High Arctic Holocene Temperature Record from the Agassiz Ice Cap and Greenland Ice Sheet Evolution

High Arctic Holocene temperature record from the Agassiz ice cap and Greenland ice sheet evolution

Benoit S. Lecavaliera,1, David A. Fisherb, Glenn A. Milneb, Bo M. Vintherc, Lev Tarasova, Philippe Huybrechtsd, Denis Lacellee, Brittany Maine, James Zhengf, Jocelyne Bourgeoisg, and Arthur S. Dykeh,i

aDepartment of Physics and Physical Oceanography, Memorial University, St. John’s, Canada, A1B 3X7; bDepartment of Earth and Environmental Sciences, University of Ottawa, Ottawa, Canada, K1N 6N5; cCentre for Ice and Climate, Niels Bohr Institute, University of Copenhagen, Copenhagen, Denmark, 2100; dEarth System Science and Departement Geografie, Vrije Universiteit Brussel, Brussels, Belgium, 1050; eDepartment of Geography, University of Ottawa, Ottawa, Canada, K1N 6N5; fGeological Survey of Canada, Natural Resources Canada, Ottawa, Canada, K1A 0E8; gConsorminex Inc., Gatineau, Canada, J8R 3Y3; hDepartment of Earth Sciences, Dalhousie University, Halifax, Canada, B3H 4R2; and iDepartment of Anthropology, McGill University, Montreal, Canada, H3A 2T7

Edited by Jeffrey P. Severinghaus, Scripps Institution of Oceanography, La Jolla, CA, and approved April 18, 2017 (received for review October 2, 2016)

We present a revised and extended high Arctic air temperature leading the authors to adopt a spatially homogeneous change in reconstruction from a single proxy that spans the past ∼12,000 y air temperature across the region spanned by these two ice caps. 18 (up to 2009 CE). Our reconstruction from the Agassiz ice cap (Elles- By removing the temperature signal from the δ O record of mere Island, Canada) indicates an earlier and warmer Holocene other Greenland ice cores (Fig. 1A), the residual was used to thermal maximum with early Holocene temperatures that are estimate altitude changes of the ice surface through time. These 4–5 °C warmer compared with a previous reconstruction, and reg- so-called thinning curves provide a valuable constraint on model ularly exceed contemporary values for a period of ∼3,000 y. Our reconstructions of the Greenland ice sheet (5). A key conclusion results show that air temperatures in this region are now at their of the study was that the current generation of 3D thermo- warmest in the past 6,800–7,800 y, and that the recent rate of tem- mechanical ice-sheet models fail to capture the large thinning perature change is unprecedented over the entire Holocene. The inferred at sites located closer to the ice margin, particularly in warmer early Holocene inferred from the Agassiz ice core leads to northwest Greenland (Camp Century drill site; Fig. 1A). How- an estimated ∼1 km of ice thinning in northwest Greenland during ever, the veracity of the results in ref. 5 have been brought into the early Holocene using the Camp Century ice core. Ice modeling question due to the possible influence of the Innuitian ice sheet results show that this large thinning is consistent with our air tem- across the Canadian Arctic on the altitude correction required to perature reconstruction. The modeling results also demonstrate the infer temperature from Agassiz ice during the early Holocene broader significance of the enhanced warming, with a retreat of the (6). A second issue is that the temperature record estimated northern ice margin behind its present position in the mid Holocene from Agassiz ice using two different proxies [ice melt percent (7) and a ∼25% increase in total Greenland ice sheet mass loss (∼1.4 m and oxygen isotope content (5); see next section] gives incon- sea-level equivalent) during the last deglaciation, both of which have sistent results in the early Holocene. Here, we address these is- implications for interpreting geodetic measurements of land uplift sues by considering the influence of Innuitian ice sheet thinning δ18 and gravity changes in northern Greenland. on the O temperature reconstruction from Agassiz ice, and applying the revised reconstruction to force a model of the ice core | temperature reconstruction | Holocene climate | Greenland ice sheet Greenland ice sheet. Results and Discussion nstrumented records of temperature and environmental change Reconstructing Holocene Air Temperatures. Previous air temperature Iextend for a few centuries at most. Although these records reconstructions inferred from Agassiz ice using observations of the provide evidence of climate warming, the time span covered is relatively short compared with the centuries to millennia response Significance times of some climate system components (1). In this respect, reconstructions of temperature and environmental changes obtained from climate proxies (e.g., sediment cores, ice cores) play a com- Reconstructions of past environmental changes are important for placing recent climate change in context and testing climate plementary role to the instrumented records by providing a longer models. Periods of past climates warmer than today provide temporal context within which to interpret the magnitude and rate insight on how components of the climate system might re- of recent changes (2). Furthermore, the relatively large spatial and spond in the future. Here, we report on an Arctic climate record temporal variability captured in these reconstructions represents a from the Agassiz ice cap. Our results show that early Holocene useful dataset to test models of the climate system (3). Of par- ’ air temperatures exceed present values by a few degrees Cel- ticular interest are periods during Earth s history when the climate sius, and that industrial era rates of temperature change are was warmer than at present, as these provide information that is unprecedented over the Holocene period (∼12,000 y). We also potentially more relevant to changes in the future. demonstrate that the enhanced warming leads to a large re- In this study, we focus on the reconstruction of past climate sponse of the Greenland ice sheet; providing information on using ice cores from the Agassiz ice cap, located on Ellesmere ’ A the ice sheet s sensitivity to elevated temperatures and thus Island in the Canadian Arctic Archipelago (Fig. 1 ). This site is helping to better estimate its future evolution. of particular interest as it is located in the high Arctic, and temperature reconstructions can be compared with those from Author contributions: B.S.L., D.A.F., G.A.M., and B.M.V. designed research; B.S.L., D.A.F., more southerly locations to estimate polar amplification of cli- and L.T. performed research; B.S.L., L.T., P.H., J.Z., J.B., and A.S.D. contributed new re- mate in the past (4). Furthermore, it is located proximal to the agents/analytic tools; B.S.L., D.A.F., D.L., B.M., and J.Z. analyzed data; and B.S.L., D.A.F., Greenland ice sheet, and so can be used to better constrain the G.A.M., and D.L. wrote the paper. climate forcing used to model the past evolution of this ice sheet. The authors declare no conflict of interest. 18 In a recent study (5), δ O measurements in ice from the This article is a PNAS Direct Submission. Agassiz (81°N) and Renland (70°N) ice caps (Fig. 1A) were used 1To whom correspondence should be addressed. Email: [email protected]. to estimate temperature records for these locations throughout This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10. the Holocene. The two time series were remarkably similar, 1073/pnas.1616287114/-/DCSupplemental.

