Winter amplification of the European Little Ice Age cooling by the subpolar gyre

Climate reconstructions reveal a strong winter amplification of the cooling over central and northern continental Europe during the Little Ice Age period (LIA, here defined as c. 16th–18th centuries) via persistent, blocked atmospheric conditions. Although various potential drivers have been suggested to explain the LIA cooling, no coherent mechanism has yet been proposed for this seasonal contrast. Here we demonstrate that such exceptional wintertime conditions arose from sea ice expansion and reduced ocean heat losses in the Nordic and Barents seas, driven by a multicentennial reduction in the northward heat transport by the subpolar gyre (SPG). However, these anomalous oceanic conditions were largely decoupled from the European atmospheric variability in summer. Our novel dynamical explanation is derived from analysis of an ensemble of last millennium climate simulations, and is supported by reconstructions of European temperatures and atmospheric circulation variability and North Atlantic/Arctic paleoceanographic conditions. We conclude that SPG-related internal climate feedbacks were responsible for the winter amplification of the European LIA cooling. Thus, characterization of SPG dynamics is essential for understanding multicentennial variations of the seasonal cycle in the European/North Atlantic sector.

Suppl. Figure 1: European temperature anomaly (in K) in winter (DJF; light blue) and summer (JJA; red), simulated in the Past1000 ensemble (a) and reconstructed (b), as in Figure 1a. Also in (a), simulated Northern Hemisphere (NH) mean temperature anomaly (in K) in summer (dark gray) and winter (light gray), smoothed with an 11-year running mean. Also in (b), reconstructed Northern Hemisphere mean temperature anomaly (in K), with gray shading illustrating the overlap of available NH temperature reconstructions (in percent), from Masson-Delmotte et al. [2013]. All anomalies are calculated with respect to 1500-1850 CE.
Suppl. Figure 2: a, European summer (JJA) temperature anomaly (in K), as in Figure 1a, simulated in the Past1000 ensemble (red), and reconstructed in Luterbacher et al. [2004]  Suppl. Figure 3: Core sites of the reconstructions of (a) sea-surface temperature shown in Figure 2c Suppl. Figure 4: In Past1000-R2, regression coefficients of (a) sea-surface temperature (in K), (b) near-surface (2 m) air temperature (SAT; shading, in K) and sea-level pressure (SLP; contours at 20-Pa intervals), and (c) sea ice concentration (shading, in percent of area) and ocean surface downward heat flux (contours at 10-W/m 2 intervals) onto the standardized Iceland-Scotland Ridge oceanic heat transport (ISR-OHT) in winter (DJF; left panels) and summer (JJA; right panels). Sea ice concentration is in late winter (March) and late summer (September), when it reaches its climatological maximum and minimum extension respectively. Note the same color scale for both panels in a-c, which is also adapted for a better view of the values over Europe in b. Only statistically significant values at the 5 % level are shown, based on two-tailed Student's t test in the PiControl climatology. Calculations are performed after applying an 11-year running mean and for the preindustrial period, 850-1849 CE, to avoid possible spurious effects due to anthropogenic influences. Maps were generated in Pyferret v. 7.0. (Information is available at http://ferret.pmel.noaa.gov/Ferret/).
Suppl. Figure 5: a, Wintertime (DJF) blocking frequency for the period 1575-1724 CE (shading) in Past1000-R2, calculated using the indicator described in Scherrer et al. [2006]. Note that each increase of one unit in the blocking index corresponds to five or more additional days under blocked atmospheric situations. Contours represent the climatological mean in PiControl. Stippling masks statistically significant anomalies between the two at the 5 % level, based on the likelihood of occurrence of the signal in PiControl. Map was generated in Pyferret v. 7.0. (Information is available at http://ferret.pmel.noaa.gov/Ferret/).
Suppl. Figure  Suppl. Figure 7: (Previous page.) Wavelet coherency between the European temperature (from Figure  1a), SPG strength, MOI, and NAO (first time series; from Suppl. Figure 6a) and the ISR-OHT (second time series; from Figure 2a) in winter (DJF; a) and summer (JJA; b) for the preindustrial period, 850-1849 CE. Wavelet coherence analysis between two time series extracts regions in the time-frequency space with large common power and consistent phase relationship and thus suggests times when there might exist causality between processes underlying the two time series. Note, however, that significant coherency does not necessarily subtend causality between the two; this requires identification of a valid physical mechanism. Color shading indicates the strength of the coherence, with warm colors within black contours significant at the 5 % level against red noise. The direction of the arrows indicates the phase relationship between the two wavelet transform: right arrows indicate that the two series are in co-phase; left indicates that the two are in anti-phase (π phase angle); and down/up arrows indicate that the second series lags/leads the first one in quadrature (one fourth of the cycle at that period), if the two series are positively correlated. The cone of influence (white dashed line) defines the area in which the border effect does not influence the wavelet spectra. We note that coherent AMOC and ISR-OHT variability is diagnosed on centennial timescales in winter for the period 1200-1700 CE; the phasing indicates that the ISR-OHT robustly leads the AMOC by roughly 30 years.
The NAO shows broadband coherency with the ISR-OHT on decadal to multidecadal time scales in winter, when the NAO tends to lead ISR-OHT changes, although with remarkably variable phasing. This multidecadal coherency, however, does not persist beyond 1600 CE, when we find the coldest centuries over Europe (Figure 1a). In summer, coherency between the NAO and the ISR-OHT occurs mostly on centennial time scales between c. 1250-1600 CE, with phasing indicating that the ISR-OHT leads the NAO by about 30 years. There is, however, overall less agreement in summer than in winter between the the ISR-OHT and the NAO.
Coherency between the SPG and the ISR-OHT is particularly strong on centennial and multicentennial time scales in both winter and summer, with both series in co-phase or with the SPG leading. Figure produced in Matlab R2013a, using the algorithm described in Grinsted et al. [2004].
Suppl. Figure   Suppl. Figure 9: (Previous page.) In Past1000-R2: a, SAT anomaly (in K) for the period 1575-1724 CE, with respect to 1901-1990 CE, in winter (DJF; left) and summer (JJA; right), as in Figure 1b but for the entire North Atlantic, Arctic, and European regions. Stippling masks statistically non-significant anomalies at the 5 % level. Note the same color scale for both seasons, which is also adapted for a better view of the values over both Europe and the subpolar North Atlantic. b-d, Regression coefficients of the SAT (in K per standard deviation, or std) onto the standardized indices for the SPG strength (b), MOI (c), and NAO (d) respectively, in winter (left) and summer (right). Calculations are performed after applying an 11-year running mean and for the preindustrial period, 850-1849 CE. Note the same color scale for all panels, which has additionally been reversed for a better comparison with panels in a. Maps were generated in Pyferret v. 7.0. (Information is available at http://ferret.pmel.noaa.gov/Ferret/).
Suppl. Figure 12: Land temperature anomalies (in K) for the three coldest 10-year winter periods reconstructed in Luterbacher et al. [2004] (left) and simulated in Past1000-R2 (right). 1st indicates the coldest decade, 2nd the second coldest one, and so on. Note that the simulated 2nd decade is shown next to the reconstructed 1st one, since they cover the same period approximately. Maps were generated in Pyferret v. 7.0. (Information is available at http://ferret.pmel.noaa.gov/Ferret/).