Add general realistic-global forward runs#643
Draft
xylar wants to merge 49 commits into
Draft
Conversation
Collaborator
Author
|
I have successfully run short (1-day) forward runs for the 240 km unified mesh for both MPAS-Ocean and Omega. Omega requires manually changing the time integrator to RK4. |
xylar
force-pushed
the
add-realistic-global-ocean-general-forward
branch
2 times, most recently
from
July 5, 2026 15:46
1fe3f28 to
64a105c
Compare
6 tasks
xylar
force-pushed
the
add-realistic-global-ocean-general-forward
branch
4 times, most recently
from
July 21, 2026 11:10
dffb364 to
4493593
Compare
Introduces docs/design_docs/global_ocean_init.md for the realistic global ocean initialization workflow. Key changes from the earlier draft: - Each task initializes the single model selected by `[ocean] model`; no single task produces both MPAS-Ocean and Omega outputs. - The p-star iteration step is model-independent; running the task for different models on the same mesh and hydrography yields the same geometric vertical grid, enabling comparison (by running separately). - Renames the intermediate step and its output from `shared_vert_coord` / `shared_vert_coord.nc` to `pstar_init` / `pstar_init.nc`; the implementing class is `RealisticPStarInitStep(PStarInitStep)`. - Replaces all `ztilde_init.md` cross-references with `pstar_init.md`. - Updates pstar_init.md with three cross-references to global_ocean_init.md and renames the planned realistic subclass to `RealisticPStarInitStep`. - Adds global_ocean_init to docs/design_docs/index.md. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Introduces docs/design_docs/global_ocean_dynamic_adjustment.md for the staged global-ocean dynamic adjustment workflow. Key changes from the earlier draft: - Each task targets exactly one ocean model, selected by `[ocean] model`; the workflow does not attempt to run both MPAS-Ocean and Omega in a single task. - The requirement, algorithm design, implementation, and testing sections for "supports both models" are updated to "supports either model", reflecting that model-specific differences are isolated in the translation layer and each task instance runs for the configured model only. - The summary and diagnostics descriptions are updated to match. - Adds global_ocean_dynamic_adjustment to docs/design_docs/index.md. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Introduces docs/design_docs/e3sm_init_component_inputs.md describing the e3sm/init component_inputs task family. The document already uses a model-gated product matrix (ocean_model = mpas-ocean | omega | none, seaice_model = mpas-seaice | none) that naturally supports single-model tasks. Two targeted updates reflect the single-model ocean init design: - Summary: clarifies that the upstream ocean init and dynamic-adjustment tasks each target a single ocean model (selected by [ocean] model); the component-inputs workflow consumes that model's products and does not build initial conditions for multiple models in one task. - Implementation: notes that the upstream ocean init output (step 2) and dynamic-adjustment restart (step 3) correspond to the single configured model. Adds e3sm_init_component_inputs to docs/design_docs/index.md. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Set default_cached = True on CombineStep and ExtrapolateStep so that the expensive WOA23 preprocessing steps are cached by default and not re-run when re-running tasks that depend on them. Refactor get_woa23_steps() to self-contain the topo combine step lookup (removing the separate get_woa23_topography_step() helper) and add an include_viz parameter so callers can opt into the visualization step. The standalone Woa23 task keeps include_viz=True and marks combine and extrapolate as free-running so the standalone task always runs them rather than restoring from cache. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Add polaris/tasks/ocean/realistic_global/init/CullTopoStep, which reindexes remapped topography from the base mesh onto the culled ocean mesh cells using the cell-index map produced by the upstream e3sm/init CullMeshStep. Inputs are topography_remapped.nc (from RemapTopoStep) and ocean_map_culled_to_base.nc (from CullMeshStep); output is topography_culled.nc. Also adds the package __init__.py files for polaris/tasks/ocean/realistic_global/init/ and tests/ocean/realistic_global/init/. