The final set of ISMIP6 experiments will use standalone ice-sheet models
driven by climate model output and other datasets. Groups and models that
have expressed an interest in participating in this aspect of ISMIP6 are
listed in Table 4. The models participating in this effort will likely be
configured differently from those in the ism-XXX-self simulations
described in Sect. 3.2. For example, an ice-sheet model that is spun up to
quasi-equilibrium with a climate model will likely have a thickness and
extent that differ appreciably from observed values, whereas standalone
models can be initialized more realistically. Also, an ISM in a climate model
might use a coarse resolution or a simple approximation of ice dynamics in
order to be more computationally efficient, while the same model used
strictly for projections would likely have a finer resolution, at least in
regions of fast flow (e.g., Aschwanden et al., 2016), and could incorporate
more complex ice-flow dynamics. Similarly, ice-sheet models that are used for
paleoclimate studies are often distinct from those used for projections of a
few hundred years.
initMIP
The initMIP ice-sheet experiments are designed to explore uncertainties in
sea-level projections associated with model initialization and spinup. Such
uncertainties have been identified by previous model intercomparison efforts
(e.g., Bindschadler et al., 2013; Nowicki et al., 2013a, b; Edwards et al., 2014a, Shannon et al.,
2013; Goelzer et al., 2013; Gillet-Chaulet et al., 2012) and include the
impacts of model initial conditions, sub-grid-scale processes, and poorly
known parameters. The initMIP project aims to evaluate initialization
procedures, to estimate trends caused by model initializations, and to
investigate the impact of choices in numerical and physical parameters (e.g.,
stress balance approximation or model resolution). Results of the initMIP
project are expected to point to specific aspects of ice-sheet initialization
that have a crucial impact on sea-level projections and that may be improved.
ISM initialization methods to present-day conditions range from running
paleo-climate spinup for thousands of years (e.g., Martin et al., 2011; Sato
and Greve, 2012; Aschwanden et al., 2013; Fürst et al., 2015; Saito et
al., 2016) to assimilating present-day observations (e.g., Morlighem et al.,
2010; Gillet-Chaulet et al., 2012; Seroussi et al., 2013, Arthern et al.,
2015). The choices made in this procedure affect ice-sheet extent, flow
rates, volume, and volume trends, which can have substantial effects on
estimates of ice-sheet contribution to sea-level rise (e.g.,
Aðalgeirsdóttir et al., 2014). Improving ISM initial conditions is an
active area of research and a multidisciplinary effort. It requires
acquisition of additional data with high spatial coverage over entire ice
sheets and at increased resolution (e.g., Bamber et al., 2013; Rignot et al.,
2011b; Joughin et al., 2010a; Howat et al., 2014). Ideally, all datasets used
in the data assimilation are from the same period, as initializing an
ice-sheet model with datasets taken at different times can cause the ice-flow
model to artificially redistribute the glacier mass in unrealistic ways that
serve only to reconcile these inconsistencies (Seroussi et al., 2011). This
also implies that the date associated with the initial state can differ
between models based on the datasets used. New algorithms that reconcile
initialization datasets are being developed, most notably for bedrock
elevation (e.g., Morlighem et al., 2011, 2014), which is notoriously poorly
constrained.
The initMIP project consists of a Greenland component and an Antarctic
component. Following initialization, there is a set of two forward
experiments for the Greenland Ice Sheet and three forward experiments for the
Antarctic Ice Sheet, each run for at least 100 years: (i) a control run
(ism-ctrl-std), (ii) a surface mass balance anomaly run
(ism-asmb-std), and (iii) a basal melt anomaly run
(ism-abmb-std) in which anomalous melt is applied beneath the
floating portion of the Antarctic Ice Sheet. All other model parameters and
forcing in the forward runs are the same as those used for initialization.
The ism-ctrl-std is an unforced forward experiment designed to
evaluate the initialization procedure and characterize model drift, the
surface mass balance remaining identical to the one used during the
initialization procedure. In ism-asmb-std, a prescribed SMB anomaly
is applied to test the model response to a large perturbation. The schematic
perturbation anomaly mimics outputs of several SMB models of different
complexity between the end of the 20th century and the end of the 21th
century, and is designed to capture the first-order pattern of SMB changes
expected from climate models. The schematic SMB anomalies are defined
everywhere on the model grid, and are therefore applicable for models with
varying ice-sheet extent. In ism-abmb-std, a prescribed anomaly of
basal melting rate under floating ice is applied while SMB is kept the same
as in ism-ctrl-std. Because of the difference in ice-shelf extent
between the different models, the basal melt anomaly is prescribed to be
constant for each basin. This scalar value is different for each basin and
derived from the mean values of the ice-shelf melt observed by Rignot et
al. (2013) and Depoorter et al. (2013). The applied anomaly simulates a
doubling of sub-ice-shelf melting after 40 years of simulation for models
with initial melting rates close to today's observations.
