Suppression of Arctic sea ice growth by winter clouds and snowfall

7 The ongoing Arctic warming has been pronounced in winter and has been associated with 8 an increase in downward longwave radiation. While previous studies have demonstrated 9 that poleward moisture flux into the Arctic strengthens downward longwave radiation, 10 less attention has been given to the impact of the accompanying increase in snowfall. Here, 11 utilizing state-of-the art sea ice models, we show that typical winter snowfall anomalies of 12 1.0 cm, accompanied by positive downward longwave radiation anomalies of ~5 W m -2 13 can decrease sea ice thickness by around 5 cm in the following spring over the Eurasian 14 Seas. This basin-wide ice thinning is followed by a shrinking of summer ice extent in 15 extreme cases. In the winter of 2016–17, anomalously strong warm/moist air transport 16 combined with ~2.5 cm increase in snowfall decreased spring ice thickness by ~10 cm and 17 decreased the following summer sea ice extent by 5–30%. Projected future reductions in 18 the thickness of Arctic sea ice and snow will amplify the impact of anomalous winter 19 snowfall events on winter sea ice growth and seasonal sea ice thickness.


Introduction 22
The multi-decadal retreat in Arctic sea ice has been superposed upon pronounced interannual 23 variability, which has motivated efforts to understand year-to-variability in the winter sea ice 24 growth season 1-3 . For example, previous studies have shown that the initial sea ice thickness 25 in late autumn-early winter preconditions the heat conductivity of the sea ice, and thereby 26 strongly influences sea ice growth through the winter 2,3 . Autumn-winter variations in poleward 27 moisture also modulate winter sea ice growth via changes in downward longwave radiation 4,5 , 28 and are predicted to become increasingly influential during the coming decades 3 . This study 29 considers an additional direct effect of interannual variations in moisture transport into the 30 Arctic on sea ice growth: increased winter snowfall. Over the Eurasian Seas, such as the Laptev, 31 East Siberian, and Chukchi Seas, snowfall makes up more than 60% of the annual precipitation 6 . 32 Because the thermal conductivity of snow is about 7 times lower than ice, it may be expected 33 to insulate the sea ice in these sectors from the atmosphere, and thus suppress winter ice 34 growth 7,8 . This insulation should be particularly effective in the Eurasian Seas, where relatively 35 thin first-year ice is becoming increasingly dominant 9 . This raises the possibility that a small 36 increase in snowfall associated with atmospheric moisture flux convergence may suppress sea 37 ice growth throughout the winter. While previous studies have pointed out the close linkage 38 between poleward moisture flux into the Arctic and increased downward longwave 39 radiation 4,5,10 , relatively little attention has been given to the accompanying increase in snowfall 40 and its potential suppression of sea ice growth. 41 In this study, the impact of winter snowfall on the wintertime seasonal cycle of sea ice thickness However, the mean snow depth and the amplitude of the interannual variability simulated by 67 PIOMAS and NESOSIM are about 30% larger than those of CICE6. Reconstruction of snow 68 depth over Arctic sea ice is a challenging issue because in-situ observations of snow on sea ice 69 have been sparse and the retrieval of snow depth from satellite measurements is still in the early 70 stage 17 . Moreover, estimating the snow depth over the eastern Arctic is more difficult than other 71 regions 18 , probably because the eastern Arctic is mostly covered by first-year sea ice and snow 72 depth is generally thinner than other regions 9 . 75 To what extent is the wintertime sea ice growth controlled by snow? Snow is a relatively poor 76 conductor of heat, compared with sea ice, because a substantial fraction of its volume is trapped 77 air. In winter, the insulating effect of snow decreases the conductive heat flux ↑ , through the 78 sea ice and snow, and thus decreases the rate at which seawater freezes to the base of the sea 79 ice. 80 The insulating effect of snow may be understood with the aid of a one-dimensional conceptual 81 model of the sea ice/snow heat budget. Assuming that the sea ice is composed by a single 82 homogeneous layer of ice for simplicity, and that the sea ice temperature instantaneously 83 equilibrates to the heat fluxes at its base and to the atmospheric conditions above the ice and 84 snow, the heat balance at the ice-atmosphere interface can be written as

