Two-slope tests find no clear low-to-high summer difference in 46 sea ice seasons per hemisphere

We test whether low and high summer sea ice extent anomalies have different relations to later winter extent. We use NSIDC Sea Ice Index Version 4 daily records for summer years 1979-2024. Each hemisphere has 46 seasonal pairs. After separate linear trends are removed, a continuous two-slope model gives a high-minus-low slope difference of 0.392 in the Arctic and -0.666 in the Antarctic. Conditional 97.5% three-year block intervals are -0.148 to 0.799 and -1.983 to 1.099. Both include zero. In 26 rolling forecasts per hemisphere, the two-slope model changes winter RMSE from 0.321 to 0.331 million square kilometres in the Arctic and from 0.615 to 0.572 in the Antarctic. Both intervals for mean squared error reduction include zero. An earlier campaign study already found weak linear Arctic endpoint memory and the shared-value problem in winter gain. We reproduce that result and add a test for a low-to-high difference. The null result does not prove equal slopes or absence of physical recovery. Short records, fixed trend removal, sensor changes, simple models, and extent-only data limit inference.

sea ice · seasonal recovery · nonlinear relation · Arctic · Antarctic · null result · replication