Unprecedented recent summer warmth in Arctic Canada

Abstract

[1] Arctic air temperatures have increased in recent decades, along with documented reductions in sea ice, glacier size, and snow cover. However, the extent to which recent Arctic warming has been anomalous with respect to long‐term natural climate variability remains uncertain. Here we use 145 radiocarbon dates on rooted tundra plants revealed by receding cold‐based ice caps in the eastern Canadian Arctic to show that 5000 years of regional summertime cooling has been reversed, with average summer temperatures of the last ~100 years now higher than during any century in more than 44,000 years, including the peak warmth of the early Holocene when high‐latitude summer insolation was 9% greater than present. Reconstructed changes in snowline elevation suggest that summers cooled ~2.7°C over the past 5000 years, approximately twice the response predicted by Coupled Model Intercomparison Project Phase 5 climate models. Our results indicate that anthropogenic increases in greenhouse gases have led to unprecedented regional warmth.

1 Introduction

[2] The instrumental record of global mean annual surface air temperature documents warming from 1970 to 2010 of ~0.6°C, while estimates of coeval trends over Arctic land areas are 3–4 times larger [Hansen et al., 2010], reflecting the strong positive feedbacks unique to polar regions [Serreze and Francis, 2006]. However, the degree of Arctic amplification and the extent to which recent Arctic warming is anomalous with respect to natural climate variations remain difficult to evaluate because of the limited temporal and spatial coverage of the instrumental record within the Arctic, as well as the presence of multidecade temperature trends that may be related to internal modes of climate system variability [Steele et al., 2008; Chylek et al., 2009]. Here we provide a millennial perspective on recent Arctic warming based on radiocarbon ages of tundra plants emerging from beneath nonerosive, cold‐based ice that is substantially longer than earlier Arctic temperature reconstructions [Overpeck et al., 1997; Kaufman et al., 2010].

[3] Arctic glaciers have receded over the past century, with rates of recession increasing in recent decades [Dyurgerov and Meier, 2000; Gardner et al., 2011b]. In the eastern Canadian Arctic, changes in glacier dimensions are a direct response to summer melt; over the past half‐century, summer temperature has determined >90% of the observed variance in glacier annual mass balance [Koerner, 2005]. On millennial time scales, orbital changes that alter the flux of summer short‐wave radiation can influence summer melt and mass balance independently of temperature [Paterson, 1994], largely due to changes in ice sheet albedo that control the absorption of short‐wave energy. However, the contribution of anomalous short‐wave radiation and the associated melt‐albedo feedback to summer melting can be independently quantified [van de Berg et al., 2011]. Accounting for this relatively small contribution permits the use of long‐term, residual changes in glacier dimensions and snowline elevation as a proxy for summer temperature change prior to the instrumental record.

[4] Although glaciers are frequently associated with deep and widespread erosion, small, cold‐based ice caps that mantle relatively flat terrain typically advance by lateral accretion rather than by basal flow and are thus capable of preserving even the most delicate features of the landscape. As these ice caps recede, they often reveal rooted tundra plants that were living at the time snow and ice last covered the site [Falconer, 1966]. Our field observations, and the presence of extensive vegetation‐free regions surrounding most retreating ice caps [Locke and Locke, 1976], indicate that most long‐dead tundra plants exposed by ice recession are rapidly removed from the landscape by wind‐blown winter snow or by runoff during the melt season. Recolonization may begin within a few years of exposure, and the few moss clumps that escape rapid erosion commonly regrow [Yashina et al., 2012; La Farge et al., 2013]. We demonstrated previously that the ¹⁴C activity of growing moss reflects that of the contemporary atmosphere and that dates on different strands of recently exposed individual moss clumps and from different nearby ice‐edge moss collections can be replicated within measurement uncertainties [Miller et al., 2013]. In order to avoid age bias from recolonization and regrowth, we restrict our sampling to rooted moss within 1 m of the ice margin, which have been exposed during the year of their collection according to observed rates of recent ice retreat [Miller et al., 2013]. The radiocarbon dates of our samples thus constrain two complementary paleoclimate indices. They record when, following any earlier ice‐free intervals, snowline last dropped below the collection site, killing the plants. With equal confidence, they also indicate when a particular site was last ice free prior to its modern exposure [Thompson et al., 2013; Miller et al., 2013] and, by extension, that recent summer temperatures at the site have been as high or higher than when those plants were alive.

