After more than a century of geological research, the Cordilleran ice sheet of North America remains among the least understood in terms of its former extent, volume, and dynamics. Because of the mountainous topography on which the ice sheet formed, geological studies have often had only local or regional relevance and shown such a complexity that ice-sheet-wide spatial reconstructions of advance and retreat patterns are lacking. Here we use a numerical ice sheet model calibrated against field-based evidence to attempt a quantitative reconstruction of the Cordilleran ice sheet history through the last glacial cycle. A series of simulations is driven by time-dependent temperature offsets from six proxy records located around the globe. Although this approach reveals large variations in model response to evolving climate forcing, all simulations produce two major glaciations during marine oxygen isotope stages 4 (62.2–56.9 ka) and 2 (23.2–16.9 ka). The timing of glaciation is better reproduced using temperature reconstructions from Greenland and Antarctic ice cores than from regional oceanic sediment cores. During most of the last glacial cycle, the modelled ice cover is discontinuous and restricted to high mountain areas. However, widespread precipitation over the Skeena Mountains favours the persistence of a central ice dome throughout the glacial cycle. It acts as a nucleation centre before the Last Glacial Maximum and hosts the last remains of Cordilleran ice until the middle Holocene (6.7 ka).
During the last glacial cycle, glaciers and ice caps of the North American
Cordillera have been more extensive than today. At the Last Glacial Maximum
(LGM), a continuous blanket of ice, the Cordilleran ice sheet
Relief map of the northern American Cordillera
showing cumulative last glacial maximum ice cover between 21.4 and
16.8
More than a century of exploration and geological investigation of the
Cordilleran mountains have led to many observations in support of the former
ice sheet
Our understanding of the Cordilleran glaciation history prior to the LGM is
even more fragmentary
In contrast, evidence for the deglaciation history of the Cordilleran ice
sheet since the LGM is considerable, albeit mostly at a regional scale.
Geomorphological evidence from south-central British Columbia indicates
a rapid deglaciation, including an early emergence of elevated areas while
thin, stagnant ice still covered the surrounding lowlands
Ice softness parameter,
Default parameter values used in the ice sheet model.
In general, the topographic complexity of the North American Cordillera and
its effect on glacial history have inhibited the reconstruction of ice-sheet-wide glacial advance and retreat patterns such as those available for
the Fennoscandian and Laurentide ice sheets How much ice was locked in the Cordilleran ice sheet during the LGM? What was the scale of glaciation prior to the LGM? Which were the primary dispersal centres? Do they reflect stable or
ephemeral configurations? How rapid was the last deglaciation? Did it include Late Glacial
standstills or readvances?
Although numerical ice sheet modelling has been established as a useful tool
to improve our understanding of the Cordilleran ice sheet
(
Our palaeoclimate forcing therefore includes spatial temperature and
precipitation grids derived from a present-day atmospheric reanalysis
Palaeotemperature proxy records and scaling
parameters yielding temperature offset time series used to force the ice
sheet model through the last glacial cycle (Fig.
Although these proxy records were all obtained outside the model domain, more
regional palaeotemperature reconstructions spanning over the last glacial
cycle are lacking. For instance, the Mount Logan ice core oxygen isotope
record covers only the last 30 000
Parameter values used in the sensitivity test.
After
After testing the model sensitivity to these climate forcing time series
(Sect.
The simulations presented here were run using the Parallel Ice Sheet Model
(PISM, version 0.7.2), an open source, finite difference, shallow ice sheet
model
Basal topography is bilinearly interpolated from the ETOPO1 combined
topography and bathymetry data set with a resolution of 1 arcmin
Ice deformation follows temperature and water-content-dependent creep
(Sect.
Each simulation starts from assumed ice-free conditions at 120
Ice sheet dynamics are typically modelled using a combination of internal
deformation and basal sliding. PISM is a shallow ice sheet model, which
implies that the balance of stresses is approximated based on their
predominant components. The shallow shelf approximation (SSA) is combined
with the shallow ice approximation (SIA) by adding velocity solutions of the
two approximations
Ice deformation is governed by the constitutive law for ice
In all our simulations, we set constant the power-law exponent,
Contrarily, we test the model sensitivity (Sect. Our Our Our
An additional simulation using the ice rheology from
Actual parameter values for
Surface air temperature derived from the climate forcing
(Sect.
A pseudo-plastic sliding law,
In all our simulations, we set constant the pseudo-plastic sliding exponent,
We also use a constant spatial distribution of the till friction angle,
An additional simulation with a constant till friction angle,
Additionally, we test the model sensitivity (Sect. Our Our Our
The effect of the three different parametrisations on the effective pressure
on the till,
Ice shelf calving is computed using a double criterion. First,
a physically realistic calving flux is computed based on eigenvalues of the
horizontal strain rate tensor
Effective pressure,
Ice surface accumulation and ablation are computed from monthly mean
near-surface air temperature,
The PDD computation accounts for stochastic temperature variations by
assuming a normal temperature distribution of standard deviation
Extremes in Cordilleran ice sheet grounded ice
extent and sea-level relevant ice volume corresponding to MIS 4, 3, and 2 for
each of the six low-resolution simulations (Fig.
