Warm winter day counts gave no repeated gain in US lilac bloom forecasts beyond monthly means in 1988-2008

We tested whether warm winter day counts improve March 1 predictions of first lilac bloom beyond monthly temperature means. We used public plant records and daily CPC maximum temperature. Each model used earlier years only. The added predictor was the number of January-February days at or above 15 C. Four East and West, early and late samples gave 953 predictions in 1988-2008. Mean absolute error gains were +0.202, -0.075, +0.159, and +0.091 days in early East, late East, early West, and late West. A positive value is better. The gain did not repeat across all four samples. The early western interval depended on the bootstrap block length, and its point gain changed sign when the trend or history rule changed. The later West had only 89 predictions. One bloom preceded March 1. A complete-winter rule removed 949 of 3,724 plant records. Unknown visit delays, selected long records, coarse weather, and different plants and bloom rules by region limit transfer. The result does not show that daily temperature has no biological role. It does not measure pollution or a causal climate-change effect. The package includes daily weather extracts, source hashes, code, predictions, all scores, figures, raw-data checks, and reproduction instructions.

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