Which forecast you are reading
Every high, low, rain chance and hourly number on the Today, Hour by hour, 7 days and Why pages is our statistical forecast: the average of four global weather models at the station's exact spot, corrected with what this station's own record says those models get wrong here. It is rebuilt every hour. The build made just before 7 AM is the forecast of record for the day's high, and the one just before 7 PM is the forecast of record for the night's low (6:05 AM and 6:05 PM today); those two are what the How we're doing grades. Every other hour is an update, and the page header always says which one you are looking at.
Running beside it, and not yet the forecast you read, is our physics forecast: a model of the air, grass and soil at the mast, started from the station's own readings, then corrected with the station's record. It is made at the same times, graded on the same Scoreboard, and if it wins its test it becomes the forecast. Its numbers for today and tomorrow are further down this page and on the Physics page.
The forecast on the site now was made at 12:05 PM, Wednesday: an hourly update. Our statistical model is version 2026-10-14, trained on the station's record through 2026-10-13.
The three kinds of model
Since October 21, 2026 the project keeps exactly three kinds of model, so that every change can be tested against one thing and either merged into it or deleted.
The benchmark: our statistical forecast
The forecast you read today. The four models' average (GFS, ECMWF, ICON, GEM) corrected by a formula learned from the station's record: clouds, wind, humidity, season, how much the models disagree, and their recent errors here. Rain chances come from the same inputs and the National Weather Service's written discussion.
Version2026-10-14, trained through 2026-10-13; retrained four times a year.
AlsoIt supplies the correction that calibrates the physics forecast, and it is the yardstick every candidate is graded against.
The main model: our physics forecast On trial
One simulation of the column of air over the mast, the grass and the soil, stepped every 5 minutes through night and day in one run. At night: the tested version 1 of the night physics with its frozen constants and the snow insulation that passed its test. By day: the tested day model with the crop calendar. It starts from the station's own reading at the latest sunrise or sunset, with the soil probes, and is driven by the models' clouds, wind and the air at 925 mb for what arrives from upstream.
Then correctedIts high and low are corrected by a formula fitted on a replay of the models' real past runs: a share from the physics, a share from the four models' average. That share is the progress meter below.
Versionv0.1, this run started at sunrise, Wednesday 7:25 AM, from 35.1° at the station.
Treatments: the main model plus one change
Each treatment is the main model with exactly one thing changed, asking one question, with its plan and passing mark written before the first night is scored. It is issued every hour beside the main model and graded on the same nights. When one passes, that change goes into the main model and the old term is removed; when one fails, it is deleted and the lesson is written down.
Live now5 treatments: L-039, L-040, L-041, L-042, L-043. Each is described below.
Where they come fromEvery treatment starts from a documented miss of the main model (the notebook's research thread R-011 keeps the list), never from an idea in the abstract.
Today and tomorrow, every forecast side by side hourly update, Wednesday 12:05 PM
The same four highs and lows from every model on the site. Our forecast is the statistical forecast as issued this hour; once a period has begun the Today page also adjusts it for how the station is running, so the two can differ by a degree. Physics, corrected is the main model after its calibration; physics alone is the simulation's own number before any correction. Each treatment's own corrected number follows, so you can see what its one change does to today's weather.