Climate forcing driving ice sheet simulations consists of a present-day
monthly climatology,
Monthly mean near-surface air temperature,
precipitation, and standard deviation of daily mean temperature (PDD SD) for
January and July from the North American Regional Reanalysis
Temperature offset time series from ice core and
ocean records (Table
Snapshots of modelled surface topography
(500
Modelled sea-level relevant ice volume through
the last 120 ka in the simulation forced by the GRIP palaeoclimate record,
using default parameters (black curves), different ice rheology parameters
(top panel), and different basal sliding parameters (bottom panel). Grey
fields indicate marine oxygen isotope stage (MIS) boundaries for MIS 2 and
MIS 4 according to a global compilation of benthic
Temperature offset time series,
Finally, lapse-rate corrections,
Despite large differences in the input climate forcing (Fig.
Simulations forced by the Greenland ice core palaeotemperature records (GRIP, NGRIP) produce the highest variability in modelled ice volume throughout the last glacial cycle. In contrast, simulations driven by oceanic (ODP 1012, ODP 1020) and Antarctic (EPICA, Vostok) palaeotemperature records generally result in lower ice volume variability throughout the simulation length, resulting in lower modelled ice volumes during MIS 4 and larger ice volumes during MIS 3. The NGRIP climate forcing is the only one that results in a larger ice volume during MIS 4 than during MIS 2.
Extremes in Cordilleran ice sheet grounded
ice extent and sea-level
relevant ice volume corresponding to
MIS 4, 3, and 2 using the GRIP palaeoclimate forcing with each
parameter configuration (Fig. 3). Relative differences (R. diff.) give
rough error estimates related to varying selected ice rheology and
basal sliding parameters (Table
While simulations driven by the GRIP and the two Antarctic palaeotemperature
records attain a last ice volume maximum between 19.1 and 16.9
Despite large differences in the timing of attained volume extrema
(Table
During MIS 3 ice volume minimum reconstructions, a central ice cap persists
over the Skeena Mountains (Fig.
Modelled ice sheet geometries during the LGM (MIS 2; Fig.
Using the GRIP ice core palaeotemperature record as climate forcing
time series, the model shows a significant sensitivity to selected ice
rheological and basal sliding parameters (Fig.
As a direct result of the different scaling factors applied
(Table
The modelled sea-level potentials show a stronger variability than the
modelled glaciated areas (Table
Large variations in the model responses to evolving climate forcing reveal its sensitivity to the choice of palaeotemperature proxy record. To distinguish between different records, geological evidence of former glaciations provides a basis for validation of our runs, while the results from numerical modelling can perhaps help to analyse some of the complexity of this evidence. In this section, we compare model outputs to the geologic record, in terms of timing and configuration of the maximum stages, location and lifetime of major nucleation centres, and patterns of ice retreat during the last deglaciation.
Independently of the palaeotemperature records used to force the ice sheet
model, our simulations consistently produce two glacial maxima during the
last glacial cycle. The first maximum configuration is obtained during MIS 4
(62.2–56.9
The exact timing of modelled MIS 2 maximum ice volume depends strongly on the
choice of applied palaeotemperature record, which allows for a more in-depth
comparison with geological evidence for the timing of the maximum Cordilleran
ice sheet extent. In the Puget Lowland (Fig.
Among the simulations presented here, only those forced with the GRIP, EPICA,
and Vostok palaeotemperature records yield Cordilleran ice sheet maximum
extents that may be compatible with these field constraints
(Fig.
Modelled surface topography (200
Because most of the marine margin of the Cordilleran ice sheet terminated in
a sector of the Pacific Ocean unaffected by variations in the California
current, it probably remained insensitive to this local phenomenon. However,
the above paradox illustrates the complexity of ice-sheet feedbacks on
regional climate and demonstrates that, although located in the
neighbourhood of the modelling domain, the ODP 1012 and ODP 1020
palaeotemperature records cannot be used as a realistic forcing to model the
Cordilleran ice sheet through the last glacial cycle. Similarly, the
simulation using the NGRIP palaeotemperature record depicts an early onset
of deglaciation (Fig.
During maximum glaciation, both simulations position the main meridional ice
divide over the western flank of the Rocky Mountains
(Figs.