[5] Over three field seasons (2005–2010), we made 365 collections following protocols above and outlined in Miller et al. [2013] from ~110 different ice caps, ranging from 470 to 1440 m above sea level (asl), in a 1000 km transect along the highlands of Baffin Island (Figure 1). We obtained 145 ¹⁴C dates from these collections, most commonly on rooted clumps of Polytrichum moss. The ¹⁴C dates were calibrated using OxCal 6.0 [Bronk Ramsey, 2009], with the resultant median estimates of calendar age expressed on the b2k time scale (i.e., years before 2000 A.D.; Table S1 in the supporting information). Using a subset of these dates, all from the past ~800 years, we argued previously that tight clustering of vegetation kill dates defined intervals of abrupt summer cooling leading to regional snowline depression across much of our field area [Miller et al., 2012].

Figure 1

Eastern Canadian Arctic digital elevation model showing collection sites, with an expanded map of the southern field area. Circles (color‐coded by their ¹⁴C age) identify the 135 sites where rooted plants were collected the year of their exposure along receding ice cap margins that have been ¹⁴C dated <5 ka; diamonds identify the sites dated >44 ka. The margins of the Laurentide Ice Sheet (LIS) at the Last Glacial Maximum (LGM) and 9 ka [Dyke, 2004] (solid lines) and the 1 km LGM surface elevation contour (dotted line) [Dyke et al., 2002] show that sites dated >44 ka were elevationally above the LIS and supported only local ice caps. P: Penny Ice Cap; B: Barnes Ice Cap. A tight cluster of five coastal sites (circled) exhibits anomalously low changes in their equilibrium line altitudes (ELAs) and is excluded from the regressions in Figure 2a.

2 Results

[6] The vast majority of vegetation kill dates presented here are from the last 5 ka, with at least five dated sites from each of the past five millennia (Figures 1 and 2b). The oldest dates are typically from the highest elevations (Table S1); low‐elevation outliers occur because winter accumulation and summer melt can be influenced by local factors such as wind‐drifted snow, solar shading, proximity to the ocean, and frequency of summer fog, allowing ice bodies in optimal settings to persist at elevations well below the regional snowline.

Figure 2

(a) The 135 ¹⁴C dates calibrated on the b2k time scale for recently ice‐free rooted vegetation (Figure 1) plotted against the difference between the elevation of their collection and the snowline at that site in the middle twentieth century interpolated from maps in Andrews and Miller [1972]. Dark circles are from the northern field area; lighter symbols are from the southern field area (Figure 1). Elevation error bars reflect the coarse contours of the maps in Andrews and Miller [1972]; calibrated age uncertainties are contained within the diameter of their symbol. The five sites without error bars are contained within the circle shown in Figure 1 and are excluded from the regressions. Solid line denotes the least squares linear regression through all data; the dashed line denotes the regression through the highest 20% of all data in each 1 ka age bin. Solid square denotes the minimum 2000–2005 snowline rise relative to the middle twentieth century snowline for the Penny Ice Cap (PIC; Figure 1); dotted line shows the average rate of ELA change since the middle twentieth century. (b) Histograms in 1 ka and 100 year bins for ¹⁴C dates <5 ka.

[7] The oldest Holocene dates in our data set are ~5 ka old, indicating that snowline fell and ice caps expanded at a number of sites at that time, consistent with estimates for the onset of Neoglaciation in this region [Miller et al., 2005]. Because ice cover persisted at each site until the year of sample collection, summer warming of recent decades that led to ice melt and sample exposure must have been greater, on average, than any interval of comparable length during the past ~5 ka.

[8] The period over which recent warmth and melting may be considered exceptional is greatly extended by the surprise finding of four small, summit ice caps, from which all 10 ¹⁴C ages on rooted vegetation exposed by recent ice recession (Figure 3) predate the Holocene (Table 1). For two of the ice caps (“a” and “c,” Table 1), we were able to obtain and date two independent collections. In addition, we made replicate measurements in individual collections from ice caps “a” and “d.” For ice cap “a,” three replicate measurements ranged from 23.9 ± 0.1 to 44.3 ± 1.3 ¹⁴C ka BP, suggesting that small amounts (3–5%) of wind‐blown modern plant debris were incorporated with the ancient plants at this site and were not always removed by our pretreatments (modern contamination disproportionately influences the apparent age of older materials). The oldest of the three dates is likely the least contaminated and the most reliable. This, together with the high and variable ages from the other ice caps (ranging from 34.3 ± 3.6 to 50.7 ± 3.1 ¹⁴C ka BP), suggests that all or most of the ¹⁴C dates should be regarded as minimum‐limiting ages. Associated estimates of calendrical age (which, in this case, would also likely represent minimum age limits) range from >43.8 to >51 ka (Table 1).