5952–5957 | PNAS | June 6, 2017 | vol. 114 | no. 23 www.pnas.org/cgi/doi/10.1073/pnas.1616287114 Downloaded by guest on September 27, 2021 −80˚ −60˚ 80˚ −40˚ −20˚ 0˚ substantially higher temperatures during the early Holocene 9 9 ABAgassiz 8 8 compared with preindustrial values. Although there are few 80˚ 7 7 quantitative reconstructions of high Arctic air temperatures for C) o Camp Century 6 6 5 5 4 4 GRIP/GISP2 3 3 70˚ HTM 2 2 70˚ Renland 1 1 0 0 A

Temperature anomaly ( 7 C) −1 −1 6 o −2 −2 5 60˚ 60˚ −12 −10 −8 −6 −4 −2 0 4 −60˚ −40˚ Time (kyr before present) 3 2 Fig. 1. Location map and Agassiz proxy temperature records: (A) Map 1 showing the study area with the names and locations of ice core borehole 0 sites mentioned in the text. (B) The 25-y resolution, elevation-corrected −1 Agassiz δ18O temperature reconstruction (dark red) with 2σ uncertainty −2 Temperature anomaly ( (light red) and the elevation-corrected Agassiz melt record (green), both −3 C) extended to 2009 CE. Ref. 5’s δ18O Agassiz–Renland temperature re- o −4 B construction is also shown (blue) for comparison. Each record is referenced to its preindustrial temperature value at 1750 CE. −5 −6 −7 18 melt layers (7) and the δ O record (5, 8) are inconsistent in the −8

B ) early Holocene (Fig. 1 ). The melt record indicates temperatures −9 peaking in the early Holocene (∼11 ky) with a steady decline until 188 −2 Summer temperature ( −10 C 8 ky (7, 9); whereas, the earlier reconstruction (5) shows air tem- 186 (W m

peratures reaching a maximum between 8 and 9 ky. The melt- N 184 o record reconstruction is a proxy for summer (June, July, August) 80 temperatures, and is derived using a linear transfer function relating 182 melt percent to summer air temperatures along with the present- 180 Methods 14 day lapse rate correction at the surface of the ice cap ( and D 178 Fig. S1). This technique can only be used to quantify summer air 12 temperatures in the range −8 °C (no melt below this temperature) 176

10 Annual average I to about −3 °C (100% melt above this temperature). In contrast, the δ18O record is a proxy for mean annual air temperature and 8 spans a much larger temperature range. However, these differences 6

between the two proxies do not reconcile the discrepancies between Pollen (grains/L) 4 the earlier δ18O-based temperature reconstructions and the melt reconstruction shown in Fig. 1B. 2 To accurately infer air temperature from ice melt and δ18O records, the effect of elevation changes must be quantified and E 40 removed (Methods). In the case of the melt record, this correction is applied at the location of the ice core to remove the 32 contributions from vertical land motion and changes in ice

thickness (Fig. S2). The elevation correction for the isotope re- 24 (ppb) cords is more complex, as the correction for vertical land motion Na SS and thickness changes of the Innuitian ice sheet is applied at the 16 location where the air moisture that precipitates over the ice cap 18 reaches a fixed maximum elevation of condensation (5, 6, 10, 11) F 8 (Fig. S3). The original δ18O-based reconstruction (5) (Fig. 1B) 15 does not account for this latter effect. When correcting for 12 thickness changes of the Innuitian ice sheet using results from a recent study (12), the δ18O-based temperature reconstruction is 9 significantly altered (Fig. 1B), with peak temperatures occurring 6 earlier (ca.10.5 ky) and a gradual decline in temperature since this time until around 1700 CE. Our reconstruction gives warmer 3 air temperatures during the early Holocene (11.7–8.5 ky) relative to the original (6), and is now more consistent with that de- Bowhead whales (frequency %) 0 termined from the melt record. The offset between maxima of −12 −10 −8 −6 −4 −2 0 the melt and δ18O proxy temperature reconstructions is likely a Time (kyr before present)

