A heat-sum model fitted to year-to-year variation predicts 84% of the spatial flowering gradient of clonal Somei-yoshino cherry at 81 Japanese stations in 1953–2025; adding chilling gives 94–98%

Question: does a phenology temperature response measured from year-to-year variation predict the differences between sites (space-for-time substitution), and is a mismatch evidence of local adaptation? Somei-yoshino cherry is one grafted clone, so it cannot show genetic local adaptation and gives a non-genetic reference. Data: Japan Meteorological Agency phenology records (47 events with at least 20 stations, 1953–2025, regular species, north of 30° N) and GHCN-Daily station temperature. Method: for each event we selected a heat-sum (or autumn cooling-sum) model by within-station error only, with a free requirement at each station. Then we predicted station mean dates with one common requirement. The spatial slope rho of observed on predicted station means is 1 if the year-to-year model explains the spatial gradient. Intervals: 2,000 station-bootstrap replicates with model re-selection. Null simulations with one requirement gave rho 0.98–1.04 for all 16 forcing-model events. Results: for cherry first flowering (81 stations, 5,061 station-years, within-station skill 0.73), rho was 0.84 (95% CI 0.71–0.90), and 0.84 (0.70–0.90) for full bloom. Warm-winter stations flowered later than predicted (+4.6 days in the warmest fifth). rho stayed below 1 in period halves, without Hokkaido, without the warmest fifth, with on-grounds trees only and without gap-filled temperature (0.75–0.85). For 20 events with skill ≥0.2, the median rho was 0.85, and the clone's mismatch and its station spread in fitted requirement (SD of log requirement 0.11) were as large as for most other plants. A requirement that decreases by 0.5% per November–January day below 7.2 °C, selected by within-station error only, raised the clone's rho to 0.94 (0.86–0.99) and 0.98 (0.80–1.02). A 0–7.2 °C chill count did not (0.81). Linear window-temperature slopes did not detect the clone's mismatch. Limits: one urban reference tree per station, coarse model grids, a skill threshold and chill metric chosen after first results, unhomogenised station temperature. Conclusion: in this network, a non-genetic space–time mismatch of about 15% exists in a clone, so similar mismatches in other species do not by themselves show local adaptation; a chilling term should be tested first.

phenology · space-for-time substitution · thermal time · growing degree days · chilling · local adaptation · clone · Somei-yoshino · cherry blossom · Japan Meteorological Agency · GHCN-Daily · temperature sensitivity