MIDDLE KINGS · good between 1,800 and 900 cfs at Rogers Crossing
Wrapped: Jun 21 → Jul 01 — a 10-day window
Gauge read 901 on Jul 04 — it hovered in the 850–890s
for a few days, then let go. 64% snow bought 10 days; the median year gets 21.
Our call
April 1 said
mid-June — from the snow survey alone, 11 weeks out
June 1 said
Jun 15 — with spring temps in hand, 3 weeks out
River did
Jun 21 — 6 days behind the call; the cold spring pushed
harder than we modeled
How this season played out
The gray dash is the median year on this gauge (1988–2025) — still up near 3,700 cfs on days
this season was already dropping into the band. That gap is the thin snowpack; the slow, even slope
is the cold spring. (Daily wiggle = the melt cycle, high each evening, low each morning —
simulated here from the daily gauge; the live page runs on hourly data.)
Why we called it
Snowpack was 64% of normal on April 1 — thin. Everything pointed early.
But spring ran cold, slowing the melt to ~100 cfs a day — that pushed the date
back and stretched the window to 10 days instead of letting a warm week collapse it to 3.
Dry winter soil soaks up the first melt, so low-precip winters come in a touch
earlier still.
What goes into this call — and where to check it yourself
Every spring 1988–2026 on this gauge; 38 of them train the model.
Track record — every call since 2016
year
we said
river did
miss
2016
Jun 30
Jul 03
3 late
2017
Aug 19
Aug 11
8 early
2018
Jun 23
Jun 19
4 early
2019
Aug 09
Aug 08
1 early
2020
Jun 14
Jun 09
5 early
2021*
May 28
May 16
12 early
2022
Jun 14
Jun 12
2 early
2023
Aug 31
Aug 29
2 early
2024
Jul 07
Jul 01
6 early
2025
Jun 28
Jun 23
5 early
2026
Jun 15
Jun 21
6 late
Misses run about 4 days, 8 at the worst — except the asterisk year, which gets
its own paragraph. Every "we said" is the June 1 call, the one you'd book flights on (2021's is
the April call — by June there was nothing left to call). "River did" = under 1,800 and staying
under. May doesn't count: the Kings head-fakes in May like clockwork, and 2022 faked it twice —
the second fake fooled the gauge, not the paddlers.
And yes, we see the lean — the river usually shows up a few days before we say
it will. Noted. The 2027 model gets retrained on it. The two years it ran late on us (2016, 2026)
both caught a heat spike at the finish: late heat keeps the melt pumping and props the river up
past the call. Heat spikes live inside a 10-day forecast, so that joins the model too.
*2021: dust on the snow. Dirty snow drinks sun, and the whole
pack went at once — in May 16, under 900 by May 28, gone before the season normally starts.
Every forecast in California died that spring, including the state's. It stays on the board and
out of the math.
Wrapped for 2026 — the first 2027 call lands ~April 1
Snow survey drops, we run the model, you get the date. Four emails a year.
These are forecasts, not gauge readings — the river gets the last word.
Check the live gauge before you burn vacation days. Seen a run come in? Tell us —
every confirmation makes next year's call better.