Charlotte, North Carolina · 35.23°N, 80.84°W
One paragraph a day about Piedmont weather, learned from eight decades of records and scored against itself in public.
Normal for the date is 66° / 42°. Open-Meteo's own forecast: 72° / 55°. Expected rain 0.17 in.
Today should top out near 74° and drop to about 59°, about 8° above the normal 66° for the date (normal low 42°). The past week has averaged 72° at the high.
There is a 41% chance of rain, so about 0.17 in — worth knowing, not worth cancelling plans over.
| Scored on the last 365 days | This model | Yesterday repeats | Normal for the date |
|---|---|---|---|
| High temperature | 2.18° | 5.57° | 7.40° |
| Low temperature | 1.99° | 4.69° | 6.79° |
| Rain (Brier, lower is better) | 0.160 | — | 0.184 |
Mean absolute error in °F. The model is retrained for every day in the window using only what was known that morning. Rain occurred on 24% of those days.
| Day | High | Low | Rain |
|---|---|---|---|
| Sat Oct 31 | 68° | 56° | 0.00 in |
| Sun Nov 1 | 65° | 54° | 0.00 in |
| Mon Nov 2 | 72° | 51° | 0.00 in |
| Tue Nov 3 | 73° | 49° | 0.00 in |
| Wed Nov 4 | 69° | 53° | 0.00 in |
| Thu Nov 5 | 76° | 66° | 0.00 in |
| Fri Nov 6 · forecast | 74° | 59° | 0.17 in |
Data. NOAA's GHCN-Daily station record for Charlotte Douglas (USW00013881), which runs from 1939 to now, with Open-Meteo filling the last few days before the station record catches up and supplying the forward forecast used for comparison. The air upwind comes from the hourly airport records at Atlanta, Greer, Columbia and Asheville, by way of the Iowa Environmental Mesonet. All of it is free and keyless.
The pattern model. Daily features — yesterday's high, low and rain, lags out to two weeks, how far the last few days sat from normal, the wet-ground signal, the position of the year, the overnight air mass upwind and the day-ahead forecast high and low — fed to ridge regression for temperatures and logistic regression for rain. It retrains every morning on everything before today and predicts today.
The writer. A free z.ai GLM model turns the numbers into the note above. It is given the figures and told not to invent any; every number on this page comes from the statistical model or the weather archive, never from the language model.
The rules. The code that scores the model is the same code that produces the forecast, so the backtest cannot cheat: on each day it sees only what was known that morning. Predictions are logged before they resolve and scored afterwards.