Upgrading from v3.1.0 to v3.2.0

LaMEM v3.2.0 is a feature release for surface processes. It adds an optional coupling to the FastScape landscape-evolution library, a new slope-dependent erosion law, and a surf_mode switch between the built-in surface processes and FastScape. Almost all of it is opt-in. An unmodified v3.1.0 .dat file runs under v3.2.0 and gives the same numbers, with one narrow exception: topo_diff = 1 combined with units = none. The other breaking changes are in the Julia test harness, and in source patches that touch the time loop or the free surface.

What this guide reflects. The v3.2.0 tag, which is also upstream master at commit 406b6444 (2026‑09‑22). No commits have landed after the tag. The baseline is 7e7a012e (2026‑08‑12), the v3.1.0 tag, which is the exact commit that Upgrading from v3.0.0 to v3.1.0 was written against. The two guides therefore cover consecutive ranges with no gap and no overlap. The range spans 89 commits and 7 merged PRs: #76 (FastScape coupling), #80 (slope-dependent erosion), #81 (the previous upgrade guide), #84 (PETSc_jll 3.25.4 in CI), #85 (README), #86 (single-threaded BLAS/OpenMP in tests) and #87 (version bump).

Every claim below was verified by direct grep in both trees. The pitfalls in §8 were confirmed on three binaries built for this guide against PETSc 3.22.5 on aarch64 Linux: a v3.1.0 build, a default v3.2.0 build, and a v3.2.0 build with surf=scape. Unmodified v3.1.0 input files were run on all three.

0. TL;DR — the v3.2.0 upgrade checklist

If you run LaMEM from .dat files and never set topo_diff under units = none, your existing inputs need no changes. Items 1–4 are about input files; the rest concern building, testing and patching.

1. topo_diff = 1 now requires units = geo or units = si. Under units = none, v3.1.0 ran anyway. It used the value as a raw non-dimensional diffusivity while printing the unit label [m^2/s]. v3.2.0 stops with an error. The new slope_dependent_erosion has the same rule. → §5.1
2. ⚠️ Never set surf_mode = 0. It is accepted, the run exits 0, and the free surface never moves: advection, erosion and sedimentation are all skipped. The only valid values are 1 (the default, the built-in processes) and 2 (FastScape). → §4.1
3. ⚠️ With surf_mode = 2, the built-in surface keys are silently ignored. These are erosion_model, sediment_model, slope_dependent_erosion, topo_diff and all their parameters. They are not even checked for missing required companions. → §4.3
4. ⚠️ vel_boundary is documented backwards. FastScape.md and info/options/input_file.dat say 0 = zero velocity. In the code, 1 sets the boundary velocity to zero and 0 keeps the velocity from LaMEM. → §4.4
5. FastScape has to be enabled at compile time, with make mode=opt surf=scape all and FASTSCAPE_LIB set. A default binary given surf_mode = 2 stops at startup. Check any binary with LaMEM -fastscape_info. → §1.2
6. Test harness: the opt= keyword is gone from perform_lamem_test, run_lamem_local_test and CreatePartitioningFile_local, so passing it raises a MethodError. deb=true now really runs the debug binary; in v3.1.0 it silently ran the optimized one. → §6.1
7. make test in test/ builds with FastScape whenever FASTSCAPE_LIB is exported in your shell. → §6.3
8. make format and make check in src/ now require exactly astyle 3.1 and refuse any other version. → §1.3
9. PETSc 3.25.4 is now the recommended version and the CI version. The code still compiles against 3.22–3.24. → §1.1
10. If you patched the time loop or the free surface, the surface-process calls moved inside if(lm->surf.SurfMode == 1), and FreeSurf and LaMEMLib gained members. → §7