Since these experiments are designed to allow comparison among the different
models, some simplifications are imposed. Neither SMB nor bedrock topography
should be adjusted in response to ice-sheet geometric changes in these
forward experiments. However, to sample the uncertainty in sea level due to
initialization, groups are encouraged to submit multiple variations of the
experiment, for example by changing the sliding law, stress balance
approximation, model resolution, or datasets (such as using different
bedrocks). While the initialization procedures used by the different
participating groups are not prescribed by ISMIP6, it is expected that
individual groups will take advantage of the initMIP results to improve their
initialization procedures. initMIP is also intended to give ice-sheet
modelers an opportunity to get involved in ISMIP6 at an early stage, before
outputs of CMIP6 AOGCM become available; hence our prescription of simplified
anomalies. We refer interested readers to the initMIP webpage
(http://www.climate-cryosphere.org/wiki/index.php?title=_InitMIP) for
more information.
<italic>ism-XXX-std</italic> configuration
The ism-XXX-std experiments target primarily the glaciology
community and seek to obtain realistic ice-sheet evolution to inform
estimates of past, present, and future sea levels. ISMIP6 will supply forcing
data from CMIP6 that allow standalone ISMs to simulate the evolution of both
the Greenland and Antarctic ice sheets. ISMIP6 seeks to assess the
uncertainty in sea-level change arising from both the ice-sheet models and
the climate forcing. A key concern is that ISMIP6 assess uncertainty
associated with emission scenarios and the AOGCMs' simulation of these
scenarios: for a given emission scenario, the AOGCMs' simulation of this
scenario will result in a range of atmospheric and oceanic forcings. Clearly,
there is a tension between the range of potential ice-sheet forcing, the need
to explore uncertainty associated purely with ISMs (e.g., related to initial
conditions, bedrock topography, and parametric uncertainty), and the
computing requirements of specific ISMs (some of which may only be able to
perform a small number of experiments). To this end, we anticipate
identifying a subset of forcing from the CMIP6 AOGCM ensemble based on the
analysis of AOGCM simulations of ice-sheet climate (Sect. 3.1). The subset
will be chosen to capture the full range of potential ice-sheet forcing for a
given emission scenario, using metrics of the SMB and ocean forcing to
investigate that range. Within the selected subset of forcing, we plan to
identify a small number of simulations that all ISMs must perform. Groups
that are able to perform numerous simulations will be encouraged to
participate in all experiments. Shannon et al. (2013) is an example of this
approach.
The forcing data can naturally be divided into atmospheric and oceanic
forcing. Central to the former is the means to determine SMB associated with
a particular CMIP6 experiment. Several methods have previously been employed
to do this. Until we can assess the quality of the climate simulated by CMIP6
AOGCMs above and around the ice sheets (after the analysis of the CMIP6 DECK
and Historical simulations), a definitive choice cannot be made. However, we
list the options in order of preference.
Use the SMB calculated by the AOGCM directly. This has the advantage that
the SMB will be entirely consistent with other parts of that AOGCM's
simulation of climate. There is concern, however, that the quality of the
SMB computed by the AOGCMs will make this approach unrealistic due primarily
to the mismatch between the spatial resolution of AOGCMs and the
characteristic length scale of variations in SMB. Several groups have,
however, made recent progress in this area (e.g., Vizcaíno et al.,
2013; Lipscomb et al., 2013). The use of anomalies should also be considered
in this context.
In the event that AOGCM-determined SMB is shown to be inadequate, an
intermediate step is required. Previously, this has been the use of Regional
Climate Models (RCMs) to simulate SMB. For example, the ice2sea effort chose
to generate SMB from an RCM (Edwards et al., 2014a, b; Fettweis et al.,
2013). This approach, however, introduces a further link into the processing
chain that may lead to delay in the production of sea-level projections. It
also introduces the issue of choice of RCM and whether results from a number
of RCMs should be used (further complicating the design of the ISM
ensemble). Furthermore, the use of RCMs as intermediaries between AOGCMs and
ISMs adds ambiguity about which biases are introduced by the AOGCMs and
which biases are the result of the RCMs.
Use a parameterization or simplified process model to simulate SMB by
downscaling atmospheric forcing over the ice sheet from an AOGCM. This
approach was used by SeaRISE (Bindschadler et al., 2013), where the
precipitation and surface temperature from 18 AOGCMs models taking part in
the A1B scenario were combined to generate monthly mean values. These mean
precipitation and temperature values were then passed to the SMB scheme of
the ice-sheet model (generally a PDD method that accounted for the
temperature aspect of the SMB–elevation feedback) to obtain SMB anomalies
that were added to the ice-sheet surface conditions at initialization.
