Adding flight date spread raised relative squared error by 0.40% in linear forecasts for 12 UK summer butterflies, 2001–2024

We tested whether adult flight date spread improved next-year butterfly abundance forecasts after mean date and abundance history were included. We extended a prior SCI benchmark with the UKBMS 2024 source release. Separate species models used earlier targets only. The main test had 11,521 forecasts for 12 summer species at 455 sites in 2001–2024. The target was ln(1 + abundance index). Adding flight date SD raised the equal-species relative mean squared error score by 0.40%. The nominal 95% interval was +0.07% to +0.83%. It used 4,999 shared three-year blocks of test years. Among seven species with at least ten forecasts in every test year, the increase was +0.02%, with an interval of −0.25% to +0.34%. The 2015–2024 increase was +0.18%, with an interval of −0.20% to +0.85%. The added spread term gave no useful average gain in this model. The small full-sample increase does not show that wider flight periods harm populations. The intervals hold fitted forecasts, species and sites fixed. The sample mainly represents established populations in England. Dates and indices share counts and errors. This is a retrospective test from final annual indices, not an independent field validation. The package preserves source data, provenance, code, predictions, figures and exact repeat checks.

butterflies · phenology · flight date spread · population forecasts · UKBMS · temporal validation · open data