The night the Guadalupe filled Canyon Lake

You can stand at a gauge on the lower Guadalupe, watch it read calm and clear, and be completely wrong about whether it’s safe. That’s the hard truth this flood wrote down, station by station, in numbers we kept. On July 16, 2026, a single wave crossed the upper basin and did a different, dangerous thing at every town it reached. The only way to understand it is to follow the water, in the order it arrived.

Before

Canyon Lake sat low. It had drifted down to 887.91 feet, better than twenty feet under its 909-foot full pool, after a long dry stretch. It was a reservoir with a lot of room in it. Upstream, the Hill Country above Kerrville is thin soil over limestone: ground that doesn’t drink a hard rain so much as shed it, funneling the water off the rock and into the river almost as fast as it falls. That geology is the whole story. It is why a storm here becomes a wall of water in hours, not days.

The wave

It came in the dark. At Hunt, up in the headwaters, the river hit 40,400 cubic feet per second at 3:35 in the morning: most of a season’s worth of water past a single point, before dawn. Seventy-five minutes later, at 4:50 a.m., Kerrville took 49,800 cfs. The wave was moving, and building.

By 8:00 a.m. it reached Comfort and peaked at 200,000 cfs, roughly five times what had crossed Kerrville four hours earlier, the runoff of the whole upper basin stacking into one crest. Between those two towns, the gauge at Center Point simply stopped rating the water. It was pushed past what it can measure and flatlined, which is why the record there is a gap, not a number. The flood didn’t skip Center Point. The instrument just ran out of river.

Downstream at FM 474 near Bergheim, the crest arrived in the afternoon, 2:30 p.m., still carrying 196,000 cfsand standing nearly 58 feet on the gauge. Spring Branch, the last main-stem station before the lake, didn’t crest until half past midnight, almost a full day after Hunt.

That lag is the part everyone gets wrong. All day, people looked at a downstream gauge reading low and read it as reassurance: the flood must be over. It wasn’t over. The water hadn’t arrived yet. A quiet gauge between you and a crest upstream is not the all-clear. It is the calm in front of the wave. The replay on the event record lets you drag through it and watch each gauge light up in turn. It makes the thing obvious in a way no warning quite does.

What the forecast knew, and when

This is the section only the archive can tell, because the Weather Service keeps only its currentforecast. Every re-issue erases the one before it. We saved all eight issued for Canyon Lake’s inflow during the event. The first, published while the wave was still building upstream in the dark, reached for an inflow crest of 40.1 feet. The river delivered 36.75. Each later issuance walked the number down as the real readings came in.

The point isn’t that the forecast missed. The point is how brutally hard this basin is to call in real time: a crest can build and pass faster than a forecast cycle can catch it, and the people issuing those numbers were working in the same dark, off the same rising gauges, as everyone else. The overlay on the event record traces every issuance against what the river actually did. Read it as respect for how fast this country floods, not a scoreboard.

The lake

Twenty feet of empty reservoir met the whole basin’s runoff. Canyon Lake climbed 21.37 feet, from 887.91 to 909.28 feet, a hair over its 909-foot full pool, while the dam held its release near zero (it never let out more than about 150 cfs) and took the blow so the towns below it didn’t. There’s a handoff almost nobody outside the basin knows about: once the lake tops full pool, day-to-day management of the dam passes from the U.S. Army Corps of Engineers to the river authority. This flood pushed the lake across exactly that line.

After

The river came down about as fast as it went up, and by July 22 the basin had settled back under the threshold that opened this record five days earlier. What’s left is the archive itself: every 15-minute reading at every main-stem gauge, the lake’s whole climb, and all eight forecasts. The official record, kept, so this event can be checked and remembered whether or not the federal services still serve it in ten years.

And the standing note, because it matters most: this site informs, it is not the go/no-go authority. When the water’s up, the gauge nearest you is not the whole story. Confirm with a local outfitter, and when in doubt, stay off the river.