March temperature gave lower cherry bloom error than June temperature in three of six adjusted tests in 1981-2024

We tested whether March temperature gives lower cherry bloom error than June temperature, which follows all selected bloom events. Separate rolling models used 30 prior years, February temperature and a time trend at Kyoto, Liestal and Washington. Tests covered 1981-2002 and 2003-2024. March minus June mean absolute error ranged from -3.60 to -0.55 days. Three six-test adjusted block intervals were below zero: early Liestal, later Kyoto and later Washington. No site had this result in both periods. Early Liestal crossed zero when the trend was removed. Fixed sites and paired errors reduce some observation bias, but missing visit logs and reporting changes remain limits. A later-month control tests temporal detail in predictive information. It does not identify a causal effect or validate an early forecast. The package gives code, aggregate results, source hashes, a figure and retrieval steps. External records and annual derivatives stay outside the publication.

phenology · cherry bloom · negative control · temporal validation · temperature · observation bias