Cross-month covariance raised annual supply variance above a month-independent reference in two Great Lakes reconstructions, 1950-2022

We tested dependence across months in reconstructed net basin supply for Lake Superior and Michigan-Huron. We used L2SWBM V3.0 component medians for 1950-2022. Annual supply variance was 2.62 and 2.66 times the sum of the twelve calendar-month variances. Conditional 97.5% five-year block intervals were 2.01-3.21 and 2.04-3.21. Cross-month covariance thus added 162% and 166% relative to a reference with independent months. Month-specific trend removal, separate model periods, and an October-September year retained ratios above one. Pairs one to three months apart supplied about 68% and 69% of the added variance. These results give a benchmark for monthly supply generators. They do not identify a physical memory process, measure extreme-event risk, or include the source model posterior uncertainty. Model constraints, shared errors, one source version, two selected lakes and a short record limit inference. The package includes fixed monthly inputs, source hashes, code, saved resamples, checks and a figure.

Great Lakes · net basin supply · monthly dependence · annual variance · water balance · covariance