Adding bottom temperature gave no clear error reduction in fixed ridge predictions of Maizuru fish shares for 2012–2023

We tested whether bottom water temperature improved predictions of recorded fish species shares beyond season, year, and the previous survey at Maizuru Bay, Japan. The data had 113 species codes and 528 surveys in 2002–2023. Models used only earlier years for fitting and predicted 288 surveys in 2012–2023. Temperature was measured during the target survey, so these are conditional predictions. At the fixed ridge penalty of one, adding temperature reduced mean squared Hellinger error by 0.55%. The nominal 95% three-year block interval was −1.03% to +1.93%. Every planned fish-history comparison interval included zero. A check added after the main result selected penalties using past years only. That rule gave a 1.50% reduction, with an interval of +0.27% to +2.66%. Removing the seasonal and year component from temperature gave a 0.17% reduction beyond fish history at the fixed penalty. A simpler season-and-year model had lower Hellinger error than the main fish-history baseline. The last observed composition had the lowest Bray–Curtis error among the fixed-penalty rules. The gain was small and depended on the model and score. These results apply to one visual transect and do not test nonlinear temperature effects. The package contains source data, provenance, code, predictions, checks, figures, and offline repeat instructions.

fish composition · Maizuru Bay · conditional prediction · water temperature · seasonality · ecological monitoring · ridge regression · reproducibility