Ice wedges as winter temperature proxy: Principles, limitations and noise in the δ18O records (an example from high Arctic Canada)

Abstract

Over the past few decades, the δ18O records of ice wedges have been used to estimate late Quaternary winter air temperatures. This study first reviews the development of ice wedges and the potential isotopic fractionation that takes place from snow deposition to formation of ice veinlets. Then, based on the high resolution analysis of ice wedges from the Eureka Sound Lowland (Ellesmere and Axel Heiberg Islands), we evaluate the effect of sampling depth and edge effects on the δ18O records and compare a composite δ18O time-series of ice wedges to the Holocene 25-yr annual and 20-yr winter δ18O records of the nearby Agassiz Ice Cap. Radiocarbon ages of dissolved organic carbon in ice wedges showed that cracking occurs mostly near the center of the wedge, but age reversals were observed. Covariance analysis showed that two δ18O profiles collected at the same depth had high noise (c. 70–80 % uncorrelated parts). Additionally, ice wedge mean δ18O records showed a decreasing trend with depth, likely due Rayleigh fractionation during freezing of residual water in the crack. Finally, the composite δ18O time-series of ice wedges and δ18O records of the Agassiz Ice Cap showed a similar Holocene cooling trend, however the wedges δ18O records had a higher degree of variability throughout the Holocene (∼4 ‰ versus ∼2 ‰ in the Agassiz records). The higher variability in the wedge is attributed to the timing of meltwater infiltrating the wedge crack over the growth period, which from the onset to termination of snowmelt, is in the order of 3–6‰. The study highlights that, if to be used as a centennial-millennial scale proxy, ice wedges should be sampled near the surface where δ18O records would be less affected by in situ freezing and plugging, and veinlets should be dated directly and smoothed to remove the local random variance.