Figure 3

(a) Low‐angle aerial photograph of site M10‐B231v, from which rooted moss ~1 m from the ice margin returned a ¹⁴C age >47 ka (Table 1). The area of flat terrain for this and the other sites dated >47 ka is <0.2 km². Vegetation (color) trimline marks the maximum extent of the ice cap at the peak of the Little Ice Age, late in the nineteenth century. (b) Fruticose lichens (M10‐B247v) <0.5 m from an ice cap margin ¹⁴C dated >47 ka (Table 1). Photos by K.A.R., August 2010.

Table 1. Radiocarbon Dates (Calibrated Using OxCal v4.2 [Bronk Ramsey, 2009]) From Four Ice Caps Where Rooted Plants All Predate the Holocenea^a At least one sample from each ice cap potentially exceeds the upper limit of the calibration program, suggesting that all dates should be considered minimum ages.

Ice Cap	UCIAMS ID	Field ID	Material	Latitude	Longitude	Elevation (m asl)	Distance From Ice Edge (m)	^14C Age (years BP)	±1σ	Calibrated Age (years BP)	±1σ
a	85968	M10‐B231v	Rooted black moss 1 cm thick	67.54911	−64.92079	1395	1	23,920	100	28,700	230
a	84673	M10‐B231v	Rooted black moss 1 cm thick	67.54911	−64.92079	1395	1	29,100	1500	33,400	1640
a	85576	M10‐B231v	Rooted black moss 1 cm thick	67.54911	−64.92079	1395	1	44,300	1300	47,550	1290
a	84674	M10‐B232v	Rooted black moss 1 cm thick	67.54919	−64.92175	1396	<0.5	37,500	3600	42,460	3380	b^b Calibration program states that this may be a minimum age.
b	84681	M10‐B247v	Fruticose lichen attached to rock	67.70471	−64.71092	929	<0.5	45,600	2500	>47,000		b^b Calibration program states that this may be a minimum age.
c	84686	M10‐B255v	Fruticose lichen attached to rock	68.00617	−66.61796	1092	<0.5	43,200	2700	46,960	1920	b^b Calibration program states that this may be a minimum age.
c	123039	M10‐B256v	Rooted moss strands	68.00617	−66.61796	1092	<0.5	50,700	3100	>51,000		c^c Calibration program states that this is a minimum age.
d	84687	M10‐B258v	Fruticose lichen attached to rock	67.96183	−66.64522	1261	0.5	34,300	3600	39,800	3610	b^b Calibration program states that this may be a minimum age.
d	85969	M10‐B258v	Fruticose lichen attached to rock	67.96183	−66.64522	1261	0.5	37,510	490	42,220	360
d	85577	M10‐B258v	Fruticose lichen attached to rock	67.96183	−66.64522	1261	0.5	39,740	950	43,800	710

• ^a At least one sample from each ice cap potentially exceeds the upper limit of the calibration program, suggesting that all dates should be considered minimum ages.
• ^b Calibration program states that this may be a minimum age.
• ^c Calibration program states that this is a minimum age.

[9] The ancient rooted plants emerging beneath the four ice caps must have been continuously ice covered for at least 44 ka. However, because the oldest dates are near the limit of the radiocarbon age scale, substantially older ages are possible. Based on temperature reconstructions for ice cores retrieved from the nearby Greenland Ice Sheet [North Greenland Ice Core Project Members, 2004], the youngest time interval during which summer temperatures were plausibly as warm as present prior to 44 ka is ~120 ka at or near the end of the Last Interglaciation. We suggest that this is the most likely age of these samples. Regardless of the absolute age uncertainties, it remains clear that these four ice caps did not melt behind our collection sites at any time during the Holocene but did do so recently, indicating that summer warmth of recent decades exceeded that of any interval of comparable length in >44 ka.

[10] The length of the averaging interval over which summer temperatures and melting could not have exceeded that of recent decades is determined by former ice thicknesses and vertical melt rates. Surface elevation contours of the continental Laurentide Ice Sheet (LIS) show that all four ice caps with pre‐Holocene dated plants were above the surface of the LIS at its Last Glacial Maximum (LGM [21 ka]; Figure 1). These sites thus supported only local ice caps then as now. In addition, because the ice caps occupy flat summits of less than 0.2 km² surrounded by steep slopes (Figure 3a), ice thicknesses of more than 70 m could not have been sustained due to simple ice‐mechanical constraints (SOM). Current (2000–2005 A.D.) surface lowering rates at these elevations (900–1300 m asl) are ~0.5 m a⁻¹ [Webb et al., 2009]; hence, a 70 m thick ice cap would have melted to less‐than‐present dimensions in 100 years if Holocene thermal maximum (HTM) summers had been as warm as present. Thus, average summer temperatures of the past century must have exceeded those of any century in more than 44 ka, including the peak warmth of the HTM (10–5 ka) [Kaufman et al., 2004].

