Ozone and temperature gave mixed held-out floral pigment prediction gains in three regions in 1941-2017

We tested whether total-column ozone and temperature improve prediction of floral UV pigment beyond species, month, location and year. We used 1,224 complete herbarium records from three regions in two periods. Models held out five-year blocks and scored 1,198 records with species and months seen in training. Added environmental terms reduced mean square error by 2.94% and 4.36% in Australia. Error rose by 1.64% and 4.03% in Eurasia. North America changed from a 5.94% decrease to a 0.97% increase. Five of six descriptive block-bootstrap ranges included zero. Later North America changed sign with ridge strength. Thus this fixed model gave no common prediction gain across all regions and periods. This is blocked reconstruction, not a future forecast or causal test. Nonrandom collection, few blocks, shared training sets and uncertain source corrections limit inference. Total-column ozone is not surface ozone pollution. Code, source hashes, aggregate errors and a figure permit reproduction.

floral pigmentation · total-column ozone · temperature · blocked prediction · observation bias · herbarium