Snow depth and ice growth rate in winter
(1) 86 Here, ↑ and ↓ denote upward and downward longwave radiative fluxes, respectively, 87 and ↑ and ↑ denote upward sensible and latent heat fluxes, respectively. We have 88 5 neglected net shortwave radiation ↓ + ↑ which is much weaker than other heat fluxes 89 in winter. Increased snowfall suppresses the ice growth by reducing the upward conductive 90 heat flux ( ↑ ), leading to a lower snow surface temperature and decreased sensible heat flux 91 ( ↑ ) and upward longwave radiation ( ↑ ). March is strongly correlated with snow depth in winter, when averaged over the entire Arctic 100 (Fig. 2a), consistent with our expectation that the decreased conductivity of the sea ice/snow 101 layer should suppress ice growth. However, the insulating effect of snow on sea ice is 102 geographically dependent. Over the Atlantic sector of the Arctic, the accumulated winter 103 snowfall often exceeds 25 cm (Fig. 3a) and snow-ice formation is generally larger than 50 cm 104 (Fig. 3b). Anomalously large winter snowfall over the Atlantic Seas tends to produce 105 anomalously thick ice, rather than anomalously thin ice 19,20 . In this study, we focus on the snow 106 effect on sea ice in the Eurasian Seas, where the first-year sea ice is becoming increasingly 107 dominant 9 and the snow-ice formation is relatively small. Over the Eurasian Seas, the 108 correlation coefficient between the areally-averaged detrended snow depth and the detrended 109 ice growth rate is -0.80 (Fig. 2b) cover is composed of first-year ice 3 and snow-ice formation is small (Fig. 3b).

135
The regression map exhibits a basin-wide increase in snow depth (Fig. 2c) and a basin-wide 136 decrease of the ice growth rate (Fig. 2d), corroborating our earlier finding of a link between 137 snow depth and ice growth over the Eurasian sector of the Arctic. On sub-basin scales, however, 138 the spatial pattern of the reduced ice growth (Fig. 2d) does not visibly correspond to that of the 139 snow depth (Fig. 2c ice concentration is less than 15%, snow depth is generally controlled by snowfall snowfall is associated with cyclone activity 24 and many of these cyclones pass through the 181 Chukchi Sea and the Barents-Kara Seas. The snowfall in MERRA2 is about 20-25% larger 182 than in the other reanalysis products (Fig. 4a) and using MERRA2 to force sea ice models is 183 known to simulate thicker snow depth over sea ice 18 . Recent studies found that reanalysis 184 products capture the satellite-observed and in-situ observed interannual variability in Arctic 185 snowfall reasonably well 25,26 .

186
Because of the snowfall accumulation throughout the winter, the snow depth anomalies peak 187 in late winter and spring, from March to May (Fig. 4d). This regression map of ice thickness 188 anomalies exhibits a basin-wide ice thinning throughout the winter and spring (Figs. 4f -h).

189
The ice thickness anomaly is largest in the late winter and spring (Fig. 4g) and persists into the ice thickness by ~10 cm over the Eurasian Seas (Fig. 7b). This seasonally persistent ice thinning 261 was followed by a notable reduction of ice cover in August-September (Fig. 7c), which is 262 approximately 30% reduction in sea ice extent.

263
Similarly, our CICE6 simulations also indicate that anomalously small snowfall and weak 264 downward longwave radiation during the winter of 1998-99 (Figs. 8a and 8d) accelerated the 265 winter sea ice growth and increased the spring and summer sea ice thickness up to 17 cm (Fig. 266 8b). This was followed by a large increase in summer sea ice concentration -more than 15% 267 over wide areas of the Arctic Ocean in August-September (Fig. 8c) JRA55 atmospheric boundary conditions, which is one of the standard component sets.

