Daily OISST v2.1 shows a near tripling of 1-day SST memory in 1982-2025, but it comes from less day-to-day variance and a 2022 input step; other products show no common weekly or monthly rise

Question: Lee et al. (2025) used daily OISST and reported that the memory of sea surface temperature (SST) anomalies increased by about 100% from 1982 to 2021. Is this increase a property of the ocean, or of one data product? Approach: We calculated autocorrelation statistics on one ice-free ocean area (60S-60N) in daily OISST v2.1 (1-degree subsample of 0.25-degree pixels, 1982-2025), weekly OISST v2, ERSST v5 and HadISST 1.1. We used block-bootstrap intervals, a noise-robust ratio of lag correlations, a two-part (fast and slow) split, and like-for-like tests on the same 2-degree boxes and weeks. Findings: In daily OISST v2.1 the median pixel memory time (tau1 = -1/ln r1) increased from 10.6 days (1982-1991) to 31.9 days (2016-2025). This agrees with Lee et al. But the day-to-day share of variance decreased by half (0.199 to 0.092), and the decay of the slow part did not change (r60/r30 0.652 and 0.655). One AR(1) process with the observed tau1 predicts a +0.29 increase of the 30-day-mean correlation; the observed increase is +0.05. Median tau1 has a step of +31 days (SE 1.9) in 2022, after the documented change to ACSPO satellite input on 1 October 2021 (we selected this test after we saw the data). On the same 7,004 boxes, the lag-1-week correlation increased from 0.726 to 0.810 in daily OISST v2.1 but stayed at 0.757-0.760 in weekly OISST v2. From 1983-2002 to 2003-2022 the monthly lag-1 autocorrelation changed by +0.021 (95% -0.013 to +0.056) in ERSST v5, -0.075 (-0.117 to -0.036) in HadISST and -0.007 (-0.047 to +0.034) in OISST v2. Conclusion and limitations: The reported memory increase is mostly a loss of fast, day-to-day variance in one product, with a large step at an input change. It does not agree with a uniform increase of thermal inertia. The products do not agree on a common weekly or monthly change, so the observations do not yet support a global increase or the predicted decrease. A real decrease of day-to-day variance cannot be fully excluded, the daily data are subsampled, and we did not examine satellite observation counts.

sea surface temperature · ocean memory · persistence · autocorrelation · OISST · ERSST · HadISST · marine heatwaves · data homogeneity · replication