Boring Winters Are Bad Duck Winters

Sixty seconds on ENSO

One number, measured in the middle of the Pacific, that weathermen argue about all fall.

The El Niño–Southern Oscillation (ENSO) is a slow slosh of warm water across the tropical Pacific. NOAA tracks it with the Oceanic Niño Index (ONI) — the three-month running average of sea-surface temperature anomaly in a box of ocean called Niño 3.4. When that box runs +0.5 °C or warmer for a while, it's an El Niño. When it runs −0.5 °C or colder, a La Niña. In between, the winter is neutral.

Hunters care because ENSO nudges the winter storm track. The folklore version: El Niño soaks Southern California and starves the north; La Niña does roughly the opposite. Wetter marsh, colder fronts, new water — all things that feel like they should move ducks. Whether they move straps is the question this page answers with data instead of campfire memory.

Throughout, a "season" is labelled by its opening year and classified by the average ONI from October through January — so season 1997 means October 1997 through January 1998, the heart of the famous very-strong El Niño.

39 seasons of ocean

The record I'm working with: 13 El Niño winters, 15 La Niña, 11 neutral.

Harvest data comes from the same place as my other refuge work: the daily check-station results California publishes for its public hunt areas — 41 refuges, 50,154 hunt-days, 1987 through the 2025–26 season. Success is measured the way it should be: hunter-weighted ducks per hunter per day, so a 900-hunter Saturday at Sacramento counts 900 times more than a 9-hunter Tuesday at Ash Creek.

The U-shape: both extremes beat boring

Line the seasons up by phase and the pattern is not the one the argument assumes.

Statewide, a La Niña season averages 15% more ducks per hunter than a neutral one; El Niño averages 11% more. If you plot success against the raw ONI value, the straight-line correlation is basically zero (r = −0.07) — which is exactly what you'd see if the truth were a U: both ends up, middle down.

Two honesty checks before anyone gets excited:

• Is it an era artifact? No. Average calendar year is essentially identical across the three groups (2006.1 / 2005.3 / 2006.5), so neutral seasons aren't just clustered in some bad decade. Refuge success also drifts slowly upward over the record (+0.011 ducks/hunter/year), so I re-ran everything on detrended values: La Niña still beats neutral by +0.25 (permutation p = 0.002) and El Niño by +0.19 (p = 0.049).
• Is it one weird region? No. Detrended, neutral finishes last in the Northeast, the Sacramento Valley, Suisun & the Delta, the San Joaquin Valley, and Southern California. Five for five.

Treating "distance from zero" as the predictor — |ONI| against detrended success — gives r = 0.24, p = 0.14. So the cleanest statement isn't "stronger event, better season" — it's categorical: seasons with an event beat seasons without one.

The rain is real. The rain isn't the reason.

ENSO does move California's winter weather — just not the part that fills straps.

This is the part I find most useful, because it kills the lazy causal story. If ENSO's hunting effect worked through local rainfall, then rainfall itself would have to predict harvest — and across most of California, it doesn't. Refuge water is managed; the flooded acres you hunt in November were pumped and scheduled long before the storm door opened. The one place rain does carry signal is the San Joaquin Valley, where wetland water has historically been scarcest — the region where an extra-wet winter plausibly changes how much habitat exists at all.

Inside the season

The neutral-year penalty isn't a late-season fade — it's there from opening weekend.

I went in expecting a timing story — maybe El Niño's storm parade pays off late, or La Niña's cold north pushes an early flight. That story isn't here. The phase effect looks like a roughly constant multiplier on the whole season, which points away from "storms during the season" and toward something set before the season: how many ducks came south, and in what condition.

Does strength matter?

Weak events hunt fine. Strong events are a coin flip with tiny sample sizes.

The two very strong El Niños (1997, 2015) averaged an excellent 2.17 ducks per hunter — but the three merely strong ones (1991, 2009, 2023) averaged 1.68, worse than a typical neutral year. Meanwhile weak events of either flavor comfortably beat neutral. With group sizes of two and three seasons, I read this as noise around the categorical result, not a dose–response curve. Don't pay extra for a "super" El Niño forecast, and don't dread one either.

Why would boring be bad?

Honest answer: I can show you the pattern; I can only gesture at the mechanism.