Quick self-check for a v3.1.0 input file

Work through these checks from top to bottom on one .dat file:

grep -nE "^\s*units\s*=\s*none"  my_model.dat   # together with…
grep -nE "^\s*topo_diff\s*=\s*1" my_model.dat   # → both hit? now a hard error (§5.1)
grep -n  "surf_mode"             my_model.dat   # → "= 0"? free surface frozen (§4.1)
grep -rn "opt *= *\(true\|false\)" my_tests/    # → keyword removed from test_utils.jl (§6.1)
grep -rn "deb *= *true"          my_tests/      # → now really runs bin/deb (§6.1)
echo "$FASTSCAPE_LIB"                           # → non-empty? make test builds surf=scape (§6.3)
astyle --version                                # → not 3.1? make format / make check refuse (§1.3)

Nothing was removed from the .dat vocabulary. LaMEM v3.1.0 could parse 471 parameters, and all 471 still parse in v3.2.0. There are 46 new parameters, all opt-in; see the Appendix.

1. Requirements & build changes in v3.2.0

1.1 PETSc and toolchain (v3.2.0)

The code's PETSc version guards did not change. The four PETSC_VERSION_LT checks are the same in both trees (src/LaMEMLib.cpp:1003, src/JacResTemp.cpp:232, src/options.cpp:70, src/lsolve.cpp:358). What changed is the recommendation:

The claim that older versions still work was checked for this guide: v3.2.0 builds warning-free against PETSc 3.22.5, in both the default and the surf=scape configurations.

Installation.md has a new §1.2 "Compiling LaMEM against the precompiled PETSc", which is the same route CI uses. In brief:

• Deploy PETSc_jll into /workspace/destdir. The path is hard-coded in the packaged petscvariables.
• Point PETSC_OPT at lib/petsc/double_real_Int32 (or …Int64).
• Build as usual.
• From PETScjll 3.25 on, a binary built this way has to be told where BLAS is before it runs outside Julia: `LBTDEFAULT_LIBS` must list both the ILP64 and LP64 OpenBLAS.

(This route was not reproduced on the machine used for this guide. It is summarised from Installation.md and test/test_utils.jl:621.)

1.2 Optional FastScape build

FastScape is not part of the default build. src/Makefile gained a surf variable, which defaults to none:

export FASTSCAPE_LIB=/dir/containing/libfastscapelib_fortran   # new
cd src && make mode=opt surf=scape all                          # adds -DWITH_FASTSCAPE
./bin/opt/LaMEM -fastscape_info                                 # prints FASTSCAPE_ENABLED / FASTSCAPE_DISABLED

• surf=scape without FASTSCAPE_LIB stops immediately: Makefile:46: *** Environmental variable FASTSCAPE_LIB must be set to installation directory. Stop.
• The link line embeds an rpath to FASTSCAPE_LIB (src/Makefile:116–120), so the binary finds the library at run time without LD_LIBRARY_PATH.
• Switching an existing build to surf=scape does not recompile it. The object files do not depend on the surf flag, so on a tree already built without FastScape, make mode=opt surf=scape all prints Nothing to be done for 'all', and the binary stays FASTSCAPE_DISABLED. Run make mode=opt clean_all (and mode=deb) first, then rebuild. Switching back the other way needs the same step.
• -fastscape_info (src/LaMEM.cpp:30–43) prints its answer and exits before reading any input file. The test suite uses it to decide whether to skip FastScape tests.
• Upstream tests against Fastscapelib 2.8.4. The easiest source is Pkg.add("Fastscapelib_jll"), then Fastscapelib_jll.LIBPATH_list. For this guide, the surf=scape binary was linked against the Fastscapelib_jll 2.8.4 shared library.

1.3 astyle is pinned to 3.1

The new checkastyle target (src/Makefile:169) runs before make format, make checkformat and therefore make check. It refuses any astyle other than 3.1, the version shipped with Ubuntu 22.04/24.04 and used by CI. The reason is that newer astyle releases reformat files that 3.1 considers correct. With Homebrew's astyle, for example, you get:

ERROR: wrong astyle version.
  required : 3.1
  found    : Artistic Style Version 3.6.9

Install 3.1 (from distro packages or built from source) if you contribute code. Nothing changes if you only compile and run LaMEM.

