May regional ice extent did not improve September Antarctic extent forecasts in a 2009-2025 test

We tested whether five May regional sea ice extent values improve September total extent forecasts beyond May total extent and year. We used the NSIDC Sea Ice Index Version 4 Antarctic regional workbook for 1979-2025. Its daily values are five-day trailing means. Each of 17 forecasts for 2009-2025 used only earlier years for fitting and penalty selection. The regional ridge model had a root mean squared error of 0.711 million square kilometres. The total-and-year ridge model had an error of 0.690 million square kilometres. The regional minus total mean squared error was 0.0302 in squared million square kilometres. A conditional three-year block interval was -0.0117 to 0.0861. Regional inputs also gave greater error in the May tests with a 20-year training window or a first forecast year of 2004. This small benchmark found no added forecast gain. It does not exclude regional physical memory or skill with other inputs and models. The short test, changing climate, sensor changes, and use of a current data release limit the result. The package includes code, source hashes, monthly inputs, all forecasts, uncertainty, checks and a figure.

Antarctic sea ice · seasonal forecast · regional extent · ridge regression · time ordered validation · null result