The 2021 DOISST v2.1 input change causes 23-60% of the 2023 marine heatwave duration excess at 60S-60N; 2023 stays the record year, and the noise hid its peak-intensity record

Question: NOAA DOISST v2.1 became much smoother at the October 2021 input change (Navy AVHRR to ACSPO L3S). Smooth data break marine heatwave (MHW) events less often. Which 2023 MHW record statistics come from the ocean and which come from this change? Approach: we calculated Hobday MHW statistics for 29,129 one-degree cells (60S-60N, 1982-2025, 1983-2012 and 1991-2020 baselines) in published DOISST and in three noise-homogenised versions. Each version gives all input segments the same sub-32-day spectrum (2016-20 level, 2021-25 level, or noise added to the 1985-2006 level). We compared 2023 with 1982-2022 and checked the results with ESA SST CCI v3.0.1 on the same cells (to 2024-06-22) and with 151 moored buoys. Results: in published DOISST, the 2023 mean event duration is 3.16 times the 1982-2022 mean (z 5.4). Homogenisation gives 1.88-2.68 times (z 3.2-4.3). CCI gives 2.31 times, against 3.07 for DOISST on the same cells and days. Thus 23-60% (CCI: 37%) of the 2023 duration excess is artifact. The DOISST/CCI duration ratio steps by +0.31 ln units after 2021 (4.5 SD of 2007-2020); matching gives -0.01 to +0.12. At 64 buoys, the DOISST/buoy event-duration ratio increased by +0.13 ln units after October 2021 (95% interval 0.08 to 0.29, p = 9e-5); homogenised versions show no significant change. 2023 is still the record year in all versions for event duration, MHW days, area and activity. MHW days (125 per cell), ocean area with a heatwave (97%) and activity (5.7e10 C day km2) change by 15% or less. The noise hides other records: 2023 peak intensity ranks 3rd in published DOISST (z 1.7) but 1st after homogenisation (z 2.5-2.9; CCI z 2.9), and the event count ranks 3rd but becomes 1st. North Atlantic, tropical and eastern Pacific duration ratios fall the most. Area-mean (basin) events do not depend on the noise. Limitations: the true pixel-scale fast variance is not known, so the three targets give a range, not a confidence interval; buoys are DOISST inputs and sample only the tropics and US waters; CCI ends in mid-2024 and has its own sensor steps; we did not reproduce the exact method of Dong et al. (2025).