An Assessment of the Temporal Variability in the Annual Cycle of Daily Antarctic Sea Ice in the NCAR Community Earth System Model, Version 2: A Comparison of the Historical Runs With Observations

16 Understanding the variability of Antarctic sea ice is an ongoing challenge given the lim- 17 itations of observed data. Coupled climate model simulations present the opportunity 18 to examine this variability in Antarctic sea ice. Here, the daily sea ice extent simulated 19 by the newly-released National Center for Atmospheric Research Community Earth Sys- 20 tem Model Version 2 (CESM2) for the historical period (1979–2014), is compared to the 21 satellite-observed daily sea ice extent for the same period. The comparisons are made 22 using a newly-developed suite of statistical metrics that estimates the variability of the 23 sea ice extent on timescales ranging from the long-term decadal to the short term, intra- 24 day scales. Assessed are the annual cycle, trend, day-to-day change, and the volatility, 25 a new statistic that estimates the variability at the daily scale. Results show that the 26 trend in observed daily sea ice is dominated by sub-decadal variability with a weak pos- 27 itive linear trend superimposed. The CESM2 simulates comparable sub-decadal variabil- 28 ity but with a strong negative linear trend superimposed. The CESM2’s annual cycle 29 is similar in amplitude to the observed, key diﬀerences being the timing of ice advance 30 and retreat. The sea ice begins its advance later, reaches its maximum later and begins 31 retreat later in the CESM2. This is conﬁrmed by the day-to-day change. Apparent in 32 all of the sea ice regions, this behavior suggests the inﬂuence of the semi-annual oscil- 33 lation of the circumpolar trough. The volatility, which is associated with smaller scale 34 dynamics such as storms, is smaller in the CESM2 than observed. 35 (NCAR) Com- 41 munity Earth System Model Version 2 (CESM2) with satellite-observed data for the years 42 1979–2014. We examine the annual cycle, trend, day-to-day change in sea ice and the 43 volatility, a new statistic that estimates the variability at the daily scale. We show that 44 the CESM2 is able to simulate sub-decadal variability comparable to that apparent in 45 the observed sea ice but not the weak, positive, linear trend. The CESM2 also simulates 46 an annual cycle of similar amplitude to that observed but the ice starts growing later 47 and retreating later in the CESM2 than is observed. This diﬀerence in timing in the an- 48 nual cycle occurs in the sea ice all around Antarctica, which suggests that it might be 49 because of a circum-Antarctic atmospheric circulation feature called the circumpolar trough.

using a newly-developed suite of statistical metrics that estimates the variability of the 23 sea ice extent on timescales ranging from the long-term decadal to the short term, intra-24 day scales. Assessed are the annual cycle, trend, day-to-day change, and the volatility, 25 a new statistic that estimates the variability at the daily scale. Results show that the 26 trend in observed daily sea ice is dominated by sub-decadal variability with a weak pos-27 itive linear trend superimposed. The CESM2 simulates comparable sub-decadal variabil-28 ity but with a strong negative linear trend superimposed. The CESM2's annual cycle 29 is similar in amplitude to the observed, key differences being the timing of ice advance 30 and retreat. The sea ice begins its advance later, reaches its maximum later and begins 31 retreat later in the CESM2. This is confirmed by the day-to-day change. Apparent in 32 all of the sea ice regions, this behavior suggests the influence of the semi-annual oscil-33 lation of the circumpolar trough. The volatility, which is associated with smaller scale 34 dynamics such as storms, is smaller in the CESM2 than observed. as the daily standard deviation which is the intra-day variation in the sea ice extent. The 141 volatility in the observed data is considered to be due largely to factors like the ephemeral 142 dynamic effects of storms at the ice edge and wave-ice interactions. Some, smaller, por-143 tion of it may be due also to instrumentation and algorithm effects.

144
Antarctic sea ice distribution varies regionally, therefore our analysis examines the 145 total SIE as well as the regional SIE variability in order to get a comprehensive sense 146 of the model's performance. The sea ice regions used in this analysis ( Figure 1)  Another is that the observed increase in sea ice extent might be due to natural variabil-171 ity rather than external forcing in the system and therefore, that the climate models do be used for diagnosing the mechanisms that force this nonlinear behavior. 190 We also examine the simulated and observed trends by region. Shown in Figure   191 3 are the observed and ensemble mean simulated trends. The curvilinearity apparent in 192 the observed total SIE (Figure 3a) is also noted regionally as is expected. It is most pro-        began approximately three weeks before the median retreat onset. This points to the ben-270 efit of using daily data, as these differences would not be adequately resolved using monthly 271 means.

