open-meteo-guide

Scope

Modeled, gridded data from Open-Meteo, keyless: deterministic and ensemble forecasts, the reanalysis archive, marine waves, CAMS air quality, GloFAS river discharge, CMIP6 climate projections, terrain elevation, and a populated-place geocoder. Not here: official watches and warnings, station observations, and weather forecasts past the ensemble's 35-day ceiling. For next month's or next season's weather, say no tool covers it; never substitute a climate projection.

Identifiers

• Coordinates come from openmeteo_search_locations, which indexes populated places only: a surf break ("Banzai Pipeline") returns no_results. Geocode the nearest town, then move the coordinate onto the water or the river channel (Gates 5 and 6).
• The served cell. Each weather response's latitude/longitude is the grid cell that answered, not the input, and it can sit kilometres away: era5 (0.25° grid) answered Seattle from 47.5, −122.25. Once a marine or flood cell is chosen, send that served coordinate on every follow-up call so all the numbers come from one cell.
• canvas_id and table_name come only from a response with truncated: true. A window small enough to return inline is not staged.

Gates

1. Odds come from the ensemble. precipitation_probability is the chance of more than 0.1 mm in the preceding hour, and daily precipitation_probability_max is the highest hourly value of the day. Neither answers "more than 20 mm" or "a wet day". For a threshold, count ensemble values over it (Workflows). In one probe the default forecast gave 21.8 mm and a 74 % maximum hourly probability for a day on which 5 of 51 ensemble values passed 20 mm.
2. Frost is a surface question. temperature_2m_min is air at 2 m. Request hourly soil_temperature_0cm (the surface temperature on land) beside it before saying "no frost": one probe returned daily minima of 0.6 °C and 1.0 °C while the surface fell to −0.7 °C and −1.8 °C. For the odds of an air frost, count ensemble temperature_2m_min members below 0.
3. A projection is not a forecast. openmeteo_get_climate returns a plausible value for any date to 2050 (it answered 2027-07-01 with no warning), and its models disagree by several degrees on a single day (5.1 to 12.7 °C at one point on 2041-01-01). Never read a date, a month, or a season from it. Compare multi-year windows only, take each model's change against its own baseline window, and report the range across models.
4. One model across a comparison. The archive's default blend changes source by date, and sources differ by more than a degree on the same day (era5 20.5 °C, ecmwf_ifs 18.8 °C). Pin one archive model for the baseline and the period you compare to it. An ERA5-family pin leaves the last week or so null; on 2026-10-28, era5 ended at 2026-10-21. End the comparison at the date the notice names; never splice in another model or forecast past_days. Not every archive model reaches 1940: era5_land returned nulls for July 1945 where era5 had values.
5. Check the marine cell. A coastal land point does not come back null. It is answered from a nearby sea cell: an inland point in Anaheim (33.80, −117.85) returned real waves from 33.708, −117.958. Only points far inland return nulls. Close to shore, neighbouring cells differ: two cells about 8 km apart off Huntington Beach gave swell of 10.3 s from 224° and 7.4 s from 274°. Place the coordinate just offshore of the break, confirm the served cell is the one you meant, and compare one neighbouring offshore cell before you commit.
6. Check the river's magnitude. GloFAS cells sit on a 0.05° grid (snapped values such as 50.925, 6.975), and the cell that answers is not reliably the biggest river nearby. Downtown Cologne (50.94, 6.96) is answered by a cell carrying a few m³/s, while the cell 0.05° west carries over 1,000 m³/s, the Rhine. If the discharge is orders of magnitude below what the named river carries, try neighbouring cells in 0.05° steps across the channel and keep the one that fits.
7. The AQI looks backward. us_aqi and the us_aqi_* sub-indices average over preceding hours: 24 h for particulates, and for ozone the higher of the hourly and 8 h values. In one probe us_aqi_pm2_5 fell from 30 to 22 while hourly pm2_5 rose from 1.4 to 2.6 μg/m³. For conditions at a given hour, read the hourly concentrations (pm2_5, ozone) for that hour alongside the index. The scales also differ: us_aqi 0–50 is good, while european_aqi 20–40 is only "fair".