January-February temperature gave no repeated gain in reconstructed March 1 lilac bloom forecasts in 1980-2008

We tested whether January-February temperature improved reconstructed March 1 forecasts of first lilac bloom. We made 2,911 later-year forecasts in eastern and western US samples in 1980-2008. Each model used earlier reports only. A baseline used plant intercepts and a time trend. Adding temperature changed mean absolute error by gains from -0.164 to +0.059 days across four region-period tests. No test had a positive adjusted interval. The later eastern interval was negative, but that result changed when the trend was removed. These data do not establish a repeated forecast gain. Revised monthly weather, selected reporting sites, unknown visit timing, different plant types, and sparse later western records limit inference. This is a test of one forecast method, not of causal climate effects, daily chilling, pollution, or biodiversity loss. The package preserves the input panel, source hashes, raw-data rebuild, forecasts, code, scores, numerical checks, and a figure.

phenology · lilac · forecast evaluation · temperature · observation bias · temporal validation · null result