Sinking streams in karst

Why does a stream in limestone country sink into the ground and rise again in a valley it never flowed toward?

Walk down a stream in limestone country and it may simply stop. The water disappears into a hole at the foot of a slope, and below that the channel continues as a dry, boulder-strewn valley. Then a few kilometres away a large spring emerges at full flow — often in a different valley, on the far side of a ridge the stream was flowing away from.

The obvious objection is that water cannot go uphill, so it cannot have crossed the ridge. It did not. Something in the assumption behind the objection is wrong, and finding it is the whole problem.

The faulty assumption is that a surface divide is a divide for water. On a hillside of clay or granite it is: water runs downhill on the surface, and the ridge between two valleys does separate two catchments. But a ridge is a statement about the shape of the land, and underground water does not consult that shape. It moves down a hydraulic gradient — higher to lower water pressure — along whatever routes exist.

So what routes exist? This collection's walk-through of cave formation supplies them and I will not re-derive the chemistry: rainwater charged with soil carbon dioxide dissolves calcium carbonate, limestone is dense rock cut by joints and bedding planes rather than porous like sandstone, and a fracture carrying slightly more flow is dissolved slightly wider, so it carries more still. A few fissures dominate and the rest are abandoned. The fixed point of difference: cave formation explains how a hole gets made, this why the hole ends up somewhere the surface would never have predicted.

So ask what the winning fissures are aiming at. A conduit only grows by carrying water, and water only flows through it if there is somewhere lower to come out. That outlet is set by where the water table meets the land surface — typically the floor of the nearest deeply incised valley, whichever that is. Meanwhile the fissures run along joints and bedding planes whose orientation was set by geology long before the present hills were carved. The route is fixed by the lowest available outlet and the direction the rock happens to be cracked, and neither knows anything about the ridge overhead.

So the water goes under the divide, not over it — leaving the surface high in one valley and emerging low in another, the whole path downhill in pressure even beneath a hill.

The feedback then closes the argument, and it is the same one that makes the caves. Once a route to a low outlet exists it captures flow; capturing flow makes it wider; wider makes it capture more, including drainage from adjacent fissures heading somewhere less efficient. That is stream capture underground: the groundwater catchment annexes territory from its neighbours without leaving any mark on the surface. Given time, the underground and surface catchments are simply different shapes.

And corroborating evidence lies in plain sight. That dry valley below the swallet is the fossil of the surface route — the channel the stream cut when it still ran on top, before the conduit could swallow the whole flow. Not a puzzle to explain away but the record of the capture, and it revives in flood, when the swallet takes all it can and the surplus runs down the old channel.

How would one test this rather than merely tell a good story? By dye tracing, the standard method: put dye into the swallet and watch which springs it emerges from, and when. If drainage followed the surface, dye would only ever appear at springs inside the surface catchment. In practice it routinely appears across divides, and one swallet often feeds several springs, the split changing between low flow and flood as overflow routes activate.

The timing is the second half of the test, and it is decisive. Groundwater moving through the pores of a sandstone aquifer travels something like metres per year. Karst tracer velocities are commonly of the order of kilometres per day — recalled as an order of magnitude, varying enormously with flow. That is open-channel flow in a pipe, not seepage through rock. A slow, diffuse dye arrival would refute the conduit picture outright; a sharp arrival hours later at a spring in the wrong valley cannot be produced any other way.

A sinking stream is not an endorheic basin, though both look like water vanishing. An endorheic basin, treated elsewhere here, has genuinely no outlet: its water leaves as vapour and salts accumulate. A swallet has an outlet that simply happens to be invisible, and water and dissolved load are exported — which is why karst uplands are not salt flats.

Nor is the divide-crossing universal. Plenty of karst drainage resurges within its own surface catchment. The point is that the surface divide has no authority, so agreement where it occurs is coincidence rather than rule — which is why a spring's catchment boundary cannot be drawn from a contour map and must be established by tracing.

It is also why karst groundwater is so vulnerable: the filtering that makes an ordinary aquifer forgiving depends on slow flow through fine pores, and a conduit provides none.

One honest simplification: real karst is rarely purely conduit-fed. Most systems are dual, a fast conduit network embedded in a slower fractured rock mass that stores water between storms and sustains the spring in dry weather. The clean story above describes what dominates during and after rain; baseflow comes from the slower part.

• Why some resurgences reverse and become estavelles, taking water in during flood and giving it out in dry weather.

Every term the collection defines is gathered in the glossary.