It's not just the rain

Sewage in English rivers has been a story going round the news for years. Channel 4's Dirty Business, broadcast earlier in 2026, pulled together a version of the story most viewers will already have seen pieces of: chronic under-investment, ageing combined sewers, a regulator that struggles to keep up.

Storm overflows are a deliberate part of the network. Ofwat, the regulator for water companies, describes them as “relief valves when the sewerage system is at risk of being overwhelmed” — releasing excess rainwater and wastewater into rivers or seas to prevent sewage backing up into homes during heavy rainfall. Anything else, by the regulator's own definition, is outside their design intent. On the same page, Ofwat acknowledges growing public concern that overflows are operating when there has not been significant rainfall. That is the question this project sets out to test.

Every water company in England runs storm overflows. The standard reasoning, when challenged on a bad year, points to rainfall as the explanation. A wetter year mechanically produces more spills. This defence is partly true.

The question we set ourselves is to check whether the defence based on rainfall is sufficient. Once you control for rainfall, and not just total rainfall, but the kind of rainfall the sewage system is actually stressed by, how much of the variation in spills is left unexplained?

The Environment Act 2021 sets the floor for what water companies have to disclose about sewage discharges. Two sections do most of the work, and neither is shaped to help an outsider analyse the system.

Section 81 requires near-real-time reporting of spill events as they happen, but says nothing about preserving the historical record. A company has to tell you a spill is happening; nobody is obliged to keep the diary once it is over.

Section 80 requires an annual return: a single yearly count and total duration of spills per overflow location. For most of the country, that annual aggregate is the most granular long-run data anyone outside the water company is allowed to see.

The sensors that record those spills, called Event Duration Monitors (EDM), were only fully rolled out across the network by the end of 2023. Earlier years exist in the same aggregate format but with sparse sensor coverage.

And the entire spill datasets are self-reported. Water companies have been fined in the past for deliberately misreporting their performance data. Auditing them sits well outside what is possible from outside the regulator, therefore we've assumed their self-reported data is truthful in our analysis.

There is one exception to the Section 80 ceiling. Thames Water publishes a detailed historical record of its own storm overflow events through a public API: start time, end time, location, going well beyond what the annual return contains. It is a voluntary disclosure, made in the middle of a long public row about the company's finances and environmental record.

The disclosure gives outsiders a resolution of data that the regulatory floor does not require of anyone. That happens to make Thames Water, the company often held up as a cautionary tale, the only one we can actually study at the event level.

Annual totals tell us whether a year was bad. Event-level data tells us why. With start and end times for every individual spill, we can see whether a given discharge happened during a storm, after a storm, or during a stretch of dry weather. The annual return doesn't have the resolution to answer any of that.

Thames Water's operating area runs from Gloucestershire to Kent, taking in urban and rural, chalk and clay. That internal range is enough for the rainfall-spill relationship to vary meaningfully within it, which is what the analysis depends on. The fuller question of how representative Thames Water is comes later, in the limitations.

The rainfall mechanism is real. Across the four-year Thames Water panel, days with new spills are nearly four times wetter on average than quiet days, and the day-level correlation between discharge counts and antecedent rainfall is r ≈ 0.66. Rainfall is the dominant immediate driver of any individual spill event. Companies are not lying when they invoke the weather.

But it is not the whole story. For Thames Water, the only company we can study at event level, 59% of 2025 spills started below the wet-anomaly threshold, with a stable 4–9% across recent years beginning after fully dry weeks. And after stripping out the rainfall each network actually faced, the gap between the best- and worst-performing operators is still around 130 percentage points of rainfall-unexplained residual. That spread is too large to attribute to weather.

What is left sits with things this analysis cannot measure directly. Combined sewers built for Victorian populations clog more often than they used to. Blockages of fatbergs, wipes, and debris trigger discharges with no weather signal at all. Maintenance regimes vary by operator. Networks age. Pumping stations fail. Some companies appear to be running their systems harder than others, in ways that show up as systematically more discharges per unit of rainfall.

None of this is hard evidence in the strict sense. Event-level data exists only for Thames Water; the other nine companies give us a single annual count per outflow, with no event timing. Rainfall is measured on a ~9km Open-Meteo grid, not at each individual overflow. The spill counts themselves are self-reported, and have, on occasion, been deliberately misreported. The 1σ threshold and the regression specification are both analytical choices among several plausible ones. The findings should be read as directional, not as precise apportionments of blame.

Back at Ufford Hole on the Deben, the bathing-water notices are still going up. On this analysis, Anglian Water spills roughly a quarter less than rainfall alone would predict. They are not the standout bad actor of the national dataset. But "not the worst" is not the same as "fine". No amount of raw sewage pollution can reasonably be called "good" when it has such detrimental impacts on the environment and on people's lives.