Such deviation from the geological inferences could reflect the fact that the
NARR has difficulties with simulating orographic processes in some areas of
steep topography
Since the model does not include feedback mechanisms between the ice sheet topography and the regional climate, the modelled easterly positions of the ice divide and eastern ice sheet margin may also be sensitive to the assumption of fixed modern-day spatial patterns of air temperature and precipitation. In fact, it is reasonable to think that the cooling during the last glacial cycle was greater inland than near the coast, prohibiting melt at the eastern margin. However, our simulations already produce an excess of ice inland. Including such a temperature continentality gradient in the model while keeping the precipitation pattern constant would thus cause an even greater mismatch between the model results and the geologically reconstructed ice margins during the LGM.
Consequently, the mismatch between the modelled and reconstructed LGM ice
margins is likely due to the assumption of the fixed modern-day precipitation
patterns rather than the assumption of the fixed modern-day temperature
patterns. Firstly, during the build-up phase preceding the LGM, rapid
accumulation over the Coast Mountains enhanced the topographic barrier formed
by these mountain ranges, which likely resulted in a decrease of
precipitation and, therefore, a decrease of accumulation in the interior.
Secondly, latent warming of the moisture-depleted air parcels flowing over
this enhanced topography could have resulted in an inflow of potentially
warmer air over the eastern flank of the ice sheet, thereby counterbalancing
the potential continentality gradient discussed above through increasing melt
along the advancing margin
However, field-based palaeoglaciological reconstructions have struggled to
reconcile the more westerly centred ice divide in south-central British
Columbia with evidence in the Rocky Mountains and beyond that indicates the
Cordilleran ice sheet invaded the western Interior Plains, where it merged
with the south-western margin of the Laurentide ice sheet and was deflected to
the south
During MIS 2, the modelled sea-level potential peaks at
8.62
Modelled surface topography (200
The modelled ice sheet configurations corresponding to ice volume maxima
during MIS 4 are more sensitive to the choice of atmospheric forcing than
those corresponding to ice volume maxima during MIS 2
(Figs.
During MIS 4, the modelled sea-level potential peaks at
7.43
Palaeoglaciological reconstructions are generally more robust for maximum
ice sheet extents and late ice sheet configurations than for intermediate or
minimum ice sheet extents and older ice sheet configurations
Modelled duration of ice cover during
the last 120
For the Cordilleran ice sheet, geological evidence from radiocarbon dating
The resulting maps show that, during most of the glacial cycle, modelled ice
cover is restricted to disjoint ice caps centred on major mountain ranges of
the North American Cordillera (Fig.
A notable exception to the transient character of the maximum extent of the
Cordilleran ice sheet is the northern slope of the Alaska Range, where
modelled glaciers are confined to its foothills during the entire simulation
period (Fig.
It is generally believed that the Cordilleran ice sheet formed by the
coalescence of several mountain-centred ice caps
The location of the modelled ice-dispersal centres is potentially biased by
the present-day ice volumes contained in the ETOPO1 basal topography data.
The most problematic part of the model domain in this respect is that of the
Wrangell and Saint Elias mountains, where ice thicknesses of up to 1200 m
have been measured by a low-frequency radar
Although the Coast, Skeena, Columbia, and Rocky mountains are
covered by mountain glaciers for most of the last glacial cycle, providing
durable nucleation centres for an ice sheet, this is not the case for the
Selwyn and Mackenzie mountains, where ice cover on the highest peaks
is limited to a small fraction of the last glacial cycle. In other words, the
Selwyn and Mackenzie mountains only appear as a secondary ice-dispersal
centre during the coldest periods of the last glacial cycle. The Northern
Rocky Mountains (Fig.
Perhaps the most striking feature displayed by the distributions of modelled
ice cover is the persistence of the Skeena Mountains ice cap throughout the
entire last glacial period (ca. 100–10
The presence of a Skeena Mountains ice cap during most of the last glacial
cycle can be explained by meteorological conditions more favourable for ice
growth there than elsewhere. In fact, reanalysed atmospheric fields used to
force the surface mass balance model show that high winter precipitations are
mainly confined to the western slope of the Coast Mountains, except in the
centre of the modelling domain where they also occur further inland than
along other east–west transects (Fig.
The modelled sea-level potential corresponding to these persistent
ice-dispersal centres attains a minimum of 1.54
Modelled duration of warm-based ice
cover during the last 120
A correlation is observed between the modelled duration of warm-based ice
cover (Fig.
The modelled distribution of warm-based ice cover
(Figs.
Modelled fraction of warm-based ice cover during the ice-covered period. Note the dominance of warm-based conditions on the continental shelf and major glacial troughs of the coastal ranges. Hatches indicate areas that were covered by cold ice only.
