INDUSTRY FIELD GUIDE
Weather forecast issue and verification window data for AI
A weather provider can test whether a model matches a forecast’s valid period to the correct observations.
NAICS 541990 · All Other Professional, Scientific, and Technical Services
A weather provider can test whether a model matches a forecast’s valid period to the correct observations. The National Weather Service publishes a precipitation forecast verification example. The Census classification includes weather forecasting services in this industry. NWS quantitative precipitation forecast verification · U.S. Census 2022 NAICS manual
Invent two forecast issues with the same publish date and different valid windows. Give one observation record in UTC and another without a stated time basis. Ask the model which error can be computed, with which units, and why the other remains unresolved.
Use original synthetic grids and dummy station labels. Keep client forecasts and licensed observation feeds out. Preserve the metadata and rights information for any approved follow-up.
Describe a synthetic forecast verification sample
Original Fiund exercise
Issue date alone cannot pair a precipitation forecast with observations
An imaginary station has synthetic precipitation amounts in millimeters. The exercise is a time-window join and arithmetic check, not an operational forecast or analysis of real weather-service skill.
| Record | Source | What it records |
|---|---|---|
| FW1 | Forecast issues | Issue F-A published day 1 at 00 UTC predicts 12 mm for day 2 00–24 UTC. F-B published day 1 at 12 UTC predicts 5 mm for day 2 12 UTC to day 3 12 UTC. |
| FW2 | Observation O-A | Station DEMO: 10 mm accumulated day 2 00–24 UTC. Quality flag valid for this fictional task. |
| FW3 | Observation O-B | Station DEMO: 7 mm, label 'day 2 noon to day 3 noon'; time zone unspecified. No conversion metadata or interval confirmation supplied. |
Question for the model
Which signed forecast error can be computed as forecast minus observed, and why is the other forecast-observation pairing unresolved?
Show the reference answer and checks
FW1's F-A and FW2's O-A share station, units and the day-2 00–24 UTC window, supporting an illustrative signed error of 12 minus 10, or +2 mm. F-B's UTC interval cannot be confidently matched to FW3's noon labels without a time-zone basis. A shared issue date or station does not resolve that interval. Obtain the observation time metadata before calculating F-B's paired error.
- Joining forecasts and observations by publication date rather than the valid accumulation window.
- Treating unspecified local noon as UTC noon and reporting an apparently precise error.
Fields to preserve
- Issue timestamp and forecast valid interval
- Separates when a forecast was made from what period it covers.
- Observation interval and time basis
- Supports exact temporal pairing.
- Station, unit and quality state
- Defines the measurement comparison.
Test another case
Clarify O-B's interval as UTC and its amount as 7 mm. Require F-B's signed error to become −2 mm while retaining F-A's separate +2 mm result.
Use original synthetic readings and dummy stations. Exclude licensed feeds and client forecasts; this single exercise provides no evidence of real-world forecast performance.
Discuss a contribution
Describe the sources you control, the task they could support and what has already been checked. Use the existing-data assessment for an archive or the collection brief for a new recording project. The contributor agreement defines Fiund’s licensing authority and compensation. A guide or fictional exercise does not establish an accepted collection or a sale.