Honeysuckle dates lowered mean lilac reconstruction error by 1.97-3.84 days in four US samples in 1972-2008

We tested whether a same-site Arnold Red honeysuckle bloom date improved retrospective estimates of lilac bloom beyond monthly weather. Each model used only earlier years within its network and period for training. Four samples had 765 test site-years in 1972-2008. Adding honeysuckle dates reduced mean absolute error by 1.97-3.84 days. Four-test adjusted intervals excluded zero in East later years and West earlier years, but included zero in the other two samples. Shared visits, coarse weather, selected complete pairs, small cluster counts, and training dependence limit inference. Honeysuckle was on or after lilac bloom in 36.7-43.0 percent of test pairs, so this is not a real-time forecast test. The study reuses a verified published panel. It does not establish a species interaction, an effect of air pollution, or a causal climate-change response. Code, source extracts, provenance, predictions, training cutoffs, and numerical checks are included.

phenology · lilac · honeysuckle · retrospective reconstruction · temperature · observation bias · time-ordered validation