How the dry-spill test works.

Everything here is built from public data and republished as public data. This page is the whole method, written plainly: what we read, how we decide a discharge happened with no rainfall to point to, and where the data can mislead.

What we measure

The record is a monitor activation, not a verified discharge. Water companies operate Event Duration Monitoring (EDM) on storm overflows and publish status in near real time. We cross that status against rainfall recorded at the nearest gauges. When an overflow is discharging and no rainfall has been recorded, we describe it as discharging outside the permit's stated basis — rainfall and snowmelt. We never assert why, and we never assert legality.

Sources & licences

The dry-spill definition

The regulator's own bar comes first. The Environment Agency defines a dry day as one where less than 0.25 mm of rain fell on that day and on the preceding day, and from 2025 it requires water companies to report spills on such days. Every discharge this site classifies as dry meets that definition — our thresholds are set at or inside the EA's, never looser — and then has to clear three further guards the EA's definition does not impose.

A discharge is classified as a dry spill only when rainfall at the assigned gauges stays below every one of these thresholds. Two of them look back well beyond the last day, because saturated ground keeps overflows running long after gauges flatten.

For reference, a discharge is treated as rainfall-associated when the trailing 6-hour total reaches 2 mm or the trailing 24-hour total reaches 6 mm. Historical instants are walked at 15-minute granularity. Rainfall between the dry and rain bands classifies the discharge with the rain, never against it. Where this site goes beyond the EA definition: the 7-day antecedent guard, the sustained-dry hold, and the rule that thin or stale gauge data means unverified, never dry.

The three guards, in plain English

The 7-day antecedent guard. A wet week leaves the ground saturated. Overflows can then run for days on genuine infiltration after the rain has stopped and the gauge reads zero. So a discharge shortly after a wet week is never called dry, even if the last 24 hours were dry.

The sustained-dry (drain-down) guard. Immediately after real rain, catchments drain down and overflows tail off. To avoid catching that tail, the dry condition must hold for the whole discharge, or for the trailing 12 hours, whichever is shorter, before we call a discharge dry.

The "we must actually have the week" guard. The antecedent test above is only meaningful if our rainfall record reaches back a full week. When it doesn't — the first days of monitoring a new gauge, or after a long gap in a gauge's readings — the seven-day total is small simply because the data isn't there, and a discharge could be called dry without the question ever really being asked. So a dry classification also requires that at least one nearby gauge supplied 75% of its expected readings across those seven days. Missing history means unverified, never dry.

When the test cannot run

A false dry-spill is the one mistake this product will not make. If the rainfall evidence is thin, the discharge is marked unverified rather than dry. The test does not run when:

• the nearest usable gauge is more than 10 km from the outfall;
• the freshest reading is older than 90 minutes (the gauge has gone quiet);
• no assigned gauge supplied at least 80% of its expected readings over the trailing 24 hours.

Brief activations — the 15-minute display rule

EDM monitors are documented — by the Environment Agency and by the companies operating them — to false-trigger on debris, weed growth and telemetry noise, and some monitors in this region log many activations of around two minutes in a single day. An activation shorter than 15 minutes is therefore recorded, published on its own permalink, and included in the raw CSV export in full — but it is labelled a brief activation, and it is never counted in headline dry-spill figures: not in the front-page count, not in a per-outfall dry-spill tally. Duration totals (hours discharged, hours with no rain recorded) are unaffected — a two-minute activation contributes two minutes, not a spill. Outfalls whose monitors log many brief activations carry a monitor-behaviour note on their page. This is a display threshold, not a change to the data: nothing is deleted, and you can count every activation yourself from the export.

What the front page’s activity list leaves out

The activity list on the front page is a display of the most recent monitor activations, not the full record. Any single outfall takes at most 3 lines in it: one monitor switching on and off every few minutes was holding around half the list and burying every other site in the region. Discharges with no rainfall recorded at the nearest gauge are exempt from that cap — the signal the site exists for is never trimmed away. Nothing is deleted: every activation, including the ones the list does not show, is on its own permalink, in the CSV export, in the feed and on the outfall’s own page.

How close the gauges are

The dry test only ever runs against rainfall gauges within 10 km of the outfall. Nearest assigned gauge, across the 1345 monitored outfalls:

Known failure modes

• Monitors misfire. Companies note that even weed growth can trigger an activation. An activation is not by itself a verified discharge; where the feed provides a confidence caveat we surface it, and activations under 15 minutes are excluded from headline counts (see above).
• Missed endings. The feed carries only each asset's latest event. If a new event starts before we observed the old one's end, the older event's duration is unknown — it is recorded with end not observed and contributes nothing to duration totals. Unknown is never presented as measured.
• Gauges are sparse. West of Exeter the gauge network thins, so some genuine discharges sit in the unverified bucket rather than being called either way.
• Emergency overflows. Permits separately allow emergency operation on pump or power failure, and the EDM feed does not distinguish those from storm activations. We acknowledge that defence before it is made: a dry classification describes the rainfall record, not the cause.
• Two methodologies. Audited annual returns count spills on a 12/24-hour basis; the live feed reports raw durations. We never mix the two in one figure without labelling which is which.

Correction policy

Classifications can change as late gauge data arrives. When that happens the event is reclassified and the change is recorded on its permalink — every per-event page carries its full reclassification history in public. We would rather show a correction than hide one.

Feed health

A source going quiet is a data point in itself. Last successful poll and the share of polls that succeeded over the trailing seven days, per source:

← Back to the live readout