A fixed June profile model showed no clear error reduction against recent August oxygen means at seven Wisconsin lakes in 2006–2025

We tested whether early lake profiles improve a forecast of deep-water oxygen in August. We used open NTL-LTER records from seven Wisconsin lakes. Forecasts for 2006–2025 used only earlier measurements and years. We compared four issue dates on the same 90 lake-years. The planned 1 June ridge model used oxygen and a temperature difference. Its equal-lake mean squared error was 1.212 (mg/L)². A mean of the last five eligible August observations had an error of 1.234 (mg/L)². The reduction was 1.8%. The nominal 95% error-ratio interval was 0.550–1.343 under shared three-year block draws. Thus, the test did not establish added skill over this recent-mean rule. The June model had 17.3% less error than a mean of all eligible past August values. A deeper band reversed the gain over the recent mean. Training coverage also changed the seasonal error curve. One large error occurred at Crystal Lake in 2012, during an artificial mixing experiment. Omitting that lake gave a 19.1% error reduction. This result applies to the fixed model and selected records. It does not show that spring oxygen has no ecological effect. The package has source data, provenance, code, all selections, fitted forecasts, results, figures, and independent checks. A full repeat matched 52 output files.

lake oxygen · ecological forecasting · seasonal forecasts · reference models · Wisconsin · NTL-LTER · reproducibility