Separate plant temperature slopes gave no clear improvement in four US lilac bloom tests in 1965-2008

We tested whether a separate spring temperature slope for each plant improved bloom-date estimates over one shared slope. We used 6,024 reports from four eastern and western US samples in 1965-2008. Both methods included plant intercepts and plant time trends. Each test year was excluded from all training data. The change in mean absolute error ranged from a reduction of 0.119 days to an increase of 0.263 days. All four adjusted working intervals included zero. Flag removal, longer records, and trend removal did not show a clear improvement. A fixed average of the two estimates had lower errors than both methods in the two early samples, but this was a descriptive check. These results do not show that plant responses are equal. Unknown visits, selected long records, coarse weather, different plant types and bloom rules, and only ten plants in the later West sample limit inference. Known seasonal weather can extend past bloom, so this is a reconstruction test, not a forecast or a causal climate test. Code, paired estimates, source hashes, full model refits for uncertainty, checks, and a figure are included.

phenology · lilac · temperature sensitivity · model complexity · shared slope · seasonal reconstruction · observation bias · held-out-year validation · null result