product of the noise levels in the melt record, particularly for SCIENCES Fig. 2. Various records related to high Arctic climate change: (A)Agassizδ18O earlier times. The A87 melt series has a double peak centered ENVIRONMENTAL temperature reconstruction (dark red) with its 2σ uncertainty (light red). (B)Agassiz on 10 and 10.75 ky; whereas, the younger 10-ky peak in the melt record summer (JJA) temperature reconstruction with the potential trend A84 melt series is less pronounced. Therefore, the resulting stack (dashed black) for temperatures below −8°Candabove−3°C(horizontalgray has a diminished 10-ky peak, which emphasizes the role of noise σ = dash; see main text and Methods). (C) Mean annual insolation at 80°N. (D)Agassiz in a two-series stack. Applying a Gaussian low pass filter ( tree (purple) and herb (pink) pollen record (15), (E) sea salt sodium in the GISP2 50 y) to the 25-y resolution temperature time series shows a rapid (doubled scaling; cyan; ref. 17) and Penny ice cores (green; ref. 16), and (F)numbers early Holocene warming with a peak of 6.1 °C at 10 ky (2σ un- of bowhead whale bones from Eastern (gray; 74.3–76.1°N, 83.3–90.5°W; n = 116) certainty 4.3–8.3 °C) followed by a gradual cooling to 1700 CE and Central (black; 71.5–73.7°N, 89.4–99.0°W; n = 96) Queen Elizabeth Islands (18, (Fig. 2A; temperatures are defined relative to the value at 1750 CE). 19). The gray area denotes the local Holocene thermal maximum, a period when 18 Together, the Agassiz δ O and melt-layer records point to Agassiz temperatures were regularly above peak contemporary values.

Lecavalier et al. PNAS | June 6, 2017 | vol. 114 | no. 23 | 5953 Downloaded by guest on September 27, 2021 the early Holocene (13), the occurrence of endostromatolites on record, exceeding 1.5 °C/century, which is consistent with data Devon Island dated to the early-mid Holocene indicates that from nearby weather stations in the Canadian high Arctic where temperatures were 5–8 °C warmer (14), providing evidence of an mean annual air temperatures have increased at a rate of early Holocene thermal maximum in this area and supporting ∼0.1 °C/decade since the 1970s (7). Therefore, although air our revised δ18O temperature reconstruction. temperatures were warmer than today in other parts of the Orbital variations are considered to be the primary driver of a Holocene, the rate of climate warming during the industrial era warm early Holocene in the high Arctic, and this is supported by is unprecedented over the past ∼12,000 y. the good correlation between peak annual average insolation at The Agassiz ice cap provides the most northerly paleoclimate 80° latitude (Fig. 2C) and temperatures at the Agassiz ice cap. A record (∼80°N) of the entire Holocene, and our δ18Otemperature maximum in pollen concentrations (Fig. 2D; ref. 15) indicates reconstruction provides important information for quantifying the that, as the climate warmed, atmospheric moisture content in- strength and timing of polar warming. A variety of feedback creased alongside a strengthening of meridional heat and mois- mechanisms are responsible for this warming, and there have been ture transport to the Arctic. As a consequence of the warmer significant advances in identifying the processes responsible for the temperatures, sea-ice cover was likely at a minimum during the modern polar amplification of climate warming (4, 20); however, early Holocene as inferred from the higher concentration of sea-salt there remains considerable uncertainty in projections of future sodium found in the Greenland ice sheet and Penny ice cap (Fig. Arctic climate change (21) and thus a need for improved observa- 2E; refs. 16, 17). Furthermore, inferences of sea-ice cover in the tions of past and present climate conditions in the high Arctic. Al- archipelago from whale bone remains in raised marine deposits though it is beyond the scope of this study to perform the regional- suggest maximum breakup following peak annual insolation and scale analysis required to rigorously examine polar amplification in temperatures at the Agassiz ice cap (Fig. 2F; refs. 18, 19). the Holocene, it is of interest to compare the magnitude of early A shallow ice core was collected in the Agassiz ice cap in 2010, Holocene warming from the Agassiz and Renland ice caps. The extending the δ18O time series and thus temperature recon- corrected Agassiz early Holocene temperatures warmed by ∼4°C struction to 2009 (Fig. S4 and Methods). The extended record more relative to those inferred at the Renland ice core (eastern indicates that modern air temperatures are ∼4 °C warmer rela- Greenland; ref. 5) during the 11–9.5 ky interval. Furthermore, peak tive to preindustrial values and at their highest in the past warmth at Agassiz and Renland differ by ∼2 ky, with Agassiz ex- ∼7000 y (2σ uncertainty 6.8–7.8 ky; Fig. 3A). We calculated rates periencing an earlier and more distinct Holocene thermal maxi- of temperature change during the Holocene using a Gaussian mum indicating a pronounced reduction in temperature filtered temperature reconstruction and linear regression (Fig. gradient between these two sites during the early Holocene. 3C). The latter approach was applied to provide a rate that is accurate for the past 200 y (a contemporary rate is not possible Implications of Agassiz δ18O Temperature Reconstruction for Greenland using the filtered data due to the edge effect associated with the Ice Sheet Evolution. We apply our δ18O temperature record to end of the time series). Rates of temperature change have consider its impact on the Holocene evolution of the Greenland fluctuated throughout the Holocene, but have generally been ice sheet. As outlined above, elevation histories at Greenland ice less than ∼1 °C/century. However, the last few centuries have core locations (Fig. 1A) were estimated based on the similarities experienced the highest rates of temperature change in our between the Agassiz and Renland δ18O records (5). Our analysis reveals that these records are different during the early Holo- cene, and so the adoption of a spatially uniform temperature change across Greenland (5) is no longer supported. Using our δ18 C) 9 O temperature record from Agassiz, and assuming that it ABo 4 8 remains valid regionally, variations in ice elevation at Camp 2 Century were isolated and subsequently corrected for upstream

7 C) 0 o effects (6), yielding a Camp Century Holocene elevation curve 6 −2 (Fig. 4B; Methods). Our estimated thinning curve at Camp

Temperature ( −0.6 −0.4 −0.2 0.0 5 Century displays an elevation reduction during the early Holo- Time (kyr) 4 cene that is a factor of 1.6 larger than the original estimate (5) 3 (Fig. S5). As a result, the model–data discrepancies noted pre- 2 viously (5, 6, 22) are enhanced (Fig. S5). The accuracy of the 1 resulting elevation history relies on the accuracy of our as-

