05. Nesting and multi-grid

Two ways to combine grids of different resolution. They are genuinely different, not two spellings of the same thing.

One-way nesting: chained ww3_shel runs

coarse run  →  ww3_ounp at points along the fine grid's boundary
            →  ww3_bounc  →  nest.ww3
            →  fine run

Simple, debuggable, embarrassingly parallel across the fine domains, and the standard approach when the parent is somebody else's operational output. The cost: no feedback. The fine grid cannot tell the coarse grid that a storm just generated a lot of energy.

Boundary setup lives in ww3_grid.nml for the fine grid:

&INBND_COUNT_NML
  INBND_COUNT%N_POINT = 2
/
&INBND_POINT_NML
  INBND_POINT(1) = 2  2  F
  INBND_POINT(2) = 2  80 T     ! CONNECT=T fills in everything between
/

Rules, both of which catch everyone:

• Boundary points must be inside the grid. Never on row/column 1 or N. The propagation stencil needs a cell outside them. This is the most frequently asked question on the WW3 Discussions board.
• CONNECT = T fills in every point on the straight line or diagonal from the previous point. Two entries define a whole edge.

Reference configuration: regtest ww3_tp2.17.

Two-way: ww3_multi

One executable runs a mosaic of grids simultaneously, exchanging spectra both directions each timestep. Driven by ww3_multi.nml instead of ww3_shel.nml, with one mod_def.<grd> per grid.

The grid definition line carries a rank and a group number. Ranks control which grid wins where they overlap; groups control which grids are computed concurrently. Getting this right is the actual skill in multi-grid WW3, and the reference material is the mww3_test_01 through mww3_test_08 regression tests plus the UFS documentation, which describes the NFGRIDS / WINDLINE / ICELINE / CURRLINE / UNIPOINTS / WW3GRIDLINE structure.

UNIPOINTS is worth knowing about: it gathers point output from all grids into one unified file, which otherwise is a genuine annoyance.

Two-way is what NOAA runs operationally. It is also substantially harder to debug, because a problem in one grid propagates into all of them.

Which to use

Start one-way. Move to ww3_multi when you have a specific reason.

Unstructured as an alternative

A single unstructured mesh with 10 km elements offshore and 200 m in the inlet sidesteps nesting entirely. The trade: implicit schemes, PDLIB + ParMetis for parallel decomposition, and mesh generation becomes its own craft (OceanMesh2D, GMSH, SMS). Regtest ww3_tp2.7 is the reference.

→ 06-output.md