One of the most common questions we receive is some variation of: “has this site gotten better?” It's a completely reasonable thing to want to know. It's also a question we are very careful about answering.
Saying a site has “improved” requires more than a lower number this year than last. It requires ruling out rainfall as the cause. It requires enough years of data to distinguish a trend from noise. It requires honest accounting of sensor gaps. And it requires being transparent about what the data actually measures — and what it doesn't.
This article explains the methodology WaterWatch uses to assess whether a site's discharge pattern is genuinely changing, the limitations of that assessment, and why we deliberately avoid making strong improvement claims on insufficient evidence.
What we actually measure
WaterWatch tracks four core metrics for every CSO site. These come directly from the Event Duration Monitoring sensor data — nothing is estimated or derived from secondary sources:
| Metric | What it tells you | Limitation |
|---|---|---|
| Annual spill hours | Total time discharging in a calendar year | Highly sensitive to rainfall — a dry year always looks better |
| Annual event count | Number of distinct discharge episodes | Short events and long events count the same |
| Average event duration | How long each discharge episode typically lasts | Skewed by a single very long event |
| Active months | How many months of the year saw at least one discharge | Does not reflect severity within each month |
No single metric is sufficient. A site can reduce its annual spill hours while increasing its event count — which might indicate more frequent but shorter overflow activations (possibly a sign of infrastructure changes) or could simply reflect a drier summer. WaterWatch looks at all four together, across multiple years, before forming any view.
The rainfall problem
This is the central confounding variable in all CSO analysis, and it is almost never adequately accounted for in media coverage. Combined sewer overflows exist precisely because sewers carry both sewage and rainwater. When it rains heavily, flow surpasses treatment capacity and the overflow discharges. That is the designed function of a CSO.
A site that discharged for 300 hours in a wet year and 180 hours in a dry year has not necessarily improved. A site that discharged for 300 hours in a dry year and 180 hours in a wet year almost certainly has. Year-on-year comparisons without rainfall context are nearly meaningless — but they are the figures most commonly cited.
WaterWatch cross-references discharge trends against Environment Agency rainfall data from monitoring stations in the same catchment. A falling spill-hour trend that correlates with falling rainfall is not treated as improvement evidence. Only trends that persist across years with comparable or higher-than-average rainfall carry meaningful weight.
Site A shows a consistent downward trend independent of rainfall. Site B's numbers track closely with rainfall variation — a dry 2024 flatters the figures.
How many years of data do we need?
WaterWatch has discharge records back to April 2022. That gives us roughly three and a half full calendar years to work with. For some purposes — like identifying the worst sites in a given year — that is entirely sufficient. For genuine trend analysis, it is borderline.
A single year's reduction in spill hours is not a trend. It is a data point. Two years of reduction is encouraging but could still be rainfall-driven. Three or more consecutive years of reduction, in rainfall-comparable conditions, starts to constitute credible evidence of structural change.
We are currently able to make meaningful trend assessments for sites that have been consistently monitored since 2022 — subject to the sensor gap caveats below. For sites where significant sensor gaps exist in early years, our assessable baseline is shorter and we treat any improvement signal accordingly.
“A single year's improvement is a data point. Three years of improvement in comparable rainfall conditions is a trend. We don't conflate the two.”
The sensor gap problem
EDM sensors are not perfectly reliable. Periods of sensor maintenance, failure, or API unavailability create gaps in the record. WaterWatch surfaces these gaps explicitly — they appear as distinct offline periods in site histories, not as zero-discharge periods.
This matters enormously for trend analysis. A site that appears to have discharged for 200 hours one year and 150 the next might simply have had more sensor downtime in the second year. A gap in the data is not evidence of improvement. It is a gap — and we treat it as one.
WaterWatch does not draw improvement conclusions for any year where a site has more than 30 days of sensor offline time unless the sensor gap period can be demonstrably attributed to a non-discharge period (e.g., scheduled maintenance in dry summer months with corroborating rainfall data).
The three-signal framework
When WaterWatch assesses improvement at a site level, we require evidence across three independent signal types. Only sites where all three signals point in the same direction are classified as showing credible improvement:
What we currently show on site pages
The WaterWatch site page for each CSO currently shows annual spill hours, event counts, and full discharge history back to 2022. We do not currently display a trend verdict or improvement indicator — deliberately. Our dataset is three years old. We are building towards the four-year baseline we consider the minimum for making confident trend assessments at a network-wide level.
What we do show is the raw data that lets users form their own view. If a site had 600 hours in 2022, 580 in 2023, 420 in 2024, and 390 in 2025 — and you can see rainfall was broadly similar across those years — you can draw your own conclusions. We're not withholding the trend; we're being honest about the fact that three data points is a nascent trend, not a proven one.
What ‘improvement’ does and doesn't mean
Even a genuine, structural improvement in spill hours does not necessarily mean the ecological situation in the receiving watercourse has improved. That depends on factors entirely outside our data: dilution ratios, river flow, season, species composition, nutrient loading, and cumulative impacts from multiple sites discharging into the same catchment.
A site that reduced from 800 to 400 annual spill hours into a small chalk stream may still be causing significant ecological harm. A site that discharged for 200 hours into a large tidal estuary with high dilution capacity may cause minimal harm at those volumes. WaterWatch does not have the data to make those comparisons, and we will not pretend that reduced spill hours automatically equal reduced harm.
| What ‘improving’ means in our framework | What it does NOT mean |
|---|---|
| Fewer spill hours in rainfall-comparable conditions over ≥3 years | The watercourse is ecologically healthier |
| Fewer discharge events and shorter average duration | The site is compliant with its permit conditions |
| Infrastructure investment has demonstrably changed discharge behaviour | Downstream pollution incidents have reduced |
| A multi-year trend persisting across wet and dry years | Thames Water has ‘fixed’ the site |
The regulatory context
The Storm Overflow Discharge Reduction Plan, published by Defra in 2022, sets targets for water companies to eliminate environmentally harmful CSO discharges at the majority of storm overflow sites by 2035, with interim milestones in 2027. Thames Water is required to produce annual reports on progress.
WaterWatch's data is independent of these reports. We do not rely on self-reported improvement claims — we track the sensor data directly and apply our own methodology. Where our analysis differs from official reporting, we will say so plainly and explain why.
WaterWatch has no relationship with Thames Water, the Environment Agency, or any water industry body. Our improvement assessments are based solely on our own analysis of publicly available sensor data. We will not moderate those assessments based on commercial or reputational considerations.
What's coming next
We are building towards a more complete improvement analysis layer. Planned additions include rainfall-normalised trend scores at the site level, catchment-level discharge comparisons that aggregate multiple sites discharging into the same watercourse, and year-on-year delta indicators on site pages that are clearly labelled with their confidence level and caveated with rainfall context.
We are also in the process of integrating historical rainfall and river level data to provide richer context for discharge events — enabling a more complete picture of whether a given overflow was triggered by genuine extreme rainfall or by infrastructure that should have coped.
Until those features are live, we ask you to treat annual spill hour comparisons as what they are: useful context, not verdicts. The data is there. The trend signals are forming. We'll tell you when we're confident enough to call them.
Sources: Thames Water EDM open data API · Environment Agency rainfall monitoring network · Defra Storm Overflow Discharge Reduction Plan (2022) · Water Industry Act 1991, Schedule 22 · Environmental Audit Committee: ‘Water Quality in Rivers’ (2022)