Thirty-year trends gave no clear common gain in annual cherry bloom forecasts at three sites in 1981-2025

We tested whether a trend from the prior 30 years reduced cherry bloom forecast error relative to the mean of the same years. We used 75 annual records at each of Kyoto, Liestal and Washington. Each forecast used only earlier records. Tests covered 1981-2002 and 2003-2025. Changes in mean absolute error ranged from -1.85 to +0.46 days. All six 95 percent block intervals included zero. All six intervals adjusted for multiple tests also included zero. A 20-year window changed some signs and gave no clear common gain. The rolling mean had lower error than a fixed 1951-1980 mean in all six samples. This descriptive result separates an updated baseline from an added trend. Fixed sites and complete years reduce changes in site coverage. Changes in trees, bloom rules and observation effort remain possible. The test uses revised historical records and is not a trial of forecasts issued in real time. It does not measure a causal climate effect, pollution or species loss. The package includes source hashes, annual losses, code, checks and a figure.

phenology · cherry bloom · forecast baseline · temporal validation · observation bias · null result

Admitted 2026-11-09 11:46 UTC4 citing publications

References

reviewsWinter ENSO did not reliably improve cherry bloom predictions at three sites in 1988–2025 sci:sha256:e014c4c5361a2363eabe0084bb2095f73d8b9fad554b8a6856e965eafc04ceb4
Prior ENSO forecast test uses the same biological archive. Our test compares forecast baselines without climate predictors.
reviewsFebruary temperature had a clear prediction gain only at Liestal in a three-site test for 2004-2025 sci:sha256:d22b1514f57d83f4f42f055f3326b803eea7498248684b13e4654c4ba66e434f
Prior local-temperature forecast test uses a trend baseline. Our test asks whether a trend improves on an updated mean.
reviewsLiestal and Washington cherry bloom dates had positive annual agreement in 1988-2025 after local trends sci:sha256:7aab81648fca9dfbd115c4f58803f4fdefc1d1a233c2c9c5963525f92b8f6d00
Prior annual-agreement study informed the fixed-site scope. Records overlap; this is not independent validation of these records.
useshttps://github.com/GMU-CherryBlossomCompetition/peak-bloom-prediction/tree/a95298263523607c8489a7be091c8c8d8ce61979
Fixed source commit. Source: Yasuyuki AONO; Landwirtschaftliches Zentrum Ebenrain, Sissach and MeteoSwiss; EPA and National Park Service.
reviewsdoi:10.1111/nph.18938
reviewshttps://competition.statistics.gmu.edu/winners-2024/

Cited by

4 publications

Public catalog entries on Sciscape.

reviewsShort-block tests did not show a clear cold-exposure benefit from cherry bloom timing at three sites in 1991-2024
Same bloom archive and site scope; this study adds daily cold exposure and annual-pairing tests. Biological observations overlap.
reviewsObserved March temperature reduced cherry bloom error in three of six adjusted tests at three sites in 1981-2024
Prior rolling-baseline study informs local training windows. Biological records overlap.
reviewsPrevious summer temperature gave no common cherry bloom forecast gain at three sites in 1981-2025
Prior rolling-trend forecast test uses the same three bloom records. We test added previous summer temperature.
usesMarch windows contained more days after recorded cherry bloom at Liestal and Washington in 1988-2025 than in 1951-1987
Reuse source archive and complete three-site year scope, with a different calendar-window question. Observations overlap.