Recent counts reduced next-survey presence forecast error for 33 Maizuru fish species in 2012–2023

We tested whether recent fish counts and water temperature improve forecasts of recorded species presence beyond the seasonal calendar. We used open twice-monthly visual census data from Maizuru Bay, Japan. The first ten years, 2002–2011, selected 33 species and supplied initial training data. Logistic models used only information available at each issue slot and were updated once per year. Tests covered 288 later surveys in 2012–2023. Recent presence and log count reduced mean Brier error from 0.121391 to 0.100399, a 17.29% gain. The nominal 95% two-year block interval was 14.25% to 20.03%. The gain occurred in every test year and in 32 of 33 species. Temperature at the preceding survey reduced calendar error by 2.00%. Beyond fish history, its gain was 0.42%, with an interval of −0.25% to +1.07%. Planned model checks preserved the history gain. In checks added after the main scores, history reduced error by 9.43% beyond a calendar with a time trend, 7.85% with five-year training windows, and 11.47% beyond previous-year same-slot records. These gains had intervals above zero. The gain size therefore depends on the baseline. The result supports recent counts as a forecast baseline at this site. It does not identify a cause, model detection, test rare species, or limit the value of nonlinear temperature models. The package preserves source tables, all probabilities, code, provenance, figures, plans, and separate numeric checks.

ecological forecasting · fish occurrence · Maizuru Bay · seasonality · forecast baseline · water temperature · open data · chronological validation