The influence of autumnal Eurasian snow cover on climate and its links with Arctic sea ice cover Guillaume Gastineau* 1, Javier García- Serrano 2 and Claude Frankignoul 1 1 Sorbonne Universités, UPMC/CNRS/IRD/MNHN, LOCEAN/IPSL, Paris, France 2 Earth Sciences Dept., Barcelona Supercomputing Center (BSC-CNS), Barcelona, Spain August 24th, 2016 *Corresponding author address: Dr Guillaume Gastineau, Sorbonne Universités, UPMC/CNRS/IRD/MNHN, LOCEAN/IPSL, 4 place Jussieu, 75005 Paris, France. E- mail: guillaume.gastineau@upmc.fr
Supplemental Material: ENSO removal in the Maximum Covariance Analysis and in Regression/correlation analysis First, we calculate the tropical SST variability using the first three empirical orthogonal functions (EOFs) of the monthly mean tropical Indo- Pacific SST. In observations, the SST domain is 15 N- 15 S 100 E- 80 W. The first EOF (53% of variance) has large loadings over the Eastern Pacific and depicts the canonical ENSO variability. The second EOF (13%) shows a dipole of opposite polarity over the central and the eastern Pacific and indicates the central ENSO variability. The third EOF (4.3%) has a large loading over a narrow latitudinal band at the Equator and opposite polarity off Indonesia, so that it resembles the Indian Ocean Dipole. In models, a larger SST domain is used to obtain similar EOFs 20 N- 20 S 30 E- 80 W. The first EOF describes ENSO (between 36% and 53% of variance), while the second and third EOFs (between 9% and 4% of variance) are more variables, with at least one associated to the central ENSO variability. When the atmosphere leads the snow cover, or when the two fields are simultaneous, the atmospheric forcing dominates, so that the fields are not substantially modified by ENSO. In that case, we left the two fields unchanged. When the snow cover leads, we calculated the regressions of the atmospheric and snow variables onto the positive and negative values of the Principal Components (PCs) with a lag of two months, the PCs leading both fields. The multivariate regression is calculated separately for positive and negative values of the Principal Components (PCs), to account for the asymmetry of the teleconnections. The period of two months corresponds to the time
taken by the extra- tropical atmosphere to fully develop as response to tropical SST anomaly, as discussed and used by Frankignoul et al. (2011). For instance, when using November snow cover and December atmosphere (one month later), we remove the regression calculated from the PCs using the values in October. As demonstrated in Appendix of Frankignoul et al. (2011), such removal leads to a proper estimate of the covariability between the two fields. However, non- linear effects are not accounted for. The MCA uses the snow cover and atmospheric fields after ENSO removal. The time series associated with the snow cover is then used to calculate linear regressions applied to other atmospheric fields with the same processing to remove ENSO, with a lag identical to the one used for the atmosphere in the MCA. Therefore, all regressions illustrated are obtained using November Snow/December SLP MCA snow times series, and ENSO calculated with the monthly tropical SST PCs using October.
TABLE S1. Statistics of the MCA for different methodology to remove ENSO teleconnections. The last two lines provide the statistics when ENSO is not removed. The level of statistical significance is given in parenthesis. Period SLP Month Snow Month ENSO lag NSC R 79-14 NOV OCT 2 1.3 (10%) 0.70 (11%) 79-14 DEC NOV 2 2.5 (0%) 0.82 (1%) 79-14 NOV OCT 1 1.2 (11%) 0.70 (19%) 79-14 DEC NOV 1 2.1 (0%) 0.82 (0%) 79-14 NOV OCT 0 1.0 (35%) 0.64 (35%) 79-14 DEC NOV 0 2.4 (0%) 0.82 (0%) 79-14 NOV OCT - 1.0 (34%) 0.62 (63%) 79-14 DEC NOV - 2.0 (0%) 0.79 (2%) TABLE S2. Same as Table 2, but using October and September SIC instead of November SIC. OBS NSC R NSC R SIC(Oct) 3.1 (6%) 0.56 (25%) 0.11 (1/4) 0.14 (1/4) SCE+SIC(Oct) 2.3 (0%) 0.72 (6%) 0.08 (1/4) 0.16 (1/4) SIC(Sep) 1.3 (49%) 0.50 (76%) 0.14 (1/4) 0.14 (0/4) SCE+SIC(Sep) 1.4 (19%) 0.78 (1%) 0.07 (1/4) 0.16 (1/4) TABLE S3. Same as Table 2, but using January and February SLP instead of November SLP. OBS NSC R NSC R SLP(Jan)/SIC 1.8 (15%) 0.59 (48%) 0.19 (1/4) 0.14 (2/4) SLP(Jan)/SCE+SIC 1.5 (4%) 0.65 (57%) 0.10 (1/4) 0.15 (2/4) SLP(Feb)/SIC 1.2 (66%) 0.52 (69%) 0.13 (1/4) 0.15 (2/4) SLP(Feb)/SCE+SIC 1.3 (14%) 0.69 (10%) 0.07 (1/4) 0.16 (1/4)
Fig. S1 : Mean November snow cover fraction and auto- correlation at lag 1- month (, i.e. correlation between October and November snow cover) in observations and models. The black contours in the right panels indicate the interannual standard deviation (in %, contour interval 10%) of the snow cover extent.
Obs b) NOV a) c) d) CED Fig. S2 : Regression of the heat flux, positive upward, in W m- 2, onto the ATM index, in (left) ERA- Interim and (right) the four models, multiplied by the correlation between ATM and MCA- snow, to illustrate the dynamical heat flux component corresponding to one standard deviation of the MCA- snow index. The ATM index is calculated using a projection onto the SLP pattern obtained by regression onto the snow cover index. Note that the ATM index has a correlation of 0.87 (0.89) with the SCA (AO) index in November (December) for the observations, and 0.78 (0.92) in average for models.
Obs b) NOV a) d) c) CED Fig. S3 : Regression of the heat flux, positive upward, in W m- 2, onto the snow cover MCA index in (left) ERA- Interim and (right) the four models.
a) Obs. 79-14 b) c) d) Fig. S4 : Same as Fig. 9 but for the heat flux in December.
SIC lags SIC leads Significance NSC Fig. S5 : Same as Fig. 1, but using the Barents and Kara sea ice concentration instead of the Eurasian snow cover. Month for SLP