Latest available level
—Awaiting dataRIVER SEVERN / WORCESTER
Watch the river.
Explore what’s next.
Start with observations. Ask better questions.
Explore the Severn at Diglis, one reading at a time.
THREE POINTS OF VIEW
The river, in context
FROM OBSERVATION TO EXPERIMENT
Does upstream data
give us a head start?
Explore how the river changes at two locations, see when rain falls, and inspect our first small forecasting test.
THE INTERACTIVE FIELD NOTEBOOK
One river. More ways to see it.
Loading the research snapshot…
Every view uses the saved research snapshot. Live readings have their own refresh control below. Keyboard controls and tap-to-inspect are supported.
OBSERVATION DESK
Diglis river levels
Loading Environment Agency observations…
Change over approximately 6 hours
—Measured change, not a forecastTypical range at this gauge
—Historical context, not a flood-warning thresholdHow the level has changed
Metres above this station’s local datum · times in UK local time
See readings as a table
| Time (UK) | Level (m) |
|---|
HISTORICAL SNAPSHOT
Rain above. River below.
Loading recorded observations…
● Diglis● Bewdley▥ Local rainfall
River lines show change from each gauge’s first reading in the selected window, not absolute height. Rain bars are hourly totals only where all four 15-minute readings are present. Missing hours stay blank.
Download the research snapshot (JSON) ↓FIRST EXPERIMENT / NOT AN OPERATIONAL FORECAST
Does the extra data help?
Average error on the later, held-out observations. Shorter bars are better.
Observed level (teal) and experimental prediction (orange) on the held-out period. Every prediction uses earlier observations. This is a historical replay, not a forecast of what happens next.
How this experiment works & what it cannot prove
This short record is not evidence of reliable flood warnings. A single local rain gauge does not represent the Severn catchment. Historical publication delays and data revisions are not reconstructed, so real-world performance may be worse. Overlapping forecast horizons make errors dependent; these are descriptive scores, not confidence estimates.
Next: longer quality-checked archives, several flood events, catchment rainfall, reporting-delay checks and rolling time-based validation.
Download results and predictions ↓THE QUESTION WORTH TESTING
Can we see a rise
coming sooner?
Can rainfall and upstream river levels improve forecasts at Diglis, compared with simply assuming the latest level will stay the same?
We’ll measure the improvement before making any claims.
- 01 / NOW
Build a trustworthy observation desk
Real readings, timestamps, missing-data checks and downloadable evidence.
- 02 / NEXT
Assemble the historical record
Upstream and local rainfall snapshots are connected. Longer quality-checked archives are the next step.
- 03 / THEN
Test a small forecasting experiment
Inspect the first 1, 3 and 6-hour holdout comparison above. High-water event validation is still outstanding.
Small scope. Clear evidence.
What the numbers mean
Stage is height relative to a local reference point. It is not river depth or height above sea level. Levels from different gauges cannot be directly compared as if their reference points were identical.
What’s working today
A live Diglis view plus saved multi-station history and an export. An exploratory historical comparison now includes upstream and local rainfall observations. No background collection, notifications or validated operational predictions. Refresh requests the latest available data while this page is open.
How we’ll judge progress
First, forecast error against a simple baseline. Later, missed high-water events, false alarms and lead time using an explicitly defined research threshold. No accuracy claims until we have tested them.