September ice extent improved October forecasts in 26 Arctic autumns, but later forecast gains were uncertain

We tested whether September ice extent improved forecasts of October, November, and December extent beyond a time trend. We used Sea Ice Index Version 4 and 46 common years from 1979 to 2025. The test excluded 1987 because December had only two daily observations. After linear trend removal, endpoint slopes fell from 0.824 in October to 0.182 in December. For 26 later-year forecasts from 2000 to 2025, adding September extent reduced October mean squared error by 67.4 percent with an expanding training window and 66.2 percent with a 20-year window. Three-year block intervals for the paired loss benefit excluded zero for October. The November and December point gains were positive, but their intervals included zero. A strong negative December gain correlation of -0.893 did not imply a separate physical recovery feedback, because gain subtracts September extent. These retrospective tests use final data and cannot identify heat, thickness, or transport mechanisms. The package extends a prior campaign March endpoint test with a reproducible autumn forecast benchmark.

Arctic · sea ice · autumn forecast · seasonal memory · mathematical coupling