On the discrepancy between observed and CMIP5 multi-model simulated Barents Sea winter sea ice decline

Abstract

This study aims to understand the relative roles of external forcing versus internal climate variability in causing the observed Barents Sea winter sea ice extent (SIE) decline since 1979. We identify major discrepancies in the spatial patterns of winter Northern Hemisphere sea ice concentration trends over the satellite period between observations and CMIP5 multi-model mean externally forced response. The CMIP5 externally forced decline in Barents Sea winter SIE is much weaker than that observed. Across CMIP5 ensemble members, March Barents Sea SIE trends have little correlation with global mean surface air temperature trends, but are strongly anti-correlated with trends in Atlantic heat transport across the Barents Sea Opening (BSO). Further comparison with control simulations from coupled climate models suggests that enhanced Atlantic heat transport across the BSO associated with regional internal variability may have played a leading role in the observed decline in winter Barents Sea SIE since 1979.

Figure 1. Observed and CMIP5 multi-model-simulated time series.

(a) March NH SIE anomaly (b) March Barents Sea SIE anomaly (c) Annual global mean SAT anomaly. Blue solid line: CMIP5 multi-model ensemble mean; light blue shading: spread (±1 s.d.) across multi-model ensemble members; blue dashed line: linear trend. Observations (SIE from NSIDC and SAT from the average of three reanalysis datasets—NCEP/NCAR, ERA-Interim and MERRA) and their linear trends are shown in red solid lines with circle markers and red dashed lines, respectively.

Figure 2. Spatial pattern of observed and CMIP5 multi-model-simulated March SIC trend over 1979–2015.

(a,c) Observations (NSIDC), (b,d) CMIP5 multi-model ensemble mean. The lower panels zoom in over the Barents Sea. Black lines mark the climatological March sea ice edge, and the red box outlines the Barents Sea. SIC from individual CMIP5 models are regridded onto a common polar-stereographic grid so that a multi-model ensemble mean is applicable. Grid points with a mean SIC less than 1% are masked white. In a,c, grid points poleward of 85°N are also masked due to missing data. The unhatched regions in all panels have statistically significant trends at 95% confidence level.

Figure 3. Scatter plots of trends over 1979–2015 from CMIP5 multi-model ensemble members.

(a) March Barents Sea SIE trend versus annual global mean SAT trend (b) March Barents Sea SIE trend versus trend of annual mean Atlantic heat transport across the BSO (HT_BSO). In parentheses (model list) is the number of ensemble members for each model. Solid red circles: CMIP5 multi-model ensemble mean trends. Thick red crosses: spread (±1 s.d.) across multi-model ensemble members. Grey solid circle and thick grey cross in b: adjusted multi-model mean trends and spread (see text). The solid black circle in a denotes the observed trends. The solid black line in b marks the observed March Barents Sea SIE trend. Red dashed lines: least-square linear fits.

Figure 4. Time series of CMIP5 multi-model-simulated annual mean HT_BSO anomaly.

Blue solid line: CMIP5 multi-model ensemble mean; light blue shading: spread (±1 s.d.) across multi-model ensemble members; blue dashed line: linear trend. Red dashed line: estimated annual mean HT_BSO trend (∼35 TW per 37 years) implied from the observed SIE trend in Fig. 3b.

Figure 5. Regression of 37-year March SIC trends on 37-year HT_BSO trends in three long control simulations.

(a) GFDL CM2.1 (b) GFDL CM3 (c) NCAR CESM. Red boxes outline the Barents Sea region, and black lines denote the climatological March sea ice edge in each model. The regression maps are scaled to correspond to a positive trend of 30 TW per 37 years in HT_BSO. White gaps in the upper right corners of a,b are due to the polar projection of SIC simulated on tri-polar grids in GFDL models.