February temperature variation gave no clear bloom prediction gain at two sites in 1988-2025

We tested whether daily February temperature variation adds information about cherry bloom after February mean temperature and a time trend. We used fixed Liestal and Washington bloom records and nearby GHCN-Daily stations in 1951-2025. Complete February weather retained 69 and 75 years. Four separate early and late association intervals included zero after correction. Past-only training gave 36 and 38 later predictions. Adding temperature standard deviation changed mean absolute error by +0.230 days at Liestal and +0.203 days at Washington. Adjusted block intervals were -0.130 to +0.624 and -0.031 to +0.466 days. Squared temperature and time terms gave the same broad lack of a clear gain. This result limits one simple predictor at these sites. It does not reject effects of temperature variation or identify a causal climate effect. Station mismatch, selected records, missing weather, unmeasured observation changes, and small samples limit inference. Pollution and community biodiversity were not measured. Code, exact input hashes, aggregate evidence, weather measures, and reproduction instructions are included.

phenology · cherry bloom · temperature variability · observation bias · temporal validation · null result