Somerset Weather · Hyper-local Forecast

The station against what was expected, tonight. The numbers first; the same things in sentences are under More detail, and the record of every graded day is further down.

Tonight reading 7 min ago

The station now50.2° −0.8° against the forecast curve for 7:35 PM
Published low tonight38.3° so far down to 50.2° at 7:35 PM
The othersNWS 39° WxSim 34.4° · four-model average 39.7° · main model 37.1° (physics alone 34.8°)
The 51 runs36 to 39°8 in 10 of them land here
Last night at this hour47.7° tonight is running 2.5° warmer; last night bottomed at 37.0° against 38.5° forecast
Main model from here34.7°its own run, restarted from the latest reading, bottoms around 7:50 AM
The grass37.6° −12.9 12.9° under the air; dew forming
Up the mast, 115 inches−1.0° colder than the station
Air above its dew point7.0°humidity 77%; 97% reached on 2 of the last 6 mornings
Soil, 2 in / 8 in46.2 / 45.6°warmer at the top: heat flowing down
Sunrise7:35 AMthe night window ends at 8:00 AM

The forecast curve is the hour-by-hour forecast as published at 7:05 PM; last night from the same readings file; the main model's run restarted each hour from the latest reading; the grass and the mast from the ground sensor.

What happened last 12 hours

  • The grass reached 32° at 7:45 AM.
  • A gust of 15 mph at 1:15 PM.
  • Sunset at 5:22 PM: the station at 59.2°, the grass 11.1° under it.

The last verdict

Thu Nov 19's high was 66.2°.

ForecastSaidMiss
Ours66.9°+0.7°
NWS65.0°−1.2°
Four-model average64.2°−2.0°
WxSim68.4°+2.2°

The most recent completed period in the record below; the closest forecast in bold.

More detail

In words

At 7:35 PM the station reads 50.2°. The hour-by-hour forecast, as published at 7:05 PM, had 51.0° for this hour: the station is 0.8° under it.

The published low for tonight is 38.3°; the NWS says 39°, WxSim 34.4°, the four-model average 39.7°.

The main model has 37.1°, its physics alone 34.8°.

The 51 runs put 8 in 10 between 36 and 39°: 51 clear runs on 37.1°.

So far tonight the station has been down to 50.2° (at 7:35 PM).

In the last hour it has fallen 4.1°.

Last night at this hour the station read 47.7°, so tonight is running 2.5° warmer; last night bottomed at 37.0° against a forecast of 38.5°.

From here the main model's own run, restarted at 7:05 PM from the station's reading, bottoms at 34.7° around 7:50 AM.

The station's air is 7.0° above its dew point (humidity 77%).

The station reached 97% humidity on 2 of the last 6 mornings.

Sunrise is at 7:35 AM.

The levels

The air 115 inches up the mast is 1.0° colder than the station's.

The grass is 12.9° colder than the station's air; the widest gap tonight was 15.4° at 6:15 PM, against a median of 11.0° over the 9 complete nights on record. It sits 5.9° below the dew point, so dew is forming.

The soil is 0.6° warmer at 2 inches than at 8, so the day's heat is still soaking down.

Over the last week the top foot of soil has been losing heat, 0.8° a day.

From the ground sensor's latest reading and its record of nights (the Station's Ground page).

Background

All 30 parts of the Tower report normal.

The ground sensor sent 286 of 288 readings in the last 24 hours; battery 3.98 V, radio -69 dBm.

The Studio's latest research report, relayed through the Tower Thu 10:16 AM: The night model with a proper soil column under the grass fits the 474 training nights 1 percent worse than the current night physics (and 6 percent worse than its best rival, 1G). (all of them: the Engine page).

Events are rules over the 5-minute readings: the first wind after half an hour of calm at night, a warm-up or a dew point rise within 30 minutes, a fall of 5° within 30 minutes away from sunset, a pressure jump of 0.7 mb in 10 minutes, cooling that stalls, the mast crossing the station, saturation, the grass at 32°, rain, gusts of 15 mph.

The record

The record so far

For the forecasts made the day before, over the last 36 days:

Daytime highs

1.6°
On average, our forecast high was off by 1.6 degrees. That tied with Our physics forecast, corrected (the candidate, on trial, L-038) for the closest of the 5 forecasts graded.
Our physics forecast, corrected (the candidate, on trial, L-038) was off by 1.6 degrees. The National Weather Service was off by 2.2 degrees. The average of the four weather models was off by 2.8 degrees. WxSim was off by 2.8 degrees.

Nighttime lows

1.6°
On average, our forecast low was off by 1.6 degrees. That tied with Our physics forecast, corrected (the candidate, on trial, L-038) for the closest of the 5 forecasts graded.
Our physics forecast, corrected (the candidate, on trial, L-038) was off by 1.6 degrees. The average of the four weather models was off by 2.2 degrees. The National Weather Service was off by 2.7 degrees. WxSim was off by 4.0 degrees.