Temperature anomaly ( sumption that the Agassiz climate history reflects climatic 0 changes occurring at Camp Century. These locations are rela- −1 tively proximal (∼500 km apart), although the possibility of sig- 2.5 −2 nificant climatic deviations at Camp Century from those on 2.0 1.5 C Ellesmere Island cannot be disregarded; particularly during the 1.0 early Holocene when large changes in ice-surface elevation and 0.5 0.0 sea-ice extent could have produced significant climate variability −0.5 in the Nares Strait region via orographic changes and enhanced

o −1.0

( C/century) – −1.5 atmosphere ocean interaction as this area became ice free. −2.0 The majority of Greenland ice sheet models are forced with a −2.5 −10−8−6−4−20 temperature history that is inferred from summit ice cores [e.g., Rate of temperature change Time (kyr before present) Greenland Ice Core Project (GRIP), Greenland Ice Sheet Project 2 (GISP2)] and then extrapolated across the ice sheet Fig. 3. Low-pass-filtered Agassiz δ18O temperature reconstruction: (A)Agassiz δ18 – δ18 σ based on O T relationships and lapse rates (e.g., ref. 22). As O temperature reconstruction (dark red) with its 2 uncertainty (light red) shown in Fig. 4A, this extrapolated curve greatly underestimates with the Gaussian filtered (σ = 50 y) reconstruction (black). (B) The Agassiz δ18 the amplitude of early Holocene warmth compared with that O temperature reconstruction over the past 1,000 y. The gray lines denote δ18 the linear regression results for three periods: the cooling leading into the inferred here from the Agassiz ice O record. To address this preindustrial period, the preindustrial era, and industrial era. (C) Rate of issue, we performed a model sensitivity analysis using a revised δ18 temperature change (black) based on the Gaussian filtered Agassiz tem- temperature history, based on the Agassiz ice O temperature perature reconstruction. The gray circles represent the rates of change time series (Fig. 4A; Methods), to force the northern sector of a re- obtained from the linear regression segments shown in B. The vertical gray cent Greenland ice sheet reconstruction (ref. 22; referred to here- band denotes the local Holocene thermal maximum, a period when Agassiz after as Huy3). The revised climate forcing yields rapid thinning temperatures were regularly above peak contemporary values. across North Greenland, particularly northwest Greenland as presented

5954 | www.pnas.org/cgi/doi/10.1073/pnas.1616287114 Lecavalier et al. Downloaded by guest on September 27, 2021 A with past ice-sheet changes was isolated by estimating and removing 8 the signal caused by ice-sheet changes and the corresponding elastic 6 earth deformation during the GPS monitoring period. Table S1 C) o provides a comparison of observed and modeled uplift rates that 4 includes values for both Huy3 and the revised model. At sites in the 2 northwest, the revised model produces an improved fit to the observed rates. However, we note that the model rates underpredict 0 those observed. In particular, we note that the uplift rates at the two −2 GPS sites nearest to Camp Century are considerably larger than those predicted by the two models, suggesting that a greater

−4 Temperature anomaly ( amount of regional thinning is required, in contrast to that suggested by the RSL reconstructions at Saunders and Thule. −6 The above-noted data–model discrepancies with respect to the RSL and GPS observations have different possible interpretations. B1200 For example, the northwest region could have a different earth 1000 viscosity structure to that of the Greenland-wide optimum used here 800 (22; determined using the Huy3 ice history). Alternatively, the magnitude and/or timing of ice unloading could be incorrect, in- 600 dicating that extrapolating the Agassiz temperature curve to Camp 400 Century is not applicable; or other processes such as changes in the δ18 200 moisture pathway are influencing the O at Camp Century, thus complicating the inferences of elevation changes at this site. De- 0 termining which of these interpretations is correct, and whether all −200 three data types (Camp Century thinning curve, RSL curves, GPS rates) can be reconciled with a single model parameter set, will re-

