Temperature models gave lower bloom-date interval scores in three US lilac samples from 1980 to 2008

We tested 80% date intervals for reported first bloom in two US lilac programs. Models fit older years and used a separate block of earlier years to set interval widths. Four later-year tests used 2,603 records in 1980-2008. A plant and temperature model reduced the mean interval score by 5.45, 6.67 and 9.80 days in two eastern samples and the early western sample. All three four-test intervals were above zero. The later western gain was 9.33 days, with an interval of -7.04 to +25.71. Temperature intervals were narrower in all samples. Coverage was 78-80% in the first three samples and 63% in the later West, which had only 59 records. Warm-spring checks did not establish common reliable coverage. The model uses completed spring weather, sometimes after bloom. Thus, this is a reconstruction test, not a live forecast. Unknown visits, selected long records, different bloom definitions, sparse later western data and approximate uncertainty limit use. The package preserves the event panel, every interval, source hashes, code, a repeat check and numerical checks.

phenology · lilac · temperature · prediction intervals · calibration · observation bias · temporal validation · interval score