On the predictability of sea ice

Abstract

We investigate the persistence and predictability of sea ice in numerical models and observations. We first use the 3rd generation Community Climate System Model (CCSM3) General Circulation Model (GCM) to investigate the inherent persistence of sea-ice area and thickness. We find that sea-ice area anomalies have a seasonal decay timescale, exhibiting an initial decorrelation similar to a first order auto-regressive (AR1, or red noise) process. Beyond this initial loss of memory, there is a re-emergence of memory at certain times of the year. There are two distinct modes of re-emergence in the model, one driven by the seasonal coupling of area and thickness anomalies in the summer, the other by the persistence of upper ocean temperature anomalies that originate from ice anomalies in the melt season and then influence ice anomalies in the growth season. Comparison with satellite observations where available indicate these processes appear in nature. We then use the 4th generation CCSM (CCSM4) to investigate the partition of Arctic sea-ice predictability into its initial-value and boundary forced components under present day forcing conditions. We find that initial-value predictability lasts for 1-2 years for sea-ice area, and 3-4 years for sea-ice volume. Forced predictability arises after just 4-5 years for both area and volume. Initial-value predictability of sea-ice area during the summer hinges on the coupling between thickness and area anomalies during that season. We find that the loss of initial-value predictability with time is not uniform --- there is a rapid loss of predictability of sea-ice volume during the late spring early summer associated with snow melt and albedo feedbacks. At the same time, loss of predictability is not uniform across different regions. Given the usefulness of ice thickness as a predictor of summer sea-ice area, we obtain a hindcast of September sea-ice area initializing the GCM on May 1with an estimate of observed sea-ice thickness anomalies. We run the GCM in a slab-ocean model configuration and obtain predictability that is lower than expected from the perfect model fully coupled GCM. We next make use of models submitted to the CMIP5 archive to investigate the spatial and temporal characteristics of ice thickness anomalies, together with the CCSM3, CCSM4 and two forced ice-ocean models, PIOMAS and CCSM4 in ice-ocean mode. We find that there is a wide spread in the characteristics of ice thickness anomalies across models, partially explained by biases in mean thickness. Additionally, forced ice-ocean models show reduced ice-thickness variability. These results have significant implications for the initialization of fully-coupled GCMs from forced GCM output. Finally we investigate the initial-value predictability of Antarctic sea ice in the CCSM3. We find that Antarctic sea-ice anomaly persistence is comparable to that of Arctic sea-ice anomalies. High values of initial-value predictability of sea-ice area can last for up to two years, and tend to advect eastward in time. We also find memory re-emergence that is driven by upper ocean heat anomalies from the melt to the growth season. Unlike the Arctic, we do not find evidence for an ice-thickness driven mechanism of memory re-emergence