In NS-IBTS winter surveys, 14 of 34 North Sea demersal fish species tracked 5-year mean SST in 1980–2001, as Perry et al. (2005) reported, but these relations did not predict 2002–2026 latitudes

Question: Perry et al. (2005) found that the mean latitude of many North Sea demersal fishes changed with winter temperature in 1977–2001. Do relations of this type, fitted on data up to 2001, predict the latitudes of the same species in 2002–2026? Data: ICES DATRAS NS-IBTS quarter-1 records, 16,769 valid GOV hauls in 1980–2026 in 162 fixed ICES rectangles, and HadISST1 January–March SST over the survey area (ERSSTv5 as a check). Method: for 34 demersal species, we calculated the CPUE-weighted centroid latitude for each year. We fitted latitude on SST in 1980–2001 and predicted 2002–2026 from the observed SST. Skill is 1 − MSE(model)/MSE(training mean). Same-year SST was the pre-specified predictor. We added the trailing 5-year mean SST, as in Perry et al., as a labelled post hoc analysis. Results: with 5-year SST, 14 of 34 species had p < 0.05 in 1980–2001 and 11 of the 14 had a northward relation, near Perry's 15 of 36 species with 13 northward. With autocorrelation-corrected sample sizes, only 5 species stayed at p < 0.05. Surrogate SST series gave 14 or more significant species in 5.2% of cases. Out of sample, the median skill was −0.08 (95% block-bootstrap CI −0.15 to 0.003) with 5-year SST and −0.02 (CI −0.03 to 0.002) with same-year SST. The model was better than no change for 15 of 34 species. For the 14 responsive species, the median skill was −0.70, and only 6 kept their slope sign in 2002–2026. Training slopes did not predict test slopes (Spearman −0.22, p = 0.21). A community index of the responsive species had r = 0.87 with SST in training and r = 0.08 in the test period. Linear trend extrapolation was worse (median skill −0.57). Depth centroids, ERSSTv5, the 1977–2001 all-gear window, log weights, an abundance covariate and five other variants gave the same result. Cod moved north more than predicted. Sole moved south although its model predicted a northward shift. Conclusion: the in-sample Perry-type pattern reproduces, but species-level temperature–latitude relations from 1980–2001 did not transfer to the next 25 years. Limitations: SST, not bottom temperature; fixed survey footprint; fishing and recruitment effects are not modelled; the 22-year training period gives imprecise slopes; the 5-year analysis is post hoc.

North Sea · IBTS · DATRAS · fish distribution · range shift · climate change · out-of-sample test · replication · sea surface temperature · centroid latitude · Perry et al. 2005