Noise steps cause 30-55% of the 1982-2025 marine heatwave duration rise in OISST v2.1, hide a peak-intensity rise and reverse the cold-spell duration trend at 60S-60N; heatwave-day trends are robust

Question: Daily NOAA OISST v2.1 has step changes in fast (sub-monthly) variance at satellite input changes. Do these steps change the standard marine heatwave (MHW) and marine cold spell (MCS) statistics, and what trends remain after they are removed? Approach: For 29,129 ice-free native pixels (60S-60N, 1982-2025), we measured the anomaly spectrum in six input segments and compared it with 173 moored buoys. We made two noise-homogenised versions that change only periods shorter than about 32 days: spectral matching to the 2016-2020 segment (closest to the buoys), and noise addition up to the 1985-2006 level. We calculated Hobday et al. (2016) statistics (1983-2012 baseline) with code that agrees with the reference code to better than 2e-11. We tested the methods on synthetic data and at 127 long-record buoys on matched days. Results: Relative to 2016-2020, OISST had 1.8 times the 4-8 day power in 1985-2006 and 0.2-0.3 times in 2021-2025; segments agree at 64 days and more. The published MHW duration trend (+4.45 days per decade) falls to +3.12 (matching) or +1.99 (noise addition); the frequency trend rises from +0.66 to +0.85 or +0.97 events per year per decade. The MHW max-intensity trend changes from +0.006 to +0.053 C per decade (tropics: -0.044 to +0.025). The MCS duration trend changes from +0.75 to -0.28 days per decade. MHW days, area and cumulative intensity change by less than 5% (matching). tau10 falls from +2.33 to +0.34 days per decade. In synthetic tests the matching method recovered the homogeneous-target MHW trends within 10%. At the buoys, homogenisation removed the published biases in peak intensity (-0.055 C per decade, p = 6e-5) and memory, and the duration medians agree (buoy +1.45, homogenised +1.56, published +2.11 days per decade). Limitations: the method corrects only spectral properties, assumes the sub-monthly power changes are not real, keeps only about one third of a real tau10 change, and does not correct a 0.03 C per decade mean-warming offset relative to the buoys; buoys are OISST inputs and sample only the tropics and US waters.