Overview: Experimental protocol for Antarctic projections

This page describes the experimental protocol for the ISMIP6 projections that target the upcoming IPCC AR6 assessment. Due to the delay in CMIP6 climate simulations, the initial set of ISMIP6 simulations are based on CMIP5 projections. These CMIP5 models were chosen following the assessment presented in Barthel et al. (2020). As CMIP6 model outputs became available, ISMIP6 included simulations based on these new models. The experimental framework was revised in September 2018 during an ISMIP6 workshop held in Sassenheim (NL).

The revised protocol described in Nowicki et al. (2020) and summarized in Fig. 1 allows for:

• Sampling CMIP scenarios: main focus is on the high emission RCP8.5, but ice sheet evolution in response to low emission RCP2.6 is also investigated.

• Sampling CMIP models: 6 AOGCMs have been selected from the CMIP5 model ensemble: CCSM4, CSIRO-Mk3-6-0, HadGEM2-ES, IPSL-CM5A-MR, MIROC-ESM-CHEM and NorESM1-M. The AOGCMs were identified based on the following steps:

1. Present-day climate near Antarctica in agreement with observations (evaluated by model biases over the historical period);
2. Sample a diversity of forcing (evaluated by differences in projections and code similarities); and
3. Allow only models with RCP8.5 and RCP2.6. The sampling methodology for the CMIP5 models is described in Barthel et al. (2020). As CMIP6 models became available, ISMIP6 prepared dataset for CNRM-CM6 (ssp585 and ssp126), CESM2 (ssp585) and UKEM1-CM6 (ssp585). Unlike for the rigorous analysis for the CMIP5 models, the CMIP6 models were selected because of their availability.

• Sampling ice sheet model uncertainty: “standard” and “open” experiments. The “standard” experiments are based on parameterizations developed by the ocean groups, while “open” experiments utilize the parameterizations already in use by respective ice sheet models. The open experiments are important as they allow to sample the uncertainty in processes that are poorly known, as reflected by different parameterizations.

• Sampling ocean forcing uncertainty: the standard experiments include “high”, “mid” and “low” values for the forcing parameters, as well as two different calibration methods for the basal melt rate parameters (see Jourdain et al., 2020).

• Experiment ranking: This experimental framework results in a series of projections (divided into core and targeted experiments), a historical run and a control run. Not every ice sheet model will be able to carry out the full set of experiments, but they are strongly encouraged to participate in the full suite of core experiments (Table 1, which uses both the “open” and “standard” experiments/parameterizations). Given that the standard experiments requires new implementations, it is OK to only participate with only the open experiments. Similarly, it is OK to only participate with the “standard” experiments. Groups are encouraged to work through the lists presented below (see Table 1 for core experiments, and targeted experiments will follow soon), starting from the top, and complete as many experiments as possible. This approach of combining core and targeted experiments is based on Shannon et al. (2013): it ensures that all groups do a subset of identical experiments, while it also allows faster models to explore the targeted experiment space more fully.

Figure 1: Overview of the Antarctic experimental framework

List of Projections

With the help of the atmosphere and ocean focus groups, a number of CMIP5 AOGCMs have been selected for ISMIP6 standalone ice sheet model projections. Table 1 lists the core experiments which are the minimum contribution expected from ISMIP6 models, in addition to the initialization experiments listed in Table 2.

All groups are encouraged to contribute to the “standard” experiments, but this is not a requirement for participation in ISMIP6. Groups that have their own methods for implementing ocean and atmosphere forcing, are encouraged to do the suite with “open” experiments (1-4), but these are not compulsory. Models that perform the “open” experiments can use the parameterization of their choice to simulate atmospheric and oceanic forcings, but these parameterizations must use the given CMIP5 AOGCM outputs.