2. What's new in v3.2.0 (the "why")

• FastScape coupling (#76). On each time step LaMEM passes the surface velocities to FastScape. FastScape advances the topography, in sub-steps of at most max_fs_dt, solving for:

  • fluvial incision (stream-power law);
  • hillslope diffusion;
  • sediment transport and deposition;
  • optionally, marine transport with compaction.

  LaMEM then takes the topography back and converts markers between air, rock and the sediment phase. The whole coupling (src/fastscape.cpp, about 3 100 lines) is compiled only with -DWITH_FASTSCAPE. It works in 2D and 3D, on uniform and non-uniform grids. The user documentation is the new page Surface processes with FastScape.

• Slope-dependent erosion (#80). E [m/yr] = prefactor_slope · |∇h|^n_slope is applied on the internal free surface. It sits on top of any erosion_model, after erosion and before sedimentation and topographic diffusion. It can mimic stream-power incision in profile models, where full FastScape coupling would be excessive.

• Unit-safety checks. Both topo_diffusivity (m²/s) and prefactor_slope (m/yr) are now rejected under units = none, because there they would be silently reinterpreted as raw non-dimensional numbers.

• CI modernisation (#84, #86). PETScjll 3.25.4 on Julia 1.12.7. The harness now pins BLAS and OpenMP to one thread. This fixes a SuperLUDIST/OpenBLAS race that made tests fail intermittently, and sometimes return wrong residuals, about two runs in three.

3. Repository layout changes in v3.2.0

Added:

Renamed: test/t02_FB2_MG/FB2_a_CoupledMG_opt.expected → FB2_a_CoupledMG_deb.expected. The name now matches what the test actually runs; see §6.1.

No directories were moved or removed. No files were deleted.

4. New surface-process options

All of these belong to the free-surface section and are read only when surf_use = 1.

4.1 surf_mode (v3.2.0)

    surf_mode = 1     # 1 - built-in erosion/sedimentation (default), 2 - FastScape

Read at src/surf.cpp:55, default 1 (src/surf.cpp:38), upper bound 2. The time loop (src/LaMEMLib.cpp:700, :723) has branches only for == 1 and == 2.

surf_mode = 3 fails loudly: Entry 1 in parameter "[-]surf_mode" is larger than allowed : val=3, max=2.

4.2 Slope-dependent erosion

    slope_dependent_erosion = 1     # default 0
    prefactor_slope         = 0.05  # [m/yr] — NOTE m/yr, unlike er_rates/sed_rates in cm/yr (default 1.0)
    n_slope                 = 2.0   # [-] (default 1.0)

• The parameters are read at src/surf.cpp:123–141 and the law is implemented in FreeSurfAppSlopeErosion (src/surf.cpp:1103).
• The slope is a centred difference, one-sided at the global boundaries.
• The update is explicit, with automatic sub-stepping: the drop per sub-step is capped at half the minimum horizontal spacing (src/surf.cpp:1208). This keeps it well defined for any n_slope, including n_slope < 1.
• Each step prints Applying slope-dependent erosion (E = … [m/yr] * slope^…) in N sub-step(s).
• It works only with surf_mode = 1. It requires units = geo or si (§5.1).

4.3 FastScape coupling (surf_mode = 2)

The requirements are checked in this order. Each one fails loudly if it is not met:

A minimal block that ran successfully here on a 2D geo-unit model (fs2D = 1):

<FastScapeStart>
    fs2D          = 1        # 2D LaMEM model → extend the profile in y
    extendedRange = 100      # [km]
    extendedNodes = 33
    max_fs_dt     = 0.001    # [Myr]
    topo_boundary = 1111
    vel_boundary  = 0000     # see §4.4 — 0 KEEPS the LaMEM velocity
    random_noise  = 1
    sed_phases    = 1        # phase ID given to new sediment (single integer here)
    kf = 1e-6   kfsed = -1.0   m = 0.4   n = 1.0
    kd = 1e-2   kdsed = -1.0   g = 0.0   gsed = 0.0   p = -2.0
<FastScapeEnd>

(Written one key per line in a real file.)

