Oxygen on the 26.5 isopycnal of the core CalCOFI grid stopped its 1984-2006 decline and stayed low in 2007-2025; trend, salinity-proxy and new-decline forecasts failed (bottle data)

Question: did the 1984-2006 oxygen decline in the subsurface southern California Current continue, and did published forecasts work? Approach: we used Winkler bottle oxygen from the CalCOFI bottle database (to 2021-05) and from the bottle rows of CalCOFI CTD product files (2021-07 to 2025-04). The data are 9,705 casts at 57 core stations. We calculated anomalies from the 1984-2006 station x season means and tested 2007-2025 out of sample. Results: on sigma0 = 26.5, oxygen fell at -1.04 umol/kg per year in 1984-2006 (95% interval -1.39 to -0.69). In 2007-2025 the slope was +0.10 (-0.08 to +0.28), and the mean anomaly was -16.2 umol/kg. The extended 1984-2006 trend (RMSE 8.7) forecast worse than the held 2002-2006 level (RMSE 3.9), but this loss difference is marginal. A new strong decline by 2020, as Pozo Buil and Di Lorenzo (2017) predicted, did not occur: 2018-2025 oxygen was 1.8 umol/kg higher than 2008-2017 (0.1 to 3.5), and the 60 umol/kg hypoxic boundary was 11 m deeper (2 to 21). The isopycnal-salinity proxy (r = -0.92 in 1984-2006) failed after 2006 (r = -0.15). Salinity went back toward its mean, but oxygen stayed 11.5 umol/kg below the proxy (-14.4 to -9.6). Solubility explains 4% of the oxygen loss. The AOU gain matched the phosphate gain at the remineralization ratio (ratio 186, interval 158-230), and N* fell by 0.67 uM. All 57 stations lost oxygen. The results hold nearshore and offshore, on sigma0 = 26.7, without 2023, and in the bottle database alone. Limitations: post-2021 values are preliminary. We tested the local CalCOFI salinity relation, not the reanalysis index of the 2017 paper. The test record has 19 autocorrelated years. The process attribution is descriptive.

ocean deoxygenation · California Current · CalCOFI · hypoxic boundary · dissolved oxygen · isopycnal salinity · spiciness · apparent oxygen utilization · forecast verification · out-of-sample test · N* · Southern California Bight · decadal variability