Modeling groups that can run many simulations will be encouraged to further explore the ice sheet response using targeted experiments (See Table). These include three additional CMIP5 AOGCMs under RCP8.5, and experiments that explore the ocean forcing uncertainty. Depending on the results of experiments 3 and 7, which consider RCP2.6, additional AOGCMs may be suggested with RCP2.6 for models that are able to do many simulations, but these would be a lower priority than the completing the set of experiments with the 6 AOGCMs for the RCP8.5 scenario. As CMIP6 AOGCMs are becoming available, we are preparing these datasets. The spreadsheet will be updated as new dataset become available. Because there is value in both completing the 6 CMIP5 AOGCMs (to sample the uncertainty in CMIP) and simulations with CMIP6 models, we encourage groups do to as many experiments as possible.

‘Note’: All datasets needed for the core experiments (Table 1) are available on Ghub Globus, as well as the dataset for the additional targeted CMIP5 and CMIP6 models. (See Table)

Initial state, control run, historical run and projections set up

The core and targeted experiments all start on January 2015 and end in December 2100. The start date follows the CMIP6 protocol for projections, while the end date is constrained by the availability of forcing. In many cases, modelers will need to run a short historical run to bring their models from the “initialization date” to the “projection start date” of January 2015 (see Table 2).

The “initialization date” (or initial state) is left to the modeling groups discretion and can be any time prior to January 2015. The “initialization date” corresponds to the date assigned to the initialization procedure. Groups can reuse their initMIP initialization configuration or generate a new initial state. In the later case, it is important to redo the initMIP schematic experiments 'asmb' and 'abmb' (see initMIP Antarctica and Seroussi et al. 2019), as it will help understanding how a novel initial state contributes to the uncertainty in ice sheet evolution.

A control run (‘ctrl’) is also needed to evaluate model drift. As for initMIP, the control run is obtained by running the model forward, keeping the surface mass balance and ocean forcing used in the initialization technique unchanged. The control run starts from the initial state (typically before 2015) and should last a minimum of 100 years, the same duration as the schematic initMIP experiments. The control run should also be sufficiently long to reach 2100. (See examples in Table 2).

Note that in the event that initMIP schematic experiments and control run are redone as part of the projection setup, then consistency with the projections protocol is more important than consistency with the original initMIP setup. For example, in initMIP bedrock was not allowed to evolve. However, an ISM planning to run the projection with evolving bedrock AND planning to redo the initial state, would also rerun initMIP (ctrl, asmb. abmb) with bedrock change. Similarly, the original initMIP requested 2D output every 5 years, whereas the projections protocol request for yearly values. Therefore if rerunning initMIP, the schematic experiments should be saved yearly.

A single “historical run” is required from each ice sheet model, from which all the projections will branch off. The historical run starts at the initMIP state and ends in December 2014. Groups are free to choose how to run the “historical run” using:

1. A reanalysis,
2. A historical run from an RCM,
3. A historical run from an AOGCM,
4. And/or combination of multiple datasets

However, we reject multiple historical runs for each individual AOGCM, because this would very much complicate the forcing strategy and interpretation. As a consequence, in case AOGCM data is used, please decide for one. ISMIP6 provides a climatology for the SMB and surface temperature for each of the AOGCMs used to generate the projection dataset, as well as anomalies.

• For Antarctica, the SMB and temperature climatology corresponds to 1995-2014, to align with the reference period used by AR6. The Antarctic SMB and temperature anomalies are available from 1950. For the Antarctic ocean, the datasets start from 1850 and the climate model climatology corresponds to 1995-2014. Groups that would prefer to use an Antarctica dataset provided by ISMIP6 are recommended to use NorESM-M climatology and anomalies for SMB and surface temperature (in the directory Atmosphere_Forcing/noresm1-m_rcp8.5 directory). See Appendix A2.2 for accessing the forcing data and directories.

• For the ocean, modelers can use observational climatology (Ocean_Forcing/climatology_from_obs_1995-2017 directory), and/or anomalies (Ocean_Forcing/noresm1-m_rcp8.5/1850-1994) directory.

