Abstract
In Arctic winter, the boundary layer occupies two preferred radiative states, a cold, clear state and a heat-trapping, cloudy state, that strongly regulate surface energy loss over sea ice. Yet, reanalyses and climate models are often biased toward a single intermediate state. Using a stochastic differential equation framework applied to 27 years of high-resolution observations, we show that the observed bistability is not sustained by multiple deterministic equilibria: The mean restoring tendency is effectively single-well, while background meteorology alone cannot explain rapid transitions. Instead, the two regimes are sustained by sharply localized, state-dependent noise in the transition range. This noise structure emerges when subhourly fluctuations in liquid water path are radiatively amplified by the nonlinear saturation of longwave emissivity. Our results provide a dynamical explanation for this model bias and identify unresolved fast cloud variability as a key source of uncertainty in Arctic winter climate projections.