Bampton Flood MonitorShill Brook · Oxfordshire
Try it yourself

Beat the model

Two models make every prediction on this site. Below are their real settings. Change them, replay real storms on Shill Brook, and see how your version would have done over six winters.

What the colours mean

Our prediction for how high Shill Brook will peak:

  • Green — under 1.06 m. Normal.
  • Yellow — 1.06–1.40 m. Brook up, nothing to do.
  • Orange — 1.40–1.78 m. Low-lying land may flood. Get ready.
  • Red — 1.78 m and over. Property may flood. Act now.

How to play: Move the sliders, replay real storms, and try to 1) warn of every real flood, 2) warn as early as possible, 3) raise as few false alarms as possible. You'll find that getting better at one usually makes another worse. That's the whole problem.

Remember: a warning doesn't stop a flood. It buys time — to move the car, lift the rugs, check on the neighbours. The best model gives the most time with the fewest false alarms.

Replay a real storm

Loading six winters of river data…

Early model

The live model (v3.1). Believes the rain forecast, so it warns early — and jumps.

+0.00 m
+0.05 m
×1.8
More settings (7)
Wetness under 20 (dry)
×0.5
+0.00 m
Wetness 20–30 (moist)
×0.8
Wetness 30–38 (wet)
×1.2
Wetness 38–45 (saturated)
+0.15 m
Wetness 45 and over (very saturated)
×2.5
+0.25 m

Peak cap: 2.50 m (fixed)

Red only when both agree

On — the site’s rule today. Red needs both models; if only one says red, the site shows orange.

Your model over six winters

2019/20 – 2024/25 · 16 real floods, 2 of them red

Scores replay past winters. They're not a promise about the next one — only 16 real floods exist in the record, so a model tuned too tightly to them can do worse on new storms.

The same for both models (fixed)

  • Effective rain = rain in the last 24 hours + 0.4 × the forecast for the next 24 hours + 0.15 × the forecast for the 24 hours after that.
  • Rise = effective rain × 0.015 m per mm × the rain multiplier for today’s wetness band.
  • Predicted peak = brook level now + rise + wetness bonus, capped at 2.50 m. Then the colour.
  • Wetness is the site’s wetness index — a running score of recent rain, worked out once a day at midnight. Its band edges stretch outside winter, when the ground dries faster: ×1.5 in April, May, September and October; ×2 in June to August.

In the live code these are getApiMultiplier and getSaturationBonus in lib/prediction.ts; the cautious model is the same sum with its own table.

How the two models work →