[matthewhoffman]

This PR adds FastIsostasy as an alternate GIA model to the existing global 1d Sea Level Model. The implementation makes use of a lot of the existing coupling code and builds off of FastIsostasy documentation on coupling here:
https://github.com/palma-ice/FastIsostasy/blob/main/README.md#coupling-fastisostasy-to-your-favorite-ice-sheet-model

[matthewhoffman]

Status: Code compiles with FASTISOSTASY=true but has not been run or tested in any way. Code was written by Copilot w/ GPT 5.3-Codex using these instructions:

Plan instructions

Plan: Couple MALI with FastIsostasy (Fortran)

TL;DR: Use the Fortran version of FastIsostasy (palma-ice/FastIsostasy) — not the Julia version — and follow the existing SLM coupling pattern already in mpas_li_bedtopo.F. The Fortran library has a clean 4-call API and the SLM code provides a near-complete template for the MPI gather/interpolate/solve/scatter workflow.

---

Key Discovery: A Fortran version already exists

FastIsostasy includes a pure Fortran implementation specifically designed for coupling with ice-sheet models. Its API is:

call isos_init(isos, path_par, "isos", nx, ny, dx, dy)
call isos_init_ref(isos, z_bed_ref, H_ice_ref)
call isos_init_state(isos, z_bed, H_ice, time, bsl)
! In time loop:
call isos_update(isos, H_ice, time, bsl)
z_bed = isos%out%z_bed

The docs for it include specific guidance for coupling to an ice-sheet model: https://github.com/palma-ice/FastIsostasy#coupling-fastisostasy-to-your-favorite-ice-sheet-model

---

Why this is tractable

MALI already has a complete coupling template in mpas_li_bedtopo.F via the SLM (sea-level model) coupling. That code handles:

• Alarm-based coupling intervals
• MPI gather of ice thickness from all ranks to rank 0
• Interpolation from unstructured MPAS mesh → regular grid (sparse matrix multiply using weight files)
• External solver call on rank 0
• Interpolation back to MPAS mesh
• MPI scatter of bed topography changes
• Halo exchange + geometry/mask update

The FastIsostasy coupling would replicate this pattern almost exactly, substituting FastIsostasy's isos_update() for the SLM solver calls.

---

Phases

Phase 1: Build System

• Add palma-ice/FastIsostasy as a submodule or vendored source
• Add USE_FASTISOSTASY conditional compilation flag (mirrors USE_SEALEVELMODEL)
• Link against libisostasy.a + FFTW dependency

Phase 2: Registry & Config

• New config_uplift_method = 'fastisostasy' option
• Config params: coupling interval, namelist path, weight file paths

Phase 3: Core Coupling — most complex, but heavily templated

• fastisostasy_init(): setup alarm, read weights, MPI layout, call isos_init/isos_init_state
• fastisostasy_solve(): gather → interpolate → isos_update() → interpolate back → scatter → apply
• New branch in li_bedtopo_solve() dispatch

Phase 4: Grid/Weight Files

• Generate ESMF/TempestRemap interpolation weights between MPAS mesh and FastIsostasy's regular Cartesian grid
• FastIsostasy needs a square domain with ≥512 km padding beyond ice extent

Phase 5: Restart

• FastIsostasy has built-in isos_restart_write/read; wire into MALI checkpoint

Phase 6: Testing

• Idealized dome test with known analytic GIA solution
• MPI correctness (1/4/16 procs)
• Bit-for-bit restart verification

---

Relevant Files

• mpas_li_bedtopo.F — primary file to modify (SLM coupling template is here)
• Registry.xml — add config options
• Makefile — build system (has USE_SEALEVELMODEL pattern to copy)
• mpas_li_time_integration_fe_rk.F — calls li_bedtopo_solve() at line 634 (no changes needed here)

---

Main Challenge: Grid Interpolation

MALI uses an unstructured MPAS mesh; FastIsostasy uses a regular Cartesian (stereographic) grid. The SLM coupling already solves this exact problem using pre-computed sparse interpolation weight files (generated by ESMF_RegridWeightGen or TempestRemap). The same tooling applies — you just need to generate weights for the FastIsostasy target grid rather than the SLM's Gauss-Legendre grid. This will be done outside of MALI in a separate preprocessing step and does not need to be implemented here. However, the MALI implementation should assume a similar format for mapping files as for the existing SLM.

Build and test loop

• I want the work divided into logical commits with descriptive commit messages.

• To get a usable environment for compiling MALI, do:
  source /global/cfs/cdirs/fanssie/users/hoffman2/compass/ismip7-devs//load_compass_pm-cpu_gnu_mpich.sh

• Before each commit, make sure the code builds:
  cd /global/cfs/cdirs/fanssie/users/hoffman2/compass/ismip7-devs/MALI-Dev/components/mpas-albany-landice; make clean; make -j 4 gnu-cray DEBUG=true

• If code compiles, you may commit and move on to next step.

• If you run into any environment problems or have any questions, stop and ask for more instructions. Otherwise continue working through the implementation steps until complete.

