Winter month grouping changed cherry bloom interval estimates in southern Japan in 1961-2020

We tested whether one December-February temperature mean gave an adequate account of cherry opening-to-full-bloom intervals. We used 2,810 paired records at 47 Japanese stations in 1961-2020. We compared a grouped winter model with separate December-January and February terms on the same records, with station effects, station trends and March-April temperature. The implied month-response restriction was rejected in both southern periods after a four-test adjustment. Southern February slopes were +0.914 and +1.045 days per degree C; December-January slopes were small and uncertain. Northern restriction tests were uncertain. All four restriction tests were uncertain after year effects removed shared annual conditions. On-site and setting checks retained the southern result. This shows sensitivity to month grouping under a stated seasonal model. It does not identify chilling, causal warming effects, pollution or biodiversity loss. The question was selected after exploratory fits. Paired-record selection, missing visits and tree histories, coarse weather and post-event weather windows limit inference. The package includes code, a derived panel, source hashes, all results, a figure and offline reproduction instructions.

phenology · cherry · seasonal windows · observation bias · Japan · model restriction