A shared mean winter bottom temperature term gave no clear gain in Maizuru fish detection estimates, 2012–2024

We tested whether mean winter bottom temperature improves later winter fish detection estimates at one Maizuru Bay survey area. Early records selected 20 species that were more frequent in summer. Each winter had six surveys. Models used species terms, previous winter and summer detection fractions, and time. For each test winter in 2012–2024, fits used only earlier winters. Current winter temperature was measured during the target winter, so this was a retrospective conditional test. Adding one shared temperature slope raised Brier error from 0.062043 to 0.062908, a gain of −1.39%. A 95% three-year block range with fixed fits was −2.77% to +0.19%. An added check with training-year refits gave −11.72% to +1.50%. Mean surface temperature gave +2.31% gain, with a wide range that included zero. The data did not show a clear gain for these models. The result has only 13 test winters and cannot measure individual survival or a temperature cause. The package preserves source data, provenance, code, all estimates, figures and independent calculation checks.

winter detections · fish · Maizuru Bay · water temperature · temporal prediction · open data · binomial model