A further consideration is that the AOGCM models assume a fixed ice-sheet
elevation: i.e., they neglect the effect of ice-sheet elevation change on the
atmosphere and hence omit the SMB–elevation feedback. Standalone ISMs will
need to include this effect by parameterizing the SMB lapse rate (Edwards et
al., 2014a, b; Fettweis et al., 2013; Goelzer et al., 2013). This approach
may be less of an issue for method 3 above because SMB is determined
interactively within the ISM rather than being prescribed as forcing.
A second way in which the atmosphere could force dynamic change in ice sheets
is through the production of large quantities of meltwater. Mechanisms have
been proposed that link meltwater to both ice-shelf collapse (Banwell et al.,
2013) and enhanced lubrication of ice flow (Zwally et al., 2002) (although
recent modeling studies suggest a minor influence of the latter on
large-scale ice flow; e.g., Shannon et al., 2013). Surface air temperature
and runoff forcing will therefore also be made available.
Both Antarctica and Greenland are thought to respond to changes in proximal
ocean temperatures, which affect the melt rates of floating ice shelves and
the vertical faces of outlet glaciers. Obtaining suitable oceanic forcing
from CMIP6 climate models will be a major challenge. Few CMIP6 models will
calculate the appropriate melt rates, and even these results are likely to
be inaccurate because of issues of model resolution and the unique physics
of ocean circulation adjacent to melting ice. Melt rates will therefore need
to be determined outside the climate model using an index for proximal ocean
temperature. This index is most likely to be water temperature (and
salinity) at the continental shelf break at an intermediate range of depths
(equivalent to the base of ice shelves or the depth of ice grounded on
bedrock). This quantity will be included in our evaluation of CMIP6 forcing
(see Sect. 3.1).
A wide range of approaches has been used to calculate the required melt rate
from prescribed ocean-temperature forcing. The simplest method is to
calculate melt rate anomalies from changes in the nearest ocean temperature
using an observationally derived relation of
10 m yr-1 ∘C-1 (Rignot and Jacobs, 2002). However, this
linear relation between ocean temperature and melt rates is calibrated for
melt rates at the grounding line, and likely is missing important
nonlinearities (Holland et al., 2008). An alternative approach is to
parameterize melt rates as proportional to the difference between ocean
temperature at the shelf break and the freezing temperature at the ice-shelf
base. Beckman and Goosse (2003) developed such a scheme for ocean models, and
similar schemes have been applied in offline ice-sheet model simulations with
idealized ocean forcing (e.g., Martin et al., 2011; Pollard and DeConto,
2012; DeConto and Pollard, 2016). In those studies, the ocean temperature is
set to the average temperature between 200 and 600 m depth (Martin et al.,
2011) or the temperature at 400 m depth (DeConto and Pollard, 2016), or
specified differently for specific Antarctic sectors (Pollard and DeConto,
2012). Depending on the evaluation of the CMIP6 models, ISMIP6 may adapt one
of these choices, or could prescribe depth-varying profiles of ocean
temperature (and possibly salinity). The dependence of melt rates on thermal
driving ranges from linear (Martin et al., 2011) to quadratic (Pollard and
DeConto, 2012; DeConto and Pollard, 2016). Since the freezing temperature at
the ice base decreases with depth, the melt rates in all schemes tend to be
higher near grounding lines, as found from observations.
If none of the CMIP6 ocean models can accurately capture the broad-scale
polar ocean circulation or produce realistic near-shelf temperatures, an
alternative is to prescribe a melt rate that simply depends on the ice-shelf
draft (e.g., Joughin et al., 2010b; Favier et al., 2014). This approach is
less satisfactory, however, as it ignores temporal changes in ocean
conditions, and typically uses coefficients calibrated to local thermal
conditions. If ISMIP6 uses this approach, the provided coefficients would not
be uniform but would take into account the fact that ocean waters reaching
ice-shelf cavities or fronts differ regionally. In Antarctica, for example,
the ice shelves of Pine Island Glacier and Thwaites Glacier lie in “warm”
water, while the Filchner-Ronne or Ross ice shelves reside in “cold” water.
Ocean temperatures reflect the dominant water sources, with warm waters
dominated by circumpolar deep waters (Jacobs et al., 2011), while cold waters
typically correspond to high-salinity shelf water (Nichols et al., 2001).
Ice–ocean interactions are an active area of research, and more
sophisticated parameterizations of melt are becoming available (e.g.,
Jenkins, 2016; Asay-Davis et al., 2016). Simplified models of the system
could be used (e.g., Payne et al., 2007), as could high-resolution ocean
models that resolve ice-shelf cavities and fjords. Given this wide range of
methods, ISMIP6 will leave the detailed choice of the parameterization to
individual ice-sheet modelers, but will issue guidance on what constitutes an
acceptable parameterization. We will organize workshops with the polar ocean
community to investigate how to best derive oceanic forcing for ice-sheet
models, so that by the time the CMIP6 ocean models are evaluated, a clearer
protocol will be in place. The calculated melt rate will be part of the
standard data request for ice-sheet models (see Appendix A), and part of our
evaluation will be to determine how well the applied forcing compares to
observed melt rates of Rignot et al. (2013) and Depoorter et
al. (2013).
