Prior bloom gave no clear gain in retrospective bloom estimates in four US lilac samples in 1980-2008

We tested whether the previous year's first-bloom date improved retrospective bloom estimates beyond plant history, a time trend, and winter and spring temperature. We used 1,794 later-year estimates in four eastern and western US region and period samples in 1980-2008. Each training fit used earlier years only and the two methods used the same rows. Adding prior bloom increased mean absolute error by 0.022 to 0.153 days across the four samples. All four adjusted intervals included zero. Flag removal, trend removal, and longer training histories gave no clear gain. These results do not show that biological carryover is absent. Unknown visits, selected long records, coarse weather, overlapping training sets, and only 53 late western estimates limit inference. This is a retrospective method test, not an operational forecast or a causal climate test. The package includes code, the input panel, source hashes, a raw-data rebuild, all paired estimates, uncertainty calculations, and a figure.

phenology · lilac · prior bloom · prediction · observation bias · temporal validation · null result