| Our forecast | Physics, corrected | Physics alone | Four models' average | National Weather Service | L-039 corrected / alone | L-040 corrected / alone | L-041 corrected / alone | L-042 corrected / alone | L-043 corrected / alone | |
|---|---|---|---|---|---|---|---|---|---|---|
| High today | 63.4° | 63.6° | 64.6° | 61.8° | 62.0° | 63.6° / 64.6° | 63.7° / 64.6° | 63.6° / 64.6° | 63.7° / 64.6° | 63.6° / 64.6° |
| Low tonight | 40.8° | 41.0° | 44.5° | 40.9° | 41.0° | 40.9° / 43.9° | 40.8° / 43.3° | 40.8° / 43.5° | 41.4° / 45.1° | 41.0° / 44.4° |
| High tomorrow | 46.9° | 48.1° | 49.3° | 46.7° | 49.0° | 48.1° / 49.2° | 48.2° / 49.1° | 48.1° / 49.1° | 48.2° / 49.4° | 48.1° / 49.3° |
| Low tomorrow night | 33.4° | 33.8° | 39.4° | 32.5° | 30.0° | 33.6° / 38.4° | 33.7° / 38.1° | 33.8° / 38.8° | 33.8° / 38.5° | 33.8° / 39.3° |
| What the models say for each period | Cloud | Wind | Rain | Air at 925 mb | Regime |
|---|---|---|---|---|---|
| High today | 89% | 3.7 mph | 0.04 in | 54.7° | ordinary |
| Low tonight | 86% | 8.8 mph | 0.23 in | 45.4° | rain |
| High tomorrow | 90% | 10.4 mph | 0.02 in | 46.0° | ordinary |
| Low tomorrow night | 92% | 8.6 mph | 0.04 in | 39.1° | ordinary |
Cloud, wind and rain are the models' means over the period, and the air at 925 mb (about 2,500 feet up) is what the physics brings down to the surface. A regime is a kind of night or day the main model's test watches separately: a clear and calm night (under 30% cloud and under 5 mph), snow on the ground or forecast, or at least a tenth of an inch of rain forecast. Regimes are decided from the forecast at issue time, never from what happened.
The treatments
L-039: the cloud deck emits at the 925 mb air temperature, not the 2 m air (R-011 miss M-1) On trial
The one question. On nights the models forecast cloudy, does the main model's night physics miss the station's low by less when the cloud part of the downwelling longwave is emitted at the 925 mb air temperature instead of the model's own 2 m air, with nothing else changed?
Status. running (treatment of the main model, with L-040, L-041 and L-042 on the same nights from their first issue after the next Tower install to April 30, 2027; evaluate in May 2027). Frozen 2026-10-21 by the output of its replay, so the code that runs tonight is the code that was frozen.
Its calibration this month. Physics share 34% for the low and 15% for the high (the main model: 32% and 14%).
Plan, passing mark and results: L-039 in the Notebook.
L-040: the 925 mb air comes down the moist adiabat where the models saturate the low column (R-011 miss M-1) On trial
The one question. On nights the models forecast cloudy, does the main model's night physics miss the station's low by less when the 925 mb air it relaxes toward is brought to the surface along the moist adiabat in the share of the low column the models saturate (their low cloud fraction), and along the dry adiabat elsewhere, with nothing else changed?
Status. running (treatment of the main model, with L-039, L-041 and L-042 on the same nights from their first issue after the next Tower install to April 30, 2027; evaluate in May 2027). Frozen 2026-10-21 by the output of its replay, so the code that runs tonight is the code that was frozen.
Its calibration this month. Physics share 36% for the low and 18% for the high (the main model: 32% and 14%).
Plan, passing mark and results: L-040 in the Notebook.
L-041: the reference air comes down from the models' cloud base when it is below 925 mb (R-011 miss M-1) On trial
The one question. On nights the models forecast cloudy, does the main model's night physics miss the station's low by less when, in the share of the column the models cloud (their low cloud fraction) and wherever their cloud base is below 925 mb, the reference air the night air mixes toward is the air at the cloud base brought to the surface, instead of the 925 mb air brought down from above the cloud, with nothing else changed?
Status. running (treatment of the main model, with L-039, L-040 and L-042 on the same nights from their first issue after the next Tower install to April 30, 2027; evaluate in May 2027). Frozen 2026-10-21 by the output of its replay, so the code that runs tonight is the code that was frozen.
Its calibration this month. Physics share 35% for the low and 15% for the high (the main model: 32% and 14%).
Plan, passing mark and results: L-041 in the Notebook.
L-042: rain cooling from the physics, with no fitted constant (R-011 miss M-1) On trial
The one question. On nights the models forecast cloudy, does the main model's night physics miss the station's low by less when the forecast rain falling through the night layer evaporates into it, cooling it by the latent heat of that rain spread over the layer and never past its wet bulb, with nothing else changed?