Other constraints. Surface processes with FastScape was written for the PR and may lag the code, so each point below says whether it was checked against the source:

• FastScape runs serially on rank 0. Checked: the Fortran solve is inside if(ISRankZero(...)) (src/fastscape.cpp:1542).
• Its surface grid is built once and never regenerated, so models whose LaMEM grid stretches (a background strain rate) are not supported. Documented only: nothing in fastscape.cpp detects a background strain rate, so such a model is not rejected. Treat the limitation as real, but do not expect an error message to warn you.
• With surf_mode = 2, FastScape state is appended to restart files (src/surf.cpp:314, :335). Restart files from surf_mode = 1 runs are unchanged. Checked in the source; a restart was not run for this guide.
• The page states max_fs_dt in [Myr]. That is only true for units = geo. The value is read in LaMEM time units (src/fastscape.cpp:102, :306, :324), so under units = si it is in seconds.

4.4 vel_boundary: the docs are inverted

vel_boundary is a four-digit string with one digit per boundary, in the order bottom, right, top, left. The implementation (src/fastscape.cpp:2826–2851) zeroes the velocity on a boundary whose digit is '1' and leaves it untouched otherwise:

if (FSLib->FS_VELBC[0] == '1' && j == 0) { vx_pass[ind] = 0; vy_pass[ind] = 0; vz_pass[ind] = 0; }

The comment at src/fastscape.cpp:109 agrees ("1 – boundary velocity == 0; 0 – boundary velocity from LaMEM"). The shipped documentation, including the published Surface processes with FastScape page, says the opposite:

• doc/src/man/FastScape.md:102: "0 sets the boundary velocity to zero and 1 keeps the velocity transferred from LaMEM"
• info/options/input_file.dat:206: "0-zero velocity, 1-keep velocity transferred from LaMEM"

A correction of both files is proposed in UniMainzGeo/LaMEM#88. Until it is merged, go by the code: 0000 (as used in t37) keeps the LaMEM velocity on all four sides, and 1111 pins all four to zero. topo_boundary is documented correctly (0 reflective, 1 fixed height).

4.5 FastScape output

A coupled run also writes <out_file_name>_fs.pvd. This is on by default: out_surf_fs, out_fs_pvd and every field flag default to 1 (src/fastscape.cpp:800–814), so you set a flag to 0 to drop that field. The fields are selected with the 12 other out_surf_* / out_fs_pvd flags listed in the Appendix. The write frequency is set with surf_out_nstep in the FastScape block.

5. Input-file breaking changes in v3.2.0

There is one, and it is loud.

5.1 topo_diff now requires dimensional units

src/surf.cpp:148–155 (commit cabca8c7). topo_diffusivity is specified in m²/s and scaled by length_si² / time_si. Under units = none both of those are 1, so v3.1.0 used the value as a raw non-dimensional diffusivity while printing it as [m^2/s].

Observed: test/t15_RTI/RTI.dat (units = none) with topo_diff = 1 and topo_diffusivity = 1e-5 appended. Under v3.1.0 it exits 0 and prints Topographic diffusion : active (K = 1e-05 [m^2/s]). Under v3.2.0 it stops with:

topo_diff defines topo_diffusivity in [m^2/s] and requires a dimensional unit system. Set units = geo or units = si, or deactivate topo_diff.

Fix: switch the model to units = geo/si. If you really meant a non-dimensional diffusivity, drop topo_diff: there is no longer a way to express that.

