July ice extent reduced pooled autumn mean squared forecast error by 43% across six Arctic seas in 2005-2024

We tested whether July ice extent improves forecasts of mean October-November extent beyond a time trend. We used Sea Ice Index Version 4 for six Arctic seas in 1989-2024. Each forecast used only earlier years to fit its coefficients. In 20 test years from 2005 to 2024, adding July extent reduced pooled normalized mean squared error by 42.6%. A conditional 95% three-year block interval was 26.9% to 53.9%. All six regional point gains were positive. Approximate six-region family intervals were above zero for Laptev, Kara and Barents. Pooled point gains stayed near 40% to 44% with later test starts, rolling fits and quadratic trends. East Siberian gains changed sign in later test periods. Separate-month checks showed that the Laptev gain was concentrated in October. The result supplies a fixed-date, threshold-free forecast benchmark. It does not identify ocean heat storage as the cause or establish operational skill. Only 20 test years, final data versions, selected regions and uncertainty conditional on fitted forecasts limit inference. Code, monthly inputs, exact source hashes, all forecasts, bootstrap draws and an independent numerical audit are included.

Arctic · sea ice · July · autumn forecast · regional predictability · Sea Ice Index