What a "100-year flood" actually means

One percent, every year, forever - unwrapping hydrology's most misunderstood phrase, and how it differs from our own percentiles.

Every time a river makes the news, the phrase arrives with it: a 100-year flood. And every time, a reasonable person draws the reasonable conclusion — that this is a flood that happens once every hundred years, that the last one has been used up, that there's a century of quiet in the bank.

Almost none of that is true. The term is one of hydrology's most useful ideas wrapped in one of its worst names, and unwrapping it changes how you read every flood story — including this summer's.

It's a probability, not a schedule

A 100-year flood is a flood with a 1% chance of being equalled or exceeded in any given year. That's the entire definition. Hydrologists call it the 1% annual exceedance probability (AEP) flood, and the "100-year" label is just the inverse of that probability dressed up as a timespan — 1-in-100 per year, so "100-year."

The river does not keep a calendar. Each year is an independent draw, and the dice have no memory: a basin that took a 100-year flood last spring faces the same 1% chance this spring. Back-to-back major floods aren't a violation of the statistics — they're an ordinary, if unlucky, outcome of them. And the quiet corollary surprises people even more: over any given hundred years, there's a decent chance the "100-year" flood doesn't happen at all, and a roughly equal chance it happens two or three times.

The framing that actually lands: over the life of a 30-year mortgage, a property in the 1% zone faces about a 26% chance of seeing that flood at least once. One-in-four over a mortgage sounds — correctly — like a very different risk than once-a-century.

Where the number comes from

The 100-year figure isn't observed; it's estimated. For each gauged river, hydrologists take the annual peak series — the single highest flow from every year of record — fit a statistical distribution to those peaks, and read off the flow that sits at the 1% exceedance level. Engineering standards, floodplain maps, insurance requirements, and the height of the dike protecting a town all flow downstream from that estimate.

Which means the number inherits every limitation of its inputs. A river with 50 years of records is being asked to characterize a 1-in-100 event from a sample that may contain zero examples of one. Two gauges away, a different record length can produce a meaningfully different estimate for the same river. This is one reason long station records are treasure — every additional year of peaks sharpens an estimate that important decisions lean on. When you see "50 years on record" on a Records & Range card, that number is quietly one of the most valuable things about the station.

Why the goalposts move

Here's the uncomfortable part: the 100-year flood is estimated from the past, and rivers are increasingly refusing to behave like their past.

The classical method assumes stationarity — that the statistical character of floods isn't changing over time. But land use changes (pavement sheds water that prairie absorbed), channels get engineered, and a warming atmosphere holds and delivers more moisture per storm. When the underlying flood behavior shifts, yesterday's 100-year estimate can quietly become today's 40- or 60-year reality. Agencies re-run the analyses and redraw the maps, but the maps always trail the water. It's why "we've had three 100-year floods in a decade" is less a statistical scandal than a signal — the estimates were calibrated to a river that no longer quite exists.

The 2026 lesson

This summer offered a live case study in why the phrase misleads. In July, the Nueces basin in Texas produced levels unseen in half a century of records — at Cotulla, the river climbed past the highest daily level in the station's entire archive. Weeks earlier, the Assiniboine in Manitoba crested metres above bankfull in July, a month the historical charts show as the river's calm season, after storms delivered a winter's worth of water in days.

Were these "100-year floods"? That determination belongs to the agencies who run the frequency analyses, and the honest answer is that the label matters less than what the gauges plainly showed: water at or beyond the edge of everything the record contains. Which is really the only fact a person standing near the river needs.

What Nibi Metrics shows instead — and why it's different

A note of honesty about our own numbers, because the concepts are easy to blur.

When a Nibi Metrics page says a river is at the 100th percentile or shows "record level for the date," that is not a return-period statement. Our percentiles compare today's level against the distribution of levels on this same calendar date across the station's history — a measure of how unusual today is for late July, not an estimate of annual flood frequency. A river can sit at its 100th percentile for the date during a modest event, and a true 1%-AEP flood will blow past every percentile we have.

The two views are complements, not competitors. Return periods answer the engineering question: how should we build for this river? Date-percentiles and records answer the situational one: how unusual is what I'm looking at right now? When our charts showed Cotulla surpassing its all-time high in real time, no return-period label could have said it more clearly.

So the next time the phrase arrives with the flood: translate it. Not "once a century" — one percent, every year, forever. The river isn't counting. The gauges are.

Records, percentiles, and typical-for-the-date context on every station — the situational half of the story, live.