Camp Century thinning curve (m) −12 −10 −8 −6 −4 −2 0 Time (kyr before present) quire improved observational constraints and a more detailed model sensitivity analysis that explores parameter uncertainties more fully. Fig. 4. Temperature and thinning curves for northwest Greenland: (A) Agassiz Even though we have focused on the north of Greenland in δ18O temperature reconstruction (dark red) with 2σ uncertainty (light red). this analysis, the differences between the original Huy3 model The temperature time series at Camp Century inferred from the GRIP ice core and the revised version are large enough to be evident in their using a δ18O–temperature relationship and lapse rate correction (dashed respective volume time series (Fig. S8). Compared with Huy3, black) and the revised temperature time series based on the Agassiz re- the revised North Greenland reconstruction delivers an addi- construction (solid black; Methods). (B) Camp Century thinning curve (green) tional 1.38 m of ice-equivalent sea level to the global oceans and 2σ uncertainty (light green) compared with model output: Huy3 (dashed during the most recent deglaciation. Although this is a relatively black) and our variant of this model reconstruction (solid black). small amount compared with the global ice volume loss (∼130 m), it is 27% of the loss in the Huy3 Greenland model. The larger tem- B perature forcing in the early Holocene also results in a modeled in Fig. 4 . As a consequence of this enhanced thinning, it was nec- retreat of the ice margin interior of its current position by 20–80 km essary to produce a considerably thicker ice sheet at the last glacial (as early as 8 ky before present in some places) followed by a maximum to match present-day ice extent and thickness in this re- B regrowth to present. The regrowth in North Greenland repre- gion. As indicated in Fig. 4 , it was possible to produce a good fit to sents a net drop of 0.18 m of ice-equivalent sea level, similar to the Camp Century thinning curve using our revised temperature that estimated for the southwest part of the ice sheet (22). Our forcing, suggesting that the failure of previous models to capture this revised model shows that rates of surface thinning reached values Methods signal (ref. 5; Fig. S5; ) reflects inaccuracy in the adopted of 36.7 m/century at Camp Century and the Greenland ice sheet Methods climate forcing. As discussed in , our simulations do not experienced a peak centennial rate of Holocene mass loss of capture the full buttressing effect of the Innuitian ice sheet on the 1,075 gigaton per year (Gt/y) during the period when recon- Greenland ice sheet across Nares Strait. Deglaciation of this area structed temperatures were greater than those at present (Fig. around 10 ky before present would also have contributed to the rapid S9). This rate of model mass loss reflects the “memory” of the ice ice thinning in the region during the early Holocene (23). sheet to past air temperature changes, notably the large increase at The larger deglaciation predicted for North Greenland will the Pleistocene–Holocene transition shortly after 12 ky, as well as influence predictions of relative sea level in this region, so we the contemporary response to the peak temperatures during the input our revision of the Huy3 ice-sheet reconstruction to a glacial early Holocene. Before 10 ky, marine retreat of the ice model isostatic adjustment (GIA) model (Methods) and performed a data– from rising sea levels (22) is a significant contributor to the rate model comparison to test whether the relative sea-level (RSL) of mass loss. After this time, the ice margin was largely land observations support the large thinning suggested by the Camp based and so the mass loss rates can be compared more directly to Century ice core record. Comparison with observations in North those estimated for the ice sheet at present using geodetic methods; and northeast Greenland suggest that the revised ice model is more for example 142 ± 49 Gt/y for the period 1992–2011 (25). compatible with the majority of the RSL reconstructions compared with the Huy3 reconstruction (Figs. S6 and S7; Methods), although Concluding Remarks we note that the quality of RSL data is low in North Greenland. In As demonstrated in the previous section, the δ18O air tempera-

northwest Greenland, nearest to Camp Century, there are some ture reconstruction resulting from this analysis has implications SCIENCES – data model discrepancies with both the original and revised for model-derived regional ice-sheet reconstructions. As the ice ENVIRONMENTAL Huy3 reconstruction. At Qeqertat, the models fit the RSL ob- history is a necessary input to arrive at the corrections applied to servations only if one takes into consideration uncertainties in the Agassiz ice-core data, the ideal approach would involve it- the North American ice complex (see figure S6 from ref. 22). At eration to ensure that the resulting temperature reconstruction Saunders and Thule, the Huy3 model underpredicts the RSL and regional ice model reconstruction are consistent. Such an observations with a mistimed fall in sea level, and the variant approach was not applied here because it would require a sig- Huy3 model over predicts the observations significantly. nificantly more complex analysis to model the Greenland and The accuracy of our revised ice model can also be tested against Innuitian ice sheets simultaneously and capture the interactions Global Positioning System (GPS) estimates of vertical land motion. between them. Of the two corrections applied to the δ18O tem- In a recent study (24), the component of this motion associated perature reconstruction—one associated with height changes of

