Do One-Minute Observations (OMO) Give You an Edge? We Backtested a Year.

This is an educational article about weather data and how a daily-high model uses it. It is not financial, investment, or trading advice, and nothing here is a recommendation to do anything. It's just meteorology and data.

There's a persistent belief in temperature markets: the edge is faster, finer data. One-minute observations (OMO). The 5-minute high-frequency METAR feed. The old ASOS one-minute phone line. The logic is intuitive — if you see the temperature more often than everyone else, you'll catch the daily peak first.

We wanted to know if that's actually true, so we tested it: a full year of one-minute observations for seven cities, millions of readings, checked against the official settled highs. The answer surprised us twice, and it's worth walking through — because it changes what "an edge" even means here.

What OMO actually is

One-minute observations (OMO) are exactly what they sound like: the ASOS station's temperature reported every minute, instead of the roughly hourly cadence you see in a standard METAR. It sounds like a strict upgrade — 60× more resolution on the temperature trace.

There's a catch that matters a lot: OMO temperature is reported in whole degrees Fahrenheit. More often, but coarser. Hold that thought — it's the whole story.

First test: OMO looks like a cheat code

Compare one-minute obs against a plain hourly METAR and OMO looks incredible. Because it samples so much more often, it "catches" sub-hourly peaks the hourly reading walks right past — often a degree or two of daily high that an hourly-only view never sees.

If you stop here, you conclude OMO is a must-have and go chase the feed.

But hourly METAR is the wrong thing to compare against. A good daily-high model doesn't run on hourly obs. It already ingests the 5-minute ASOS feed, the METAR 6-hour maximum group (the precise sub-hourly record), and the eventual CLI. Against that, the picture changes.

Second test: so is precision the edge? Also no.

Our next guess was that precision would be the real winner. The METAR remark T-group reports temperature to a tenth of a degree Celsius, every hour. OMO is only whole degrees. Surely precise-but-slower beats fast-but-coarse?

The year of data said no. The official daily high is itself a whole-degree number, and the coarse-but-frequent OMO feed predicted that settled number about as often as the precise feed did — sometimes more. Neither "faster" nor "more precise" was an edge on its own. Both tidy stories were wrong.

What actually held up

Two findings survived the full year:

1. The precise 5-minute observations a good pipeline already ingests predict the settled high with essentially zero bias. For practical purposes, they already are the settlement. There isn't much headroom above that for a new feed to claim.
2. Every single raw feed has a failure mode. The OMO archive has multi-week gaps and the occasional bogus sensor spike (one city's file had a 53°F jump that never happened). The hourly feed systematically under-reads. The precise T-group is unbiased but fuzzy right at the rounding boundary — where markets settle. Lean on any one feed and you inherit its blind spot.

Put those together and the conclusion isn't "OMO is useless." It's subtler: no single feed wins, because each one is wrong in its own way.

The real edge: fusion and calibration

If no feed is the answer, what is? The thing that fuses them.

A daily-high model that's worth anything doesn't pick a feed — it reconciles them. It takes the tenth-of-a-degree T-group, the whole-degree obs, the 6-hour maximum group, and the official climate report, notices where they disagree, and resolves the conflict. Then it does the part that actually matters for a market that pays out on a bracket: it turns that reconciled picture into a calibrated probability — how likely the high crosses a given threshold, whether the peak is already in, and how confident to be — tuned so that "90%" means it happens about nine times in ten.

That's the same reason a model firms up in the afternoon rather than the morning (When Daily-High Temperature Models Firm Up During the Day), and the same machinery that explains oddball settlements like Chicago Midway settling 88°F when the report said 87°F. A single fast feed can't do any of that. Reconciliation and calibration can.

The trader's read

• A faster feed is not automatically an edge. More frequent readings of a coarser signal don't beat a model that already fuses the precise ones.
• A more precise feed isn't automatically an edge either. We tried to prove it and the data didn't cooperate.
• Every feed lies to you somehow — gaps, spikes, under-reads, boundary fuzz. The value is in combining them, not in owning the fastest one.
• The edge is the honest probability on top of the data, not the data itself. Feeds are a commodity; a calibrated model is not.

A quick note on intellectual honesty, because it matters: our cleanest test against the exact number the market settles on is still a small sample — good archived history is short. So read this as "no single feed beat a fused model in our data," not a mathematical law. We kept the backtests, and we'll revisit as more settled days accumulate. If a feed ever earns its keep, we'll write that up too.

See it live

WeatherEdge runs a fused, calibrated daily-high model in real time — combining the precise and coarse feeds, tracking the 6-hour maximum record, and publishing a live uncertainty range for KMIA, KLAX, KNYC, KMDW, and KHOU. You can watch it tighten through the day and check it against the final CLI yourself. The free tier shows KHOU.

Once more: this is an explainer about weather data and modeling, not financial or trading advice. It describes how a model uses observations — nothing more.