Our physics forecast On trial hourly update, Thursday 7:05 PM
A model of the column of air over the mast, the grass and the soil, stepped every 5 minutes through night and day in one run. It starts from the station's own latest reading, with the soil probes, and is driven by the weather models' clouds, wind and the air about 2,500 feet up for what arrives from upstream. Its high and low are then corrected with what the station's record says the physics and the models get wrong here. It is made every hour beside the forecast you read on the other pages, graded on the same How we're doing, and if it wins its test it becomes the forecast.
Today and tomorrow
| Physics, corrected | Physics alone | Our forecast | National Weather Service | The station so far | |
|---|---|---|---|---|---|
| Low tonight | 37.1° | 34.8° | 38.3° | 39.0° | 52.2° so far |
| High tomorrow | 67.9° | 64.3° | 68.9° | 68.0° | not started |
| Low tomorrow night | 39.2° | 38.9° | 39.2° | 41.0° | not started |
"Our forecast" is the statistical forecast as issued this hour; the Today page also adjusts it for how the station is running once a period has begun, so the two can differ by a degree. The physics forecasts today and tomorrow only; further out the statistics do better. This month the calibration puts a weight of 32% on the physics for the low and 14% for the high, with the four weather models' average carrying most of the rest: the progress meter, explained with the calibration and the treatments on The models.
The next two days, minute by minute started at sunset, Thursday 5:22 PM; reset to the station at 7:05 PM
The violet line is the main model's own curve, uncorrected, through four halves: the half in progress and the three after it, each starting where the one before ended. Shaded bands are nights. The dot at the start of the half in progress is where the station's latest reading reset the run; the dots at each peak and trough are the corrected highs and lows from the table above. Tick a treatment to draw its line: each is the main model with exactly one change (what each changes). Blue dashed is our forecast, orange the station.
The latest night run, with the station's own readings drawn over it. Every run as images (60 days).
Tonight, sunset to sunrise running through the night
The physics alone now expects 34.8° at the station by 7:40 AM, 3.5° colder than our forecast; corrected with the station's record, 37.1°. Restarts from the station's latest reading and the grass probe (52.2°, dew point 44.1°).
The station's measured temperature and the grass probe are drawn as the night goes on, so you can watch the physics and the real night side by side.
This is the night run alone (the same night physics the main model carries), before any correction. Air is the temperature at the height the station measures (5 feet). The grass surface cools faster and further than the air above it on a clear night; that is where frost forms. The dew point is how low the air can cool before dew (or frost, below 32°) forms. The faint violet line is the run made at sunset; the bold one restarts every hour from the latest readings. The thin line marks 32°.
How tonight's run got its bearings
Before it runs, the model takes stock of the station, the ground and the forecasts. These are the values this run started from, and what it is being fed through the night.
| What it starts from | Value | Source |
|---|---|---|
| Air at the start | 52.2° | Restarts from the station's latest reading and the grass probe |
| Dew point at the start | 44.1° | the same reading |
| Grass surface at the start | 40.9° | the grass probe |
| Soil below (warmth from underneath) | 45.5° | the 4 and 8 inch soil probes |
| Air pressure | 998 millibars | the station's barometer, at the station's height |
| Snow on the ground | none | From the forecast models (no recent report) |
What the forecasts feed it, every two hours
| Time | Air 2,500 feet up, brought to the ground | Cloud: low, middle and high | Wind, miles per hour (models) | Rain in the next 2 hours, inches |
|---|---|---|---|---|
| 7:05 PM | 65.5° | 0%, 0%, 1% | 6.0 | - |
| 9:05 PM | 65.3° | 0%, 0%, 0% | 6.2 | - |
| 11:05 PM | 65.0° | 0%, 0%, 0% | 5.6 | - |
| 1:05 AM | 64.8° | 0%, 0%, 0% | 5.0 | - |
| 3:05 AM | 64.6° | 0%, 0%, 0% | 4.3 | - |
| 5:05 AM | 64.3° | 0%, 0%, 0% | 3.5 | - |
| 7:05 AM | 64.1° | 0%, 0%, 0% | 3.3 | - |
Air 2,500 feet up: the temperature at the 925 millibar level in the European and German weather models, worked out as it would be if that air were stirred down to the ground (it warms as it sinks; along the saturated rate when it rains). The breeze pulls the night toward it; on a calm, clear night the ground lets go of it and cools on its own. Cloud sends heat back down, low cloud the most. Rain cools the air toward its wet-bulb.
The odds: 51 runs of the main model Being graded
The main model run once on each of the ECMWF ensemble's 51 members through the same two days: each member's own clouds by layer, wind, air aloft and sunshine; the same station start and the same reset to the station's latest reading. One run gives one number; 51 runs show how sure that number is. Each member's departure from the members' middle is laid on the corrected forecast, so the spread is read against the number the site shows.
