Recent water balance components gave no clear added three-month supply prediction gain at two Great Lakes in 1985-2021

We tested whether recent precipitation and runoff improve three-month component net basin supply predictions beyond recent total supply. We used fixed L2SWBM V3.0 monthly medians for Superior and Michigan-Huron and 444 chronological predictions per lake in 1985-2021. Adding two prior-three-month component means raised the equal-lake mean squared error by 0.98%. The conditional 95% five-year block interval for error reduction was -2.80% to +0.63%. All planned model settings gave negative mean point gains. The test gives no clear added gain for these component terms. It is a benchmark within a revised reconstruction, not an operational forecast test or evidence that physical memory is absent. Source balance constraints can transmit later information into predictors. Short dependent records, one source version, and omitted source posterior uncertainty limit the result. The package includes monthly inputs, exact source hashes, all predictions, code, uncertainty draws, a figure and a separate numerical audit.

Great Lakes · net basin supply · water balance · component history · chronological prediction · null result