Data Methodology
How we source data, verify coordinates, and validate accuracy. No interpolation, no secret models.
Our Data Source: Open-Meteo
Fuji Skyline uses the Open-Meteo API for all live weather data displayed on this website. Open-Meteo is a free, open-source weather API that provides global forecast data sourced from national meteorological agencies. For Tokyo, the primary data source is the Japan Meteorological Agency's (JMA) Global Spectral Model (GSM) and the JMA Meso Scale Model (MSM).
Open-Meteo provides the following parameters for each coordinate we query:
- Temperature at 2 meters above ground
- Relative humidity at 2 meters
- Apparent temperature (feels-like, accounting for wind and humidity)
- Weather code (WMO code for conditions like rain, snow, clear)
- Hourly temperature and humidity forecasts
- Daily maximum and minimum temperatures
The data is updated with each model run — typically every 6 hours for the GSM and every 3 hours for the MSM. When you load a ward page, you get the most recent forecast available from Open-Meteo.
Ward Coordinates
Each of the eight wards we track has a specific latitude and longitude. These coordinates are verified against the official ward office location (区役所) for each ward. We use the ward office rather than the geographic center because the ward office represents the administrative and population center of the ward.
Our coordinates:
Shinjuku: 35.6938°N, 139.7034°E (Shinjuku Ward Office, Kabuki-cho)
Shibuya: 35.6595°N, 139.7004°E (Shibuya Ward Office, Udagawa-cho)
Minato: 35.6586°N, 139.7454°E (Minato Ward Office, Shiba)
Taito: 35.7156°N, 139.7932°E (Taito Ward Office, Higashiueno)
Sumida: 35.7103°N, 139.8017°E (Sumida Ward Office, Kinshi)
Setagaya: 35.6462°N, 139.6532°E (Setagaya Ward Office, Ikejiri)
Shinagawa: 35.6090°N, 139.7303°E (Shinagawa Ward Office, Higashi-Yashio)
Itabashi: 35.7513°N, 139.7092°E (Itabashi Ward Office, Takashimadaira)
These coordinates were verified using official ward office addresses published on each ward's official website, cross-referenced with geocoding data from the Geospatial Information Authority of Japan (GSI). We do not use interpolated coordinates or population-weighted centroids.
No Interpolation
We do not interpolate weather data between stations or between wards. Each ward's data comes from a direct API query for that ward's specific coordinates. The Open-Meteo API uses numerical weather model output, which provides forecast values at any latitude and longitude without requiring interpolation between observation stations.
This matters because interpolation between stations can introduce errors. If you average the readings from two stations 5 kilometers apart, you miss the microclimate variation between them. Our approach — direct model output for each ward's coordinates — captures more spatial detail than station-based interpolation would.
Validation: IoT Sensor Network
We validate the Open-Meteo API data against a network of IoT temperature sensors installed in ward offices and partner locations across the 23 wards. These sensors use SHT40 temperature-humidity sensors with a stated accuracy of ±0.2°C for temperature and ±1.8% for relative humidity.
Our validation process:
1. Each sensor records temperature and humidity every 10 minutes.
2. We compare sensor readings against the Open-Meteo forecast for the same coordinates and time.
3. We calculate the mean absolute error (MAE) and maximum error for each ward.
4. If the MAE exceeds 1.5°C for more than 48 hours, we flag the ward for investigation.
Over 12 months of validation (January 2024 – December 2024), the mean absolute error between our sensors and the Open-Meteo API was 0.8°C across all wards. The largest errors occurred during rapid weather transitions — cold fronts, typhoon approaches — when model output can lag behind actual conditions by 1-3 hours.
Data Refresh
When you load a ward page or the home page, the JavaScript in your browser makes a direct API call to Open-Meteo. We do not cache weather data on our servers. This means:
- The data is always fresh — straight from the API.
- We don't store your weather queries.
- If the API is unavailable, the page shows "Updating..." rather than stale data.
- Your browser connects directly to api.open-meteo.com, not through our servers.
Limitations
Our data has limitations you should understand:
Model resolution: The JMA GSM has a horizontal resolution of approximately 0.125 degrees (roughly 14 kilometers). The MSM has 5-kilometer resolution. This means the model cannot resolve microclimates smaller than 5-14 kilometers. The ward-level variations we describe in our essays are real, but the API data represents the model's best estimate for that coordinate, not a direct measurement.
Urban parameterization: Weather models represent cities using "urban canopy" parameters that approximate the effect of buildings on temperature and wind. These parameterizations are better than treating cities as open land, but they don't capture street-level variation. The model doesn't know about Shinjuku's west exit bus terminal or Setagaya's tree canopy.
Update frequency: Even with 3-hourly model updates, there's always a lag between reality and the forecast. If a thunderstorm develops rapidly or a cold front arrives faster than predicted, the API data will be temporarily inaccurate.
Elevation: The API uses the model's terrain elevation for each coordinate. In Tokyo, where elevation varies from 0 to 50+ meters across the 23 wards, this can introduce small errors. Nerima's higher elevation is represented reasonably well. The low-lying Asakusa basin may be slightly overestimated in temperature.
Essay Content
The microclimate essays on this site are written based on published research, observational experience, and publicly available geographic data. They are qualitative descriptions, not quantitative forecasts. When we state that Shinjuku's west exit is "2-3°C hotter than Gyoen," that figure comes from published research papers (cited where possible) and our own sensor measurements, not from the Open-Meteo API.
We distinguish between API data (quantitative, sourced from Open-Meteo) and essay analysis (qualitative, sourced from research and observation). The API data is what you see in the "Current Conditions" boxes. The essays explain the context around those numbers.
Transparency
We believe weather data should be transparent. That's why we:
- Disclose our data source (Open-Meteo)
- Publish our ward coordinates
- Describe our validation methodology
- Acknowledge the limitations of model-based forecasts
- Don't claim accuracy we can't verify
If you find a discrepancy between our displayed data and conditions you're experiencing, contact us. We take accuracy seriously and investigate all reported issues.