281
The interannual variability of winter snowfall over the Eurasian Seas in JRA55 is very similar 282 to that of ERA5 (Fig. 4a), except that the wintertime mean snowfall is about 10% smaller than 283 that of ERA5. While using a full ocean model has merit in realistically simulating the 284 interaction between sea ice growth/melting and the ocean mixed layer, it is difficult to control 285 the SSTs over the marginal seas of the Arctic, which strongly influence sea ice extent 32 .

288
Using CESM2 with JRA55 atmospheric boundary conditions, we performed the same summer sea ice thickness by ~10 cm (Fig. 7e). These sea ice thickness anomalies are similar to 294 14 those simulated in our CICE6-slab ocean model (compare Figs. 7b and 7e). This seasonally 295 persistent ice thinning is followed by a reduction of ice cover in August and September (Fig. 296 7f), which is approximately 5% reduction in sea ice extent. Note that direct comparisons of  Fig. 1a and Supplementary Fig. 4).

302
Consistent with our CICE6-slab ocean model simulations, the anomalously small snowfall and 303 weak downward longwave radiation during the winter of 1998-99 substantially increased sea 304 ice thickness throughout the seasons (Fig. 8e). The sea ice thickening was followed by an In summary, our model simulations demonstrate that the Arctic winter snowfall serves as one 312 of the key controls of winter sea ice growth. A key finding of this study is that the effect of 313 winter snowfall on winter and spring sea ice thickness is comparable to that of downward 314 longwave radiation combined with surface air warming/moistening. The combined impacts on 315 sea ice are not limited to winter, but rather persist through the ensuing spring and summer. In 316 extreme cases, the basin-wide ice thinning is followed by a shrinking of summer ice extent.

317
This indicates that winter snowfall anomalies, along with accompanying anomalies in 318 downward longwave radiation and surface air warming/moistening, may serve as a useful 319 predictor of the following summer sea ice extent.

320
Arctic sea ice is projected to become thinner with future climate change, and snow depth is 321 likely to decline continuously 33,34 . As the idealized 1D model demonstrates, snow can be more 322 effective in suppressing the winter sea ice growth when the snow depth and sea ice thickness 323 are relatively thin (Fig. 9), suggesting that snowfall will more strongly influence the seasonal 324 sea ice growth and thickness in coming decades. This effect will be compounded by the and downward longwave radiation, which is also accompanied by surface air warming and 425 moistening, we followed a similar procedure as in our CICE6-slab ocean model experiments. 426 We configured CESM2 with historical atmospheric forcing, but replaced the downward 427 longwave radiation, surface air temperature, surface specific humidity, and snowfall with their respectively. is wind speed at 10 m and is the specific air humidity at 2 m. is the 445 saturation specific humidity. is Stefan-Boltzmann constant and is turbulent transfer 446 coefficient over sea ice.

447
Following ref. 48 , which assumes a linear temperature gradient through snow and sea ice, the 448 conductive heat flux ( ) ↑ is: where is the freezing temperature of sea water, ℎ and ℎ are thicknesses of ice and snow, 451 respectively, and and are thermal conductivities of ice and snow, respectively. Note 452 that snow is an effective thermal insulator: is about seven times smaller than . In winter, 453 sea ice grows by conducting heat upward from the bottom of ice to the surface. Assuming that 454 the ocean surface is at the freezing temperature, the freezing rate at the bottom of ice is 455 simplified as: where is density of ice and is latent heat of fusion. Here, we calculate and ↑ by 458 solving equations (2) and (3) with prescribed thicknesses of ice and snow, ℎ and ℎ . Then, 459 the ice growth rate Φ ℎ can be estimated from equation (4).

460
In this study, we estimated typical values of these parameters from ERA5, specifically  Nonnewton. Fluid Mech. 138, 22-32 (2006    The year-to-year variations of (a) late summer (Aug-Sep) Arctic sea ice extent simulated by 701