Similarly to other glaciated regions, most glacial traces in the North
American Cordillera relate to the last few millennia of glaciation, because
most of the older evidence has been overprinted by warm-based ice retreat
during the last deglaciation
In the North American Cordillera, the presence of lateral meltwater channels
at high elevation
Temperature offset time series from the GRIP
and EPICA ice core records (Table
In our simulations, the timing of peak ice volume during the LGM and the
pacing of deglaciation depend critically on the choice of climate forcing
(Table
Snapshots of modelled surface topography
(200
Modelled age of the last deglaciation.
Areas that have been covered only before the last glacial maximum are marked
in green. Hatches denote readvance of mountain-centred ice caps and the
decaying ice sheet between 14 and 10
Hence, the two model runs, while similar in overall timing compared to the
runs with other climate drivers, differ in detail. On the one hand, the EPICA
run depicts peak glaciation about 2
Modelled patterns of ice sheet retreat are relatively consistent between the
two simulations (Figs.
Modelled deglacial basal ice velocities.
Hatches indicate areas that remain non-sliding throughout deglaciation
(22.0–8.0
The possibility of late-glacial readvances in the North American Cordillera
has been debated for some time
Although further work is needed to constrain the timing of the late-glacial
readvance, to assess its extents and geographical distribution and to
identify the potential climatic triggers
Because a general conjecture in glacial geomorphology is that the majority of
landforms (lineations and eskers) are part of the deglacial envelope
Patterns of glacial lineations formed in the northern and southern sectors of
the Cordilleran ice sheet (
On the Interior Plateau of south-central British Columbia, both simulations
produce a retreat of the ice margin towards the north-east
(Fig.
The modelled deglaciation of the Interior Plateau of British Columbia
consists of a rapid northwards retreat (Fig.
Modelled bedrock (black) and ice surface
(blue) topography profiles during deglaciation (22.0–8.0
Modelled bedrock (black) and ice surface
(red) topography profiles during deglaciation (22.0–8.0
Numerical simulations of the Cordilleran ice sheet through the last glacial
cycle presented in this study consistently produce two glacial maxima during
MIS 4 (62.2–56.9
In all simulations presented here, ice cover is limited to disjoint mountain ice caps during most of the glacial cycle. The most persistent nucleation centres are located in the Coast Mountains, the Columbia and Rocky mountains, the Selwyn and Mackenzie mountains, and most importantly, in the Skeena Mountains. Throughout the modelled last glacial cycle, the Skeena Mountains host an ice cap which appears to be fed by the moisture intruding inland from the west through a topographic breach in the Coast Mountains. The Skeena ice cap acts as the main nucleation centre for the glacial readvance towards the LGM configuration. As indicated by persistent, warm-based ice in the model, this ice cap perhaps explains the distinct glacial erosional imprint observed on the landscape of the Skeena Mountains.
During deglaciation, none of the climate records used can be selected as
producing an optimal agreement between the model results and the geological
evidence. Although the EPICA-driven simulation yields the most realistic
timing of the LGM and, therefore, start of deglaciation, only the GRIP-driven
simulation produces late-glacial readvances in areas where these have been
documented. Nonetheless, the patterns of ice sheet retreat are consistent
between the two simulations and show a rapid deglaciation of the southern
sector of the ice sheet, including a rapid northwards retreat across the
Interior Plateau of central British Columbia. The GRIP-driven simulation then
produces a late-glacial readvance of local ice caps and of the main body of
the decaying Cordilleran ice sheet primarily in the Coast and the Columbia
and Rocky Mountains. In both simulations, this is followed by an opening of
the Liard Lowland and a final retreat of the remaining ice caps towards the
Selwyn and, finally, the Skeena mountains, which host the last remnant of
the ice sheet during the middle Holocene (6.7
J. Seguinot ran the simulations; I. Rogozhina guided experiment design; A. P. Stroeven, M. Margold and J. Kleman took part in the interpretation and comparison of model results against geological evidence. All authors contributed to the text.
Foremost, we would like to thank Shawn Marshall for providing a detailed, constructive analysis of this study during J. Seguinot's PhD defence (September 2014). His comments were used to improve the model set-up. We are very thankful to Constantine Khroulev, Ed Bueler, and Andy Aschwanden for providing constant help and development with PISM. We thank Shawn Marshall, Alexander Jarosch, and Andrew Stumpf for their constructive and complementary reviews. This work was supported by the Swedish Research Council (VR) grant no. 2008-3449 to A. P. Stroeven and by the German Academic Exchange Service (DAAD) grant no. 50015537 and a Knut and Alice Wallenberg Foundation grant to J. Seguinot. Computer resources were provided by the Swedish National Infrastructure for Computing (SNIC) allocation no. 2013/1-159 and 2014/1-159 to A. P. Stroeven at the National Supercomputing Center (NSC) and by the Swiss National Supercomputing Centre (CSCS) allocation no. s573 to J. Seguinot.Edited by: G. Hilmar Gudmundsson