Image time series analysis of Arctic sea ice

Describes the application of per-pixel time series modelling to a sequence of 108 monthly images of Arctic sea ice concentrations. The authors outline a semi-automated model fitting procedure which they used to fit autoregressive-moving average (ARMA) models to the time series at each pixel in a hypertemporal image stack. After deseasonalizing each series, they found that most of the variability in sea ice concentrations in the Arctic follow a first-order autoregressive (i.e., AR(1)) process. The strongest autoregressive autocorrelations are found in the central Arctic Basin and generally become weaker closer to the southern limit of the seasonal ice extent. In the marginal sea ice zone the mean ice concentration for a given month is shown to become less dependent on past month´s values