3 Summer Temperature Change

[11] Snowline, the elevation at which winter snow accumulation is matched by summer snowmelt, is determined currently by summer temperature [Koerner, 2005]. Consequently, changes in regional snowline provide a quantitative basis for reconstructing past summer temperatures. Each dated plant sample defines when snowline last dropped and remained largely below the collection site until recent decades. To reconstruct changes in snowline through time, we account for spatial gradients of snowline across the study region by comparing the elevations of our dated samples to an estimate of the middle twentieth century snowline at each site from previous mapping of equilibrium line altitudes (ELAs) based on aerial photographs from 1957 to 1961 AD (Figure S1). For the slow‐flowing ice caps we collect from, snowline and glacier ELA are approximately coincident, with the minor caveat that the ELA defined by middle twentieth century glacier morphometry may not reflect complete equilibration to climate at that time [Falconer, 1966]. The estimated snowline elevation change plotted as a function of sample age describes the evolution of regional snowline over the past 5 ka (Figure 2a).

[12] Fitting of a least squares linear regression to all of the data indicates average snowline lowering of 108 ± 20 m a⁻¹ (r² = 0.45) at the 95% confidence level. The five largest low‐elevation outliers with respect to this trend are a tight cluster of coastal sites (circled in Figure 1), where sea ice persists longest in summer [Moore, 2006]; hence, these sites are unlikely to be representative of the regional snowline. An alternative and likely more reliable regression through all of the data, except those from the five anomalous sites, yields a slope of 129 ± 18 m a⁻¹ (r² = 0.61; Figure 2a). A second alternative fit to the highest 20% of observations in each 1 ka age bin, which largely eliminates the influence of low‐elevation outliers that exist below the regional snowline, yields a similar slope (131 ± 12 m a⁻¹, r² = 0.78). Both alternative regressions yield a 40 b2k (1960 AD) intercept that lies above the middle twentieth century reference snowline (i.e., 0 m elevation change). Hence, both slopes are likely conservative estimates of the millennial‐scale average rate of middle to late Holocene snowline lowering. Although our dates do not constrain the magnitude of snowline lowering at the peak of the Little Ice Age (LIA), vegetation trimlines suggest that the snowline was nearly 200 m lower during the nineteenth century than in the middle twentieth century [Locke and Locke, 1976], emphasizing the anomalous character of peak LIA summer cold with respect to long‐term trends.

[13] The first‐order trend of snowline lowering over the past 5 ka (Figure 2a) is an expected response to orbitally driven reductions in Northern Hemisphere (NH) summer insolation. Precession of the equinoxes reduced summer insolation across the NH, particularly in the Arctic, where June–July insolation at 65°N has decreased by 9% since 11 ka [Berger and Loutre, 1991]. Consistent with this change, most NH paleoclimate reconstructions document peak summer warmth early in the Holocene, followed by widespread, although irregular cooling, through the late Holocene, with intermittent glacier advance and retreat during Neoglaciation [Wanner et al., 2011; Marcott et al., 2013]. The decrease in reconstructed regional snowline between ~5 ka and the early twentieth century (Figure 2a) was 650 ± 90 m and is associated with a ~5% decrease in June–July insolation at 65°N [Berger and Loutre, 1991].

[14] Information on the snowline of recent decades is derived from repeat NASA lidar altimetry for the Penny Ice Cap (Figure 1) and demonstrates that snowline rose above the Penny Ice Cap (i.e., >1980 m asl) between 2000 and 2005 A.D. [Webb et al., 2009], a snowline increase of >830 m since the middle twentieth century (square, Figure 2a). A similarly large rise in the ELA over west Greenland (660 m) occurred between 1997 and 2011 A.D. [McGrath et al., 2013].

[15] To reconstruct summer temperature change associated with the observed lowering of snowline elevation since 5 ka (Figure 2a), we first account for effects of anomalous summer (June–July) short‐wave radiation associated with insolation 5 ka ago that was 24 W m⁻² higher than present. To do this, we scale recent model estimates of the short‐wave contribution to snowline change on the Greenland Ice Sheet during the Last Interglaciation [van de Berg et al., 2011] to anomalous Arctic radiation 5 ka ago. This implies snowline lowering of ~95 m since 5 ka independent of surface air temperature change, which we apply as a linear correction to the observed average rate of snowline descent (SOM).

