HELPERG Data · Guide · Weather
Weather forecast vs observations: what a model value is, and is not
A weather forecast value comes from a numerical model for a grid cell; an observation is measured by an instrument at one place. Why the two differ, and what HELPERG “current conditions” are.
The short answer
A weather forecast value is computed by a numerical weather model. It describes a grid cell at a given time, from a model run that started earlier. An observation is measured by an instrument at one place and one time. The two answer different questions. “What does the model expect here at 15:00?” is not the same as “what did the thermometer read at 15:00?”, so they will rarely agree exactly.
Everything HELPERG shows as weather is model output from MET Norway. That includes the “current conditions”: they are the first value of the model’s series as HELPERG last fetched it, which can be hours old, and never a measurement. HELPERG takes no observations.
Side by side
| Weather forecast value (model) | Observation (measurement) | |
|---|---|---|
| Made by | a numerical model, starting from an analysis of earlier observations | an instrument: a thermometer, an anemometer, a rain gauge |
| Describes | a grid cell, interpolated to your coordinate | the spot where the sensor stands |
| Time | a valid time, from a run issued earlier | the time of the measurement |
| Changes later? | yes, every new run replaces it | it stays as measured, apart from quality corrections |
| Typical error | model physics, grid resolution, terrain the grid smooths away | siting and exposure of the instrument |
Why the two differ
- Area and point. A grid cell averages the terrain, so a valley, a coast or a city can sit inside one cell with one value. A single station feels its own spot.
- Height. Model temperature is defined 2 m above the ground and wind 10 m above it. A rooftop sensor, or a station at a different altitude from the model’s smoothed terrain, measures something else.
- Timing. A shower that the model places an hour late, or a few kilometres away, misses the gauge entirely. The amount over the area can still be close.
- Instant and period. Precipitation in a weather forecast is a total over a period. A reading of the rain rate at one minute is a different quantity.
- Age. A weather forecast value comes from a run that started hours earlier. A newer run can change it, and an observation cannot be changed by a run.
Comparing them fairly
- Match the time: compare the value valid at the observation’s time, from the run that was available then.
- Match the quantity: compare an hourly total with an hourly total, and an instant with an instant.
- Match the place: use the station’s coordinate and keep its altitude in mind.
- Judge many hours, not one. A single miss says little, while a bias that repeats over weeks is informative.
Illustrative, not current data. If a station reads 16.2 °C at 15:00 and the model value valid at 15:00 was 17 °C, the difference is −0.8 °C. Whether that is typical can only be seen across many such pairs.
HELPERG’s weather, in these terms
- The API field
observedAtis the time of the first entry of the series the provider returned when the payload was fetched. Despite its name, it is never the time of a measurement, because none was taken. providerModelUpdatedAtnames the model run.meta.fetchedAtandmeta.cacheAgeSecondssay how old HELPERG’s copy is. Check them before calling a value “now”.- If you need measurements from stations, HELPERG’s weather is not the source for them. Use a national meteorological service’s observation data.
How each value in a row is defined is covered inhow to read an hourly weather forecast. The freshness rules are in the weather API quickstart.