1. The three readings
Every morning between 06:30 and 07:00 CET we open three windows side by side.
- Météo-France Paris bulletin. The 06:00 UTC run of the ARPEGE model, as presented on the Météo-France Paris page. We read the 12-hour forecast for wind, cloud base, and horizontal visibility over Île-de-France.
- Observatoire de Paris–Montsouris. The last twelve hours of surface readings from the Montsouris station: pressure, temperature, dew-point spread, and wind. The dew-point spread is a useful early warning for haze — a spread of less than 3 °C in the morning often means a hazy afternoon.
- SETE anemometer. When the Eiffel Tower's public page shows a wind reading from the summit, we record it. If no reading is available (early morning, planned maintenance) we interpolate from the ARPEGE forecast adjusted for the known increase of wind with altitude in Paris of roughly 1.4× between 50 m and 300 m on a stable day.
2. From readings to the panel
We average the three visibility figures with a weighting of 40 % Météo-France, 40 % Montsouris, 20 % SETE anemometer condition. The result is rounded to the nearest kilometre and then banded.
| Visibility band | Horizontal distance | Typical landmarks visible |
|---|---|---|
| Excellent | > 50 km | Forest of Fontainebleau, Rambouillet plain |
| Good | 30 – 50 km | Sacré-Cœur, La Défense, Saint-Germain-en-Laye |
| Fair | 20 – 30 km | Centre of Paris, Bois de Vincennes |
| Poor | 10 – 20 km | Only inner arrondissements |
| Very poor | < 10 km | Nothing beyond the Champ-de-Mars |
3. When we are wrong
We are wrong on roughly one morning in six. The two most common reasons are (a) a Saharan dust layer that arrives after the 06:00 UTC run of ARPEGE and (b) a low cloud base that drops below 300 metres between our morning reading and midday. We publish an update at 13:00 CET to correct both.
4. What we do not do
- We do not run a model. We do not have the computing power or the training.
- We do not accept data from private aggregators (AccuWeather, Weather.com) because we cannot audit their modelling.
- We do not forecast specific photographs — colour, cloud types, the sparkle. These depend on light and timing in ways no forecast can anticipate.
5. Reproducibility
Everything we read is publicly available and free of charge. Anyone with an interest can reproduce our note by opening the same three sources in order. We encourage this. If your reading differs from ours on a given day, we would like to hear why — the editor's address is on the imprint page.