Note that the climate model oceanic climatologies (Ocean_Forcing/noresm1-m_rcp8.5/climatology_1995-2014 directory) are not intended for use by modelers, but are simply provided so that user can see what was subtracted during the datasets preparation, as the ocean forcing data (Ocean_Forcing/noresm1-m_rcp8.5/1850-1994 directory) is the sum of the observational climatology and anomalies.

The “projection control” (ctrl_proj) is an unforced simulation that starts at the same ice sheet state as the projections. It will run until 2100, and is implemented with zero anomalies. It is meant to capture how much drift arises from the historical run.

In most cases the same historical run (and potentially spin-up) can be used for both the standard and open experiments. However, if the implementation of the open and standard experiments requires changes to the ISM that are substantially different (in terms of physics in the ISM), then modelers are allowed to carry out an historical run for the open experiments and an historical run for the standard experiments.

Atmospheric forcing: SMB and temperature anomalies

ISMIP6 provides surface forcing datasets for the Antarctic ice sheet (AIS) based on CMIP AOGCM simulations. Two approaches are possible: using AOGCM output directly, or re-interpreting the GCM climates through higher-resolution regional climate models (RCMs). The later allows to capture narrow regions at the periphery of the ice sheet with large surface mass balance (SMB) gradients, which are not captured by CMIP5 AOGCMs, and is the technique used for the Greenland ice sheet.

For the Antarctic CMIP5 based projections, RCMs are not used, so SMB anomalies based on AOGCM are directly applied. For CMIP6, many of the AOGCMs that have indicated participation in ISMIP6, now use multiple elevation classes to downscale SMB to finer grid resolution. Once these models have completed the CMIP6 projections, our goal is to include additional ISMIP6 projections using SMB downscaled via elevation class.

For the ISMIP6 projections based on CMIP5 AOGCMs, the surface forcing consists of anomalies in SMB and surface temperature (Fig. 2). SMB is needed by ISMs to compute mass changes at the surface, and surface temperature (i.e., the ice temperature at the base of the snow, as distinct from the 2-m air temperature or skin temperature) is used by many ISMs as an upper boundary condition. The following remarks refer mostly to SMB, but the same comments would generally apply to surface temperature as well.

ISMIP6 provides yearly averaged surface mass balance anomalies, aSMB(x,y,t), along with its components (precipitation, evaporation and runoff), along with SMB climatologies used to compute the anomalies:

aSMB_AOGCM(x,y,t) = SMB_AOGCM(x,y,t) - SMB_CLIM_AOGCM(x,y)

where SMB_AOGCM is the SMB for a given AOGCM and SMB_CLIM_AOGCM is the climatology for that AOGCM. The SMB_CLIM_AOGCM were computed by taking the mean value of all SMB_AOGCM over the reference period (from January 1995 to December 2014). ISMs can use these climatologies for spin-up, if desired, but are free to use their own preferred SMB forcing for spin-up and historical run.

During the run, modelers need to reintroduce the climatology that best fit their simulations. SMB is computed as:

SMB(x,y,t) = SMB_ref(x,y) + aSMB(x,y,t)

where SMB_ref is the SMB that the ice sheet model would have used over the reference period (from January 1995 to December 2014 for Antarctica) and should be the same for all the core and targeted experiments. If a time-dependent SMB is used, then SMB_ref(x,y) is the average over the reference period. If an SMB climatology is used, then SMB_ref(x,y) is simply the climatology. ISMIP6 accept that the use of existing climatologies (or dataset of SMB averaged over many year) may not align with the time period for the AR6 reference period.

