Summer input gave no clear added cold-season evaporation forecast skill in two Great Lakes at 1985-2021 origins

We tested whether May-September precipitation plus runoff improves prediction of October-March evaporation in Superior and Michigan-Huron. We used L2SWBM V3.0 monthly medians and 37 expanding-fit forecasts at 1985-2021 summer origins. Adding summer input to year and previous cold-season evaporation reduced mean squared error by 6.3% and 12.4%. Conditional 97.5% five-year block intervals were -10.7% to 21.9% and -7.4% to 29.8%. Thus, neither lake gave a clear forecast gain. Full-record adjusted slopes were negative: -13.1 and -14.9 mm evaporation per 100 mm summer input. Adding summer evaporation to the baseline removed the point gains from summer input. All planned forecast checks had intervals that included zero. These results show a seasonal association in reconstructed fluxes, but no resolved added forecast skill. Short records, model balance constraints, final-version inputs and omitted source posterior uncertainty limit inference. The test does not identify a physical cause or operational forecast skill. Code, monthly inputs, hashes, all forecasts, saved resamples, an independent audit and a figure are included.

Great Lakes · evaporation · seasonal prediction · water balance · forecast baseline · null result