Methodology Notes

This page documents the specific methods we use to observe, detect, and analyze Tokyo's sea-breeze front. We publish this for transparency and reproducibility. If you disagree with our approach, spot a methodological flaw, or want to replicate our work, this is the place to start.

1. Open-Meteo API Integration

The current weather boxes displayed on our site pull data from the Open-Meteo forecast API. We chose Open-Meteo because it's free, requires no API key, and sources its output from multiple NWP models including JMA's GSM. The API endpoint we use is the standard forecast endpoint with the following parameters:

• current=temperature_2m,relative_humidity_2m,apparent_temperature,pressure_msl — current conditions at 2 meters above ground
• daily=temperature_2m_max,temperature_2m_min — daily extremes for context
• timezone=Asia/Tokyo — local time
• forecast_days=3 — minimal horizon, we only display the current interval

The API is called directly from the client's browser using JavaScript's Fetch API. Our server is not involved in the request or response. This means your browser sends the HTTP request to Open-Meteo's servers, not ours. Open-Meteo may log your IP address; see their privacy policy for details.

Spatial resolution caveat: Open-Meteo's underlying model data has a horizontal resolution of approximately 0.25° (~25 km) for the global models. Tokyo's 23 wards span roughly 15 km east-west, which means the entire metropolitan area falls within a single grid cell of the coarsest model. The API output for Koto and Shinjuku may therefore return very similar values, as both points are resolved by the same grid cell. We acknowledge this limitation and display the data as "current conditions at [location]" rather than implying ward-scale precision.

2. JMA AMeDAS Station Pair Method

Our historical analysis and long-term trends rely on the JMA AMeDAS network. The core of this method is the "coastal-inland pair" — two AMeDAS stations with similar elevation but different distances from the bay, whose temperature difference proxies the sea-breeze strength.

Primary pair: Haneda (station 4426) and Fuchu (station 4413).

• Haneda: 35.5489°N, 139.7836°E, elevation 6 m, distance from bay ~0.5 km
• Fuchu: 35.6689°N, 139.4781°E, elevation 39 m, distance from bay ~25 km

Both stations are at low elevation (the elevation difference of 33 m introduces a temperature bias of approximately 0.2°C, which we correct for using the standard lapse rate of 6.5°C/km). The key difference is their position relative to the bay: Haneda is directly on the shore, while Fuchu is far enough inland to be outside the sea-breeze influence on typical days.

We classify sea-breeze days using the Haneda-Fuchu pair as follows:

• Strong sea-breeze day: Temperature differential (Fuchu minus Haneda, lapse-rate corrected) ≥ 3.0°C at 14:00 JST, with Haneda wind direction in the onshore sector (45°-180°).
• Moderate sea-breeze day: Differential ≥ 2.0°C at 14:00 JST, with Haneda wind direction in the onshore sector.
• Weak or no sea-breeze day: Differential < 2.0°C, or onshore wind criterion not met.

The 14:00 JST timestamp is chosen because it represents the typical peak of the sea-breeze circulation in August. We have verified that using 13:00 or 15:00 produces similar classifications in 92% of cases.

3. Frontal Observation by Temperature Differential and Wind Direction

For detecting the sea-breeze front's inland position on a given day, we use a multi-criteria approach:

Step 1: Check the Haneda-Fuchu pair. If the differential is below 2°C, classify as "no front detected" and stop.

Step 2: Check the Koto-Shinjuku pair (using Open-Meteo data as a proxy, since there are no AMeDAS stations at exactly these locations). If the differential exceeds 2°C, the front has penetrated at least to central Koto.

Step 3: Check the Odaiba-Shimbashi gradient. If the temperature at Odaiba is more than 2°C cooler than Shimbashi with an easterly wind at Odaiba, the front has reached the 5-km mark.

Step 4: During field deployment periods, consult the instrumented transect data (100-m spacing) for precise frontal position within the Ginza-Shimbashi corridor.

This hierarchical approach — coarse pair first, finer pairs next, transect for precision — allows us to classify every day even when fine-scale data is unavailable. The coarse Haneda-Fuchu pair catches the large majority of sea-breeze days. The finer pairs add detail about penetration depth. The transect, available only during deployment windows, adds the precise inland position.

4. Instrumented Transect Protocol

Our highest-resolution data comes from a line of portable temperature loggers deployed across the Ginza-Shimbashi corridor. Here's how it works:

• Equipment: Onset HOBO U23 Pro v2 temperature loggers, ±0.2°C accuracy, ±0.1°C resolution.
• Spacing: Loggers placed at 100-meter intervals along a roughly east-west line from Ginza Crossing to Shimbashi Station.
• Duration: Loggers deployed for 2-week periods during peak sea-breeze season (late July through mid-August).
• Logging interval: 1 minute, providing high temporal resolution to catch rapid frontal passages.
• Positioning: Loggers mounted at 1.5 meters above ground on streetlight poles, shaded from direct sun, with radiation shields.

Data is downloaded after retrieval and processed through a Python pipeline that: (1) applies a quality control filter to remove obvious sensor errors, (2) identifies frontal passages using the temperature-gradient criterion (>2°C per 100 m sustained over 200 m), and (3) outputs timestamped frontal positions for each detected event.

5. Satellite Imagery Analysis

We use Himawari-8 Band 3 (visible, 0.64 μm) imagery at 10-minute temporal resolution and 500-meter spatial resolution. The sea-breeze front cloud line is identified visually by one analyst (Marina Kondo) and confirmed by a second (Tetsu Yamamoto) when possible. The cloud line's inland position is estimated by overlaying the imagery on a geographic reference and measuring the distance from the bay shore to the leading edge of the cloud line along a transect through central Tokyo.

This method is subjective — cloud lines can be ambiguous, and their correlation with the surface front is not perfect. We estimate our satellite-derived frontal position accuracy at ±2 km. The cloud line is used as supplementary evidence, not as a primary detection method.

6. Known Limitations

Our methods have several limitations that we acknowledge openly:

• The Open-Meteo API provides model estimates, not ground truth. Local variations of 1-3°C are possible.
• The Haneda-Fuchu pair has a 25-km separation, which smooths over fine-scale frontal structure.
• The instrumented transect covers only the Ginza-Shimbashi corridor. Frontal behavior north or south of this line may differ.
• Our satellite cloud-line detection is manual and subjective.
• Building-level wind channeling is not captured by any of our methods.