The gap between observed bloom variance and frozen model variance increased at 46 Japan sites in 1954-2021

We tested whether changes in year-to-year cherry bloom variance exceeded changes from a frozen forcing model. We used the same sites in 1954-1987 and 1988-2021 and removed local linear trends. The samples had 3,098 records at 46 Japan sites and 1,811 records at 29 Swiss sites. In Japan, observed mean residual variance changed from 16.12 to 21.59 days squared. Model variance changed from 9.73 to 7.90. The excess change was 7.30 days squared, with a 97.5% five-year block interval of 0.98 to 13.35. All planned Japanese checks kept positive intervals. The Swiss excess was 5.51, with an interval of -17.88 to 34.18. A common excess change was not established. The Japanese ratio of variance ratios had an adjusted interval that included one. The result thus depends on the additive variance scale. The observed Japanese variance change alone also had an interval that included zero. We reuse a published panel and hold model parameters fixed. Selected records, unknown observation changes, monthly Japanese weather, and omitted model uncertainty limit inference. The result does not identify a cause. The package includes code, source hashes, all bootstrap draws, raw-date checks, numerical checks, and figures.

cherry bloom · phenology · variance change · thermal time model · observation bias · Japan · Switzerland · block bootstrap · reproducible analysis