However, it is assumed that the differences between climatologies will be less than the inter annual variability from the SMB resulting from the AOGCMs, and thus changes in aSMB. What is important is that the SMB climatology (or SMB_ref) is computed over many years. This assumption also allows for ISM to work with their favorite SMB.

aSMB(x,y,t) is constant over the entire year and changes stepwise at the beginning of every year. SMB climatologies and SMB anomalies are given in units of kg m-2 s-1 (water equivalent), and surface temperature in units of deg K. To convert aSMB to units [m yr^-1] typically used in an ice sheet model, multiply the netcdf variable by 31,556,926 s/yr, 1/1000 m³/kg and by the density ratio rho_w/rho_i:

aSMB [m yr^-1^] = aSMB [kg m^-2^ s^-1^] * 31,556,926 / 1000 * (1000/rho_i)

where rho_i is your specific ice density (typically 917.0 or similar).

The datasets can be obtained via Ghub Globus (see instructions in the section above). Files are provided for several resolutions (1 km, 2 km, 4 km, 8 km, 16 km, and 32 km). Modeling groups should use the resolution closest to their native grid to conservatively interpolate data to model (see Appendix 1, below).

Figure 2: SMB and surface temperature anomalies for CCSM4, MIROC-ESM-CHEM, and NorESM1-M under RCP8.5 and 2.6 (top). SMB climatology for the reference period (January 1995-December 2014) for these models under RCP8.5 (middle), along with difference in SMB climatology between RCP8.5 and 2.6 (bottom).

Oceanic forcing: temperature, salinity, thermal forcing and melt rate parameterization

ISMIP6 provides datasets of extrapolated ocean “ambient” temperature (T), salinity (S) and thermal forcing (TF) from 1850-2100 that are appropriate for present and future ice-shelf cavities. These datasets originate from CMIP models and have been extrapolated under ice shelves, using rules that account for sills and troughs by Xylar Asay-Davis (Fig. 3). The datasets are on the 8 km ISMIP6 Antarctic grid.

Figure 3: Bathymetry and IMBIE2 basins (left) used in the sub-ice shelf extrapolation of ocean temperature (right).

ISMIP6 standard approach

The ISMIP6 standard approach was developed by the Antarctic ocean focus group, and consist of two approaches for the parameterization of basal melt (Jourdain et al., 2020). These ocean melting parameterizations are evaluated for an idealized geometry of the Pine Island glacier in Favier et al. (2019).

The first approach, a non-local quadratic melting parameterization, is the preferred method for ISMIP6 simulations for obtaining melt rates m(x,y) in meters of pure water/yr:

However, an alternative (and easier to implement) takes the form of a local quadratic melting parameterization:

The forcing dataset (for example in the /Ocean_Forcing/noresm1-m_rcp8.5/1995-2100 directory) consist of annual anomalies from the climate models, which were added to the observed climatology (/Ocean_Forcing/climatology_from_obs_1995-2017 directory). Modelers should simply use the data as they are to compute the melt rates using either (1) or (3). In addition to the annual forcing datasets needed for use with these parameterizations, parameters needed to sample the uncertainty in the basal melt are also provided in the /Ocean_Forcing/parameterizations directory.

The files with names *median.nc, *5th_percentiles.nc, *95th_percentiles.nc correspond to the median, low and high cases of the gamma_0 and DeltaT values in the core experiments listed in Table 1.

We test the impact of the calibration method by doing a last experiment in which only a subset of the observation data is used to calibrate the gamma0 coefficient. This choice is motivated by the large impact of melt rate values close to ice shelf grounding lines on the evolution the ice streams feeding them. We choose Pine Island Ice Shelf as it both experiences the largest melt values observed and direct measurements of ocean conditions in the ice shelf cavity are available. We therefore calibrate gamma0 using only the highest 10 basal melt values in Pine Island Ice Shelf, all located close to Pine Island grounding line, in order to assess the sensitivity of the projections to the calibration method. The calibration of the deltaT values is similar to the first calibration method, so that average melt rates for each sector are similar to the melt rates used in the first calibration method and only their spatial distribution varies.

The files with names *_PIGL_gamma_calibration.nc correspond to the experiment 13 listed in Table 1.