The same 51 runs, sorted by what each one does with the sky. No run is picked; each keeps its own sky, wind and low together. Look out after dark and read the line that matches.
| Scenario | Tonight | Tomorrow night | ||||
|---|---|---|---|---|---|---|
| Runs | Low, 8 in 10 | Middle | Runs | Low, 8 in 10 | Middle | |
| Clear most of the night | 51 of 51 | 36 to 39° | 37.1° | 51 of 51 | 37 to 43° | 39.2° |
| Calm, under 5 mph | 51 of 51 | 36 to 39° | 37.1° | 30 of 51 | 36 to 39° | 38.0° |
| A breeze, 5 to 10 mph | none | 21 of 51 | 40 to 43° | 41.6° | ||
Sky is each member's cloud through the night weighted the way the physics weights it (low cloud counts most); wind is the member's 10 m wind averaged through the night. A group under 3 runs is not shown. When the sky thermometer is on the roof, the Tower can read the sky itself at dusk and set the lines that do not match aside.
The shaded band holds 8 runs in 10 of the main model's air through the two days, the line is the middle run, orange the station as the hours pass. Being graded (L-044, May 2027): the station's low should land inside the 8-in-10 range on about 8 nights in 10, frost should come on about as many nights as the chance says, the band should be wider on the nights the middle run misses by more, and by day the physics' spread should catch the station's high better than the members' own spread does. The odds of fog wait for the moisture budget treatment (L-043) to pass.
Night by night the run made at sunset
| Night of | Physics low (alone) | Our forecast low | Actual low | Physics: grass low | Grass probe low | Hours at 97% humidity or more |
|---|---|---|---|---|---|---|
| Wednesday, November 18 | 41.0° | 38.5° | 37.0° | 40.2° | 27.7° | 0.0 hours |
| Tuesday, November 17 | 29.7° | 32.0° | 33.3° | 28.8° | 23.9° | 0.0 hours |
| Monday, November 16 | 27.4° | 29.6° | 29.3° | 26.4° | 22.4° | 0.6 hours |
| Sunday, November 15 | 26.2° | 31.6° | 27.3° | 24.7° | 22.2° | 0.0 hours |
| Friday, November 13 | 26.3° | 29.0° | 29.3° | 25.9° | 26.0° | 0.0 hours |
| Thursday, November 12 | 41.2° | 36.8° | 38.5° | 40.6° | 36.8° | 0.0 hours |
| Wednesday, November 11 | 42.9° | 40.3° | 43.9° | 42.6° | 39.8° | 0.0 hours |
| Tuesday, November 10 | 34.8° | 36.1° | 34.2° | 34.2° | 29.6° | 0.2 hours |
| Monday, November 9 | 42.5° | 42.4° | 40.5° | 41.8° | 32.3° | 0.0 hours |
| Sunday, November 8 | 35.7° | 36.2° | 36.9° | 34.8° | 30.1° | 0.0 hours |
| Saturday, November 7 | 33.7° | 33.4° | 31.6° | 33.0° | 24.8° | 0.0 hours |
| Friday, November 6 | 40.4° | 37.6° | 37.0° | 40.0° | 34.4° | 0.0 hours |
| Thursday, November 5 | 50.8° | 45.8° | 45.1° | 49.2° | 43.0° | 0.0 hours |
| Wednesday, November 4 | 51.1° | 50.2° | 51.3° | 50.7° | 48.1° | 0.0 hours |
| Tuesday, November 3 | 65.1° | 62.2° | 61.5° | 64.8° | 59.7° | 0.0 hours |
| Monday, November 2 | 58.6° | 57.6° | 58.1° | 58.1° | 56.4° | 0.0 hours |
| Sunday, November 1 | 54.8° | 54.8° | 55.2° | 54.5° | 53.4° | 0.0 hours |
| Saturday, October 31 | 51.5° | 50.6° | 51.4° | 50.8° | 49.0° | 0.0 hours |
| Friday, October 30 | 53.8° | 49.9° | 52.5° | 53.2° | 49.0° | 0.0 hours |
| Thursday, October 29 | 43.4° | 42.7° | 40.6° | 42.4° | 35.0° | 2.4 hours |
Blue marks the closer of the two to the actual low, the station's lowest reading between sunset and sunrise. The physics here is the night run alone, before any correction; the corrected forecasts of record and their grades are on the How we're doing and, month by month, on The models. Hours at 97% humidity or more: how long the air sat at saturation (the sensor tops out at 98%), the local sign of fog or a soaking dew.