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Add RemapWoa23Step, which uses pyremap to horizontally remap the extrapolated WOA23 conservative temperature and absolute salinity product from the 0.25-degree lat-lon grid to the culled MPAS ocean mesh. The step scales its task count from the approximate mesh cell count configured in [unified_mesh] approximate_cell_count. Output is woa23_on_mesh.nc. Includes unit tests in tests/ocean/realistic_global/init/test_remap_woa23.py. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Add polaris/tasks/ocean/realistic_global/init/pstar_init.py with RealisticPStarInitStep(PStarInitStep), which jointly initializes the p-star vertical coordinate and WOA23-derived tracer fields for a global MPAS mesh. RealisticPStarInitStep subclasses PStarInitStep (from polaris/ocean/vertical/pstar_init.py) and implements init_tracers() by interpolating conservative temperature and absolute salinity from the pre-remapped WOA23 product (woa23_on_mesh.nc) to the current p-star midpoints at each fixed-point iteration. The run() method calls run_pstar_init() and writes the model-neutral intermediate product pstar_init.nc. The step is renamed from the earlier SharedVertCoordStep design; the output file is renamed from shared_vert_coord.nc to pstar_init.nc. Includes unit tests in tests/ocean/realistic_global/init/test_pstar_init.py covering _geom_z_bot_from_topo, _fill_tracer_columns constant/linear/valid- level/deep-extrapolation behavior. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- No `model` constructor arg; reads `config.get('ocean', 'model')` at
run time (concrete after Ocean.setup() resolves 'detect').
- Inputs: `pstar_init.nc` (was `shared_vert_coord.nc`) from
`RealisticPStarInitStep`, plus culled mesh and graph files.
- Calls `add_output_files_for_ocean_model_input` to register
model-gated outputs (vert_coord.nc for Omega, culled_graph.info for
MPAS-Ocean).
- Tracer conversion (CT/SA -> pot-T/prac-S via GSW) still only applies
for MPAS-Ocean.
- Tests drop the obsolete `model`-constructor and `_effective_model()`
tests; keep helper-function tests for `_add_layer_thickness`,
`_add_normal_velocity`, and `_convert_tracers_mpas_ocean`.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
- Add get_realistic_init_steps(component, mesh_name) in steps.py that
composes the full shared-step dependency chain:
get_cull_topo_steps -> WOA23 steps -> CullTopoStep -> RemapWoa23Step
-> RealisticPStarInitStep -> InitialStateStep; all new steps use
get_or_create_shared_step for reuse across tasks on the same mesh
- Add RealisticGlobalInit task in task.py with subdir
'spherical/realistic_global/init/{mesh_name}' (no /{model} segment);
calls set_shared_config and adds all returned steps
- Add add_realistic_global_init_tasks in tasks.py: one task per mesh,
no model loop; target model resolved from config at run time
- Register add_realistic_global_init_tasks in realistic_global/__init__.py
- Add realistic_global_init.cfg placeholder config
- Add approximate_cell_count to all four unified mesh cfg files for use
by RemapWoa23Step._estimate_cell_count
- Update tests to expect one task per mesh with pstar_init/initial_state
step names; drop per-model and shared-step-reuse tests
- Remove stale get_woa23_topography_step from api.md; add init package
entries; update realistic_global.md to describe the init step chain
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Co-authored-by: Carolyn Begeman <cbegeman@lanl.gov>
- Add VizInitStep in viz.py, a model-agnostic OceanIOStep that reads
through open_model_dataset / open_vert_coord_dataset (mapping Omega
names to MPAS-Ocean and reconstructing geometric layerThickness) so
the maps and transects work for both models. It produces:
* initial_state_summary.png: histograms ported from Compass'
plot_initial_state with the Haney (rx1Edge) panel dropped; the
prognostic-thickness panel shows each model's native variable
(layerThickness for MPAS-Ocean, PseudoThickness for Omega) read
from the raw output file
* vertical_coordinate.png: vertical structure from the geometric
restingThickness of the deepest column (no ref* profiles exist in
this workflow)
* global native-mesh maps (plot_global_mpas_field) of temperature and
salinity at configured depths plus surface/seafloor, and
bottomDepth, ssh, maxLevelCell and column thickness; for Omega also
native surfacePressure/bottomPressure and a TEOS-10 in-situ Density
(stratification) check
* multi-basin vertical transects (compute_transect / plot_transect)
of temperature and salinity (and Density for Omega)
* xdmf/init and (Omega) xdmf/vert_coord exports via MpasToXdmf,
keeping Omega variable names but renaming dimensions to MPAS-Ocean