"Off by" means how many degrees the forecast high or low was from what the station measured, averaged over the days graded. Smaller is better. Every forecast is graded on what it said at the same time ahead, so changing its mind later never improves its score. 36 days is not many yet, and these numbers will move as more days are added. The year-long test further down this page, on about 359 days, is the steadier guide.

On this page:Every range, every forecastRecent days The year-long testRainMore ways to look at it

Every range, every forecast

How many degrees each forecast was off, on average. The same day or night is graded three ways: on the forecast that was out just before it began (for the high, before 8 AM that morning; for the low, before 8 PM that evening), on the forecast made the day before, and on the one made two days before. The number in parentheses is how many days or nights each figure covers.

Daytime highs

ForecastMade just before the day or night beganMade the day beforeMade two days before
Our forecast (the statistical forecast the site shows) 1.5° (36 days) 1.6° (35 days) 2.0° (34 days)
The National Weather Service 2.0° (36 days) 2.2° (35 days) 2.2° (34 days)
WxSim 3.1° (35 days) 2.8° (34 days) 2.5° (33 days)
The average of the four weather models 2.4° (36 days) 2.8° (35 days) 3.1° (34 days)
Our physics forecast, corrected (the candidate, on trial, L-038) 1.3° (29 days) 1.6° (28 days) not made

Nighttime lows

ForecastMade just before the day or night beganMade the day beforeMade two days before
Our forecast (the statistical forecast the site shows) 1.4° (36 nights) 1.6° (35 nights) 2.2° (34 nights)
The National Weather Service 2.5° (36 nights) 2.7° (35 nights) 2.8° (34 nights)
WxSim 3.2° (34 nights) 4.0° (33 nights) 3.6° (32 nights)
The average of the four weather models 2.1° (36 nights) 2.2° (35 nights) 3.1° (34 nights)
Our physics forecast, corrected (the candidate, on trial, L-038) 1.5° (28 nights) 1.6° (27 nights) not made

The closest forecast in each column is in bold. Since November 1, 2026 a second complete forecast built on our physics, corrected with the station's own record, is made at the same times as ours and graded the same way; if it wins it becomes the forecast (the plan and what counts as winning: L-038). It forecasts today and tomorrow only, so it has no "two days before" forecast. Which model is which, with today's numbers from every one of them side by side: The models. How the physics works: Physics.

Recent days

Day or nightThe station measuredOur forecastThe National Weather ServiceWxSimThe average of the four models
Thursday, November 19, the high66.2° 66.9°+0.7 65.0°−1.2 68.4°+2.2 64.2°−2.0
Wednesday, November 18, the low that night37.0° 38.5°+1.5 39.0°+2.0 31.2°−5.8 39.2°+2.2
Wednesday, November 18, the high63.5° 63.3°−0.2 61.0°−2.5 66.3°+2.8 60.4°−3.1
Tuesday, November 17, the low that night33.3° 32.0°−1.3 33.0°−0.3 29.1°−4.2 33.2°−0.1
Tuesday, November 17, the high56.7° 57.7°+1.0 55.0°−1.7 57.4°+0.7 54.6°−2.1
Monday, November 16, the low that night29.3° 29.6°+0.3 32.0°+2.7 27.0°−2.3 31.3°+2.0

Each forecast as it stood just before the day or night began. The number after each forecast shows how far off it was: +2.2 means 2.2 degrees too warm, and −2.3 means 2.3 degrees too cold. The closest forecast on each row is in bold.

More detail

The year-long test

Before going live, our forecast was tested on a full year of days it had never seen (October 2025 to October 2026), against the average of the four weather models on the same days.

Days aheadHighs: our forecastHighs: the modelsHighs: our forecast within 2° Lows: our forecastLows: the modelsLows: our forecast within 2°
11.6°2.0°68% of days 1.7°1.9°69% of days
22.0°2.5°61% of days 2.0°2.3°62% of days
32.4°2.8°53% of days 2.3°2.6°53% of days
42.8°3.2°49% of days 2.5°2.8°51% of days
53.2°3.6°46% of days 3.1°3.4°44% of days
64.0°4.3°37% of days 3.8°4.0°35% of days
74.8°4.9°29% of days 4.7°4.9°27% of days

"Our forecast" and "the models" are how many degrees each was off, on average. "Within 2°" is the share of days our forecast was within 2 degrees of what the station measured. "Days ahead" counts the day itself as 1, so 1 is the forecast made that morning or evening, and 2 is the one made the day before. Our forecast was closer than the models' average at every range.

Rain chances

ForecastDaytime, made the day beforeOvernight, made the day before
Our forecast 0.114 (35 days) 0.126 (35 nights)
The National Weather Service 0.115 (35 days) 0.138 (35 nights)

A score for rain chances: 0 would be perfect (saying 100% when it rained and 0% when it did not), and lower is better. With few rainy days so far, these mostly show dry days called dry.

More ways to look at it