainfall

That variability is the basis of the policy shift. Moisture on the plateau falls along a strong gradient from the wetter southeast to the arid northwest, shaped by terrain, soil texture and land use, and the same species planted at both ends produces opposite outcomes.

The emerging prescription is zoning. Forests and shrubs belong in the southeast, where the soil reservoir can carry deep-rooted vegetation through dry years. Natural grassland belongs in the northwest, where afforestation reliably triggers rapid desiccation and a dried layer within a couple of decades. Grass has shallower roots, lower water demand, and on these slopes it holds soil nearly as well as a struggling plantation does.

The broader hydrogeomorphic picture of the plateau compiled in the Annual Review of Earth and Planetary Sciences supports the same logic: the plateau is not one system but a mosaic of catchments with different water budgets, and treating it as one was always going to produce local failures.

A 2024 perspective in the Journal of Earth Science arguing for a more sustainable approach to the plateau puts water management at the centre of the next phase, ahead of further planting. Field studies point the same way, finding that forage land and shrubland do little to relieve deep desiccation once rainfall drops below roughly 100 millimetres a month during the growing season.

Some of this is already being applied. Species selection is matched to slope, aspect and rainfall band rather than applied uniformly. Stunted stands are being thinned. And on the driest ground, new planting is grass.

The hydrological trap is not unique to dry uplands. Solar Daily’s reporting on mangrove restoration found much the same failure mode at sea level, where a majority of replanting schemes collapse within five years because planners chose a species and a site without first working out where the water goes. The plateau made the more expensive version of that mistake, at continental scale, and is now spending decades unwinding it.

What the slopes look like now

Fly over the hills above Yan’an in July and the difference is not obvious from altitude. The slopes are green. The terraces still read like contour lines. What you cannot see from a plane is which of those dark patches are twenty-five-year-old locust stands that stopped growing a decade ago, standing over three metres of loess drier than it has been since the last glaciation.

The sediment policy question that dominated the last four decades has largely been answered. Check dams fill and need dredging, terraces need maintenance, and the specialists who once argued about silt loads now argue about evapotranspiration.

The dust that built this plateau took roughly two million years of Pleistocene wind. The erosion crisis took some thousands of years of farming. The greening took forty. The soil water those trees are drawing on fell as rain before any of them were planted, and at the rate the plateau recharges, nobody now working on it will see the ledger balanced.