Three candidate explanations, in the order I find them plausible:

• Supply of birds. ENSO acts on the whole continent, not just our region — prairie pothole water, breeding effort, the size and timing of the fall flight. The patchy flyway counts I have lean the right way (midwinter counts average highest in El Niño years and lowest in neutral years), but coverage gaps keep me from calling that anything more than suggestive.
• Weather volatility, not weather totals. Section 04 showed season-total rain doesn't drive harvest. But event winters bring more fronts — and my day-scale model already showed fronts move daily success. A season of churn beats a season of ridge-and-fog, even at the same total rainfall.
• Effort feedback. Event seasons draw more hunters (89–94k hunter-days vs. 80k in neutral years). More pressure keeps birds moving, which raises per-hunter success at refuges — the same dynamic behind the weekend effect. But causality could run the other way: word spreads fast when the hunting's good.

The prediction engine already knew

This isn't the first time my data has voted on ENSO — a production model tested it the hard way.

Readers of From White Paper to Duck Forecast will remember that the Filthy Spoon prediction engine doesn't take anyone's word for which factors matter: every candidate signal gets a trust dial (λ, from 0 = ignored to 1 = full face value), tuned by walk-forward backtesting on seasons the model never saw. Dozens of plausible signals died in that process — raw bird counts, snowpack, moon folklore variants — and landed in the graveyard at λ = 0.

ENSO survived. The tuner assigned it λ = 0.2 for ducks — partial but permanent trust — with extra regional weight in the Northeast (0.2) and the San Joaquin Valley (0.3), and adding it moved the engine's out-of-sample daily skill up about a tenth of a point. Two details make that result rhyme almost perfectly with this article:

• The engine found the same U. The note the tuning process left on the factor reads, verbatim: "both active phases out-shoot quiet neutral winters." A completely different method — daily multiplicative model, coordinate-descent trust tuning, hunter-weighted MAE — landed on the same categorical shape as the season-level permutation tests here.
• Geese zeroed it. The goose model, tuned by the identical procedure, set ENSO's trust to 0.0 and cranked up day-of weather instead. Ducks ride the continent-scale, season-scale signal; geese ride the day. That's exactly what you'd expect if ENSO works through the fall flight rather than through your local forecast — the supply-side mechanism from section 07.

The sizes agree too. A season-constant factor can't help a daily model much — the engine's +0.1 point of skill is small next to fronts and weekend pressure — but at season scale, where this article lives, the same signal is worth a quarter of a duck per hunter. Same story, read at two different zoom levels.

Explore every season yourself

All 39 seasons, statewide, by region, or refuge by refuge — with the ocean and the rain gauge attached.

Methods, caveats, and receipts

Short version: hunter-weighted means, detrended permutation tests, and public code.

• Harvest. CDFW daily check-station results for 41 public hunt areas, seasons 1987–2025 (50,154 refuge-days). Success = Σducks / Σhunters within each scope and season. Days without a species breakdown drop out of the duck rate. Geese are tracked separately and show the same neutral-last ordering, for what it's worth.
• ENSO. NOAA CPC Oceanic Niño Index; each season classified by its Oct–Jan mean ONI at the ±0.5 °C thresholds, with CPC strength labels.
• Weather. Open-Meteo ERA5 reanalysis at each refuge's coordinates: season window = Oct 1–Jan 31, precipitation as percent of that refuge's 1991–2020 normal, temperature as anomaly vs. the same normal, aggregated to regions weighted by hunter volume.
• Stats. Bootstrap CIs resample seasons (4,000 draws). Phase contrasts use linear detrending then two-sided permutation tests (20,000 shuffles). Correlations report permutation p-values. No model, no priors — this piece is deliberately descriptive.
• Caveats. n = 39 seasons and only 11 neutral ones; multiple regions and metrics mean some p-values will flatter themselves; harvest per hunter confounds birds present with hunter behavior; and ENSO phase is known by roughly September, so this "predictor" arrives before the season but after your reservations strategy is mostly set.

Sources & replication

• NOAA Climate Prediction Center — Oceanic Niño Index (ONI), historical episodes
• NOAA CPC — ENSO Diagnostic Discussion, Aug 13 2026 (the >90% very-strong / 69% historic outlook for winter 2026–27)
• Open-Meteo — ERA5 historical weather archive
• CDFW — public-area waterfowl hunt results (compiled in the repo below)
• Replication package — gitlab.com/duckchugger/refuge-harvest-model (the harvest, weather, and ENSO inputs this page reuses)
• Related reading — The Duck Day Model (what predicts a single day) and From White Paper to Duck Forecast (the production model)