A change in response slope at the seasonal mean gave no clear Michigan-Huron forecast gain at 1970-2023 origins

We tested whether different response slopes above and below the past seasonal mean improve Michigan-Huron lake level forecasts. We used coordinated v2025 monthly levels from 1918-2025 and 648 twelve-month forecasts at 1970-2023 origins. Fits used only training targets before the origin year. Adding absolute level anomaly to a seasonal local-history model increased mean squared error by 2.53%. RMSE rose from 22.99 to 23.28 cm. The conditional 95% five-year block interval for MSE reduction was -6.81% to +1.62%. Four secondary comparisons also had negative point gains and intervals that included zero. The result gives no clear gain for this simple change in response slope. It does not establish equal physical recovery above and below normal. One lake, revised data, one response form and conditional uncertainty limit inference. The package includes numerical inputs, source credit and hashes, code, every prediction, all score draws, a separate audit and a figure.

Great Lakes · Michigan-Huron · nonlinear response · chronological forecast · null result · lake level