I made a cursory review of the implementation. The coupling code looks correct in that it mirrors the coupling code for the SLM heavily, as intended, but I have not done a line by line review. The Makefile/build changes look reasonable, but some cleanup is needed to generalize it to any machine. Note that the FastIsostasy build is really slow for two reasons: 1. it uses config which is generally slow and 2. the FFTW library it depends on is very slow to compile and we may want to support optionally pointing to a pre-build lib of it.

Next steps are to build mapping files, set up a FI namelist, and attempt a run.

[hollyhan]

Update (09/22/2026): This issue has been resolved with commit da45530edb7f95726ddb73c15738d9b9d2bee5ad

There seems to be an issue in linking the fastisostasy libraries- I am not able to build MALI with FASTISOSTASY=true without running into errors that are sometimes solved and sometimes not by adding
override LIBS += $(FASTISOSTASY_LINK_LIBS) in the build_options.mk file.

Built attempted on Perlmutter (make gnu-cray FASTISOSTASY=true ALBANY=true DEBUG=false)

Currently, the way I make the build work spuriously is by compiling once with override LIBS += $(FASTISOSTASY_LINK_LIBS) removed, then add it back without make clean. This seems to work all the time.

[hollyhan]

Results of the testing can be seen here: MPAS-Dev/compass#970

[hollyhan]

Test results:
Setting up and running coupled MALI-FI simulations with a realistic Antarctic Earth structure (i.e., a 2D field derived by depth-lumping the actual 3D viscosity structure) requires an external NetCDF file. The weak and strong Earth-structure files (Swierczek-Jereczek et al.) can be downloaded from https://zenodo.org/records/21105061. The medium Earth model file can be derived from the other two by taking the mean of log10 effective viscosity and the mean of lithospheric thickness. The three Earth structure profiles look like this:

Coupled simulation has been tested with realistic climate forcing (ISMIP7 Experiment C007: CESM2-WACCM SSP585) and the weak Earth structure. Below results are based on the 2D MALI outputs.

The spatial pattern of bedrock response is coherent with changes in ice thickness.

The restart capability has been tested to the extent possible – by code inspection config_do_restart=.true. correctly triggers FastIsostasy to read its own restart files rather than calling isos_init_ref. Based on the MALI outputs, globalStats show smooth curves in the ice state through the restart, and the bedrock elevation change rate also shows no kink at the restart boundary either. Based on this evidence, the restart capability can be judged to be sufficiently verified for production use. However, to fully close this out, more direct testing would help — but since FastIsostasy currently overwrites its restart file every year, that would need a small addition on the MALI coupling side to call FastIsostasy's existing (but currently unused) diagnostic-output routine.

[Copilot]

Copilot review overview

🟡 Changes recommended

Missing submodule gitlinks, an unguarded module import, and invalid-grid handling currently break supported build or initialization paths.

Get a fresh assessment by requesting another Copilot review.

Review effort: Balanced
Findings: 2 · 3

Open (5)

• Add pinned gitlinks for FastIsostasy dependencies · New
• Guard isos_utils import when FastIsostasy is disabled · New
• Include grid resolution in the main FastIsostasy namelist builder · New
• Include grid resolution in the alternate namelist builder · New
• Skip FastIsostasy allocation after non-square grid validation failure · New

What changed in this PR

Adds FastIsostasy as an alternative MALI glacial-isostatic-adjustment model.

Changes:

• Adds FastIsostasy coupling, remapping, and restart handling.
• Integrates FastIsostasy and fesm-utils into the build.
• Adds related namelist configuration.

File	Description
.gitmodules	Declares external dependencies.
components/​mpas-framework/​Makefile	Isolates PIO test libraries.
components/​mpas-albany-landice/​src/​Registry.xml	Registers FastIsostasy options.
components/​mpas-albany-landice/​src/​mode_forward/​mpas_li_core.F	Writes coupled restart state.
components/​mpas-albany-landice/​src/​mode_forward/​mpas_li_bedtopo.F	Implements coupling and remapping.
components/​mpas-albany-landice/​src/​Makefile	Builds external libraries.
components/​mpas-albany-landice/​src/​build_options.mk	Adds compile and link options.
components/​mpas-albany-landice/​bld/​namelist_files/​namelist_definition_mali.xml	Defines namelist entries.
components/​mpas-albany-landice/​bld/​namelist_files/​namelist_defaults_mali.xml	Adds defaults.
components/​mpas-albany-landice/​bld/​build-namelist-section	Assembles section defaults.
components/​mpas-albany-landice/​bld/​build-namelist	Assembles MALI defaults.

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

[matthewhoffman]

Testing

Compass full_integration suite on pm:

• compiled using DEBUG=true FASTISOSTASY=false : All tests pass
• compiled with DEBUG=true FASTISOSTASY=true : fails to compile

[matthewhoffman]

Testing

Compass full_integration suite on pm with the two new commits modifying the FI build. Built from a fresh clone.

• compiled with DEBUG=true FASTISOSTASY=true : Compiles and all tests pass