- Create the viz step as a shared step in get_realistic_init_steps via
get_or_create_shared_step; add an include_viz flag so it is only
added to steps_to_run when requested (keeps it out of other
consumers' run lists)
- Pass include_viz=True from the standalone RealisticGlobalInit task so
viz runs by default there
- Add [realistic_global_init_viz] and
[realistic_global_init_viz_transects] config sections (projection,
central_longitude, depths, and four default basin transects)
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
- Add viz.VizInitStep entries (setup, run) to the ocean developer API - Describe the viz step in the developer guide: its shared/include_viz behavior, model-agnostic reads, and the full set of outputs (summary and vertical-coordinate figures, global maps, transects, Omega-only pressure and stratification plots, and ParaView XDMF exports) - Add a user-guide init section covering the visualization outputs and the [realistic_global_init_viz] / [realistic_global_init_viz_transects] config options Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
The realistic_global/init p-star coordinate produced a spurious non-zero ssh (down to -1590 m) and a runaway BottomPressure (up to ~2.2e8 Pa, a pseudo-depth of ~21,700 m) in ~420 deep cells. The 80layerE3SMv1 reference grid defines pseudo-thicknesses whose deepest interface is ~5549.7 m (pseudo), but run_pstar_init targets the geometric seafloor depth from base_elevation (up to 6761 m). Since geometric thickness = RhoSw * specVol * pseudoThickness and deep water is denser than RhoSw, the deepest representable geometric column is only ~5171-5475 m; where base_elevation is deeper, the geometric column saturates, the fixed-point iteration diverges, and the residual shows up as negative ssh and an inflated BottomPressure inconsistent with the clipped PseudoThickness. Clamp the target seafloor into the representable range [-D_max, -D_min] before calling run_pstar_init, so the existing loop converges to a resting state (ssh = 0), BottomPressure = P_max (no runaway), and bottomDepth = the clamped geometric thickness: - Add RealisticPStarInitStep._clamp_to_representable_depth(), which builds one saturated full-grid column (inherited base _build_pstar_coord_ds + this step's init_tracers), and the pure helper _clamp_geom_z_bot(), which computes D_max (geometric thickness of the saturated column) and D_min (the deeper of min_bottom_depth and the depth needed for min_vert_levels layers, capped at D_max) and clamps geom_z_bot. - ocean.cfg: add neutral default min_bottom_depth = 0.0 (clamp disabled; min_vert_levels stays 1) so other p-star users (e.g. horiz_press_grad) and the base run_pstar_init tests are unaffected. - realistic_global_init.cfg: set min_bottom_depth = 10.0 and min_vert_levels = 3 (three layers is the practical minimum for baroclinic dynamics). - Add unit tests for _clamp_geom_z_bot (deep/shallow/in-range clamp, min_bottom_depth floor, density-reduced D_max, attribute passthrough). No changes to the shared base run_pstar_init (pstar_init.py) or pstar.py, so horiz_press_grad's p-star path is untouched. Verified end-to-end on icos240 (Omega): max|ssh| = 1.9e-8 m (was -1590 m), BottomPressure <= P_max for all cells, bottomDepth in [10, 5475] m, maxLevelCell >= 5. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
The p-star initialization path bypassed init_vertical_coord/add_1d_grid, so pstar_init.nc lacked the 1D reference vertical coordinate variables (refBottomDepth, refZMid, refTopDepth, refInterfaces). MPAS-Ocean requires refBottomDepth as an input on its mesh stream. Reuse the existing add_1d_grid() in the base PStarInitStep output assembly so every p-star init step produces these fields, and expose a shared REF_COORD_VARS constant for downstream routing/exclusion. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
The base p-star init now emits refBottomDepth/refZMid/refTopDepth/ refInterfaces for MPAS-Ocean. Omega uses RefPseudoThickness instead and ignores these, so drop them from the Omega initial state file to keep its output minimal and model-appropriate. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Assert the base PStarInitStep output contains refBottomDepth and its companions, verify they are kept for MPAS-Ocean but dropped for Omega in write_initial_state_dataset, and guard that REF_COORD_VARS stays in sync with add_1d_grid. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Reorder the realistic_global init work-directory layout from
