Methodology and data sources

All meteorological data is retrieved through Open-Meteo, which aggregates open data from national weather services. The data is published under the Creative Commons Attribution 4.0 licence.

Numerical models in use: ECMWF IFS (European Centre for Medium-Range Weather Forecasts, 25 km), NOAA GFS (United States, 13 km), DWD ICON (German Weather Service, 11 km) and JMA (Japan Meteorological Agency, 5 km).

Monthly climate averages are computed from the ERA5 reanalysis over the last ten complete years — observations, not forecasts.

Air quality comes from CAMS, the Copernicus Atmosphere Monitoring Service.

Current conditions: at most every 15 minutes.

Hourly forecast: every 60 minutes. Daily forecast: every 3 hours.

Monthly climate averages: every 30 days.

Every page shows when its data was retrieved. If the upstream server is briefly unavailable we serve the last good copy and still show its true age rather than pretending it is fresh.

Each day the system stores the forecast it issued for 1, 2 and 3 days ahead. Once the day has passed we compare it with observed values and compute the absolute error.

Results are published on the accuracy page, including when they are unflattering. The table only appears once at least 14 days have been verified.

A forecast is a probability, not a fact. Error grows sharply after day 7; figures at days 10–14 should be read as a trend, not a commitment.

Convective storms in Cambodia are smaller than the grid cell of most global models. A thirty-minute downpour over one district can be absent from the forecast entirely.

For communes and small places the coordinates are a representative centre point; conditions can differ, especially in mountains or near the coast.

Every day the system stores the forecast it issued, then checks it against observed data. These numbers are the result of that check, not a marketing claim.

2026-08-13 → 2026-09-02 · n=60