Bottom temperature reduced error in quadratic models of observed fish richness at Maizuru in 2013–2023

We tested whether bottom water temperature predicted observed fish richness better than surface temperature at one Maizuru Bay visual survey site. We used 528 surveys from 2002–2023 and 113 fish species codes. Each model used a quadratic temperature term. For each test year in 2013–2023, we fitted the model to all earlier years. The 264 test surveys gave mean absolute errors of 2.358 species for bottom temperature and 2.707 for surface temperature. The decrease was 12.9%. The paired difference was −0.349 species, with a nominal 95% three-year block interval of −0.572 to −0.081. Bottom temperature gave less error in nine of 11 test years. The result depended on the target and model. With exact richness standardization to 20 counted fish, the quadratic difference was −0.011 expected species, with an interval of −0.030 to +0.010. A cubic check still favored bottom temperature for that target. Calendar terms removed clear support for a bottom-temperature advantage in observed richness and reversed the ranking for 20-fish richness. The comparison describes predictions conditional on temperature measured during each survey. It does not identify a causal temperature effect or establish a sensor ranking at other sites. The package preserves source records, provenance, model fits, all predictions, figures, and offline reproduction code.

fish richness · water temperature · Maizuru Bay · temporal validation · rarefaction · seasonal ecology · open data