Swath data lives on the satellite's curvilinear (n_lines, n_pixels) geometry. For spectra, derivatives, comparisons with models, or combining passes, you usually want the field on a regular grid. odsl-swot-regrid resamples Expert or Unsmoothed swaths onto a lat/lon or Cartesian grid, with sensible quality masking and corrections built in.

What you'll learn

  • Regridding one granule onto a lat/lon grid, with a quicklook plot.
  • Batch-regridding many granules onto one shared grid.
  • Cartesian (metre-spaced) regional grids.
  • What the quality/masking defaults do and how to change them.

Prerequisites: tutorial 1. Example: pass 13 over the Gulf of America box -98 18 -80 31.

1. One file, lat/lon grid

odsl-swot-regrid \
    --input "/spray/swot/versionD/cycle_020/SWOT_L2_LR_SSH_Expert_020_013_*.nc" \
    --variable ssha_karin_2 \
    --bbox -98 18 -80 31 --resolution-deg 0.02 \
    --output ~/swot/grids/c020_p013_ssha.nc \
    --plot

This defines a 0.02° grid over the box, resamples the (quality-masked, corrected) SSHA onto it with a Gaussian kernel of 20 km radius of influence (the defaults), and writes a NetCDF plus a PNG next to it. Grid cells with no swath pixels in range become NaN.

Options you'll touch most:

  • --resolution-deg — one value, or two (D_LON D_LAT).
  • --method — gaussian (default), nearest, linear, bilinear, or kriging (requires the pykrige extra).
  • --radius-of-influence — search radius in metres (default 20000). Larger = smoother and fewer gaps; smaller = crisper and gappier.
  • --plot, --cmap, --vmin/--vmax — quicklook cosmetics (default colormap RdBu_r, centred on zero; --no-center-zero to disable).

The --input glob may match one file or many — with a single --output it must resolve to one file; for many, see the next section.

2. Batch: many granules, one shared grid

To compare cycles cell-for-cell, every file must land on the same grid. Save the grid definition once and reuse it:

odsl-swot-regrid \
    --input "/spray/swot/versionD/cycle_0*/SWOT_L2_LR_SSH_Expert_*_013_*.nc" \
    --variable ssha_karin_2 \
    --bbox -98 18 -80 31 --resolution-deg 0.02 \
    --grid-file ~/swot/grids/gulf_002deg.nc \
    --output-dir ~/swot/grids/pass013 \
    --plot-dir ~/swot/grids/pass013/png
  • --grid-file — created on first use, loaded on every subsequent run (so a later run for pass 26 on the same grid just points at the same file; --overwrite-grid rebuilds it).
  • --output-dir — one output NetCDF per input granule, named after it.
  • --plot-dir — optional PNG per file.

Note the quotes around the glob — the tool expands wildcards itself, which keeps the command line short with hundreds of files.

3. Cartesian grids

For dynamics work where you want uniform metre spacing (spectra, gradients), build a local equal-area grid instead:

odsl-swot-regrid \
    --input "/spray/swot/versionD/cycle_020/SWOT_L2_LR_SSH_Expert_020_013_*.nc" \
    --grid-type cartesian --center-lon -89 --center-lat 25 \
    --width-m 1500000 --height-m 1200000 --dx-m 2000 \
    --output ~/swot/grids/c020_p013_cart.nc

The grid is centred at --center-lon/--center-lat with --dx-m spacing (default 2000 m; --dy-m defaults to --dx-m), sized either by extent (--width-m/--height-m) or by cell count (--nx/--ny). Cartesian grids work with --grid-file and batch mode exactly like lat/lon grids.

4. What happens to the data on the way through

By default, before resampling, the tool:

Step Default Switch
Quality mask keep pixels with flag ≤ 2 (ssha_karin_2_qual) --max-quality, --no-quality-mask, --quality-variable none
Crossover quality mask keep pixels with height_cor_xover_qual ≤ 1 --max-xover-quality, --no-xover-quality-mask
Ocean mask drop land/lakes via depth_or_elevation --no-ocean-mask
Crossover correction add height_cor_xover to SSHA --no-xover-correction
Internal-tide correction add internal_tide_hret to SSHA --no-internal-tide-correction

So the defaults give you corrected, ocean-only, good-quality SSHA — the same conventions as the subset cubes in tutorial 4. For non-SSHA variables (e.g. sig0_karin_2) the corrections don't apply, but quality masking does.

For the Unsmoothed product add --product-type Unsmoothed (or let auto detect it from the filename) and pick sides with --side left|right|both.

Troubleshooting

  • Output is all NaN — your --bbox doesn't overlap the granule's swath (check with odsl-swot-swath-map, tutorial 3), or the radius of influence is too small for the grid spacing.
  • Blocky/duplicated-looking field — --method nearest with a coarse grid; switch to gaussian.
  • pykrige import error — kriging is an optional extra: pip install -e ".[kriging]".
  • Regridding is slow on big batches — raise --nprocs for pyresample's parallel kd-tree, and keep the bbox tight so the source subset stays small.

For swath-aligned 1 km OI (Unsmoothed KaRIn, no geographic rotation) and the 250 m FFT-tile catalog, see tutorial 8.

Next: Tutorial 6 — The PIXC pixel cloud.