Primary networks
JMA observatories. The four Tokyo-area Japan Meteorological Agency observatories — Tokyo, Yokohama, Chiba, Saitama — supply hourly records from 1949 onward and the 1991–2020 normals used as our baseline. JMA station metadata (siting, elevation, relocation history) is consulted before any station is used as a reference; the Tokyo station's three relocations are handled by using the homogenized series and flagging trend figures with the relocation caveat.
AMeDAS. The Automated Meteorological Data Acquisition System's ten-minute interval data resolves the variation hourly records miss. We use the eight Tokyo-prefecture stations, with Hachioji as the Tama Hills reference. AMeDAS siting favors aviation and agriculture; we therefore never use a single AMeDAS station alone to characterize a zone — zone figures are station-plus-transect blends.
Tokyo Metropolitan Government heat island monitoring. The 2023 report and its 120 surface temperature survey points anchor our surface-temperature claims, supplemented by Landsat thermal imagery analysis for skin temperature at district scale. Satellite passes are snapshots; we use them for spatial structure, never for time-series trends.
XRAIN radar. JMA's radar composite at 250-meter resolution verifies baiu front position and typhoon rain-band distribution across the five zones. Radar estimates are always cross-checked against gauges before publication — radar alone overestimates orographic cells over the Tama ridges, a known bias we correct against the ground network.
Saitama cooperative gauges. Forty rain gauges hosted by agricultural cooperatives across the northern plain, maintained with contributor David Park. Tipping buckets, logged hourly, read against the JMA Saitama observatory monthly. This network is the sole source of the dry-slot mapping.
The mobile transects
Between 2019 and 2022 the survey vehicle ran fixed routes in all five zones on a repeating schedule: pre-dawn (04:00–06:00), midday (12:00–14:00), and evening (21:00–23:00). Instrumentation: aspirated, radiation-shielded thermometers at 1.5 meters, GPS-logged at ten-second intervals. Handling rules: no readings within 50 meters of running vehicle exhausts beyond our own; transects discarded if the shield aspirator fails inspection; each route repeated a minimum of eight times per season per year before its means enter the dataset.
Derived figures
Zone bias figures are five-year means of simultaneous station differences, cross-checked against transect means. Annual temperature means use the 1991–2020 normal period. Single-event figures (Hagibis, Faxai, the 2018 yamase, Kumagaya's 41.1°C) are reported as events and never folded into climatological means. The seasonal matrix values are normals, zone-adjusted using AMeDAS differentials where a JMA observatory doesn't sit inside the zone.
Live feeds
Current conditions come from the Open-Meteo API at page-load time for the five measurement coordinates, Asia/Tokyo timezone. The feed is model-interpolated point data. Failure behavior is fixed: the display reads “Updating…” and no cached or estimated values are substituted, ever.
Review
Every published figure passes a two-person check: the analyst who produced it and one other team member reproducing it from source. Disagreements over 0.2°C or 10mm go back to the instruments. What survives that process still occasionally turns out wrong — which is what the corrections page is for.