Note that the climate model climatologies (Ocean_Forcing/noresm1-m_rcp8.5/climatology_1995-2014 directory) are provided to allow ISMIP6 members to see what was subtracted in the preparation of the ocean forcing dataset, but are not intended to be used in the initialization for example. Instead it is recommended to use the climatology from the observations.

ISMIP6 open approach

The ISMIP6 open approach is used to sample a larger variety of oceanic forcing parameterizations, as it remains an active field of research. Models are free to continue applying the ocean forcing parameterization they used during the model initialization or their preferred method, but should still rely on the ocean forcing datasets provided by ISMIP6 to simulate future ocean conditions. ISMIP6 provides datasets of extrapolated ocean “ambient” temperature (T), salinity (S) and thermal forcing (TF) from 1850-2100 that are appropriate for present and future ice-shelf cavities. These datasets originate from CMIP models and have been extrapolated under ice shelves, using rules that account for sills and troughs by Xylar Asay-Davis (Fig. 3).

The temperature, salinity and thermal forcing data provided for CMIP5 models are the anomalies of each climate model with respect to its January 1995- December 2014 average, added to an observational climatology (based on WOA, EN4 and MEOP datasets). Thus, the data sets are designed to be directly usable by models without the need to compute anomalies or to select reference observations of your own. In such cases, anomalies should be computed with respect to the January 1995- December 2014 average, as this was the period used to anomalize the CMIP5 model input (and slightly different from the time period, 1995-2017, spanned by the observations).

If at all possible, groups using the open approach are encouraged to investigate the uncertainty in their melt parameterization, in a manner similar to the standard approach (where low, median and high values of the gamma_0 and DeltaT parameters are used to investigate the uncertainty in the melt rate. These were obtained by calibration to observed melt.

Antarctic ice shelf fracture

Surface melting can trigger ice shelf collapse (for example, the Larsen B ice shelf in the Antarctic Peninsula). This mechanism is separate from cliff-collapse, but is a precursor to cliff-collapse. Although the mechanisms for Larsen B-style ice shelf collapse are still poorly understood, ISMIP6 provides dataset for ice shelf collapse in the form of a time dependent mask (Figure 4). These datasets were derived from CMIP5 near surface air temperature (tas) following the method described in Trusel et al. (2015), which results in annual surface melt. For ISMIP6, Luke Trussel prepared the bias corrected annual surface melt, which were used to generate the masks. Ice shelves are assumed to collapse following a 10 year period with annual surface melt above 725 mm (Trusel et al., 2015). Some experiments require to model ice shelf collapse and the ISMIP6 masks provided should be used in this case. For the other experiments, ice shelf collapse should not be included.

Models are free to decide on the appropriate method to simulate tributary glaciers’ behavior following the collapse of ice shelves. As the masks were derived from observations, the observed ice shelf may not always corresponds to an ice shelf in the ISM. In the event that the ice shelf collapse mask corresponds to a region which an ISM considers to be grounded (ice sheet), the collapse should not be imposed. Similarly, in the event that applying the mask results in “iceberg” or regions of floating ice shelf that are now detached from the ice shelf, these floating part of the ice should be removed as well.

The datasets can be obtained via Globus under AIS Projections Ice Shelf Fracture datasets

Figure 4: Ice shelf collapse mask for CCSM4 under RCP8.5

Requirements for the projections

• Models have to be able to prescribe a given SMB anomaly
• Models have to be able to use one of the two ocean parameterizations (non-local or local) proposed for the core experiments.
• The adjustment of SMB due to geometric changes in forward experiments is encouraged.
• Bedrock adjustment in forward experiment is allowed.
• The choice of model input data is unconstrained to allow participants the use of their preferred model setup without modification. Modelers without preferred data set choice can have a look at the ISMIP6 537d9b52609d43> page for possible options.
• Participants must submit a README file along with the model outputs as an integral part of the contribution to the ISMIP6. It may be obtained here or requested by email to ismip6@gmail.com. To allow for analysis, any modeling choice needs to be well documented.

