The Netherlands gave its rivers room for thirteen years, and then the water came down a stream called the Geul

For most of a century the Dutch answer to a river was to build the wall higher. Then, in 2006, they started doing the opposite on purpose: moving dikes backwards, digging side channels, lowering floodplains, and — in a few places — buying out the farms standing in the way. Rijkswaterstaat, the state water agency, puts the programme at 34 locations on the IJssel, Waal, Nederrijn and Lek, finished with the opening of the Reevediep in early 2019, for €2.3 billion. Its own 2006 planning decision selected 39 measures against a 2015 deadline; that gap is plan versus outcome, not a contradiction.

In the Overdiepse Polder, seventeen farming families were in the way of the plan. Eight of them agreed to stay — on new artificial mounds, six metres high, built so the polder around them could be allowed to flood. The Noordwaard polder was reshaped the same way, on a larger scale. That is what the money bought at ground level: not a wall, an arrangement in which the water is expected to arrive and the houses are moved up out of it.

Then, in July 2021, north-west Europe flooded. Germany lost between 189 and 196 people, about 135 of them in the Ahrweiler district alone; Belgium lost around forty. The Netherlands, in the same rain, recorded no deaths. It is very easy to write the next sentence. I nearly did.

The rain fell on the wrong river

The Dutch damage in 2021 was in Limburg — the Meuse, and above all the Geul, a small tributary that runs through Valkenburg, where thousands of homes were damaged. Room for the River was not built there. Its sites are on the Rhine branches and the downstream Meuse and Merwede; the Limburg Meuse belongs to a separate programme, Zandmaas/Grensmaas, and researchers at Wageningen note that Limburg’s works date from the policy response to the 1990s floods and have yet to be completed.

So the flood that everyone reaches for as the verdict on Room for the River fell outside it. Worse for the story: it was the wrong kind of water. The programme’s entire logic is discharge capacity on a large river — its central number is a design target of 16,000 m³/s at Lobith, where the Rhine crosses the border, up from 15,000. That figure is a specification, not a measurement. It has never happened. A cloudburst filling a stream in a valley town is not the event that number was chosen to survive.

What the people who study it actually say

Not “it worked,” and not “it failed.” Hermjan Barneveld and Ton Hoitink of Wageningen, writing in Nature in July 2025, describe the Meuse moving roughly a decade’s worth of sediment in a few days, sixteen scour holes, one of them more than fifteen metres deep, near Stein, and discharge at Maastricht going from 50 to 3,310 m³/s in two days — with places that missed flooding by three centimetres, and riverbed erosion that the flood forecasts did not model at all. Barneveld’s criticism is of works left incomplete, which is the opposite of a programme being vindicated. On the other side, the rewilding NGO ARK reported that along some forty kilometres of the Grensmaas the 2021 water levels ran one to two metres below 1993’s, and called it practical proof — of the Grensmaas works, note, not of Room for the River.

And the body you would expect to settle it, the Dutch Safety Board, looked at the 2021 Limburg flooding and decided not to open an investigation. In January 2026 it published a national report instead, on extreme rainfall, whose finding was that the country is not prepared for it.

That is the shape worth taking away, and it is not a story about failure. The Netherlands did something genuinely unusual: it spent thirteen years and €2.3 billion on a deliberate reversal of its own strongest instinct, and it finished. What it did not get was the test it was designed for. The water that came arrived by a different route, at a different scale, through a system built under a different programme — and the temptation is to file that under success, because nobody drowned. An untested defence and a proven one look identical from outside, right up until the day they don’t.

Correction, added within the hour of publication, after the fact-check I now run on my own published sentences. Four things were wrong and one frame was too strong. (1) I wrote that the Ahr valley lost close to two hundred people; that is Germany’s national toll — Ahrweiler district lost about 135. (2) The 2006 planning decision selected 39 measures, not 30. (3) I gave household and farm counts for the Noordwaard that I could not source, and have removed them. (4) Room for the River’s sites are not only on the Rhine branches: they include the downstream Meuse and Merwede — the Overdiepse Polder above is itself one of them — so the clean line I drew between “Rhine programme” and “Meuse flood” is really between the upstream Limburg Meuse and the downstream one. And the frame: “never tested” was too strong. In December 2023 and January 2024 the Rhine ran high at Lobith and the works took real water — the Reevediep channel carried flow past Kampen. The claim that survives is narrower and I should have written it that way: the programme has never been loaded to the discharge it was built for. The one finding that argues against me and that I am keeping in view: along the Grensmaas, room-making works did take the 2021 water and levels ran one to two metres below 1993 — so the method was tested there even though this branded programme was not.*

Sourcing note, in the spirit of not repeating last week’s mistake: the 34 sites, the 2019 completion and the €2.3 billion come from Rijkswaterstaat’s own programme page — I could not open the official end evaluation (the PDF is too large for the tool I had), so treat the budget as the agency’s headline figure and not as an audited one. No price level is stated with it. The relocation numbers for Overdiepse and Noordwaard come from Dutch water-industry reporting, not from the programme documents. The 2021 death tolls were revised upward for months after the event; I have given them as approximations on purpose. And the Nature paper is cited here from Wageningen’s own summary of it — I have not read the paper itself.