ISMIP6 will not dictate the choice of ice-sheet model complexity in terms of
the ice-flow approximation, the basal sliding law, the treatment of grounding
lines, the calving law, the ice-sheet-specific boundary conditions (e.g.,
bedrock topography), or the initialization method. An exception is that
models of the Antarctic Ice Sheet should include floating ice shelves and
grounding-line migration. The spatial resolution of the ISM in the vicinity
of fast-flowing ice streams and the grounding line affects the dynamic
response (Durand et al., 2009; Pattyn et al., 2012, 2013), and the model
resolution must be fine enough to capture this response accurately. To this
end, participating models are encouraged to take part in model
intercomparison efforts that target specific aspects of ice-sheet modeling,
such as the current MISOMIP (Marine Ice Sheet–Ocean Model Intercomparison
Project; Asay-Davis et al., 2016), and are required to take part in initMIP
(initialization-focused experiments that compare and evaluate the simulated
present-day state; Sect. 3.3.1). The lack of a stricter protocol is a
reflection of the challenges in identifying which factors are the most
important when making projections, which datasets are most accurate, and how
to best capture and parameterize certain ice-sheet processes. For example,
although the choice of bedrock topography affects mass transport and is thus
likely to influence a projection, it is currently not possible to identify a
best dataset due to the difficulty in obtaining bedrock measurements. Groups
are encouraged to repeat the experiments with a variety of perturbations of
weakly constrained parameters, boundary conditions, etc. in order to test the
sensitivity of projections to these choices.
Unlike the protocol for climate models, the ism-XXX-std simulations
cannot be initiated from a spinup corresponding to year 1850. This is due to
the challenge of initializing ice-sheet models to pre-industrial conditions,
which are constrained more weakly than the contemporary state: the quantity
of accurate, high-resolution data available during the satellite era far
exceeds that available for pre-industrial and historical periods. The
majority of ice-sheet models use these data in sophisticated initialization
and assimilation procedures, such that the present-day state of the ice sheet
is simulated with high fidelity. The lack of suitable data before the
satellite era means that no such accuracy can be assumed for simulations of
the historical periods. Such inaccuracies are known to have a large effect on
projections. For instance, discrepancies between projections can often be
attributed to slight differences in the geometry (e.g., Shannon et al.,
2013). The ism-XXX-std simulations will thus be initiated from a
present-day spinup.
The first ism-XXX-std simulation is ism-pdControl-std, the
ice-sheet present-day control with constant forcing needed to evaluate model
drift. This constant forcing is based on the climate at the end of the
initialization procedure. For many models, the forcing and simulation will be
the same as ism-ctrl-std in the initMIP experiment (Sect. 3.3.1),
unless a change has been made in the initialization. The idealized climate
change experiment, ism-1pctCO2to4x-std, considers a
1 % yr-1 atmospheric CO2 concentration rise until quadrupled
concentrations and stabilization thereafter. ism-historical-std will
be an abbreviated simulation for the historical period (as it begins from the
present-day spinup) and, following the CMIP6 protocol, ends in December 2014.
ism-amip-std is a simulation for the last few decades to understand
the well-observed record of ice-sheet changes. The results from
ism-amip-std and ism-historical-std are likely to differ,
and the comparison will provide some insight into the relative importance of
biases, climate variability, and climate change. The main simulation for
projecting 21st century sea-level rise is ism-ssp585-std, which is
initiated from the ism-historical-std simulation. (As mentioned
previously, other scenarios will be considered if time permits.) If possible,
projections should continue to the end of the 23rd century.
We complement the experiments for the recent past and future with one paleo
experiment (ism-lig127k-std), to simulate Greenland ice-sheet
evolution during the Last Interglacial. The transient simulation will span
the period 135 to 115 kyr to include transitions from the preceding to
following cold periods. The climate forcing for ism-lig127k-std will
be derived from the PMIP4-CMIP6 experiment lig127k and other
(transient) LIG climate simulations (cf. Bakker et al., 2013;
Lunt et al., 2013) that will be performed by PMIP4 (Otto-Bliesner et al.,
2016). The proposed experiment builds on past efforts to study Greenland
ice-sheet stability and evolution during the LIG and constrain the Greenland
contribution to the LIG sea-level highstand (e.g., Robinson et al., 2011;
Born and Nisancioglu, 2012; Helsen et al., 2013).