03. Grids, bathymetry, masks

Four grid types

Unstructured is where coastal work is going, because the resolution mismatch between a 1000 km shelf and a 200 m inlet is brutal on a rectilinear grid. The cost: unstructured WW3 uses implicit schemes and needs PDLIB + ParMetis for domain decomposition (added v6.04), which is a build-time complication.

Grid closure

GRID%CLOS matters only for global spherical grids:

• NONE: no wrap. Regional.
• SMPL: periodic in i; (NX+1, J) → (1, J). Forces SX = 360/NX.
• TRPL: tripole. Periodic in i and folded at j = NY+1: (I, NY+1) → (NX−I+1, NY). Requires NX even and a curvilinear grid. This is how you avoid the North Pole singularity on a global grid.

Bathymetry

Sign convention: negative below mean sea level. This is opposite to how most people think about "depth" and it is the single most common grid error. GEBCO's elevation is already negative below sea level, so DEPTH%SF = 1. For a file of positive depths, DEPTH%SF = -1.

IDLA decides which way is up. 1 = bottom row first, 3 = top row first. Check by comparing mask.ww3 (ASCII, written by ww3_grid) against a map. Do it visually, once, every time you build a new grid.

ZLIM vs DMIN. ZLIM (e.g. -0.10) is the coastline: points shallower than this are permanently excluded and will never be wet even if water level rises. DMIN (e.g. 2.50) is a floor applied to the depth used in the physics, so a nearly-dry cell doesn't produce division-by-zero. They do different things and both matter.

Generating the files

ww3_grid does not read netCDF bathymetry. It reads plain ASCII arrays, one row per line in the order IDLA declares, and you are expected to produce them. The two examples each ship a small Fortran program for that, and the point of generating the files rather than committing them is that you can change NX, NY or the depth and rerun, which is exactly what the exercises ask you to do.

Each example's run.sh compiles its generator with gfortran if the binary is missing, runs it, and then runs the WW3 programs. Read the sources: they are short, and the mask legend and the IDLA convention are spelled out in the comments.

Nearest-neighbour sampling is a deliberate downgrade from interpolation. Point-sampling a 15-arc-second dataset onto a 0.1° grid throws away most of the information and can miss a whole shoal; for production work you want area-weighted (conservative) regridding, which is a job for a proper regridding tool, not a forty-line program. For a course grid it is fine, and it has one virtue interpolation lacks: every depth in bathy.inp is a number GEBCO actually reported.

Masks

 -2  excluded boundary point (ice-covered)
 -1  excluded sea point (ice-covered)
  0  excluded land point
  1  sea point
  2  active boundary point
  3  excluded grid point
  7  ice point

Note 2: boundary points are declared in the mask, and/or via &INBND_POINT_NML in ww3_grid.nml. ww3_grid will promote active points on the declared boundary segments to status 2 and report how many.

Obstruction grids: the underrated one

At 0.1° you cannot resolve a 3 km island. But that island blocks waves. WW3's answer is subgrid obstruction: a per-cell transparency in x and y, between 0 (fully blocking) and 1 (fully transparent), applied during propagation.

Controlled by &MISC FLAGTR in namelists.nml:

0  no obstruction (and then you must NOT supply an obstruction file)
1  transparency at cell boundaries
2  transparency at cell centres
3  as 1, with continuous ice
4  as 2, with continuous ice

Generate these with gridgen (MATLAB). Without them, archipelagos are invisible and your lee-side wave heights are badly wrong. With them, a coarse global grid produces surprisingly decent results in the Indonesian seas and the Aegean.

Choosing resolution

Two competing constraints:

1. Physics. Resolve the bathymetric features that refract the waves you care about.
2. CFL. Δt ∝ Δx. Halve the grid spacing and you double the timestep count and quadruple the point count: 8× the cost for a 2D grid.

The standard resolution is a nested mosaic: a coarse global grid (0.5°) feeding a regional grid (0.1°) feeding a coastal grid (0.01°). Either as separate runs chained through ww3_bounc, or as a single ww3_multi run with two-way interaction. See lesson 05.

Exercise

Build the same domain three ways (0.2°, 0.1°, 0.05°) with correctly recomputed timesteps (edit NX, NY, SX, SY in make_bathy.F90 and ww3_grid.nml together), and run example 02's forcing through all three. Pull Hs at the shelf-break point out of each ww3.nc with ncks -v hs -d longitude,<i> -d latitude,<j> (or ncdump -v hs and your eyes) and tabulate the three time series side by side. Where does it converge? Where doesn't it? That answer is specific to your coastline and it's the only honest way to pick a resolution. Lesson 06 shows the tools, and nccmp-tol is the wrong tool for this one: the grids differ, so the fields aren't comparable point by point.

→ 04-forcing.md