Observed SST anomaly persistence did not decrease detectably from 1982-2003 to 2004-2025 in OISST, ERSSTv5 and HadISST1, and it did not follow GODAS mixed-layer change

Climate models project that ocean memory, the year-to-year persistence of sea-surface temperature (SST) anomalies, will decrease in the 21st century, mainly because the mixed layer becomes shallower (Shi et al. 2022). We tested whether this decrease is already present in observations. We used OISST v2.1, ERSSTv5 and HadISST1 on a common 2-degree grid (7,291 ice-free boxes, 60S-60N) and compared 1982-2003 with 2004-2025. We removed a quadratic trend and the seasonal cycle in each box. The area-mean change in the lag-1-year autocorrelation of annual means (A1) was -0.031 (OISST), -0.023 (ERSSTv5) and +0.012 (HadISST1). All block-bootstrap 95% intervals included zero, and all values were inside the range of 59 pre-1982 window pairs. The projected forced change for this interval is only -0.022 to -0.029, and the noise SD is 0.05-0.07, so a two-window test needs windows of about 50-75 years. The lag-1-month autocorrelation increased in OISST (+0.032) and ERSSTv5 (+0.025) but decreased in HadISST1 (-0.063, with a step in 2002-2005). A noise-robust AR(1) estimate (r2/r1) did not change in OISST or ERSSTv5. GODAS mixed layers became deeper (+4%), not shallower, and the change in damping rate did not follow the change in mixed-layer depth (OISST slope +0.00, 95% interval -0.17 to +0.12; fixed-feedback slab model predicts -1). A post hoc result: in the NE Pacific, where the largest decrease is projected, A1 increased by +0.17 to +0.23 in all three products. Limitations: GODAS includes Argo data from about 2004 (the split year), the products share input data, and the reference variability comes from overlapping pre-1982 windows, not from a model control run.

ocean memory · sea surface temperature · persistence · autocorrelation · mixed layer depth · OISST · ERSSTv5 · HadISST · GODAS · detection · climate change · marine heatwaves