Lecavalier et al. PNAS | June 6, 2017 | vol. 114 | no. 23 | 5955 Downloaded by guest on September 27, 2021 the Innuitian ice sheet and the other with isostatic land motion— of shallow Arctic ice cores exhibit a robust linear relationship as shown in the former is potentially the more important in terms of being a Fig. S1 (7, 30). This linear transfer function emphasizes that summer (June– significant feedback on the estimated temperature reconstruc- August; JJA) temperatures below −8 °C yield no melt fraction in the ice, tion because it is not directly constrained via the results of this therefore, melt percent values of zero signify summer temperatures less than or equal to −8 °C. Similarly, melt percent values of 100 signify summer study. In contrast, the isostatic response was calibrated to RSL − data from Ellesmere Island and so even if the regional ice temperatures above or equal to 3 °C. The elevation changes for the melt loading history is significantly altered when using the revised record are converted to summer (JJA) temperature using the present-day lapse rate for Ellesmere Island of −0.43 °C per 100 m (7, 30). temperature reconstruction, the earth viscosity parameters would be The summer temperature to melt percent transfer function is based on varied to maintain an isostatic response that best matches the RSL present-day observations. This transfer function likely correlates with in- data. That is, the optimal viscosity model will be different but the solation among other time-varying environmental parameters through the isostatic response (and thus land uplift correction) will be similar. Holocene. However, other processes and feedbacks in the climate system Regarding possible changes to the Innuitian ice sheet via revising render the temporal calibration of the transfer function highly nontrivial, the temperature reconstruction, the impact of this on the results especially because there is a lack of constraints on key environmental pa- presented here is difficult to determine without applying an iterative rameters in the high Arctic during this period. Therefore, we assume that the approach. However, we note that a high variance subset of Innuitian present-day transfer function and its uncertainties adequately capture the ice reconstructions was adopted to partially account for not applying temperature to melt percent relationship to first order during the Holocene. an iterative approach. The rate of temperature change from the Agassiz annual temperature The consistency between the proxy records (δ18O and melt reconstruction was determined by discretized differentiation of the Gaussian records from Agassiz and those shown in Fig. 2), and the good fit filtered record presented in Fig. 3A. This provided multicentennial-scale between the modeled ice thinning and that inferred from the variations in the rate of temperature change (Fig. 3). The centennial Camp Century ice core, indicate that the feedbacks noted above trends highlight fluctuations in the rate of temperature change, which are 18 are relatively minor and our primary conclusions are accurate to not strongly influenced by noise within the high-resolution δ O record. first order. However, this is an aspect of the current analysis that However, the low-pass Gaussian filter truncates the record to 100 y before could be improved upon in future studies. present where edge effects are negligible. To supplement the truncation near present, linear regression was conducted on the raw temperature re- Methods construction on 200-y intervals starting at present and going back to 600 y before present. This yielded a rate of −0.08, 0.37, and 1.97 °C/century for the δ18 Inferring Temperature Records from Ice Core Measurements. The Agassiz O periods before present of 600–400 y, 400–200 y, 200 y to present, re- and melt records have been influenced by a thinning ice sheet and isostatic spectively (Fig. 3). rebound. Therefore, an altitude correction is required to obtain temperature The error analysis for the Agassiz δ18O temperature reconstruction ac- reconstructions that are elevation independent. Although the δ18O is sensitive counts for the uncertainties arising from: thinning of the Innuitian ice sheet to elevation changes along the southeast coast of Ellesmere Island, the melt (12); moisture pathway to the Agassiz ice cap (6); δ18O altitude relation (27); record is affected by elevation changes at the borehole site and so separate and the temperature–δ18O relationship (5, 6, 29). The uncertainties in the altitude corrections are required for each record (Figs. S2 and S3). Previous temperature reconstruction are nonparametric and for this reason conser- work has illustrated that local changes in the altitude and thickness of the vative error estimates are presented in this study. When dealing with δ18 Agassiz ice cap do not affect the O composition of the ice (9, 11). The at- Gaussian uncertainties (e.g., δ18O altitude relation based on linear regression), mospheric moisture that precipitates onto the Agassiz ice cap primarily passes traditional error analysis methods were applied. However, when dealing with through (and partly originates in) Baffin Bay. It subsequently encounters the nonparametric uncertainties, rather than conducting a Monte Carlo error anal- southeastern shores of Ellesmere Island where it is elevated, and given there ysis to estimate confidence intervals, the upper and lower bound error estimates are no further inland features capable of forcing the air masses significantly were used to evaluate the uncertainties. The choice to present conservative δ18 higher, the O composition in the ice is predominantly sensitive to altitude uncertainty estimates in these cases was made to partly compensate for a likely changes along this shoreline (5, 6, 10). Thus, the elevation correction is not underestimation of nontrivial uncertainties such as the temporal evolution of the determined at the drill site, but rather at the location where the air mass δ18O–altitude and temperature–δ18O relationships. initially encounters a major topographical feature along the eastern coastline of Ellesmere Island. Furthermore, the position where the correction is esti- The Agassiz 2009 Extension Core Series. Starting in the late 1970s, ice-core mated changes through the Holocene due to topographical changes as the drilling campaigns have been undertaken at the Agassiz Ice Cap. The stud- Innuitian ice sheet melted (6). The two processes that dominated altitude ies resulted in Holocene records of stable water isotopes, melt layers, solid changes of southeastern Ellesmere were: (i)GIAand(ii) thinning of Innuitian conductivity, and pollen (8, 9). The Agassiz A84 and A87 isotope records are ice (Figs. S2 and S3). Similar to previous studies (i.e., ref. 6), we assume neg- based on cores (obtained in those years), which were 100 m apart. In 2010, a δ18 ligible change in seasonal biases in the O relationships, with snowfall across 16-m core (A09; 80°49’N, 72°53.74’W) was obtained, located between those the Agassiz ice cap occurring throughout the year, although with increased earlier cores, to extend the melt and isotope records to 2009 CE. Eight precipitation starting in the spring and decreasing in the fall at present (8). hundred water samples were analyzed at University of Ottawa for 18O using The Agassiz altitude correction was determined using the GIA model pa- an LGR liquid water isotope analyzer. The OA-ICOS liquid water analyzer rameters from ref. 6 and Innuitian ice sheet thinning from the analysis of ref. 12. was coupled to a CTC LC-PAL autosampler for simultaneous 18O/16O ratio

Regarding the latter, a high variance subensemble of the Bayesian calibration measurements of H2O. Following analysis, all measured water samples were specifically weighted to the Innuitian ice sheet was adopted; the best scoring verified for spectral contaminants in the samples using the LGR spectral in- model was used to determine the altitude corrections and the variance from the terference contamination identifier software. Analyses were calibrated and subensemble represents the uncertainty. Over the early Holocene, Innuitian ice normalized to internal laboratory water standards that were previously thinned by ∼400 m along southeast Ellesmere and so dominates the altitude calibrated relative to Vienna Standard Mean Ocean Water (VSMOW) using a correction. In contrast, the mid to late Holocene correction is dominated by conventional isotope ratio mass spectrometer. Consequently, the results are 18 GIA of ∼100 m (Figs. S2 and S3). Note that the δ O measurements were presented using the δ-notation (δ18O), which represents the parts per also corrected to account for changes in δ18O content of sea water (26). thousand differences for 18O/16O in a sample with respect to VSMOW. An- The altitude correction for the Agassiz δ18O record was applied through alytical reproducibility for δ18Ois± 0.3‰. the δ18O altitude relation derived from shallow ice cores [−0.62 ± 0.03 ‰ per The age–depth timescale for the A09 ice core is described in ref. 7. The 100 m (27, 28)]. A forward modeling method was used to calibrate a tem- A09 core includes a period of overlap (∼30 y) with the deep core melt (ref. 7, perature–δ18O relationship with the borehole temperature profiles from their figure 2) and δ18O records from Agassiz. The δ18O overlap shows that GRIP, North GRIP (NGRIP), DYE-3, and Camp Century by numerically solving the A09 core is properly aligned with the old deep cores and that there is a the differential equation for energy conservation in moving ice, yielding a smooth transition ensuring a homogeneous extended time series. Fig. S4 degrees Celsius/δ18O slope of 2.1 ± 0.2 (5, 6, 29). Upon deriving a corrected shows the high-resolution A84, A87, and A09 δ(18O) series. δ18O record, the temperature–δ18O slope conversion produces the annual air temperature reconstruction for a fixed elevation (present-day borehole el- Greenland Ice-Model Sensitivity Analysis. The North Greenland sensitivity evation), as presented in Fig. S3D. analysis was conducted using a glaciological model and GIA model of relative The Agassiz melt record was related to summer temperatures using a sea-level change (22). This previous study produced the Huy3 Greenland transfer function (7). Summer temperatures and melt layers across a number reconstruction, which was achieved by simultaneously tuning/calibrating a

