Most of the 1982-2025 rise in daily SST memory in DOISST v2.1 is a loss of fast variance that steps at satellite input changes and is absent in 34 NDBC buoys

Question: Lee et al. (2025) reported that the memory of daily sea surface temperature (SST) anomalies in NOAA DOISST increased across the global ocean after 1982. Is this a change in the ocean or a change in the analysis? Approach: We repeated their one-lag memory metric tau10 = -10/ln(phi10) on DOISST v2.1, 1982-2025, for 29,423 ice-free 1-degree cells (60S-60N). We fitted each yearly autocorrelation function at lags 5-20 days as A exp(-k/tau_s). A is the share of slowly decaying variance and tau_s is its decay time. A synthetic test showed that this fit separates a loss of fast variance from a real slowing. We compared breaks with documented DOISST input changes, compared DOISST with hourly in situ temperature from 34 NDBC moored buoys on matched days, and compared it with OI.v2 weekly, HadISST and ERSST v5. Results: The tau10 increase repeats (11.5 to 22.6 days, 1982-1986 to 2021-2025; significant increase in 63% of the area). 66% of the ln(phi10) trend comes from A and 24% from tau_s. tau_s increased significantly in only 10% of the area. The variance of the 1-day anomaly change fell by a factor of about 6, in steps that a break search placed at 2007 and 2021. These are the dates when DOISST changed from Pathfinder to Navy AVHRR and from Navy AVHRR to ACSPO L3S. Steps at input-change dates fit the global tau10 series better than a linear trend (BIC 58.0 versus 80.5). Inside single-input segments the tau10 trends were small and not significant. At the 34 buoys, buoy day-to-day variance had no trend (median -0.014 per decade in log units) but DOISST at the same cells fell (-0.160; paired p = 1e-7). The buoy tau10 trend was -0.24 days per decade and the DOISST trend was +0.32 (paired p = 5e-4). OI.v2 weekly and HadISST monthly showed no increase in persistence in 1982-2021. Conclusion: Most of the reported increase in daily SST memory is a product artifact from a loss of fast variance. A moderate rise in slow decay time remains possible in some regions. Limitations: the buoys cover only US sites at 17N-57N and also feed DOISST; significance tests do not correct for serial correlation; the 2022-2025 period mixes an input change with the 2023-2024 warm event.

sea surface temperature · persistence · memory timescale · autocorrelation · marine heatwaves · OISST · DOISST · data homogeneity · observing system change · NDBC buoys · AVHRR · ACSPO · critical review · replication