spherical/realistic_global/init/{mesh_name}/... to
spherical/realistic_global/{mesh_name}/init/... so that a mesh's
tasks are grouped under the mesh rather than scattered under sibling
family directories as more task families (forward, dynamic_adjustment,
restart, ...) are added.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Mirror the reordered work-directory layout
spherical/realistic_global/{mesh_name}/init/... in the task and step
subdir assertions.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Reflect the reordered work-directory layout
spherical/realistic_global/{mesh_name}/init in the developers guide
and the pstar_init design doc.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
The realistic_global init step left mesh.nc as a bare symlink to the culled mesh, which lacks the Coriolis parameter (fCell/fEdge/fVertex). The ocean model requires these in its horizontal mesh stream, so Omega aborted reading FVertex/FEdge/FCell from HorzMeshIn. Add the Coriolis fields via the shared add_coriolis_to_dataset helper (as every other ocean task does) and write a real mesh.nc, reading the culled mesh from a renamed culled_mesh.nc input. A [coriolis] section with type = spherical is added to the init cfg so a realistic Earth mesh gets f = 2 * omega * sin(latitude). Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Ocean-model resource and time-step sizing for the realistic_global meshes should reflect the ocean + sea-ice culled domain that MPAS-Ocean actually runs, not the full mesh (which includes finer land/river refinement). Add ``get_unified_ocean_finest_cell_width`` which returns the finest ocean background cell width, ignoring the land and river-channel widths, and add a ``culled_ocean_cell_count`` option to the unified mesh configs. For the constant 30 km ``u.oi30.lr10`` mesh this is ~470000 cells (~0.71 * 6e8 / 30**2); the variable meshes carry a commented placeholder to be filled in from a trial run, falling back to ``approximate_cell_count`` until then. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
The private ``_estimate_cell_count()`` helper lived in the WOA23 remapping step, but sizing a step from the mesh name is generally useful and will be needed by other steps (e.g. the forward run) as well. Move it into a new ``realistic_global/mesh_info.py`` as public ``estimate_cell_count()``, and add ``estimate_ocean_cell_count()``, which reads the ``culled_ocean_cell_count`` option added to the unified mesh configs (falling back to the full-mesh estimate when it is unset). WOA23 is remapped onto the culled ocean mesh, so ``remap_woa23`` now sizes itself from the ocean-culled count rather than the full unified-mesh estimate, which includes the finer land and river-channel refinement. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Only u.oi30.lr10 had culled_ocean_cell_count set, so the other three meshes fell back to approximate_cell_count, which includes the finer land and river-channel refinement that is culled away before the ocean model runs. That badly over-estimated the ocean domain and led to unreasonable node counts for realistic_global init steps. Set the ocean-culled counts for the remaining meshes: 7500 for u.oi240.lr240, 4000000 for u.oi6to18.lr6to10 and 800000 for u.oi.so12to30.lr10. Add a test that every unified mesh defines the option and that it is smaller than the full-mesh count. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Add the standard topography variables from the remapped topography dataset as validation variables for topography_culled.nc so that the culled fields are compared against a baseline when one is provided. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Adds a package to hold per-mesh config overrides for realistic_global ocean tasks, one optional <mesh_name>.cfg per mesh. These are deliberately separate from the per-mesh configs in polaris.mesh.spherical.unified: those belong to the mesh component and describe the mesh itself, while options specific to the ocean (vertical grid, ocean-culled cell count) belong here in the ocean component. The two are joined by the mesh name. Meshes with no config file here simply use the task defaults. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Adds per-mesh configs giving qu240km, icos240km and u.oi240.lr240 a 16-level tanh_dz vertical grid with a 3000 m bottom depth, matching the Compass configuration for these meshes. The shallow bottom depth is intentional: these coarse meshes are for cheap smoke-testing rather than physical realism. Only the five options that differ are overridden; coord_type, min_vert_levels, min_bottom_depth and the rest are inherited from realistic_global_init.cfg. Wires the overrides into _get_init_config(), which now takes mesh_name and applies them last so they take precedence over the task defaults. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