Today, sunrise to sunset from the sunrise reading
Violet is the day run alone, orange the station. Below it, how deep the sun mixes the air. Ground moisture from ECMWF topsoil moisture. In the main model this same day run is reset to the station's latest reading at every issue, with the station's departure fading over three hours.
| Day | Physics high (alone) | At about | Our forecast high | Actual high | At |
|---|---|---|---|---|---|
| Thursday, November 19 | 63.7° | 3:34 PM | 66.9° | 66.2° | 2:00 PM |
| Wednesday, November 18 | 61.5° | 3:33 PM | 63.4° | 63.5° | 2:55 PM |
| Tuesday, November 17 | 55.6° | 3:37 PM | 57.4° | 56.7° | 3:25 PM |
| Monday, November 16 | 50.9° | 3:31 PM | 53.4° | 52.9° | 4:15 PM |
| Sunday, November 15 | 49.0° | 12:35 PM | 49.3° | 46.9° | 1:55 PM |
| Saturday, November 14 | 43.6° | 3:38 PM | 44.2° | 45.0° | 3:00 PM |
| Friday, November 13 | 44.5° | 2:32 PM | 44.8° | 44.8° | 2:00 PM |
| Thursday, November 12 | 48.6° | 12:06 PM | 47.7° | 48.9° | 2:45 PM |
| Wednesday, November 11 | 60.0° | 3:45 PM | 63.6° | 64.4° | 2:50 PM |
| Tuesday, November 10 | 65.3° | 2:34 PM | 66.4° | 66.0° | 3:45 PM |
| Monday, November 9 | 60.4° | 3:07 PM | 61.4° | 61.2° | 4:05 PM |
| Sunday, November 8 | 58.3° | 3:36 PM | 58.9° | 59.7° | 4:25 PM |
| Saturday, November 7 | 51.2° | 2:35 PM | 51.9° | 51.3° | 3:40 PM |
| Friday, November 6 | 48.8° | 2:34 PM | 49.9° | 50.2° | 1:15 PM |
| Thursday, November 5 | 56.3° | 1:43 PM | 56.2° | 56.5° | 10:00 AM |
| Wednesday, November 4 | 70.3° | 12:11 PM | 70.0° | 68.5° | 1:55 PM |
| Tuesday, November 3 | 74.3° | 3:05 PM | 75.6° | 76.5° | 3:20 PM |
| Monday, November 2 | 62.8° | 5:39 PM | 60.5° | 58.8° | 4:50 PM |
| Sunday, November 1 | 67.7° | 3:08 PM | 66.4° | 68.2° | 3:45 PM |
| Saturday, October 31 | 64.7° | 3:12 PM | 62.7° | 66.7° | 3:00 PM |
| Friday, October 30 | 66.4° | 3:21 PM | 66.0° | 64.8° | 4:00 PM |
How it works
One run, night and day. The main model steps the column over the mast every 5 minutes through four halves: the half in progress and the three after it, each starting where the one before ended. At every hourly issue the half in progress is reset to the station's latest reading, and the station's departure from the run fades over about three hours.
At night. How much heat the grass loses to the sky and how much comes back (clear air sends little, low cloud a lot), how much the breeze mixes warmer air down, the heat released when dew or frost forms, and the warmth rising from the soil (measured live by the ground sensor). Clouds, wind and the air at 925 millibars come from the European (ECMWF) and German (ICON) weather models.
By day. The sun heats the ground; part of that evaporates water (following a crop calendar: bare fields in spring, a transpiring canopy in midsummer, drying crops in fall), part soaks into the soil, and the rest heats a layer of air that deepens through the morning. The sunshine is the forecast models', calibrated against the station's solar sensor.
On snow. Snow insulates the ground from the soil's warmth and cannot warm past 32°; its depth comes from the snow estimate. Rain is not yet modelled in the main model: a treatment for it is being tested (L-042).
Then corrected. The run's high and low are corrected by a formula fitted on a replay of the models' real past runs since April 2026: a share from the physics, a share from the four models' average, refit monthly. That share is the progress meter on The models.
Fitted, not guessed. The night model's 13 constants were fitted to the station's minute-by-minute record on 474 nights in 2024 and 2025; the day model's 15 on 455 days. The constants are listed, with the equations, on Inside the physics.
What is in it, and what is being tried.
- The day model with the crop calendar Proven and the snow insulation Proven are in the main model.
- The main model itself is on trial against our forecast On trial; the verdict comes in May 2027.
- the cloud deck emits at the 925 mb air temperature, not the 2 m air (R-011 miss M-1) Treatment
- the 925 mb air comes down the moist adiabat where the models saturate the low column (R-011 miss M-1) Treatment
- the reference air comes down from the models' cloud base when it is below 925 mb (R-011 miss M-1) Treatment
- rain cooling from the physics, with no fitted constant (R-011 miss M-1) Treatment
- version 3's moisture budget inside the main model's night (from L-012 and L-034) Treatment
- Tried and left out: rain cooling fitted as a separate wet-night version added nothing to the low Not proven; a day model driven by the models' clouds alone was worse than our forecast Not proven.
Every change is a numbered entry with its plan and passing mark written first, in the Notebook; what each model is and how they are graded is on The models.