Stock-specific warming effects on fish surplus production from Free et al. (2019) changed with RAM Legacy v4.66 assessment revisions and did not improve one-step forecasts of later years in 91 stocks

Question: are the stock-specific sea-temperature effects (theta) on surplus production of Free et al. (2019, Science) stable enough to forecast? Approach: we refitted their Pella-Tomlinson model with AR(1) error per stock. We used the revised RAM Legacy v4.66 assessments and stock SST series emulated from ERSST v5 (median r = 0.99 with the original series). We compared theta before and after each stock's original end year (91 stocks, 742 forward stock-years, mostly 2010-2017), and one-step forecasts with and without theta. A simulation in which theta does not change gave the expected agreement. Results: our fit reproduced the published fixed-effect theta (Spearman 0.92). For the same years, the revised assessments changed theta (Spearman 0.73 between versions), and the change increased with the revision of the surplus-production series (Spearman -0.75). Of 11 published significant stocks, 6 stayed significant with the same sign. Forward theta agreed with earlier theta less than the constant-theta simulation expected (Spearman 0.18, expected 0.41, p = 0.02; median absolute change 0.62, expected 0.37, p < 0.01). Refitted theta made forecasts worse than theta = 0 (median skill -0.026, expected +0.008, p < 0.01; 45% of stocks better). The published shrunk theta was neutral (median skill +0.0007). Even with constant theta, the expected median gain was below 1%. HadISST and other sensitivity tests gave the same conclusions. Limitations: emulated SST, a fixed-effect fit, short forward periods, and assessment terminal years that are not independent observations. The test covers stock-level forecasts, not the long-term fleet average. RAM and SST data are external; code, our per-stock estimates and aggregates are bundled.