How it works

1 · Read the radar, every 5 minutes

Met Éireann's Dublin radar completes a volume scan every five minutes. We take the sweeps, cut a horizontal slice at a fixed height, remove the ground clutter the radar always sees (the same hills and masts in the same places), and convert reflectivity to rain rate with the standard Marshall–Palmer relation. The result is a 1 km map of how hard it is raining right now, over the whole island.

The interference spoke filter

On 23 September 2026 the engine fired a cluster of false alarms. The radar had painted narrow wedges of weak echo along a handful of adjacent rays — gone the next frame, back the one after. That is what an external RF emitter looks like to a weather radar. Rain never lives on isolated rays, and nothing physical fills a ray all the way to the 240 km edge where the beam sits kilometres above the ground. So the filter now runs on each sweep before anything else: a near-full ray whose neighbours are near-empty is blanked as interference. The wedge never reaches the map, the motion estimate or your phone. The latest cycle's blanked-ray count is on the status page.

2 · Track the motion

Rain cells move, and over an hour they mostly keep moving the way they were. Optical flow between the last few frames gives every part of the field a speed and a heading (today's answer is one number for the island — typically 20–40 km/h from the south-west). We then carry the current rain forward along those vectors in twelve 5-minute steps to 60 minutes ahead. This is nowcasting by advection — the same method the national services use for the first hour, because for the first hour nothing beats it.

The stability gate

Before the forecast is published it has to pass a sanity check: did the wet area swing wildly between frames, did the motion estimate hit its clamps on most of the field? If so the cycle is markedreduced confidence — you still get the picture, but the app and this site say so. If the radar feed itself is late, you see radar late with the frame's age instead of a silently old forecast.

3 · Your spot, one sentence

For each of your places we sample the thirteen fields (now plus twelve forecast steps) in a ~3 km neighbourhood around the point, so a shower drifting a kilometre sideways still counts. The first step at or above your sensitivity threshold is the onset; the run of wet steps after it is the duration; the strongest step is the peak. Those three numbersare the sentence. The alert state machine then decides whether to push — with per-place warning time, quiet hours, a cool-down after each alert and a hard cap of four notifications an hour, so a flickering shower can never turn into an alert storm.

Honest limits

• Rain that hasn't formed yet. Extrapolation moves existing rain; it cannot foresee a shower that initiates within the hour. This is the hardest case in nowcasting and it shows up as lower detection at 30 minutes. We publish that number anyway.
• Growth and decay. Cells intensify and fade as they move; advection keeps them as they are. Peaks are "~" for a reason.
• Drizzle below the beam. Far from the radar the beam sits high; very light drizzle underneath it can go unseen — by the forecast and by the verification alike.
• Radar interference. The spoke filter removes the patterns we have seen; a new kind of emitter could still paint echo we would mistake for rain for a frame or two. The latest cycle's blanked-ray count is on the status page.
• Coverage edge and detail. Scores cover the Dublin radar's ~240 km range; spots near the edge get fewer, noisier samples. The continental previews run at ~2 km, not 1 — seecoverage.
• Radar-vs-radar. We verify against the same radar we forecast from, not rain gauges. Shared radar biases cancel out of the score, but what reaches your jacket can differ from what the beam sees.

A forecast you can't check is just marketing. Every 5 minutes we issue a forecast, wait, and score it against what the radar actually did —published weekly, misses included.