How Luxembourg flood warnings went without action

Strong signs of an extreme flood event were available to authorities in Luxembourg several days before devastating floods struck in 2021, new research has found.

Despite the signals, the country's warning system was unable to translate the available information into effective action in time to prevent the disaster from unfolding. The study by University of Reading flood scientists shows that the outcome was shaped by structural features of the warning and decision-making system that remain largely unchanged.

The July 2021 floods were Luxembourg's most financially costly disaster on record, with damages exceeding €145 million and more than 6,500 homes inundated. In neighbouring Germany and Belgium, similar flooding caused catastrophic loss of life.

The study, published in the academic journal Natural Hazards and Earth System Sciences, found that forecasts showing elevated flood risk were available from 7 July onwards, a full week before the worst impacts. Yet official warnings did not escalate to the highest alert level until the evening of 14 July, by which time flooding had already begun in several locations.

Lead author Jeff Da Costa, a PhD researcher in hydrometeorology at the University of Reading, said: "The problem wasn't that forecasters couldn't see this event coming. Strong signals were there, and they were persistent across successive forecast updates. The problem was that the system was designed to wait for certainty as a trigger for action."

System constraints

The researchers found that Luxembourg’s warning system relied on fixed, deterministic decision rules that limited how probabilistic and spatially distributed forecast information could be used operationally. As a result, early signals of elevated risk did not lead to anticipatory measures before impacts occurred.

The study also found that responsibilities for forecasting, warning issuance, emergency coordination and response were distributed across multiple institutions, with no single mechanism to integrate information and initiate early, coordinated action. National-level crisis coordination was activated only after flooding was already underway. 

Da Costa said: "This is a structural issue. The system prioritises confirmation over anticipation, which makes it difficult to act on uncertain but potentially high-impact information.

“Instead, the legal and policy framework needs to make allowances for information that contains a level of uncertainty, yet can still trigger decisions. As things stand, people in Luxembourg are just not ready to interpret or act on the early warnings that they are getting."

Lessons for the future

Unlike Germany, Belgium, the Netherlands and France, which conducted independent reviews after the 2021 floods, Luxembourg did not commission an external evaluation of its warning system. The researchers found that the procedures and decision rules that shaped the 2021 response remain largely unchanged, so that similar extreme weather events could produce similar outcomes in the future. 

Co-author Dr Jess Neumann, associate professor of hydrology at the University of Reading, said: “Improving weather models or forecasting technology alone will not address this risk.

“What's needed are changes to decision rules, mandates, and how uncertainty is interpreted across institutions."

The research introduces a new conceptual model, described as the 'Waterdrop Model', to illustrate how forecast signals can be filtered out or delayed within systems not designed to process uncertain information collectively.

Professor Hannah Cloke, a co-author and flood forecasting expert at the University of Reading, said: "This research shows that having good forecasts is not enough. Countries around the world are investing heavily in better early warning technology, but these investments only reduce risk when institutions and people are able to act on uncertain information before the floods occur."

The findings have broader relevance beyond Luxembourg. Recent events, including the October 2024 floods in Valencia, Spain, have highlighted similar challenges in translating forecast information into timely public warnings and coordinated action. 

The paper 'Signals without action: a value chain analysis of Luxembourg's 2021 flood disaster' by Jeff Da Costa, Elizabeth Ebert, David Hoffmann, Hannah L. Cloke and Jessica Neumann is published in Natural Hazards and Earth System Sciences and is available at: https://doi.org/10.5194/nhess-26-343-2026

Image: Flooding on a street in Clausen, Luxembourg, July 2021. Image via Wikimedia Commons.