Winter precipitation gave no clear spring runoff forecast gain in two Great Lakes reconstructions for 1990-2022

We tested whether winter precipitation adds information about March-May runoff beyond winter runoff and a linear time trend. We used L2SWBM V3.0 monthly medians for Superior and Michigan-Huron. Expanding fits used earlier years only, with 33 test years in 1990-2022. Adding December-February over-lake precipitation reduced test mean squared error by 27.0% in Superior and 18.1% in Michigan-Huron. Conditional 97.5% three-year block intervals were -10.8% to 62.1% and -28.7% to 46.1%. Both include zero. Fixed first-period fits gave reductions of 7.4% and -13.3%. Thus, the primary test does not establish a reliable gain. These are retrospective reconstruction tests, not real-time forecasts. Over-lake precipitation is an imperfect proxy for land precipitation. Short records, model constraints, shared errors and uncertainty conditional on fitted forecasts limit inference. We do not identify snow or soil storage as a cause. The package contains code, exact source hashes and URLs, aggregate results, bootstrap draws, a figure and an independent numerical audit. Source tables and record-level outputs remain outside the signed package.

Great Lakes · runoff · seasonal forecast · winter precipitation · forecast validation · water storage