solver#

hydromodpy.solver exposes the backend abstraction (base/) plus three concrete backends: MODFLOW 6, MODFLOW-NWT, and Boussinesq. It also hosts shared helpers (modflow_common, modflow_grid) that stay backend-agnostic. Time-grid primitives now live in hydromodpy.discretization.time.

GR4J is not a fourth hydromodpy.solver backend in V1. It is a calibration-side lumped catchment model under hydromodpy.calibration.lumped; it runs in memory and only promoted trials are written through the result catalog.

Sub-modules#

  • solver/base/ – registry, SolverAdapter Protocol, Solver lifecycle. Lazy-loads built-in adapters and extractors through _BUILTIN_PATHS and _BUILTIN_EXTRACTOR_PATHS.

  • solver/boussinesq/ – in-house 2D unconfined flow solver. Triangular runtime meshes; chained resolution flow.flow_regime -> flow.runtime_backend -> flow.surface_interaction_model -> methods / engines / runtime selection.

  • solver/modflow6/ – FloPy-backed MODFLOW 6 wrapper. Flow through adapters/flow.py, transport through adapters/transport.py, output extraction through extractors/.

  • solver/modflow_nwt/ – legacy MODFLOW-NWT wrapper plus Modpath / MT3DMS transport adapters.

  • calibration/lumped/ – GR4J adapter and extractor. This lives outside solver/ because it does not use the staged flow-runner lifecycle or solver binaries.

  • solver/modflow_common/ – shared MODFLOW-family helpers (executables, boundary packages, flux translators, runtime arrays, calibration extractors).

  • solver/modflow_grid/ – backend-agnostic grid primitives: SolverMesh, SolverGridContext, DisDescriptor, DisvDescriptor, build_spatial_discretization, build_temporal_discretization.

  • discretization/time/ – TimeGrid, TmeshGenerator, TMeshConfig for stress-period generation.

Adapter contract#

Every solver implements SolverAdapter from solver/base/protocol.py:

class SolverAdapter(Protocol):
    process_type: ClassVar[str]
    solver_name: ClassVar[str]
    requires: ClassVar[tuple[tuple[str, str], ...]] = ()

    def validate(self, ctx) -> None: ...
    def execute(self, ctx) -> RunExecutionResult: ...
    def cleanup(self, ctx) -> None: ...
    def extract_calibration_series(self, ctx, store, *, variable, ...) -> "pd.Series": ...

Output adapters implement the dual contract:

class SolverOutputAdapter:
    solver_name: ClassVar[str]
    def extract(self, sim_id, solver_output_dir, store) -> None: ...

Registry resolution#

ctx.run.process.process_type plus ctx.run.process.solver_name resolves through the registry:

("flow", "modflow6")
    -> _BUILTIN_PATHS lookup or plugin entry point
    -> Modflow6FlowAdapter (instantiated by get_solver_adapter)
    -> adapter.execute(ctx)
    -> get_extractor("modflow6") -> Modflow6OutputAdapter
    -> extractor.extract(sim_id, output_dir, store)

Built-in capabilities are declared in _BUILTIN_CAPABILITIES and queryable through capabilities(process_type, solver_name).

Key public symbols#

  • hydromodpy.solver.base.protocol.SolverAdapter

  • hydromodpy.solver.base.registry.{register, get, get_solver_adapter, register_extractor, get_extractor, list_pairs, capabilities, required_bindings, load_plugins}

  • hydromodpy.solver.modflow_grid.{SolverMesh, SolverGridContext, DisDescriptor, DisvDescriptor, build_spatial_discretization, build_temporal_discretization}

  • hydromodpy.discretization.time.{TimeGrid, TmeshGenerator, TMeshConfig, load_tmesh_toml, validate_tmesh_config_data}

  • hydromodpy.solver.modflow_common.{flow_adapter_helpers, calibration_extractors, flow_translator, binaries, boundary_packages, executables, runtime_arrays}

Layer-matrix neighbours#

  • Allowed targets: core, schema, physics, spatial, discretization, solver, simulation.

  • Allowed sources: config, calibration, workflow, project and cli.

  • Each backend folder is independent: solver/modflow6/ cannot import solver/modflow_nwt/. Shared helpers live in solver/modflow_common/ and solver/modflow_grid/.

See also#

On this page