Status. pre-registered (treatment of the main model, with L-039, L-040 and L-041 on the same nights from its first issue after the next Tower install to April 30, 2027; evaluate in May 2027). Frozen 2026-10-21 by the output of its replay, so the code that runs tonight is the code that was frozen.
Its calibration this month. Physics share 35% for the low and 19% for the high (the main model: 32% and 14%).
Plan, passing mark and results: L-042 in the Notebook.
L-043: version 3's moisture budget inside the main model's night (from L-012 and L-034) On trial
The one question. On nights the models forecast cloudy, does the main model's night physics miss the station's low by less when the air's humidity is carried by a moisture budget (dew water on the grass, evaporation from it and from the soil, a local departure from the models' moisture fading over a few hours, fog's latent heat and its own longwave) instead of the drift-limited humidity, with nothing else changed? And, reported as a secondary outcome, does its saturation rank the fog nights in Brian's log better than the models' own humidity forecast?
Status. pre-registered (treatment of the main model, with L-039, L-040, L-041 and L-042 on the same nights from its first issue after the next Tower install to April 30, 2027; evaluate in May 2027). Frozen 2026-10-22 by the output of its replay, so the code that runs tonight is the code that was frozen.
Its calibration this month. Physics share 33% for the low and 14% for the high (the main model: 32% and 14%).
Plan, passing mark and results: L-043 in the Notebook.
The treatments live now form one family: they are scored on the same nights, and a stricter passing mark is applied so that running several at once cannot make a pass easier. When more than one passes, the larger change goes into the main model first and the next is re-run as a difference from the new main model before it goes in.
Older trials, and what happens to each
Before the main model, the physics was tested as separate versions and blends. On October 22, 2026 every trial still running was sorted: closed without a verdict when the structure it tested had been replaced, converted into a single-change treatment when its question was still useful, or left running when it costs nothing. A closed trial claims no pass or fail, and what it logged is exploration only.
| Entry | What it tested | Where it stands |
|---|---|---|
| L-008 | Our night physics, live: does it improve the low we issue at sunset? (research R-010) | closed (2026-10-22) without a verdict: the quarter-way blend of the night physics into our low belongs to the structure the main model replaced (the physics nudging the statistical forecast); no outcome it could reach changes what the project does, and the physics-alone and corrected numbers it compared are reported by L-038 |
| L-010 | Our day physics with the models' sunshine, on fresh days (research R-010) | closed (2026-10-22) without a verdict: the day blend at sunrise belongs to the structure the main model replaced; whether the day physics helps the high is answered inside L-038 by the main model's day half and its calibration weight; its question continues as L-038 |
| L-012 | Fog: does our night physics add to the models' own humidity forecast? (research R-010) | closed (2026-10-22) without a verdict: its question (does the night physics with a moisture budget add to the models' own humidity forecast for fog) continues as a single-change treatment of the main model: the version 3 moisture budget inside the main model's night, sized on the replay (R-011, October 22, 2026) and frozen as L-043 with the fog ranking against Brian's log as its secondary outcome; its question continues as L-043 |
| L-022 | Live trial of the midday sunshine update (follows L-021) | closed (2026-10-22) without a verdict: the midday restart from measured sunshine is now part of the main model (v0.1: the half in progress starts from the station's latest reading, the departure fading over 3 hours), chosen on this trial's logged afternoons; the remaining question, whether the hourly updates after the forecast of record beat it, goes to the update-hours entry; its question continues as L-038 |
| L-026 | Night physics 1G: an evening layer that starts shallow and deepens, for the shape of the night (research R-010) | closed (2026-10-22) without a verdict: sized on the main model's replay as two single changes (R-011, October 22, 2026): the Dilley sky formula is worse than the fitted Brutsaert sky at both scales, by whole degrees at the textbook scale, so 1G's sky is rejected; the deepening layer alone helps cloudy nights by a third to a half of a degree and hurts clear calm nights by the same, so as fitted jointly with 1G's other constants it is not a single improvement; 1G's result was carried by its layer; the layer continues as a research step in R-010 (a bounded refit of z_f and z_tau inside the main model, then a sizing), not as a running trial of a two-change fork |
