The number and its counterfactual
Matt Tomkins and colleagues at the University of Manchester modelled 2,356 canal sections in Great Britain and Ireland at 15-minute steps through 2022, and report a median daytime effect of 0.77 degrees Celsius within a 20 to 50 metre band along the banks, reaching 1.5 to 2.3 degrees Celsius during heatwaves. The quantity being reported is a difference: the model computes the energy balance of the water, computes it again for an equally large surface of asphalt, and converts the gap into a potential change in air temperature.[1]
So the comparison built into the figure is asphalt. That is a defensible choice for a national model and the authors state it in the first paragraph of their methods, but it fixes what the number can answer. It answers how much cooler the air near a canal is than it would be if that same strip were paved. A city considering where to spend an adaptation budget is usually asking a different question, which is how a canal-side street compares with a similar street that has no canal, and whose surfaces are brick, grass, water-cooled tarmac and shade in some untidy mixture.[1]
Six canals, and what they measured there
The validation section is short and honest. Modelled water temperature was compared with measured water temperature at 40 to 60 centimetres depth on six canals, using data supplied by the Canal and River Trust: the Leeds and Liverpool, the Shropshire Union, the Huddersfield, the Nottingham and Beeston, the Bridgwater and Taunton, and the Coventry. Residuals sit mostly within 1 to 2 degrees Celsius and Spearman coefficients run from 0.57 to 0.81. The authors then do something worth noting: they tell the reader to read the effect size rather than the p-value, because with 8,516 to 34,727 paired intervals even a weak correlation clears significance.[1]
What that comparison covers is the water. The air-temperature difference, which is the figure that will be quoted, never meets a measured air temperature anywhere in the paper. The model produces it as a contrast between two modelled surfaces, and the residual on the input that was checked is of the same order as the annual daytime effect being claimed. The fair reading is that the two quantities are not interchangeable in either direction: because the difference is taken between two surfaces driven by the same weather, a systematic offset in water temperature would partly cancel, which could leave the contrast steadier than its input residual suggests. That is a plausible defence and it is also untested here, which is exactly the gap a reader should keep open.[1]
The shape of the question repeats. Four days ago the same column asked where an earthquake model actually gets tested, and settled on the part of the design that had been locked before the observation arrived. Here the locked part is the water: measured independently, at six sites, over a full year. The air-temperature claim rests on the physics linking the two rather than on a second measurement, and the distinction is worth carrying into any planning document that quotes the number.[1], [2]
What would settle it
The paper fences its own claim in three places, and the fences deserve to travel with the headline. Night-time warming is reported at an annual median of 0.16 degrees Celsius, rising to about 0.4 degrees Celsius after heatwaves. The reduction in heat-risk hours, 58 per cent at moderate risk and 74 per cent at high risk, comes from adjusting air temperature alone and leaves out the humidity that evaporation adds. Sensitivity testing puts the limit of that omission at a 10 per cent rise in humidity, beyond which the comfort benefit turns.[1]
The test that would convert the modelled contrast into a measured one is cheap by the standards of climate work: paired air-temperature sensors in a 20 to 50 metre band from a canal bank and at matched inland sites with comparable built form, run through a summer that includes a heatwave. If such paired measurements are published by the end of September 2027, the observed daytime difference during heatwave hours should land inside the band Nature Communications reports, 1.5 to 2.3 degrees Celsius. A result well below that band would leave the water-temperature model intact and point instead at the asphalt counterfactual as the source of the effect.[1]