272
The amplitude adjusted annual cycles are also examined for each region alongside 273 the total SIE for comparison ( Figure 5). The regional cycles, both simulated and observed, 274 exhibit marked differences in the shape and length of the annual cycles which demon-275 strate why it is important to study Antarctic sea ice variability from a regional perspec-    Table 1 shows this delay in retreat which is also visible in Figure 5 (It is easier 291 to see in the day-to-day changes in Figure 6, the subject of the next section). This de-292 lay is longest in the Ross which begins to retreat approximately one month after the ob-293 served, and shortest in East Antarctica which experiences a delay of only six days. 294 These regional differences in the shape and length of the annual cycle are interest-295 ing to explore, and indicate that there is much to learn about Antarctic sea ice variabil-296 ity at the regional scale. Certainly the fact that each sea region is influenced by differ-   Total  125  266  352  103  282  3  16  King Haakon VII Sea  166  280  349  124  295  362  15  Ross  87  267  5  97  297  18  30  East Antarctica  125  277  323  102  283  364  6  Weddell  121  244  4  102  259  9  15  ABS  168  241  343  118  254 11 13 a Regional observed and simulated Julian day-of-the-year for the date of maximum SIE advance rate, maximum SIE and maximum SIE retreat rate. The last column is the number of days delay in the start of SIE retreat from observed to simulated.
slower than observed. The slower rate of retreat is likely linked to thicker ice that de- mer which is probably why the annual cycle for the Ross is closer in shape to the observed.

308
That the difference in phase is consistent in all of the regions around the continent 309 suggests that it is due to a large-scale rather than regional mechanism. A potential agent   This describes a well-known characteristic of the Antarctic sea ice cycle -a relatively 324 slow growth to maximum followed by a rapid retreat. This daily analysis, seen in all of 325 the regions as well as the total SIE, shows that the rate of ice advance is not monotonic, 326 but the rate of retreat is monotonic both when it is increasing and decreasing.

327
As might be expected from the analysis above, there are clear regional differences that this sector has the smallest SIE. Table 1 gives the Julian days of maximum advance 332 and maximum retreat and of maximum SIE by region.

333
As shown by the ensemble mean (Figure 6a), the simulations capture the general 334 shape of the day-to-day changes in ice but there are important differences. SIE in the 335 CESM2 starts advancing later, from a lower value, but achieves its peak growth rate ear-336 lier (see Table 1), and has a maximum growth rate that is higher than the observed. Once 337 its peak growth rate is achieved however, it continues to grow more slowly than the ob-338 served for the rest of its advance. It begins retreat later, achieving a maximum rate of 339 retreat that is faster and later in the cycle than is observed (see Table 1), continuing to 340 retreat after the observed has begun to advance. The day-to-day change in Figure 6a  consistent with the annual cycle shown in Figure 4, especially with the phase differences 342 seen in Figure 4b. Additionally, it suggests that the very low minimum SIE achieved by 343 the CESM2 is related to the high, late stage, maximum decay rate.

344
Regionally, the day-to-day changes (Figure 6b-f)   The sea ice volatility, the daily standard deviation in the sea ice simulated by the 362 coupled climate models, has not been evaluated before. However, as shown in Figure 7, 363 it can be responsible for fluctuations at the ice edge on the order of 40,000 -50,000 km 2 364 which, while small compared to the total SIE, becomes significant at the regional scale 365 and when compared to the size of the sea ice grid box. The volatility is considered to 366 be due mainly to the dynamic effects of storms, ocean circulation (eddies) and wave-ice  to-day mean and the volatility we see that the CESM2 simulates an annual cycle with 402 amplitude similar to that observed but with a retreat phase that begins later in the cy-403 cle. We also see that the simulated maximum decay rate is greater, occurs later in the 404 cycle, and is associated with the late peak in volatility. We address now a factor that 405 moderates the timing or phase of the annual cycle, the semi-annual oscillation (SAO).

406
Although it has not been fully quantified, a number of studies suggest that the timing   is that the rate of sea ice advance is not monotonic but the rate of sea ice retreat is mono-482 tonic when it is increasing and when it is decreasing ( Figure 6). This knowledge is po-483 tentially useful when considering thermodynamic vs dynamic effects on sea ice advance 484 and retreat.

485
A potential contributor to the retreat phase difference between the observed an- However, the CESM2 does simulate the peak volatility associated with the very rapid 504 rate of decay late in the ice cycle. As mid-winter sea ice variability is associated with 505 the smaller scale dynamics such as storms (e.g., Stammerjohn et al., 2003), ocean ed-506 dies and wave-ice interaction at the ice edge, it may be that the model is not simulat-507 ing these processes well, something that is common across the CMIP models. We note 508 also that the observed sea ice grid size at 25km x 25km is much smaller than that of the 509 CESM2's (1 degree) thus might be expected to exhibit more daily volatility than the CESM 510 which is a 1 degree model.

511
Finally, the focus of this analysis has been to determine the ability of the CESM2 512 to simulate the key components of the variability of Antarctic sea ice and to suggest what 513 might be the proximate cause of the differences that are seen. However, what has be-514 come even clearer in the process is that in-depth analysis of Antarctic sea ice variabil-515 ity requires a regional (or by sea ice sector) approach. Important differences in variabil-516 ity that are apparent by sector are muted or damped, when only the total SIE is con-  The CESM2 model output used in this study is available at the NCAR Digital As- area, sea ice trends, and winds in CMIP5 simulations. Journal of Geophysical