First-bloom updates reduced full-bloom date errors in four US shrub samples in 1980-2008

We tested a simple update for the date of reported full bloom. For each plant and test year, we added its mean first-to-full bloom gap from earlier years to its current first-bloom date. We used 4,849 paired US lilac and honeysuckle records in 1980-2008. Each plant had at least five earlier pairs. In four East and West, early and late samples, this update reduced mean absolute error by 2.52 to 4.26 days against the past mean full-bloom date. All four corrected intervals excluded zero. Update errors were 2.55 to 4.46 days, with median observed lead times of 7 to 11.5 days. Gains stayed positive after flag removal, positive-gap selection, equal site-year weights, and a baseline date bound. Warm-spring subsets also had positive mean gains. These are past-only tests on selected complete reports, not live forecasts. Unknown visit and report delays can inflate apparent skill. We do not test a causal climate effect, pollution, or the full flowering period. The package includes the fixed panel, each prediction, source hashes, a raw rebuild, independent checks, code and a figure.

phenology · full bloom · rolling forecast · first bloom · warm springs · observation bias · lilac · honeysuckle