Appendix 1 – Output grid definition and interpolation

All 2D data is requested on a regular grid with the following description. Polar stereo-graphic projection with standard parallel at 71° S and a central meridian of 0° W on datum WGS84. The lower left corner is at (-3,040,000 m, -3,040,000 m) and the upper right at (3,040,000 m, 3,040,000 m). This is the same grid used to provide the SMB and basal melting anomaly forcings. The output should be submitted on a resolution adapted to the resolution of the model and can be 32km, 16 km, 8 km, 4 km, 2 km or 1 km. The data will be stored on this resolution for archiving and conservatively interpolated on a 8 km resolution for diagnostic processing by ISMIP6. Output should be provided with single precision.

If interpolation is required in order to transform the SMB forcing (1 km grid data) to your native grid, and transform your model variables to the initMIP output grid (32 km, 16 km, 8 km, 4 km, 2 km, 1 km), it is required that conservative interpolation is used. The motivation for using a common method for all models is to minimize model to model differences due to the choice of interpolation method.

A1.1 Regridding Tools and Tips

 dimensions:
    time = UNLIMITED ; // (3 currently)
 variables:
    double time(time) ;
        time:units = "days since 1-1-2013" ; // This date correspond to the example basetime 
        time:calendar = "360_day" ; // Other calendars can be used... change here to relevant calendar
        time:axis = "T" ;
        time:long_name = "time" ;
        time:standard_name = "time" ;
 data:
  time = 0, 360, 720; // If you use a different calendar these values will change

=For thickness for the projection=
'''''note''' that the full time entries are not shown, only beginning and end) would be:

 dimensions:
    time = UNLIMITED ; 
 variables:
    double time(time) ;
        time:units = "days since 1-1-2013" ; // This date correspond to the example basetime
        time:calendar = "360_day" ; // Other calendars can be used... change here to relevant calendar
        time:axis = "T" ;
        time:long_name = "time" ;
        time:standard_name = "time" ;
 data:
  time = 720, 1080, 1440, …, 31320, 31680; // If you use a different calendar these values will change

=The flux variable, like SMB, would be recorded as the average over a full year, so for the historical:=

 dimensions:
    time = UNLIMITED ; // (2 currently)
    bnds = 2 ;
 variables:
    double time(time) ;
        time:bounds = "time_bnds" ;
        time:units = "days since 1-1-2013 " ; // This date correspond to the example basetime
        time:calendar = "360_day" ; // Other calendars can be used... change here to relevant calendar
        time:axis = "T" ;
        time:long_name = "time" ;
        time:standard_name = "time" ;
    double time_bnds(time, bnds) ;
 data:
  time = 180, 540 ; // If you use a different calendar these values will change. 
                              //This is the middle of the time_bnds
  time_bnds =
   0, 360, //If you use a different calendar these values will change. 
               //These are the day since basetime at the beginning and end of the year
   360, 720 ;

=For the projection=
'''''note''''' that the full time entries are not shown, only beginning and end):

 dimensions:
    time = UNLIMITED ; // 
    bnds = 2 ;
 variables:
    double time(time) ;
        time:bounds = "time_bnds" ;
        time:units = "days since 1-1-2013 " ; // This date correspond to the example basetime
        time:calendar = "360_day" ;  // Other calendars can be used... change here to relevant calendar
        time:axis = "T" ;
        time:long_name = "time" ;
        time:standard_name = "time" ;
    double time_bnds(time, bnds) ;
 variables:
    double time(time) ;
        time:bounds = "time_bnds" ;
 data:
  time = 900, 1260, 1620, ..., 31140, 31500; // If you use a different calendar these values will change. 
                                                      //This is the middle of the time_bnds
  time_bnds =
 720, 1080, //If you use a different calendar these values will change. 
                  //These are the day since basetime at the beginning and end of the year
 1080, 1440,
 1440, 1800,
 ....
 30960, 31320,
 31320, 31680;