Sea ice algorithm matches seasonal models using historical records

Researchers at NYU Abu Dhabi report that an algorithm using only historical Arctic sea ice records can forecast ice extent up to nine months ahead. In their Scientific Reports study, its September forecast error was comparable to that of 34 seasonal forecasting models. The Random Analog Predictor finds past conditions resembling current ones and uses subsequent changes to generate multiple possible forecasts. Their spread provides an estimate of uncertainty. The researchers propose it as a benchmark for physics-based and AI forecasting systems. They position it as a way to test the predictive value of more complex models, rather than replace them.