5956 | www.pnas.org/cgi/doi/10.1073/pnas.1616287114 Lecavalier et al. Downloaded by guest on September 27, 2021 3D thermomechanical ice-sheet model in series with a GIA model. Please see for ice extent and RSL. Given the sensitivity of this proxy to ice extent, the ref. 22 for further details on the models applied and methodology followed. time at which maximum RSL occurred can be difficult to constrain and so We adopt the Huy3 reconstruction here and revise it by tuning the North introduces ambiguity into the model–data comparison. At the Kronprins Greenland climate forcing to reflect our Agassiz temperature reconstruction. site, the variant reconstruction of Huy3 achieves a perfect fit to the obser- Given that the ice model does not include the adjacent Innuitian ice sheet vational constraints by remaining below the upper-limiting and above the nor the buttressing effect of potential ice shelves in the Baffin Sea (both of lower-limiting RSL dates, including the highest lower-limiting RSL date and which would result in a thicker last glacial maximum ice sheet in the marine limit, which the original Huy3 model does not reach. In contrast at northwest sector), an alternative method was required to enhance past ice Jorgen, no model prediction achieves a fit to the RSL observations. The thickness in this region, as suggested by the Camp Century thinning curve. To model curves reach the uppermost lower-limiting dates but do not capture this end, we decided to vary the input precipitation field because it is poorly the rapid sea-level fall to reach the oldest upper-limiting date. However, the constrained and the parameterization schemes previously adopted lacked rapid sea-level fall observed at Jorgen is not observed at other neighboring the necessary degrees of freedom to account for these uncertainties. The sites, which might be indicative of local effects that are not captured in temperature forcing across North Greenland was parameterized to coincide regional GIA models. At Constable, the variant reconstruction shows a with the Agassiz temperature reconstruction. The parameterization of the marginally improved fit over the Huy3 chronology; it reaches the marine climate forcing was conducted in a similar manner to that of the Holocene limit, remains above the uppermost lower-limiting dates, passes through the thermal maximum in ref. 22. A tuning of the climate forcing was achieved by sea-level index point, and passes a few meters above the late Holocene adding linear and/or quadratic empirical equation estimates of the en- upper-limiting RSL dates. However, there is some conflicting data during the hanced temperatures at Agassiz compared with the GRIP temperature de- early to mid Holocene where upper- and lower-limiting dates and sea-level rived climate fields (Fig. 4A shows an example for the Camp Century index points suggest different sea level histories. The RSL predictions gen- location). This revision to the GRIP inferred temperatures was extended prior erated by the variant reconstruction achieve a significantly improved fit at a the start of the Agassiz record given the large uncertainties in the climate number of sites—JPKoch, Nyboe, Hall East, Hall West—compared with forcing and thus maintains the necessary degrees of freedom. Huy3 by remaining within the bounds of the limiting dates. The RSL data at Taking the variant model reconstruction, we computed RSL histories for sites Lafayette and Humboldt poorly constrain sea level and therefore do data sites in North Greenland (Fig. S6) to further test the accuracy of the not strongly discriminate between the two model reconstructions. Huy3 model revision. Even though the quality of RSL data from this region is We applied the same isostatic model to compute vertical land motion at sites in relatively poor (22), these observations are a primary constraint on model northwest Greenland where GPS receivers have been installed. Because the reconstructions of the ice history. RSL model predictions for the Huy3 model observed rates of land motion have been corrected for elastic deformation as- reconstruction and its variant produced here are compared with observations sociated with ice loading over the GPS monitoring period (24), the elastic con- in Fig. S7. At a number of RSL sites—Holm Land, Ingeborg Halvo, Herluf- tribution was removed from the model output. Comparing observed and sholm, Ole Chiewitz—both Huy3 and its variant reconstruction remain modeled rates in Table S1 indicates that the revised model is an improvement at within error of the observations when considering uncertainties in the earth all sites except those in the northeast (JGBL, KMJP, LEFN, NORD). viscosity structure (see figures 12 and S6 from ref. 22). It should be noted that there is some ambiguity when interpreting the marine limits (i.e., ACKNOWLEDGMENTS. This work was funded by the Natural Sciences and maximum RSL at a given site) because this quantity represents the highest Engineering Research Council of Canada. This paper is a contribution to the point reached by sea level when it is ice free, and so represents a dual proxy PAGES/INQUA funded PALSEA2 working group.

