Past water temperature gave no clear gain in ridge forecasts of Maizuru fish count shares in 2014–2023

We tested whether past water temperature improves a prediction of relative fish counts at one Maizuru Bay transect. We used 113 fish codes and 528 twice-monthly surveys in 2002–2023. Each test year used only earlier years for fitting, input scaling, temperature means and ridge cost selection. The main test scored 240 one-survey-ahead predictions in 2014–2023. The baseline used seasonal terms, year and the previous fish composition. Adding four past bottom-temperature terms changed mean squared square-root-share error from 0.465096 to 0.465235. The error reduction was −0.030%, with a conditional 95% two-year-block interval of −0.591% to +0.363%. Surface temperature, occurrence shares and a 2009–2023 test also gave no clear gain. This is a null result for the stated linear models and lead time. It does not reject nonlinear or causal temperature effects. The package preserves source tables, all forecasts, tuning scores, code, provenance, figures and a clean replay that matched 29 output hashes.

fish community · Maizuru Bay · ecological forecasting · water temperature · temporal validation · ridge regression · relative composition · open data · null result