[16] To estimate temperature change associated with the residual snowline elevation change, we utilize summer temperature lapse rates recorded on glacier surfaces in Arctic Canada (4.9 ± 0.4°C km⁻¹) [Gardner et al., 2011a], a more relevant (and more conservative) constraint than the average free‐air moist adiabatic lapse rate (~6°C km⁻¹). The observed lapse rate indicates summer temperature lowering between 5 ka and the middle twentieth century of 2.7 ± 0.7°C (with uncertainties propagated for uncertainties in slope of the alternative snowline regressions and lapse rate, but not snowline change due to insolation), assuming no significant contribution from durable changes in precipitation. This assumption is supported by the lack of a significant trend in annual accumulation on the adjacent Greenland Ice Sheet over the past 8 ka [Meese et al., 1994], despite evidence for widespread ice sheet advances since 5 ka [Kelly and Lowell, 2009]. There has been also no trend in local snowfall amount since 1960 A.D. [Koerner, 2005] despite recent warming; hence, the snowline rise of >830 m since 1960 A.D. most likely corresponds to a summer temperature increase of >3.7°C.

[17] The magnitude of the reconstructed summer temperature decrease during the late Holocene is similar to estimates of late Holocene cooling derived from borehole temperature profiles through the adjacent Greenland Ice Sheet (2.5 ± 0.5°C) [Dahl‐Jensen et al., 1998] but exceeds that simulated by many climate models. For example, we evaluate nine different models in the Coupled Model Intercomparison Project Phase 5 (CMIP5) that include simulated summer (June–July) temperatures in response to uniformly specified changes in insolation and greenhouse gases for 6 ka and the preindustrial interval (1850 A.D.; http://cmip‐pcmdi.llnl.gov/cmip5/). They produce mean Arctic (60–90°N) land cooling from 6 ka to 1850 A.D. of 1.1 ± 0.3°C (range: 0.7 ± 0.0 to 1.4 ± 0.1°C; Table S2); restricting the region to North Atlantic Arctic land areas gives a similar result (1.0 ± 0.3°C). The complex topography of Baffin Island is not well resolved in most global climate models, which generally underrepresent the high‐elevation terrain along the east coast. To evaluate any possible elevation bias in the model estimates of temperature change since 6 ka, we determined the change in surface air temperature in 200 m elevation bins from 0 to 2600 m asl in the North Atlantic Arctic sector (Table S3 and Figure S2). There is a modest tendency for increased temperature anomalies with height (0.035°C km⁻¹), but in the absence of large, unrepresented local feedback, it is far too small to account for differences between the models and our reconstruction. The mismatch (which is reinforced by the fact that our reconstruction is 1 ka shorter than the CMIP5 simulations and that we do not resolve the LIA cold anomaly in our observations) more likely indicates that many climate models underestimate the magnitude of Arctic amplification, even on relatively long time scales.

4 Conclusions

[18] The age distribution of 145 radiocarbon‐dated rooted plants that were exposed by ice recession in the year of collection along the highlands of Baffin Island, Arctic Canada, demonstrates that several ice caps are now smaller than at any time in more than 44 ka. There has been no intervening century during which summer warmth exceeded that of the last ~100 years. This is the first direct evidence that the contemporary warmth in the eastern Canadian Arctic now exceeds the peak warmth of the early Holocene, when summer insolation across the region was ~9% greater than present. Recent laser altimetry data [Webb et al., 2009] have indicated that ice bodies are losing mass at all elevations; hence, they are still adjusting to current summer temperatures. These findings add additional evidence to the growing consensus that anthropogenic emissions of greenhouse gases have now resulted in unprecedented recent summer warmth that is well outside the range of that attributable to natural climate variability.

Acknowledgments

[19] We thank J. Briner, R. Anderson, K. Zalzal, C. Florian, C. Anderson, and S. DeVogel for assistance in the field and laboratory; W. Colgan for advice on short‐wave radiation; and the Inuit of Pond Inlet and Qikiqtarjuaq for help during fieldwork and for permission to access field sites. Field and analytical programs were supported by National Science Foundation grants ARC‐0454662, ARC‐0714074, and ARC‐0903024. Polar Continental Shelf Project assisted with field logistics. G.H.M. designed the study. G.H.M. and K.A.R. collected the samples. S.J.L. and J.R.S. carried out the radiocarbon dating. G.H.M. and S.L.J. wrote the manuscript. All authors contributed to the interpretations and conclusions.

[20] The Editor thanks John Smol and an anonymous reviewer for their assistance in evaluating this paper.