Time trends reversed spring weather prediction gains in two English Meadow Brown panels in 2013-2024

We tested whether observed March-May temperature improved later Meadow Brown mean flight-date predictions in England. Models fitted in 1980-2001 used fixed panels of 112 southern and 13 northern sites. Adding temperature to site and linear time terms raised mean squared prediction error in all four region-period tests in 2002-2012 and 2013-2024. Only the later southern primary interval excluded zero. Its gain was -44.77 days squared, with a 98.75% annual block bootstrap interval of -67.65 to -18.51. Without time terms, later gains were +25.61 and +21.55 days squared in South and North. Count controls and a high-count filter kept the primary gain signs negative. Thus the later point gains depended on the time term. This is a retrospective model-transfer result, not proof of no temperature effect. Selected sites, missing visits, shared weather, coarse grids, few northern sites and uncertainty conditional on training fits limit inference. The package gives source hashes, code, aggregate losses, checks and a figure. External source and record-level data remain outside the baseline publication.

phenology · butterfly · Meadow Brown · temperature · model transfer · prediction · observation bias