Parameters

The elsa namelist group

elsa reads seven parameters from one namelist group. The group name is chosen by the caller of elsa_init, so that one file can hold several configurations. The defaults below are those of input/elsa_defaults.nml, which is both the fallback for any omitted parameter and the schema against which the user group is validated (see the user guide).

&elsa
    n_layers_init    = 10          ! [1]  initialization layers
    layer_resolution = 200.0       ! [yr] add an isochrone this often
    layer_file       = "None"      ! [-]  or an explicit list of isochrone times
    grid_factor      = 1.0         ! [1]  coarsen the host grid by this factor
    dt_coupling      = 50.0        ! [yr] elsa update interval
    cfl              = 0.9         ! [1]  advection sub-step target
    allow_pos_bmb    = False       ! [-]  let freeze-on thicken the bottom layer
/

Reference

Parameter	Type	Default	Valid range
n_layers_init	integer	10	\(\geq 1\)
layer_resolution	real, yr	200	\(> 0\), or 0 to use layer_file
layer_file	string	"None"	a path, or "None" to use layer_resolution
grid_factor	real	1	\(\geq 1\)
dt_coupling	real, yr	50	\(> 0\)
cfl	real	0.9	\((0, 1]\)
allow_pos_bmb	logical	false

Values outside the valid range stop the program at elsa_init.

n_layers_init

The number of equal-thickness layers into which each ice column is divided at a cold start. These layers carry no age. They only resolve the vertical variation of the horizontal velocity within the ice that was present at the initial time, so that the deep initial ice is advected with the slow near-bed velocity rather than with the column mean. Basal melt removes these layers first, and basal freeze-on is added to the lowest one.

layer_resolution and layer_file

Exactly one of the two defines the isochrone schedule. Setting both, or neither, is an error.

With layer_resolution, an isochrone is laid down every layer_resolution years after the initial time, up to but excluding time_end. On a restart the schedule remains anchored on the original initial time. The number of scheduled isochrones is \(\lceil (t_\mathrm{end} - t_\mathrm{init})/\mathrm{layer\_resolution} \rceil - 1\).

With layer_file, the schedule is read from a text file with one isochrone time per line, in years, on the same time axis as the host. Blank lines are skipped. The times must be strictly increasing and later than the initial time. Times at or after time_end are skipped with a note, so that one file can serve every segment of a restarted experiment. This option allows the layers to be aligned with dated radiostratigraphic horizons.

In both modes, a layer is laid down at the first update at or after its scheduled time, and is stamped with the time of that update. The schedule and the host time steps should thus be chosen so that updates land on the scheduled times, if exact horizons are needed.

In both cases the surface at the initial time is an isochrone as well, so the number of dated isochrones is one more than the number of scheduled ones. The total number of layers allocated is n_layers_init plus the number of scheduled isochrones plus one.

The memory and the run time of elsa both scale linearly with the number of layers. Each layer holds four double-precision fields on elsa’s grid (d_iso, dsum_iso, ux_iso and uy_iso). At 16 km over Greenland (106 × 181 cells), 1000 layers thus need around 0.6 GB.

grid_factor

elsa’s grid is the host grid coarsened by this factor, which is a real number and need not be an integer. Scalar fields are remapped conservatively and the velocities bilinearly (see interp). The cost of the advection scales roughly with the inverse cube of the grid spacing, since the number of substeps grows as the cells shrink, so grid_factor is the most effective lever on run time.

Two restrictions apply. A non-integer factor can leave a strip of host cells uncovered at the high edge of the domain (see the limitations). The Yelmo coupling currently requires grid_factor = 1 (see Yelmo coupling).

dt_coupling

The interval between elsa updates. elsa_update returns without doing anything until this much time has elapsed since the last update.

The vertical thinning of the layers converges at first order in dt_coupling. At an ice divide with 3000 m of ice and an accumulation of 0.3 m yr\(^{-1}\), the maximum departure from the Nye solution is around 5.5 m at dt_coupling = 100 yr, and it halves when the period is halved (see column).

A shorter coupling period does not increase the cost of the advection, since the number of substeps is set by the CFL condition over the simulated time. It does increase the cost of the horizontal maps and of the vertical velocity average, which are computed once per update. A longer period also increases the number of margin columns in which the ablation over one period exceeds the ice thickness (see Design). In practice, dt_coupling should be no longer than layer_resolution, so that every layer receives accumulation.

cfl

The target for the explicit substep. For each layer and each update, the substep is chosen so that the outflow from every cell over one substep does not exceed the fraction cfl of the cell’s content. Any value in \((0, 1]\) keeps the layer thickness non-negative and conserves mass. Lower values increase the number of substeps proportionally, and with a first-order upwind scheme they also increase the numerical diffusion slightly. There is thus little reason to depart from the default.

allow_pos_bmb

Controls the treatment of basal freeze-on (bmb > 0). If true, the frozen-on ice is added to the lowest layer. If false, it is not added to any layer, and the subsequent normalization onto the host ice thickness distributes the corresponding thickening over the whole column in proportion to the layer thicknesses. Basal melt (bmb < 0) is always removed from the bottom of the column upward, independently of this switch.

Fixed constants

Three constants are set in the source rather than in the namelist.

Constant	Value	Meaning
H_ICE_MIN	\(10^{-6}\) m	a column thinner than this is treated as ice-free
MV	\(-9999\)	missing value, used for the deposition time of undated layers
TIME_TOL	\(10^{-6}\)	relative slack when comparing model times

Example configurations

The benchmark namelists in par/ are working examples. par/test_greenland.nml holds a native-resolution group and a group with grid_factor = 2.5. par/test_column.nml holds four groups that differ only in dt_coupling.