Where the reports come from
All are public pages. We take the facts only (spot, status, date), once or twice a day, and credit and link the source wherever we show it.
The day each city's official maple and ginkgo tree turns colour and drops its leaves.
Japan Meteorological Corporation
City colouring forecasts for 51 cities, maple and ginkgo, several rounds a season.
Status for about 750 spots from the JR System feed, with one time stamp per prefecture page.
Nikko's own tourism body: status for about 24 spots, updated every few days.
Kyoto City Official Travel Guide
Kyoto City: status for about 78 temples and gardens, from Oct 30.
Weathernews
Spot status and user reports. Coming soon.
Instagram photos
Dated visitor photos for hero spots. Coming soon.
Visitor reports
Visitor votes, shown separately. Coming soon.
Prefecture outlines on the Best time map: 「国土数値情報（行政区域データ）」（国土交通省）をもとに PeakYet 作成. Contains National Land Numerical Information (Administrative Areas) by the Ministry of Land, Infrastructure, Transport and Tourism (MLIT), CC BY 4.0, simplified by PeakYet. https://nlftp.mlit.go.jp/ksj/gml/datalist/KsjTmplt-N03-2025.html
How we label each report
Each label tells you how much weight a report can carry.
| Label | When we use it | What you see |
|---|---|---|
| Confirmed | An official tourism body in the last 3 days, or 2 or more dated photos from the last 48 hours that agree | Solid pin with a dark ring, "Seen 6h ago" |
| Reported | Another source in the last 3 days (today, the JR System feed on tenki.jp) | Solid pin zoomed out, a ring zoomed in; source + age |
| Older than 3 days | The newest report is more than 3 days old | Grey pin with its age |
| PeakYet Score | A tested forecast for the city's official tree | City circle and curve, label in words, % on tap |
| Usual window | The source's usual timing for the spot | Grey text, never colours the map |
When sources disagree
The official tourism body comes first, then the JR System feed on tenki.jp. Sources we add later will sit between them. Visitor reports, once they start, will appear separately and never change a status.
A city forecast never decides a spot's colour. Each city's forecast follows one tree that the weather agency watches, and the spots nearby keep their own timing. Kyoto shows why: the city's tree turned on Dec 10 in 2025, while the famous temples usually peak in late November. A spot page shows its city's chance only when the spot is in the same prefecture and usually peaks at the same time as that tree.
The PeakYet Score: the chance a city is at peak
For 51 cities (one per prefecture, six in Hokkaido, none yet in Okinawa), Japan's weather agency (JMA) watches one maple and one ginkgo tree, and records the day each turns colour and the day its leaves fall. The PeakYet Score is the chance that a date falls between those two days.
Until JMA records the colour, the score mixes two things equally: Japan Meteorological Corporation's forecast, widened by how far its past forecasts missed, and the last 10 or so years of JMA records for that city. Each day that passes without colour rules out the earliest dates. Kyushu maples use Kyushu's own forecast errors.
Once JMA records the colour, the start date is known and only the length of peak is uncertain. Once the leaves fall, the score drops to 0. Future days are capped at 85%, because weeks ahead nothing is certain. Cities with fewer than 5 seasons of records get no score.
How we tested it
We scored 1,001 past forecasts (527 maple, 474 ginkgo, 2018–2025) with the same code that runs the site, testing each year with a model built without that year. The table uses the Brier score, a standard measure of how far chances land from what happened: lower is better.
| Tree | PeakYet Score | JMC forecast alone |
|---|---|---|
| Maple | 0.0948 | 0.0994 |
| Ginkgo | 0.1010 | 0.1068 |
When the score gave a chance, it came true about that often: in every band from 0 to 90%, the share of days actually at peak was within 3 points of the score.
Labels
We show the chance in words. Tap it to see the %. A date range gets the chance of its best day.
- Unlikely: 0% to 25%
- Possible: 25% to 50%
- Likely: 50% to 75%
- Very likely: 75% and up