No units = none input shipped with v3.1.0 uses topo_diff. The only in-tree topo_diff test, t35, uses units = geo, so no example or test needed changing.

6. Test framework & CI in v3.2.0

6.1 The opt keyword was removed

opt was dropped from run_lamem_local_test (test/test_utils.jl:49), perform_lamem_test (:685) and CreatePartitioningFile_local. All 81 opt=true occurrences were removed from runtests.jl. The binary is now chosen by deb alone: deb=false (the default) runs bin/opt, and deb=true runs bin/deb.

Before (v3.1.0):

perform_lamem_test(dir, ParamFile, "MyTest_opt"; opt=true, keywords=keywords, ...)
perform_lamem_test(dir, ParamFile, "MyTest_deb"; deb=true, opt=false, ...)

After (v3.2.0):

perform_lamem_test(dir, ParamFile, "MyTest_opt"; keywords=keywords, ...)
perform_lamem_test(dir, ParamFile, "MyTest_deb"; deb=true, ...)

Passing opt now fails when the function is called:

MethodError: no method matching perform_lamem_test(::String, ::String, ::String; opt::Bool)
This method does not support all of the given keyword arguments (and may not support any).
  … got unsupported keyword argument "opt"

6.2 FastScape-only tests

LaMEM_has_fastscape(; bin_dir, deb) (test/test_utils.jl:165, exported) runs the binary with -fastscape_info. t37 (runtests.jl:1492) uses it to gate the opt and deb runs separately. When the binary lacks FastScape, the testset records @test_skip "FastScape is not installed" instead of failing. Use the same pattern for any test that needs an optional build feature.

6.3 FASTSCAPE_LIB switches the test build

test/start_tests.jl:29 rebuilds LaMEM with surf=scape, both opt and deb, whenever FASTSCAPE_LIB is set in the environment, and with the plain build otherwise. So an exported FASTSCAPE_LIB in your .bashrc silently turns every make test into a FastScape build. If it points at the wrong directory, the link fails. To test the default build, use:

env -u FASTSCAPE_LIB make test 38

The switch does not force a rebuild. If bin/ already holds a non-FastScape build, make reports Nothing to be done (§1.2), t37 skips itself, and the suite still reports success, with t37 counted as Broken. This was observed while preparing this guide. Run make mode=opt clean_all and make mode=deb clean_all in src/ before switching, and confirm with ../bin/opt/LaMEM -fastscape_info.

CI (test/compile_lamem.jl:41) always builds with surf=scape, against Fastscapelib_jll staged in /workspace/destdir/lib/fastscape.

6.4 Runtime environment for test runs

Every LaMEM process started by the harness now goes through add_dylibs (test/test_utils.jl:621), which does three things:

• Pins OMP_NUM_THREADS, OPENBLAS_NUM_THREADS and VECLIB_MAXIMUM_THREADS to 1, even for a locally built PETSc. SuperLUDIST's OpenMP threads were intermittently passing corrupted arguments into OpenBLAS (** On entry to DGEMM parameter number 8 had an illegal value). The corrupted solves produced wrong residuals, for example `|mRes|2off by 82 % in t02, according to PR #86. If you run a SuperLU_DIST model by hand with threaded BLAS and see that message, setOMPNUMTHREADS=1`.
• Sets the platform's real loader variable (LD_LIBRARY_PATH on Linux, DYLD_FALLBACK_LIBRARY_PATH on macOS, PATH on Windows). v3.1.0 always set the macOS variable.
• Sets LBT_DEFAULT_LIBS for PETSc_jll ≥ 3.25 (§1.1).

t23_Permeable now uses direct_solver_type = mumps (test/t23_Permeable/Permeable.dat:267), because the superlu_dist solve diverged under PETSc 3.25 with Int64. The description of PR #86 says SuperLUDIST was restored there, but the merged tree still uses MUMPS, and the `adddylibsdocstring confirms it.coupleddirect` + SuperLUDIST is therefore not covered by the v3.2.0 suite.

