DOISST v2.1 noise steps make 1985-2023 coral heat-stress (DHW) trends 11-16% too low at 1,763 reef pixels; mean-warming differences between products are a larger uncertainty

Question: NOAA DOISST v2.1 lost fast (day-to-day) SST variance in steps at satellite input changes. Degree Heating Weeks (DHW), the standard coral bleaching heat-stress index, adds only SST excesses of 1 C or more above the warmest-month mean, so it is a convex function of daily SST. Do the noise steps bias reef DHW trends? Approach: At 1,763 DOISST reef pixels (sites of a global coral-bleaching database), we calculated Coral Reef Watch statistics (CoralTemp v3.1 method) for raw DOISST and for three versions in which the spectrum at periods shorter than about 32 days is homogenised to one input segment. We tested the method on synthetic data, at 149 moorings (97 GTMBA, 52 NDBC; 2,816 buoy-years) on matched days, and against ESA SST CCI v3.0.1 on every day of 1982-2024 at 683 reef pixels in three boxes. Results: Relative to 2007-2015, reef 4-8 day power was 1.86 times larger in 1985-2006 and 0.29 times smaller after 2021. The 1985-2023 trend of annual maximum DHW was +0.92 C-weeks per decade (0.66 to 1.18) in raw DOISST and +1.04 to +1.10 after homogenisation; the trend of the reef fraction at DHW >= 4 was +9.8% and +10.4 to +11.7% per decade. Raw DOISST overstates 1985-2006 heat stress (mean annual maximum DHW 1.67 against 1.41 C-weeks) and understates 2022-2025 heat stress (6.35 against 6.66). Homogenisation recovered the synthetic truth (0.980 against 0.983). At 74 GTMBA moorings, after removal of the DOISST-minus-buoy mean warming drift (median +0.04 C per decade), the raw DOISST-minus-buoy DHW trend was -0.13 C-weeks per decade (-0.22 to -0.04) and the homogenised trend was +0.005 (-0.08 to +0.09); without this removal the two errors cancel. In the boxes, CCI gave +0.56 C-weeks per decade and DOISST +1.05; DOISST with the CCI mean warming gave +0.60, so about 90% of the product difference is the mean warming trend (DOISST +0.06 C per decade faster). Conclusion: The input steps make DOISST DHW trends too low by about 11-21%, opposite in sign to the heatwave-duration bias. Differences in mean warming between products change reef DHW trends by a factor of about two. Limitations: the true fast variance is unknown; moorings are not at reefs and are DOISST inputs; CCI has its own variance changes and covers three boxes to 2023; we did not test the CoralTemp product or bleaching observations.