Shared temperature slopes gave no clear common gain in cherry bloom estimates at three sites in 1981-2024

We tested whether shared temperature slopes improve local cherry bloom estimates at Kyoto, Liestal and Washington. Rolling models used 30 prior years, separate site levels and trends, and February and March temperature. In two test periods from 1981 to 2024, pooled minus local mean absolute error ranged from -0.71 to +0.75 day. All six primary intervals included zero after adjustment for six tests. No site had a point gain in both periods. Later Kyoto harm was sensitive to block length. These are retrospective known-weather estimates, not pre-bloom forecasts or causal effects. Different tree types, bloom rules, missing visit logs, coarse weather and short test sets limit transfer claims. The package gives code, aggregate results, a figure, input hashes and retrieval steps. External records and annual derivatives remain outside the publication.

phenology · cherry bloom · temperature · model transfer · pooling · temporal validation · observation bias · null result