Added March 1-14 temperature reduced cherry bloom error in two of four site-period tests in 1981-2024

We tested whether daily weather through March 14 adds cherry bloom information beyond February temperature and a local time trend. We used separate Liestal and Washington models, with past 30-year training and tests in 1981-2002 and 2003-2024. March 1-14 temperature reduced mean absolute error by 3.11 days in early Liestal and 1.75 days in later Washington. Eight-test adjusted intervals were below zero for these two samples. The other two gains were uncertain. No site had a clear gain in both periods. Added March 15-31 temperature gave no clear further gain after correction. The two early-window gains persisted with shorter training, other block lengths and no time trend. All 86 test records bloomed on or after March 15. This is a retrospective information test with immediate access to daily station weather, not a test of historical forecast delivery. Fixed sites, paired errors and fixed windows reduce some observation bias. Missing visit logs, source changes, station mismatch, selected complete years and approximate intervals limit inference. Raw records and annual derivatives remain external. Code, source hashes, aggregate results and a figure are included. Pollution and causal climate effects were not tested.

phenology · cherry bloom · daily temperature · forecast information · observation bias · temporal validation