The ocean-culled cell count is an ocean concept that was living in the mesh component's unified-mesh configs. It moves to the new per-mesh configs under realistic_global/mesh_configs, in a new [realistic_global_mesh] section, so the mesh component no longer carries options specific to the ocean. approximate_cell_count stays behind: it is a property of the mesh itself. estimate_ocean_cell_count() also gains an optional config argument. Its caller in remap_woa23 has the task config in hand, so passing it means a user override of culled_ocean_cell_count now affects task sizing; reading the package file directly ignored user overrides. Without a config the function still falls back to the package file. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Covers the vertical-grid override on the 240 km meshes, that unset options are still inherited from realistic_global_init.cfg, and that a mesh with no per-mesh config keeps the defaults. Also checks that every config file in the package is named after a mesh that exists. A misnamed file would otherwise be silently ignored, since a mesh without a config file is a valid and common case. Adds coverage for the new config argument to estimate_ocean_cell_count. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Adds a users guide section describing the mesh_configs override mechanism, which meshes currently override what, and the rule for deciding whether a per-mesh option belongs to the mesh component or the ocean component. Adds the new module and mesh_info to the ocean API docs. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
pyremap's Remapper defaults to use_tmp=True, which puts the SCRIP files and the MOAB .h5m files in a TemporaryDirectory() under /tmp. On Chrysalis (and many other HPC machines) /tmp is a node-local disk, so the MPI tasks running mbtempest on other nodes cannot see the files and abort during the parallel read. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
RemapWoa23Step reimplemented mapping-file creation that polaris.remap.MappingFileStep already provides, and it ran both an MPI operation (mbtempest) and a serial one (ncremap) in a single step, which will not schedule well under task parallelism. Add Woa23MapStep, a MappingFileStep subclass that builds the bilinear weights and carries the ntasks sizing from the estimated ocean cell count. RemapWoa23Step becomes a serial step that depends on it and applies the weights with ncremap. Getting the remapper from MappingFileStep also picks up parallel_executable from config, which pyremap's default of bare 'mpirun' does not honor. The bilinear method is now passed explicitly rather than relying on pyremap's default. The woa23_on_mesh.nc output is unchanged, so pstar_init needs no modification. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
MappingFileStep reads [mapping] map_tool from polaris/remap/mapping.cfg, so add that package config and drop the duplicated map_tool option from [realistic_global_init]. Raise remap_cells_per_task from 500 to 10000 and remap_min_cells_per_task from 5000 to 50000. The old values asked for 926 MPI tasks for u.oi30.lr10 (462,924 culled ocean cells), which only ran at 512 because constrain_resources clipped it to the allocation. Timings on Chrysalis show that more tasks is actively worse here: the mbtempest weight computation takes only 5-8 s at any task count, while the serial mbpart partitioning that precedes it gets more expensive as the number of partitions grows. 512 tasks took 97 s, 128 took 52 s, 93 took 45 s and 64 took 44 s. The new values give ntasks=47 and min_tasks=9 for that mesh, halving the core reservation relative to 5000 cells per task at no measured cost in runtime. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Cover that Woa23MapStep uses bilinear remapping and does not put SCRIP files in /tmp, that its task counts follow the culled ocean cell count and stay modest for a coarse mesh, and that RemapWoa23Step is serial and reaches the remapper through the step dependency mechanism. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
pyremap partitions the SCRIP files with 'mbpart <ntasks>' before running
mbtempest, and mbpart rejects a request for one partition:
