Flood models need to behave like real rivers

Global flood maps could be improved by dropping a decades-old assumption that all rivers behave the same way, according to new research.

Most global flood models assume a river fills to the top of its banks about once every two years. But a major international study published in August 2026 has found this does not hold everywhere. Rivers in the tropics reach these "bankfull" conditions far more often than rivers in arid regions.

Professor Hannah Cloke, Regius Professor of Meteorology and Climate Science at the University of Reading and a co-author of the study, said: "Real rivers are not interchangeable pipes. If we want better flood maps, our models have to represent the river that is actually carrying the water."

Published in Nature Communications, the study used 2,657 real-world measurements of bankfull river discharge, combined with data on climate, river flow, landscape and channel shape. Using machine learning, the team estimated bankfull discharge along 2.87 million kilometres of rivers worldwide, at roughly one-kilometre intervals.

The study was led by Dr Yinxue Liu, now a Vice-Chancellor Independent Research Fellow at Loughborough University, with Dr Michel Wortmann as joint first author.

The team found big differences between climates. On average, rivers reach bankfull conditions every 1.5 years in the tropics, 1.8 years in temperate regions, 2.8 years in cold regions, and 4.3 years in arid regions.

This is important because bankfull discharge tells a flood model how much water a river can hold before it spills onto the floodplain. Give a model too much capacity, and it keeps water inside the banks when the real river would already be flooding.

Using the standard two-year assumption, the modelling overestimated bankfull discharge by around 54% in tropical regions. In arid regions, it underestimated bankfull discharge by about 10%.

So the two-year rule is not wrong everywhere – it is a reasonable approximation of bankfull flow – but it hides regional differences that can bias flood simulations.

Dr Liu said: "The two-year rule was a reasonable inference from the evidence available in the 1950s. What has changed is the data."

The new dataset gives flood modellers an alternative. Instead of giving every river the same assumed capacity, models can now reflect differences in climate and, at higher resolution, the characteristics of individual river reaches.

Better modelling of river channels could improve estimates of where and how deeply floodwater will spread, helping with flood-risk assessment, infrastructure planning, insurance and climate adaptation.

Notes to editors

The paper, "Global estimation of bankfull river discharge reveals distinct flood recurrences across climate zones", is published in Nature Communications.

The research is part of EVOFLOOD: The Evolution of Global Flood Hazard and Risk, a £3.7 million research programme funded by the Natural Environment Research Council (NERC). EVOFLOOD is developing a new generation of global flood models that better capture how river channels and floodplains vary and change.

The EVOFLOOD collaboration brings together researchers from the Universities of Reading, Hull, Loughborough, Southampton, Oxford, Bristol, Birmingham, Exeter, Durham and Brighton. Some researchers and elements of the project funding have moved institutions during the lifetime of the programme.

Full reference: Liu, Y., Wortmann, M., Hawker, L. et al. Global estimation of bankfull river discharge reveals distinct flood recurrences across climate zones. Nat Communications 17, 8179 (2026). https://doi.org/10.1038/s41467-026-76433-3