August-limited retreat dates gave uneven gains in Arctic ice recovery forecasts for 2006–2025

An earlier campaign study found regional ice retreat forecast gains. We tested a retreat rule that uses no observations after August 31 and a five-day autumn recovery rule. We used Sea Ice Index Version 4 for seven Arctic regions in 1989–2025. Models used only earlier years and were tested in 2006–2025. At the central 50% winter-reference threshold, five of six scored regions had lower mean absolute error when retreat date was added to a time trend. Gains were 0.43–2.50 days. East Siberian error rose by 0.31 days. Gains changed with threshold and eligible years. Kara gains stayed positive across three thresholds on 17 common test years. These exploratory results extend a forecast benchmark. They do not establish a causal effect of ocean heat storage. Short samples, excluded event years, and final data releases limit the result. Code, daily subsets, exclusions, forecasts, and checks are included.

Arctic · sea ice · forecast verification · retreat · ice advance