Revisiting GRACE Antarctic ice mass trends and accelerations considering autocorrelation

Abstract

Previous GRACE-derived ice mass trends and accelerations have almost entirely been based on an assumption that the residuals to a regression model (including also semi-annual, annual and tidal aliasing terms) are not serially correlated. We consider ice mass change time series for Antarctica and show that significant autocorrelation is, in fact, present. We examine power-law and autoregressive models and compare them to those that assume white (uncorrelated) noise. The data do not let us separate autoregressive and power-law models but both indicate that white noise uncertainties need to be scaled up by a factor of up to 4 for accelerations and 6 for linear rates, depending on length of observations and location. For the whole of Antarctica, East Antarctica and West Antarctica the scale factors are 1.5, 1.5 and 2.2 respectively for the trends and, for the accelerations, 1.5, 1.5 and 2.1. Substantially lower scale-factors are required for offshore time series, suggesting much of the time-correlation is related to continental mass changes. Despite the higher uncertainties, we find significant (2-sigma) accelerations over much of West Antarctica (overall increasing mass loss) and Dronning Maud Land (increasing mass gain) as well as a marginally significant acceleration for the ice sheet as a whole (increasing mass loss).