| L-030 | Live trial of the night restart from the station's reading (L-023 missed narrowly) | closed (2026-10-22) without a verdict: the night restart from the station's reading is now part of the main model (v0.1), chosen on this trial's logged nights, which were its exploration set; the remaining question, whether the hourly updates after the forecast of record beat it, goes to the update-hours entry; its question continues as L-038 |
| L-032 | The physics race: does the physics' own day-ahead forecast improve ours one and two days out? | closed (2026-10-22) without a verdict: replaced outright by L-038, which issues the same chained physics corrected with the station's record, at the same cut-offs, with a plan and a bar; L-038 reports the physics-alone line this entry was; its question continues as L-038 |
| L-035 | The physics ensemble: are its odds honest? (51 runs of the night physics on the ECMWF ensemble) | closed (2026-10-22) without a verdict: its question (are the night physics' odds on the ECMWF members honest) continues as L-044, the main model on every member, re-centred on the corrected forecast and graded by the same bar; the version 1 ensemble code leaves the live physics with the install after L-044's first run is confirmed; its question continues as L-044 |
| L-036 | Version 1R (the air's own radiative cooling) in the physics ensemble: more honest odds than version 1? | closed (2026-10-22) without a verdict: version 1R's question (does the air's own radiative cooling make the odds more honest) is not carried as a second ensemble: it becomes a sizing candidate for a single-change treatment of the main model (R-011 rule 8), to be sized on the replay before any freeze; the 1R ensemble code leaves the live physics with the install after L-044's first run is confirmed; its question continues as L-044 |
| L-037 | The day ensemble: does the physics spread the high better than the ECMWF ensemble's own members do? | closed (2026-10-22) without a verdict: its question (does the day physics spread the high better than the members' own 2 m highs) continues as L-044, the main model on every member through the day halves, with the same CRPS bar; the day ensemble code leaves the live physics with the install after L-044's first run is confirmed; its question continues as L-044 |
| L-005 | Confirming the hour-by-hour forecast on fresh months (follow-up to L-003) | running (collecting fresh months to January 31, 2027) |
| L-006 | Does WxSim's forecast improve our highs and lows? (research R-009) | running (collecting forecasts to March 31, 2027) |
| L-025 | A wider +/- for the daytime high on days forecast cloudy or wet (research R-001) | running (fresh days from October 19, 2026) |
| L-027 | Standing corn in fall: does the daytime high run warm until harvest? (research R-004) | running (part A done: not met; fresh parts fall 2026, 2027 and 2028) |
| L-028 | Starting the night physics from the MEASURED grass temperature at sunset (research R-010, R-007) | running (fresh dry nights from October 19, 2026) |
| L-029 | Winter confirmation of the 6-hour nowcast (L-013), which was tested only from July to October | scheduled (fresh months November 2026 - February 2027; evaluate in March 2027) |
| L-031 | The grass's evening cooling, fitted on the grass probe (the root of the physics' evening lag) | scheduled (probe evenings from October 18, 2026; evaluate from November 16, 2026 once the conditions in the bar are met, at the latest December 15) |
| L-034 | Live trial: restart the night in the right state (follows L-033) | running (live trial, nights October 21, 2026 to April 30, 2027; evaluate in May 2027) |
The ones still running either cost no code (they log beside the benchmark) or are waiting to be converted; each conversion is sized on the replay first and frozen only if a season can resolve its effect. None of them changes the forecast you read.
The calibration, and the progress meter
The physics forecast you see is a + b × physics + c × the four models' average, with a, b and c fitted on a replay of the models' real past runs since April 2, 2026 and refit on the first day of each month. The share b is the progress meter: how much of the corrected forecast comes from the physics. Every treatment is judged partly by whether it raises that share and lowers the physics' own error.