6.5 New tests and numbering

t37 (t37_Collision_FastScape) and t38 (t38_slope_dependent_erosion) were added. PR #80 used the number t37 but was renumbered to t38 at merge. The next free number is t39.

For this guide the complete v3.2.0 suite was run locally (make test, PETSc 3.22.5, 32-bit indices, aarch64 Linux) on both builds:

The first attempt at the FastScape run silently reused a default build, which is how the pitfall in §6.3 was found.

7. If you patched the v3.1.0 source

Much smaller than the v3.1.0 upgrade. Only one prototype was added to the existing headers (FreeSurfAppSlopeErosion), and none were changed or removed. The edits that can conflict with a patch are these:

The time loop. In LaMEMLibSolve (src/LaMEMLib.cpp:699–755), the free-surface steps are now selected by mode.

Before:

PetscCall(FreeSurfAdvect(&lm->surf));
...
PetscCall(FreeSurfAppErosion(&lm->surf));
PetscCall(FreeSurfAppSedimentation(&lm->surf));
PetscCall(FreeSurfAppTopoDiffusion(&lm->surf));

After:

if(lm->surf.SurfMode == 1) PetscCall(FreeSurfAdvect(&lm->surf));
#ifdef WITH_FASTSCAPE
if(lm->surf.SurfMode == 2) PetscCall(FastScapeCopyVelocity(&lm->FSLib));
#endif
...
if(lm->surf.SurfMode == 1)
{
    PetscCall(FreeSurfAppErosion(&lm->surf));
    PetscCall(FreeSurfAppSlopeErosion(&lm->surf));   // new
    PetscCall(FreeSurfAppSedimentation(&lm->surf));
    PetscCall(FreeSurfAppTopoDiffusion(&lm->surf));
}
#ifdef WITH_FASTSCAPE
if(lm->surf.SurfMode == 2) { FastScapeRun; FreeSurfSmoothMaxAngle; FreeSurfGetAvgTopo; }
#endif

A custom surface process belongs inside the SurfMode == 1 block. Its parameters must be read inside the if(surf->SurfMode == 1) block of FreeSurfCreate (src/surf.cpp:77).

Structs.

• FreeSurf (src/surf.h) gained FastScapeLib *FSLib, SurfMode, and slope_dependent_erosion / prefactor_slope / n_slope.
• LaMEMLib gained FastScapeLib FSLib (src/LaMEMLib.h:45).
• These members are unconditional: they exist in default builds too. Only the code that uses FastScape is behind #ifdef WITH_FASTSCAPE.
• LaMEMLibSetLinks wires surf.FSLib and FSLib.{surf,pvsurf,jr,scal}.

Marker phase conversion. In ADVMarkCrossFreeSurf (src/subgrid.cpp:545) and its passive-tracer counterpart (src/passive_tracer.cpp:885), an air marker that ends up below the surface under surf_mode = 2 always becomes surf->phase, the FastScape sed_phases. The v3.1.0 SedimentModel logic now runs only in mode 1.

Unit conversions. If you add a parameter with physical units, look at the new units = none guards (src/surf.cpp:129, :150) before relying on getScalarParam's scaling argument alone.

Formatting. make check needs astyle 3.1 (§1.3).

To see exactly what moved:

git diff 7e7a012e 406b6444 -- src/surf.h src/surf.cpp src/LaMEMLib.h src/LaMEMLib.cpp src/subgrid.cpp src/passive_tracer.cpp

8. Pitfalls when upgrading to v3.2.0

Silent: surf_mode = 0 freezes the free surface

The run exits 0, the topography never changes, and no message says so. See §4.1.

Silent: built-in surface keys under surf_mode = 2

erosion_model, sediment_model, slope_dependent_erosion, topo_diff and their parameters are ignored without even a missing-parameter check. See §4.3.

