February freezing nights gave no clear cherry bloom prediction gain at two sites in 1988-2025

We tested whether February freezing-night counts improved cherry bloom predictions beyond February mean temperature and a time trend. Fixed records and nearby daily weather stations retained 69 Liestal years and 75 Washington years in 1951-2025. Models used only earlier years for training. In four site and target-period tests from 1988 to 2025, changes in mean absolute error were +0.340, +0.038, +0.001 and -0.089 days. All four 98.75% block intervals included zero. Separate early and late association intervals also included zero. Freezing counts and mean temperature had strong correlations, which limited their separation. A -2 degree threshold gave a small pooled secondary gain in Washington, but no clear gain in both separate target periods. This result limits one simple predictor. It does not show that cold weather has no effect on flowering. Small samples, station mismatch, different bloom rules, missing weather and unknown observation changes limit inference. We do not estimate air pollution effects, damage, biodiversity loss or climate causation. The package includes code, exact source hashes, weather evidence, model inputs, all results, validation reports and a figure.

phenology · cherry bloom · freezing nights · temperature · temporal validation · observation bias · null result · public data