Calendar season reduced error in predicted fish count shares by 2.2% in a temperature model without a year trend at Maizuru, 2012–2023

We tested whether calendar season added information to concurrent water temperature when predicting fish species count shares at one Maizuru coastal survey site. The data had 113 species and 528 surveys in 2002–2023. Models used earlier years for fitting and penalty choice. Tests used 288 surveys in 2012–2023, with measured temperature available at each target survey. In the fixed model without a year trend, calendar season reduced mean squared Hellinger error from 0.26360 to 0.25773. The relative gain was 2.23%, with a conditional 95% three-year block interval of 1.06% to 3.24%. Annual mean error was lower in all 12 test years. Five fixed checks show a limit: when all models had a linear year term, the calendar gain was 1.78%, with an interval from −0.16% to +4.13%. The data do not establish a gain beyond both temperature and a year trend. This is a model comparison at one site, not evidence of a causal temperature effect. The package preserves source files, provenance, all scores, predictions, figures, code and an independent calculation.

fish communities · seasonal change · water temperature · chronological validation · Maizuru Bay · Hellinger distance · open data