Accuracy Log
Data verification results, known discrepancies, and corrections.
About This Log
This page documents our ongoing data verification process. Mei Tanaka, our data verification lead, compares Open-Meteo API outputs against JMA AMeDAS (Automated Meteorological Data Acquisition System) observations daily. When we identify systematic discrepancies, unusual errors, or data quality issues, we log them here. This is part of our commitment to transparency — we don't hide when our data sources are imperfect.
The log entries below include the date of observation, the city affected, the type of discrepancy, our assessment of cause, and any user-facing impact. Entries are organized with the most recent first.
Known Systematic Patterns
These are recurring discrepancies we've identified through months of verification. They are characteristics of the underlying models, not errors we can fix. We note them so you can mentally calibrate the data.
| Pattern | Cities Affected | Description | User Impact |
|---|---|---|---|
| Summer urban heat underprediction | Tokyo, Osaka, Nagoya | Model temperatures for city-center coordinates run 1-2°C below actual readings at dense urban stations during peak summer afternoons. | Real conditions may feel hotter than displayed. Add 1-2°C mentally for Shinjuku, Umeda, Sakae districts. |
| Sapporo winter overprediction | Sapporo | Winter temperatures in Sapporo are occasionally overpredicted by 1-3°C during clear-sky cold outbreaks, likely due to model bias toward climatological normals. | Dress for conditions 1-2°C colder than displayed on the coldest winter days. |
| Coastal moderation underprediction | Yokohama, Kobe, Fukuoka | Model coastal temperatures occasionally don't fully capture sea breeze cooling effects on summer afternoons. | Waterfront areas may be 1°C cooler than displayed during afternoon sea breeze events. |
| Thunderstorm spatial mismatch | All cities | Precipitation from localized thunderstorms is inherently difficult to predict spatially. A storm predicted for central Tokyo may hit eastern Tokyo instead. | Treat precipitation forecasts as probabilities, not precise location predictions. Carry an umbrella if probability exceeds 40%. |
| Humidity overprediction in dry winter | Sapporo, Nagoya | Winter humidity readings occasionally run 5-10% higher than AMeDAS observations during extended dry periods. | Minimal practical impact. Winter comfort depends more on temperature and wind than humidity. |
Incident Log
2024-11-15
Sapporo: 3°C Temperature Overprediction
On November 14, 2024, Open-Meteo forecast a high of 12°C for Sapporo. The actual high recorded at the Sapporo District Meteorological Observatory was 9°C. This 3°C overprediction occurred during a clear-sky day following a cold front passage. The model appears to have overestimated surface heating from solar radiation. This is consistent with our known "Sapporo winter overprediction" pattern. No correction applied — data displayed as received. Users in Sapporo should note that clear winter days after frontal passages may run colder than forecast.
2024-10-22
Tokyo: Typhoon Precipitation Underestimation
Typhoon Maria's remnants passed near Tokyo on October 21-22. The Open-Meteo forecast (queried 24 hours ahead) predicted 35mm total precipitation for the 24-hour period. Actual rainfall at Tokyo Observatory was 78mm. This degree of underestimation (55mm shortfall) is typical for typhoon precipitation, which depends on precise track forecasting that remains challenging. We added a note to our Tokyo page acknowledging that typhoon rainfall forecasts should be treated as lower bounds. No data correction applied.
2024-09-08
Nagoya: Consistent Summer Overprediction Noted
Analysis of 45 days of summer data (July 15 - August 31) shows Open-Meteo consistently overpredicted Nagoya's maximum temperature by an average of 1.4°C. The model's 5km grid doesn't resolve the urban heat island over Nagoya's central wards, but the bias is larger than expected. Possible cause: the model's surface parameterization underestimates the heat absorption of Nagoya's dense urban fabric. We've added a calibration note to our Nagoya city page advising users to add 1-2°C to displayed summer afternoon temperatures for the city center.
2024-08-19
Fukuoka: Humidity Reading Discrepancy
Humidity readings for Fukuoka ran consistently 8-12% higher than AMeDAS observations during a 5-day period (August 14-19). This coincided with a persistent high-pressure system over Kyushu. The model's boundary layer humidity parameterization appears to overestimate moisture retention during stagnant conditions. Issue resolved when weather pattern changed on August 20. No permanent calibration needed — this appears to be pattern-specific rather than systematic.
2024-07-30
All Cities: API Outage (Resolved)
Open-Meteo API experienced a partial outage on July 29, 2024, from approximately 06:00-10:00 UTC. During this period, our site displayed "Updating..." for all city cards and city pages. The issue was on Open-Meteo's infrastructure, not our site. Service was fully restored by 10:30 UTC. This incident highlights why we display "Updating..." rather than stale cached data — we prefer no data to wrong data.
2024-06-14
Osaka: Tsuyu Onset Timing Error
The rainy season (tsuyu) officially began in the Kansai region on June 11, 2024 (per JMA announcement). Open-Meteo's model correctly captured the synoptic shift, but precipitation type classification initially showed "moderate rain" for June 11-12 when actual conditions were light drizzle. The total precipitation amounts were accurate (8mm over 48 hours), but the intensity classification was slightly overstated. Minor issue — no correction applied. Note that tsuyu drizzle often feels lighter than model "light rain" classifications suggest.
2024-05-03
Kobe: Orographic Enhancement Not Captured
A frontal system on May 2 produced 24mm of rain at the Kobe Marine Observatory, while the Open-Meteo forecast predicted 12mm. The 100% overprediction gap is explained by orographic enhancement from the Rokko Mountains, which forced additional uplift as the moist frontal air encountered the terrain. The 5km model grid captures the broad topography but misses the local enhancement effects. We've noted this in our Kobe city page — hillside areas in Kobe receive more rain than the forecast suggests during frontal passages.
Verification Methodology
Our verification process uses the following methodology:
- Reference data: JMA AMeDAS observations from the station nearest to our query coordinates for each city.
- Comparison metric: We compare the 24-hour maximum and minimum temperature forecasts (issued the previous day) against actual observations. For precipitation, we compare 24-hour cumulative amounts.
- Threshold for logging: Temperature discrepancies greater than 2°C or precipitation discrepancies greater than 10mm trigger a log entry.
- Frequency: Verification is performed daily for all 8 cities. Systematic patterns are assessed over rolling 30-day periods.
- Publication: Significant findings are published on this page within 48 hours of identification.
Limitations of Our Verification
Our verification has limitations you should understand:
- We compare model grid-cell forecasts against point observations. A 5km grid cell contains multiple microclimates. The station we use may not represent your specific location.
- We verify only temperature and precipitation. Humidity, wind, and other parameters are checked occasionally but not systematically.
- Our sample size is limited. We've been operating since early 2024, so we don't yet have full seasonal cycles for all patterns.
- We cannot verify the "current" conditions display in real-time because we don't have access to real-time station feeds. We verify against archived observations with a 24-hour lag.
How to Report Data Issues
If you observe a significant discrepancy between our displayed weather data and actual conditions, you can report it through our contact form. Please include: the city, the date and time, what our site showed, and what you observed (with your measurement source if available). We investigate all reports and add confirmed issues to this log.