March 2026
The most recent window in the archive, and the one that shows what the instruments look like when nothing is happening.
No quotable travel time in this window
Storm archive
What a Storm Looks Like Here
Four windows pulled out of the federal record and replayed gauge by gauge. Each answers the same question: when the water rose at the top of the creek, how long did the bottom of the creek have?
Why Past Storms Matter Here
A live reading tells a resident what the creek is doing. It does not tell them how much time they have, and time is what an evacuation runs on. The only honest way to show it is to take real storms out of the archive and run them back at their own speed.
These replays are built from the USGS instantaneous record, the same 15-minute data the gauges were writing while the storms were happening. Nothing is modelled, smoothed or filled in. Where the record is missing, and on Coyote Creek in 2017 most of it is, the page names the station, the years, and the reason.
The Windows
February 2024
A single fast pulse down Coyote Creek, with five of the eight stations reporting. The closest thing in the record to a rehearsal of 2017 on a working network.
17 hours and 30 minutes from the upstream peak to the downstream peak
January 2023
Nine atmospheric rivers in about three weeks. Valley Water opened its Emergency Operations Center on December 31 and declared a countywide flood emergency on January 4.
No quotable travel time in this window
February 2017
Coyote Creek left its channel in San Jose and about 14,000 residents were ordered out. The gauges were reading the whole time.
25 hours and 45 minutes from the upstream peak to the downstream peak
How These Storm Records Were Assembled
Each window was requested from the USGS instantaneous values service for all 14 Coyote Creek and Guadalupe River stations at once, with the daily values service as a fallback where the instantaneous record is empty. A peak is the maximum of the real fetched series, carrying the real timestamp of that maximum. An interval between two peaks is timestamp subtraction and nothing else.
An interval is quoted as a travel time only when the downstream peak follows the upstream peak inside 48 hours. Across a three-week run of storms those two peaks belong to different events, and subtracting them produces a number that looks like a lag and is not one. Those intervals are still published, with the reason.