Five days closer buys 9 to 31 points of accuracy — and past a month, forecasts change what they even promise

Asked:

How much does the accuracy of weather forecasts increase as the forecasted date gets closer? Include long-term forecasts (>=1mo in advance)

The clean like-for-like evidence is Met Office UK summer 2025 verification at roughly 120 sites: four variables scored as percent correct within a fixed tolerance at 5-day and 1-day leads. Long-range products (a month or more ahead) predict averages and anomalies, not weather on a date, and are scored with correlation and skill scores — shown here on their own scales. Data researched on 2026-09-01.

Days 5 to 1: the same forecast, measurably sharper

Percent of forecasts correct within the Met Office tolerance, UK summer 2025, about 120 sites. Colour marks the gain from 5 days to 1 day ahead.

At a month or more, the product changes — so does the yardstick

CFSv2 seasonal hindcasts: anomaly correlation for three-month-mean extratropical sea-surface temperature (poleward of 20°N/20°S). This predicts a three-month average anomaly, not weather on a date.

Hydrological illustration only: mean CRPSS of UK ensemble streamflow prediction by lead — a river-flow skill curve, not general weather accuracy.

Horizon guide: what a forecast promises at each range

Days 1–5

Exact local conditions

Temperature, wind and rain for a place and a date, verifiable within tolerances — the only range where "percent accurate" applies directly.

Weeks 2–4

Probabilistic weekly anomalies

Warmer or cooler, wetter or drier than normal for a week, stated as probabilities; experimental ML hindcasts push deterministic skill toward this range.

1 month and beyond

Monthly and seasonal averages

Broad averages, anomalies and tercile categories, scored against climatology with correlation and skill scores (ACC, HSS, BSS, CRPSS) — never percent-correct weather.

Wind direction gained the most from a shorter lead — 31 points (57% to 88%, correct within one point of a 16-point compass); wind speed the least at 9 points (84% to 93%, within 5 knots). Simple average across the four unlike measures: 19.75 points. metoffice.gov.uk

CFSv2 three-month-mean extratropical SST anomaly correlation rises from 0.27 at a 6-month lead to 0.34 at 3 months — getting 3 months closer adds 0.07, about 26% relative. cfs.ncep.noaa.gov

An experimental ML study reports practical Z500 skill (ACC 0.6) to 27.5 days and statistical significance to 33 days — optimized hindcasting, not an operational one-month exact-weather forecast. arxiv.org

NOAA's monthly Alaska outlooks are verified with the Heidke Skill Score over tercile categories — a different question from percent-correct daily weather, so the numbers cannot share an axis. cpc.ncep.noaa.gov

The rows

Forecast / variable	Metric	Farther lead	Value	Closer lead	Value	Change	Source

Met Office UK summer 2025 site verification (4 variables, 5-day vs 1-day percent correct within variable-specific tolerances, ~120 sites); CFSv2 hindcast anomaly correlation for three-month-mean extratropical SST (0.34 at 3-month lead, 0.27 at 6-month); UK ensemble streamflow CRPSS sequence shown as a hydrological illustration only. Data researched on 2026-09-01. Percentages, ACC, HSS, BSS and CRPSS are different metrics and are never combined. One incompletely extracted seasonal row was dropped; supporting search claims not shown were cut for space.

This report was generated automatically by Keenable SELECT at a user's request, from publicly available web sources linked herein. Keenable does not review, verify, or endorse its contents and makes no representation as to accuracy, completeness, or timeliness; AI-based extraction may contain errors. Nothing in this report is investment, legal, financial, or other professional advice. All trademarks and referenced content remain the property of their respective owners; no affiliation or endorsement is implied. To report an error, rights concern, or request removal: legal@keenable.ai.