** Please specify #parts = 1 to be greater than 1.
A serial test run confirmed this fails outright. With the cell counts
per task raised in the previous commit, icos480km resolved to ntasks=1
and icos240km/icos120km to min_tasks=1, all of which would have hit
this. Floor both at 2.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Add docs/design_docs/global_ocean_forward.md describing a reusable framework for realistic global-ocean forward runs with MPAS-Ocean or Omega: a Forward step built on OceanModelStep, two initial-condition sources (Polaris database file and the realistic_global/init outputs), a model-agnostic per-run settings structure mapped to the configured model in one place, and restart in/out seams that leave clean paths to restart tests and the staged dynamic-adjustment workflow. Register the doc in the design-docs index. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Add the model-agnostic pieces the forward step will use, not yet wired to a step: - ForwardStage (forward/stage.py): a dataclass of per-run settings (durations, time step, damping, restart-in fields) with from_config and a single model_replacements mapping onto MPAS-Ocean/Omega template values, plus bottom_drag_options for MPAS-Ocean Rayleigh damping. - InitialCondition sources (forward/initial_condition.py): a base class with StepInitialCondition (links mesh/vert_coord/init/graph from an upstream realistic_global/init step) and DatabaseInitialCondition (constructs a database filename; not yet registered). - realistic_global_forward.cfg: the [realistic_global_forward] section. Co-locate the DDDD_HH:MM:SS string helpers so a function and its inverse live together: move get_time_interval_string into polaris.mpas.time next to the new duration_to_seconds parser, and re-export both from polaris.mpas and polaris.ocean.model so existing imports are unchanged. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Add Forward(OceanModelStep), which wires its model inputs from an InitialCondition and renders forward.yaml from a ForwardStage in a single dynamic_model_config mapping site. It sizes resources from the exact nCells in the mesh at run time (config approx_cell_count fallback during setup) and derives the mesh minimum resolution from dcEdge for per-km time-step scaling. forward.yaml sets the neutral time block plus the MPAS-Ocean restart/ output streams and Omega History/RestartWrite streams; input streams come from the shared model_inputs.yaml. Restrict Omega to the RK4 (RungeKutta4) time integrator, raising a clear error if split_explicit_ab2 is selected for Omega, since the two are not equivalent and an Omega split-explicit integrator is still in development. Add restart_freq to the stage's model replacements for the Omega restart stream. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Add a runnable simple-forward task: RealisticGlobalForward (task.py) runs the shared realistic_global/init steps for a mesh, then a single Forward step wired to their outputs via StepInitialCondition, with its run settings coming from [realistic_global_forward]. add_realistic_global_ forward_tasks (tasks.py) registers one task per base and unified mesh, mirroring the init factory, and add_realistic_global_tasks now calls it. Make the Forward step's stage optional: when a task does not pass an explicit ForwardStage (the simple-forward case), the step builds one from config in dynamic_model_config so user config overrides take effect; dynamic adjustment will pass explicit per-stage settings instead. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Add tests/ocean/realistic_global/test_forward.py covering the ForwardStage mapping (per-km vs explicit time step, RK4->RungeKutta4, the Omega split_explicit_ab2 error, output/restart freq, Rayleigh damping options, restart_interval defaulting) and the InitialCondition sources (StepInitialCondition input/graph wiring and DatabaseInitialCondition filename construction for both models). Extend test_tasks.py to confirm a RealisticGlobalForward task is registered per mesh and that its forward step and shared init chain land in the expected work dirs. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
The forward stage previously set config_dt from dt_per_km (the long baroclinic step) for every integrator. That is only correct for split time stepping (split_explicit_ab2), which subcycles the barotropic mode on config_btr_dt. Non-split integrators (RK4, and Omega's RungeKutta4 -- the only Omega option) have no barotropic subcycling and must advance on the short barotropic step, so they now use btr_dt_per_km for config_dt. Without this, Omega runs would take the far-too-large baroclinic step and go unstable. Add a test covering the non-split short-step behavior and clarify the dt_per_km / btr_dt_per_km cfg comments. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