1. Stocker TF, et al. (2013) IPCC, 2013: Climate Change 2013: The Physical Science Basis. 17. Mayewski Pa, et al. (1997) Major features and forcing of high-latitude northern Contribution of Working Group I to the Fifth Assessment Report of the Intergovernmental hemisphere atmospheric circulation using a 110,000-year-long glaciochemical series. Panel on Climate Change (Cambridge Univ Press, Cambridge). J Geophys Res 102:26345–26366. 2. Marcott SA, Shakun JD, Clark PU, Mix AC (2013) A reconstruction of regional and 18. Dyke AS, Hooper J, Savelle JM (1996) A history of sea ice in the Canadian Arctic ar- global temperature for the past 11,300 years. Science 339:1198–1201. chipelago based on postglacial remains of the bowhead whale (Balaena mysticetus). 3. Masson-Delmotte V, et al. (2013) Information from paleoclimate archives. Climate Arctic 49:235–255. Change 2013: The Physical Science Basis. Contribution of Working Group I to the 19. Dyke AS, Hooper J, Harington CR, Savelle JM (1999) The late Wisconsinan and Ho- Fifth Assessment Report of the Intergovernmental Panel on Climate Change (Cam- locene record of Walrus (Odobenus rosmarus) from North America: A review with – bridge Univ Press, Cambridge), pp 383–464. new data from Arctic and Atlantic Canada. Arctic 52:160 181. 4. Bekryaev RV, Polyakov IV, Alexeev VA (2010) Role of polar amplification in long-term 20. Pithan F, Mauritsen T (2014) Arctic amplification dominated by temperature feed- – surface air temperature variations and modern arctic warming. J Clim 23:3888–3906. backs in contemporary climate models. Nat Geosci 7:2 5. 5. Vinther BM, et al. (2009) Holocene thinning of the Greenland ice sheet. Nature 461: 21. Collins M, et al. (2013) Long-term climate change: Projections, commitments and ir- 385–388. reversibility. Climate Change 2013: The Physical Science Basis. Contribution of 6. Lecavalier BS, et al. (2013) Revised estimates of Greenland ice sheet thinning histories Working Group I to the Fifth Assessment Report of the Intergovernmental Panel on Climate Change (Cambridge Univ Press, Cambridge), pp 1029–1136. based on ice-core records. Quat Sci Rev 63:73–82. 22. Lecavalier BS, et al. (2014) A model of Greenland ice sheet deglaciation constrained 7. Fisher D, et al. (2012) Recent melt rates of Canadian arctic ice caps are the highest in by observations of relative sea level and ice extent. Quat Sci Rev 102:54–84. four millennia. Global Planet Change 84–85:3–7. 23. MacGregor JA, et al. (2016) Holocene deceleration of the Greenland Ice Sheet. Science 8. Fisher DA, Koerner RM, Reeh N (1995) Holocene climatic records from Agassiz Ice Cap, 351:590–593. Ellesmere Island, NWT, Canada. Holocene 5:19–24. 24. Khan SA, et al. (2016) Geodetic measurements reveal similarities between post-Last 9. Koerner RM, Fisher DA (1990) A record of Holocene summer climate from a Canadian Glacial Maximum and present-day mass loss from the Greenland ice sheet. Sci Adv high-Arctic ice core. Nature 343:630–631. 2:e1600931. 10. Fisher DA (1990) A zonally-averaged stable-isotope model coupled to a regional 25. Shepherd A, et al. (2012) A reconciled estimate of ice-sheet mass balance. Science 338: variable-elevation stable-isotope model. Ann Glaciol 14:65–71. 1183–1189. 11. Fisher DA (1992) Stable isotope simulations using a regional stable isotope model 26. Waelbroeck C, et al. (2002) Sea-level and deep water temperature changes derived coupled to a zonally averaged global model. Cold Reg Sci Technol 21:61–77. from benthic foraminifera isotopic records. Quat Sci Rev 21:295–305. 12. Tarasov L, Dyke AS, Neal RM, Peltier WR (2012) A data-calibrated distribution of 27. Dahl-Jensen D, et al.; NEEM community members (2013) Eemian interglacial re- deglacial chronologies for the North American ice complex from glaciological mod- constructed from a Greenland folded ice core. Nature 493:489–494. – eling. Earth Planet Sci Lett 315:30 40. 28. Johnsen SJ, Dansgaard W, White JWC (1989) The origin of Arctic precipitation under 13. Renssen H, Seppä H, Crosta X, Goosse H, Roche DM (2012) Global characterization of present and glacial conditions. Tellus B Chem Phys Meterol 41:452–468. – the Holocene Thermal Maximum. Quat Sci Rev 48:7 19. 29. Dahl-Jensen D, et al. (1998) Past temperatures directly from the Greenland ice sheet. 14. Lacelle D, Pellerin A, Clark ID, Lauriol B, Fortin D (2009) (Micro)morphological, Science 282:268–271. SCIENCES

inorganic-organic isotope geochemisty and microbial populations in endo- 30. Marshall SJ, Sharp MJ, Burgess DO, Anslow FS (2007) Near-surface-temperature lapse ENVIRONMENTAL stromatolites (cf. fissure calcretes), Haughton impact structure, Devon Island, Canada: rates on the Prince of Wales Icefield, Ellesmere Island, Canada: Implications for re- The influence of geochemical pathways on the preservation of isotope bioma. Earth gional downscaling of temperature. Int J Climatol 27:385–398. Planet Sci Lett 281:202–214. 31. Tarasov L, Peltier WR (2002) Greenland glacial history and local geodynamic conse- 15. Bourgeois JC, Koerner RM, Gajewski K, Fisher DA (2000) A Holocene ice-core pollen quences. Geophysical Journal International, 150:198–229. record from Ellesmere Island, Nunavut, Canada. Quat Res 54:275–283. 32. Simpson MJ, Milne GA, Huybrechts P, Long AJ (2009) Calibrating a glaciological model 16. Fisher DA, et al. (1998) Penny ice cap cores, Baffin Island, Canada, and the Wisconsinan of the Greenland ice sheet from the Last Glacial Maximum to present-day using field foxe dome connection: Two states of Hudson Bay ice cover. Science 279:692–695. observations of relative sea level and ice extent. Quat Sci Rev 28:1631–1657.

Lecavalier et al. PNAS | June 6, 2017 | vol. 114 | no. 23 | 5957 Downloaded by guest on September 27, 2021