How accurate is the forecast?

Everyone has an opinion. Almost nobody checks, because checking means writing the forecast down before the day arrives — and the Weather Service only ever publishes the current one. So we started keeping the receipts: every day we archive the forecast for 386 towns, and afterwards we compare it against what the nearest weather station actually recorded.

Forecasts graded

14,310

Towns

386

Since

2026-08-15

Station distance

4.8 mi

Today’s forecast misses the daily high by 3.52°F on average. The forecast made 6 days ahead misses by 6.26°F.

0°2°4°6°8°today+1d+2d+3d+4d+5d+6ddaytime highovernight lowFORECASTERHQ.COM
Average miss, in degrees Fahrenheit, against how far ahead the forecast was made.
MadeHighLowWithin 3°FGraded
Same day3.52°F3.38°F72%3,177
1 day ahead3.96°F3.5°F67%2,808
2 days ahead4.45°F3.44°F61%2,423
3 days ahead5.16°F3.55°F53%2,039
4 days ahead6.11°F3.57°F46%1,649
5 days ahead7.3°F3.68°F38%1,261
6 days ahead6.26°F5.92°F39%861

What this does and does not say

This measures the National Weather Service, not us. We archive their forecast; we are not grading our own homework. The Weather Service is very good at this, and the numbers above are better than most people assume — a two-day-out forecast landing within 3°F most of the time is a real achievement.

Outcomes come from the nearest automated station, usually at an airport — the same instruments the official daily high and low come from. The median town here is 4.8 miles from the station that graded it, so on a day when weather varies sharply across a county, the station is describing a slightly different place.

The lead times are not measuring the same days. A forecast six days ahead can only be graded for dates at least six days after the archive began, so each column covers a slightly different stretch of weather. That is why the six-day figure can sit below the five-day one rather than continuing to climb: it is a smaller sample of different days, not evidence that forecasting further out is easier. Comparisons between adjacent columns are weaker than the overall shape, and both will firm up as the record lengthens.

The record is young. It starts on 2026-08-15, so this is one stretch of weather in one season, not a verdict on the whole year. Forecasts are easier in a settled pattern than in an unsettled one, and this page will get more honest as the archive lengthens. Lead times with too few graded days are left blank rather than filled with a number that would only get quoted without its caveat.

Forecasts from the National Weather Service. Observations from the Iowa Environmental Mesonet’s summaries of ASOS data. Updated daily.