| Model | Period | Physics share (b) | Models' weight (c) | Offset (a) | Fitted on |
|---|---|---|---|---|---|
| Main model | Low | 32% | 66% | -0.33° | 381 periods, for 2026-11 |
| Main model | High | 14% | 89% | -0.15° | 388 periods, for 2026-11 |
| L-039 | Low | 34% | 65% | -0.31° | 381 periods, for 2026-11 |
| L-039 | High | 15% | 87% | -0.11° | 388 periods, for 2026-11 |
| L-040 | Low | 36% | 63% | -0.40° | 381 periods, for 2026-11 |
| L-040 | High | 18% | 84% | -0.02° | 388 periods, for 2026-11 |
| L-041 | Low | 35% | 64% | -0.46° | 381 periods, for 2026-11 |
| L-041 | High | 15% | 87% | -0.11° | 388 periods, for 2026-11 |
| L-042 | Low | 35% | 64% | -0.59° | 381 periods, for 2026-11 |
| L-042 | High | 19% | 84% | -0.01° | 388 periods, for 2026-11 |
| L-043 | Low | 33% | 66% | -0.37° | 381 periods, for 2026-11 |
| L-043 | High | 14% | 89% | -0.13° | 386 periods, for 2026-11 |
Month by month
| Month | Physics alone | Physics, corrected | Our forecast | Highs and lows graded |
|---|---|---|---|---|
| November 2026 | 2.21° | 1.37° | 1.02° | 36 |
Average miss against the station, in degrees, for the forecasts of record. Smaller is better. The replay that fitted the calibration used a reanalysis for the deep soil; the live runs use the 8-inch probe at the mast. That difference is recorded in the plan as a known risk.
How they are issued and graded
- Every hour, five minutes past. Our forecast, the main model and every treatment are rebuilt together and logged with the time they were made.
- Two forecasts of record a day. The build made just before the cut-off for the high (12 UTC) and the one just before the cut-off for the low (0 UTC): 6:05 AM and 6:05 PM local today. The cut-offs are fixed in universal time, so they read an hour later in daylight time.
- Everything else is an update. Updates use later model runs and the station's latest readings. They are shown, labelled, and graded separately at their own range; they never count as the forecast of record.
- Two ranges for the main model. Made just before the period began (lead 1) and made the day before (lead 2). Lead 2 is the headline. Days 3 to 7, rain chances and the hour by hour stay with our statistical forecast.
- Graded as issued. If the physics cannot run, the main model issues our forecast in its place, flags it, and is graded on what it issued.
- The main model's test. From November 1, 2026 to April 30, 2027 (L-038). It replaces our forecast only if it is not materially worse overall at both ranges, better in at least one regime, and rarely degraded. Otherwise the verdict is "not yet" and it is judged again on fresh periods.
- The treatments' test. Each is scored on the nights after it first issued, never on the nights where its miss was found, against the main model on the same nights, with the family's stricter mark.
Graded so far: the main model's numbers are on the How we're doing, in the rows for the physics forecast.
What feeds each one
Our statistical forecast
The four global models (GFS, ECMWF, ICON, GEM) at the station's spot, every hour they run; about 120 ensemble runs for how far they could be off; the National Weather Service's point forecast and written discussion; the station's record since 2023 for the correction, and its readings in the last week for the models' recent error here.
The main model
The station's reading at the latest sunrise or sunset (temperature and dew point); the 4 and 8 inch soil probes; snow on the ground; the models' cloud by layer, wind, rain, sunshine and 925 mb air, hour by hour; the four models' average for the calibration. Readings whose station clock has not moved are never used.
The treatments
Exactly what the main model is fed, plus whatever their one change needs from the same model runs: the 925 mb temperature, the models' low-cloud share, the cloud base height, or the forecast rain.
The sky, the wind and the rain all come from the global models. Nothing here forecasts clouds on its own, so when the models get the clouds wrong, every forecast on this page misses together. The physics earns its place where local processes decide the outcome: clear calm nights, inversions, fog, frost timing, snow cover, and the shape of the night's cooling.
Go deeper
- Physics: tonight's run, the day run, the ensemble odds, and the physics' other tests.
- Inside the physics: the equations, the constants, and tonight's run re-run in your browser with the knobs.
- Run charts: every run drawn, with the station overlaid.
- Notebook: every test, its plan and passing mark written first, and what happened.
- Engine room: what the station computer is doing, and how many forecasts each model has issued and graded.
- How it works: the statistical forecast in plain words, how it is tested, and its year-long results.