Silent: vel_boundary means the opposite of what the docs say

If you follow the docs, 1111 looks like "keep LaMEM velocities everywhere", but it actually pins all four boundaries to zero. See §4.4.

Silent: deb=true in your own tests now runs a different binary

A test that passed in v3.1.0 was comparing the opt binary's output. It may now drift outside tolerance against an expected file generated by opt. See §6.1.

Silent: an exported FASTSCAPE_LIB changes what make test builds

See §6.3.

Silent: a stale build ignores surf=scape

make … surf=scape all on an existing default build does nothing, so the binary has no FastScape and FastScape tests are skipped rather than failed. Run clean_all first (§1.2).

Loud: topo_diff with units = none

topo_diff defines topo_diffusivity in [m^2/s] and requires a dimensional unit system. Set units = geo or units = si, or deactivate topo_diff.

The same applies to slope_dependent_erosion (slope_dependent_erosion defines prefactor_slope in [m/yr] and requires a dimensional unit system. …).

Loud: FastScape requested but not available or not configured

surf_mode = 2 (FastScape) requested, but this LaMEM binary was built without FastScape support. …
<FastScapeStart> - <FastScapeEnd> blocks must be defined
Incorrect unit type for FastScape
Makefile:46: *** Environmental variable FASTSCAPE_LIB must be set to installation directory.  Stop.

Loud: opt= in a test

A MethodError naming the unsupported keyword opt. Delete the keyword (§6.1).

Loud: wrong astyle

ERROR: wrong astyle version. Install astyle 3.1.

Not a pitfall: your v3.1.0 input files

Seven unmodified v3.1.0 free-surface inputs were run for 2 steps under both a v3.1.0 and a v3.2.0 binary: t10 Compressible1D_withSaltandBasement, t14 1D_VP, t15 RTI, t16 TimeTransition, t20 RTI_FSSA, t22 ridge_geom_2D and t32 BC_velocity_2D_LR. Every |Div|_inf, |Div|_2 and |mRes|_2 line was bit-identical, so the SurfMode refactor does not change surf_mode = 1 results. The expected files of t24 (erosion + sedimentation), t35 (topographic diffusion) and t36 (spatially limited erosion) did not change between the tags either.

Appendix: v3.2.0 parameter change reference

Extracted with the get*Param vocabulary diff. Each of the 46 additions was verified by direct grep to be absent from v3.1.0 src/. Line numbers refer to v3.2.0. No parameter was removed.

Free surface (surf_mode = 1 or general):

<FastScapeStart> block (src/fastscape.cpp:82–156), surf_mode = 2 only:

* n and sed_phases are not new names (see §4.3), so the bulk diff does not list them. Inside the block they are new keys with different meanings. That brings the count to 42 FastScape keys in the diff (29 in the block plus 13 output flags), and 2 reused ones.

FastScape output flags (src/fastscape.cpp:818–838), all optional: out_surf_fs, out_fs_pvd, out_surf_topofs, out_surf_silt_fraction, out_surf_basement, out_surf_total_erosion, out_surf_drainage_area, out_surf_erosion_rate, out_surf_slope, out_surf_curvature, out_surf_chi, out_surf_catchment, out_surf_lake_depth.

Totals: v3.1.0 parses 471 parameters and v3.2.0 parses 517. That is 471 unchanged, 0 removed and 46 added: 4 free-surface, 29 FastScape block, 13 FastScape output.

Other user-visible changes that are not .dat parameters:

Guide generated by diffing upstream 7e7a012e (v3.1.0, 2026‑08‑12) against 406b6444 (v3.2.0, 2026‑09‑22). It was checked by running unmodified v3.1.0 inputs and targeted variants under freshly built v3.1.0, v3.2.0 and v3.2.0 + FastScape binaries (PETSc 3.22.5, aarch64 Linux). Parameter claims were confirmed by direct grep in both trees rather than taken from the bulk vocabulary diff.