Winter ice had a lower slope for following than preceding autumn evaporation in Lake Superior estimates for 1973-2022

We tested whether the winter ice link to autumn evaporation differs before and after winter. We used fixed NOAA GLERL maximum ice cover and L2SWBM v2023.2 monthly net evaporation medians for 50 Lake Superior winters in 1973-2022. After adjustment for a linear trend, the ice slope was +1.43 mm before winter and -7.06 mm after winter per 10 ice percentage points. The paired after-minus-before contrast was -8.49 mm, with a conditional 95% five-year block interval of -13.16 to -5.08. The preceding-autumn slope alone was uncertain. All planned contrast checks stayed negative, including first differences, both half-periods, an earlier release, and omission of 2014-2015. A secondary Michigan-Huron test gave a contrast of -13.71 mm (-19.82 to -9.68), using the mean of two individual ice maxima as a winter proxy. The result shows a timing difference in model-based estimates. It is consistent with a delayed cold-lake signal, but common weather and model inputs can also produce it. It does not establish an ice effect or forecast gain. Source dependence, 50 winters, ice-summary choice and omitted joint posterior uncertainty limit inference. Original code, aggregate results, resampling draws, a figure, source hashes and a separate numerical audit support inspection. Source records remain external.

Great Lakes · Lake Superior · evaporation · winter ice · seasonal timing · thermal memory · paired block bootstrap · water balance