Methodology & data quality
How WeatherRecall builds historical weather results, probabilities, and confidence. Last updated: May 26, 2026.
1) Data sources
WeatherRecall combines model-based historical weather data with local station data where available. The global base layer comes from Open-Meteo historical archives, which rely on reanalysis datasets. Station data is used as a local correction layer when stations are sufficiently close to the selected coordinates.
- Global coverage: model/reanalysis data for any coordinate.
- Local enhancement: weather stations within 25 km when available.
- Common fields: temperature max/min, precipitation, wind, cloud cover, weather code.
2) Station blending logic
When multiple stations are available, WeatherRecall uses inverse-distance weighting. Closer stations receive higher weight than distant ones, reducing local representation error. If no station passes quality filters, the result falls back to model-only data.
Distance weighting
Each station contribution scales with inverse distance. This prioritizes nearby observations and limits over-influence of distant microclimates.
Data completeness
Stations with missing or inconsistent fields are down-weighted or excluded for affected variables.
Fallback behavior
If station data quality is insufficient, model output remains the primary source to preserve coverage and avoid synthetic noise.
3) Weather classification and score
Weather codes are mapped into descriptive condition labels such as clear, partly cloudy, light rain, heavy rain, snow, and thunderstorm. These labels are then translated into a normalized weather score from 0 to 10 for comparability across years.
- Scores 8 to 10: predominantly clear or favorable outdoor conditions.
- Scores 5 to 7: mixed but generally usable conditions.
- Scores below 5: unfavorable conditions driven by precipitation, storm risk, or extreme cloud/wind combinations.
The score is a planning indicator, not a forecast metric. It summarizes historical favorability and should be interpreted with precipitation and variability metrics together.
4) Probability calculation
For a selected date and location, WeatherRecall retrieves historical observations over a configurable number of years. It then computes frequency distributions for weather condition labels and derived score brackets.
Example interpretation: if 8 of 10 historical years show clear or partly cloudy conditions for the same date, the probability of favorable conditions is shown as 80% in historical terms. This is evidence of historical tendency, not a guarantee for future weather.
5) Quality controls
Range checks
Extreme outliers outside plausible physical ranges are flagged and excluded from summary statistics.
Field consistency checks
Temperature, precipitation, and weather code combinations are validated for logical consistency before aggregation.
Coverage checks
If historical year coverage is incomplete for a date/location query, the interface reports reduced confidence and avoids overstating certainty.
6) Known limitations
- Microclimate effects (urban canyons, coastlines, altitude transitions) can differ from nearby stations.
- Historical tendency is not future certainty; climate trends can shift seasonal expectations over time.
- Some regions have weaker station density, leading to greater dependence on model data.
- Daily summaries can mask short intense events; time-of-day splits should be checked for event planning.
7) Transparency commitments
WeatherRecall publishes methodology updates whenever core logic changes materially. We also maintain correction workflows for data issues and content inaccuracies. See the editorial policy for correction standards.
Questions about methods can be sent to [email protected].