For 1,200 years, someone in Kyoto has written down when the cherries bloomed

How do we know what a spring in the year 1000 was like?

At the time, almost nobody wrote "the cherries bloomed today". But people threw parties under them. The climatologist Yasuyuki Aono, of Osaka Prefecture University, built on lists by Taguchi (1939) and Sekiguchi (1969): he read the diaries of the imperial court, of nobles and of Kyoto's monasteries, and pulled out the days on which people had gone out to view the blossom. Of the 838 dates, 454 come from the description of such a party: 54%.

Some traces are even less direct. Thirty dates come from diaries that mention cherry branches sent as gifts by people who had been at a party. Another 39 come from the title of a poem. For 14 years in which nobody wrote about the cherries, Aono started from the day wisteria or kerria had flowered, noted in the same diaries, and worked out the cherry date from it; he published the wisteria method with Shizuka Saito.

The first date in the series is from 812. The official chronicle Nihon kōki describes a flower banquet given by Emperor Saga in the Shinsen-en garden. The chronicle does not say which trees they were, so it is not certain they were cherries.

The 703 springs before 1888 come from 210 diaries, chronicles and compilations. These gave the most; the names are romanised as in the series.

Which year did the cherries bloom earliest?

In 2021, Kyoto's cherry trees were in full bloom on 26 March. International coverage, citing Aono's team, called it the earliest bloom in more than 1,200 years. Two years later the trees bloomed a day earlier still, on 25 March. In the series, 2023 holds the record, yet a New York Times article of 17 April 2026 still presented 2021 as the earliest bloom in 1,200 years.

Before them, the record was 27 March, reached three times: in 1236, from the title of a poem; in 1409, from a party; and in 1612, from a diary.

A one-day ranking needs care. Aono and Keiko Kazui write that parties were held very close to the first day of full bloom, and the wild cherry stays in full bloom for only two to four days. The dates for 2021 and 2023 come from newspaper notices about Arashiyama; the one for 1236 from a poem. A day between them is less than full bloom lasts, and the old dates carry no day-level error estimate. The 11-day move in the average is of another order.

At the other end, the latest recorded bloom was in 1323, on 4 May.

Are cherry trees really blooming earlier than they used to?

Yes. Between 812 and 1850, the average of the 667 recorded springs falls on 15 April. Over the last 30 years, 1996 to 2025, it falls on 4 April, 11 days earlier. On the chart, the average line wanders up and down around the same day for almost a thousand years and drops steeply only in the 20th century.

The rises have an explanation too. Aono and Kazui used the flowering dates as a thermometer and found four cold periods: 1330–1350, 1520–1550, 1670–1700 and 1825–1830. By their reconstruction, March in those periods was 3–4 °C below the present-day normal used in the study.

How much of the recent drop is down to people? Nikolaos Christidis of the Met Office, Aono and Peter Stott answered in 2022 with climate models. Human influence has brought flowering forward by over a week, and about half of that shift comes from the warming of the city itself. A bloom as early as 2021's would be rare without human influence; it is now about 15 times more likely.

How reliable are the old dates?

• The species. The series follows the wild mountain cherry, yamazakura (Prunus jamasakura), though old documents do not always name the species. Weather stations do not routinely record it, which is why the modern dates come from elsewhere.
• The modern dates. Since the 1880s the day comes from newspapers: railway companies often advertised when the cherries at Arashiyama, a well-known blossom-viewing spot on the western edge of the city, were in flower.
• The calendar. The old documents used the Japanese lunisolar calendar. Every date was converted to its Gregorian equivalent, so 1 April 812 compares directly with 1 April 2023.
• The party as a measure. Full bloom lasts two to four days, and the party fell close to the first. Dates deduced from wisteria carry an average error of about 3.4 days.
• The gaps. 377 of 1,215 years have no date. On the chart they stay empty, with no calculated values in their place.
• 2026. The date of 29 March was added by Genki Katata, who continues the series after Aono, and is marked in his table as an estimate. On the chart it has a dotted outline. It is counted among the 838 springs but left out of the averages and the records.

Sources

• Genki Katata, Historical Series of Phenological data for Cherry Tree Flowering at Kyoto City, the full table 812–2026 with document names (accessed 28 September 2026). The same data are at Our World in Data (CC BY 4.0).
• Yasuyuki Aono and Keiko Kazui, "Phenological data series of cherry tree flowering in Kyoto, Japan, and its application to reconstruction of springtime temperatures since the 9th century", International Journal of Climatology 28 (2008): 905–914, doi:10.1002/joc.1594 (free text).
• Yasuyuki Aono and Shizuka Saito, "Clarifying springtime temperature reconstructions of the medieval period by gap-filling the cherry blossom phenological data series at Kyoto, Japan", International Journal of Biometeorology 54 (2010): 211–219, doi:10.1007/s00484-009-0272-x (free text).
• Nikolaos Christidis, Yasuyuki Aono and Peter A. Stott, "Human influence increases the likelihood of extremely early cherry tree flowering in Kyoto", Environmental Research Letters 17 (2022): 054051, doi:10.1088/1748-9326/ac6bb4.
• Aono's earlier series (to 2015) is archived at NOAA NCEI Paleoclimatology; it matches Katata's table day for day.
• The 2021 coverage: NBC News, 1 April 2021. The 2026 article: The New York Times, reprinted by CIGS. The banquet of 812: Hakushika Sake Museum.