The forward step previously read the mesh minimum resolution from a [realistic_global_forward] min_res config option (default 30 km), an unsafe guess, and otherwise from the mesh file at run time. Instead, min_res is now a property of the InitialCondition: - StepInitialCondition takes min_res, which the task looks up from the existing mesh definitions -- get_base_mesh_definition().min_res for base meshes and get_unified_finest_cell_width() for unified meshes. - DatabaseInitialCondition takes min_res as a required argument, since a database mesh has no init pipeline to derive it from and it cannot be guessed. Remove the min_res config option and the run-time dcEdge read. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
An unsupported Omega time integrator was raised at setup time, which prevented setting up the step and changing the config option before running. Thread at_setup through model_replacements so the neutral integrator name is left in place at setup and validated only at run time. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
The forward step sized setup-time resources from a fixed [realistic_global_forward] approx_cell_count = 100000 whenever the mesh file did not yet exist. That one number was used for every mesh, so it badly over-provisioned coarse meshes (e.g. u.oi240.lr240 targeted 500 / required 48 tasks instead of the mesh-appropriate count), which in turn oversized the batch job. Instead, derive the estimate per mesh, mirroring the earlier min_res refactor: - Move the base-vs-unified mesh-property lookups into a shared realistic_global/mesh_info.py, exposing public estimate_cell_count() (6e8 / min_res**2 for base meshes; [unified_mesh] approximate_cell_count for unified meshes) and min_res_for_mesh(). remap_woa23 and forward/task both use it. - Carry approx_cell_count on the InitialCondition, set from the mesh definition at task registration. - Forward.compute_cell_count() falls back to that estimate (and raises a clear error if it is unknown) instead of the config option. - Remove the now-unused approx_cell_count config option. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
The forward step lived at .../{mesh}/forward/forward: the per-mesh
"forward" group directory (holding the task at forward/task) plus a step
also named "forward". The doubled name was confusing and said nothing
about what the run is.
Rename the step (and its work-dir leaf) to "short", so the brief 1-day
smoke run lands at .../{mesh}/forward/short, beside .../{mesh}/forward/task
and consistent with the init layout (init/task + init/<step>). This
leaves room for other forward runs under the same forward/ group later.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
The realistic_global forward step sized its resources and time step from the full unified-mesh estimate, which includes finer land/river refinement that is culled away before MPAS-Ocean runs. For u.oi30.lr10 that made the setup-time estimate (2,000,000 cells, and a 10 km min resolution from the land/river width) far larger than the ocean domain the model actually runs. Add ``estimate_ocean_cell_count``, which reads the new ``culled_ocean_cell_count`` option (falling back to the full estimate when unset), and use it for the forward step. Switch ``min_res_for_mesh`` to the ocean-only finest cell width so unified meshes report their ocean resolution (30 km for u.oi30.lr10) rather than the land/river refinement. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
The forward task built its shared config from the task's own realistic_global_forward.cfg only, so per-mesh overrides in the mesh_configs package were ignored. Call add_realistic_global_mesh_config() after the task config is added, as the init steps already do. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
xylar
force-pushed
the
add-realistic-global-ocean-general-forward
branch
from
July 21, 2026 11:20
4493593 to
3c30fa5
Compare
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
(Full description to come)
This PR extends #612 and generalizes #593 to support both initial conditions from the Polaris workflow and from the Polaris database.
It also provides support for future restart tests and dynamic adjustment workflows.
Checklist
api.md) has any new or modified class, method and/or functions listedTestingcomment in the PR documents testing used to verify the changes