Antarctic surface mass balance: natural variability, noise and detecting new trends

The emergence of new, statistically robust trends in Antarctic surface mass balance (SMB) requires an understanding of the underlying SMB variability (noise). We show that simple white or AR[1] noise models do not adequately represent the variability of SMB in both the RACMO2.3p2 SMB model output (1979-2017) and composite ice core records (1800-2010), under-estimating low-frequency variability. By testing a range of noise models, we find that a Generalized Gauss Markov (GGM) model better approximates the noise around a linear trend. The general preference for GGM noise applies over spatial scales from the total ice sheet down to individual drainage basins. Over the longest timescales considered, trend uncertainties are 1.3-2.3 times larger using a GGM model compared to using an AR1 model at the ice sheet scale. Overall, our results suggest that larger trends or longer periods are required before new SMB trends can be robustly separated from background noise.