Geographic regridding (odsl-swot-regrid, tutorial 5) maps swaths onto lat/lon or Cartesian grids. This tutorial covers the swath-aligned OI products used for KaRIn noise work: a full-pass 1 km grid, and 30 km tiles at 250 m after an FFT low-pass.

Coordinates stay in the swath frame (x = cross-track metres, y = along-track geodesic metres). There is no rotation to geographic north. Output NetCDFs still carry latitude / longitude on the new grid.

Prerequisites: tutorial 1. Science cycle 3 pass 1 is a good first example.

1. Uniform 1 km coordinate grids (once per pass)

odsl-swot-build-1km-grid builds merged left+right 1 km grids from Expert nadir tracks. Science and cal/val orbits are separate. Cycle 005 is the preferred Expert source (all 584 science passes); empty nadir tracks fall back to other cycles.

python swot/build_uniform_grid.py --spacing 1000 --workers 16

Output (already built on Spray, ~24 GB, 612 passes):

/spray/swot/regrid_1km/science/pass_NNN_grid_1km.nc
/spray/swot/regrid_1km/calval/pass_NNN_grid_1km.nc

Each file has latitude, longitude, cross_track_distance, along_track_distance. Cross-track spans about −65 km to +65 km (131 pixels); along-track is ~19 731 lines. The nadir gap is kept as NaN.

2. Regrid one Unsmoothed pass to 1 km

odsl-swot-regrid-1km --cycle 3 --pass 1
# or
python swot/regrid_unsmoothed_to_1km.py --cycle 3 --pass 1 --workers 8

Default parameter sets (Gaussian half-decay: weight = 0.5 at that distance):

Label Search radius Half-decay
r1000_l500 1 km 500 m
r2000_l1000 2 km 1 km

Custom pair: --search-radius 1500 --half-decay-scale 800.

Writes /spray/swot/regrid_1km/cycle_CCC/pass_PPP_regridded_1km.nc.

Default (two parameter sets) stores ssha and uncert with a n_param dimension plus search_radius_m, half_decay_scale_m, and parameter_labels. A single custom pair writes 2-D ssha / uncert.

QC applied to Unsmoothed pixels (in order)

  1. ssha = ssha_karin_2 + height_cor_xover (additive)
  2. ssha_karin_2_qual == 0
  3. ancillary_surface_classification_flag == 0 (open ocean)
  4. |ssha| ≤ 3 m

OI then drops a target if more than 70% of raw pixels in the search radius are non-ocean, if fewer than min_obs valid observations remain, or if the estimate is outside the clip range. Rejected points are NaN in both ssha and uncert.

Lon in Unsmoothed files may be [0, 360) — normalize before geographic subsetting.

A ~19 min full pass is typical (param set 1 ~2.5 min, set 2 ~16 min).

python swot/plot_regridded_segment.py \
    --cycle 3 --pass 1 --start-lat 15 --end-lat 32

3. 30 km / 250 m FFT + OI tiles (noise catalog)

swot/unsmoothed_fft_oi.py turns each Unsmoothed left/right swath into 30 km tiles at 250 m after a 4 km FFT2 low-pass. Resume-safe: existing .h5 files are skipped.

python swot/unsmoothed_fft_oi.py --cycle 6 --pass 350
python swot/unsmoothed_fft_oi.py --workers 8          # mass job

Output:

/spray/swot/statistics/oi_250m_fft4km/cycle_XXX/pass_YYY_{left,right}.h5

HDF5 groups seg_NNNN hold ssha_native, ssha_fft, ssha_oi plus lat/lon. Sidecar CSV has seg_index, start_line, swh_m, noise_m, valid_frac.

Locked algorithm (do not “improve” without being asked):

  • Middle 150 of 240 cross-track pixels; no latitude gate (ice is Expert-flagged).
  • Pixel QC: SSHA / σ0 / xover qual == 0 and open-ocean class.
  • Expert line QC: rain_flag==0, dynamic_ice_flag==0, ice_conc<0.5%, rain_rate<0.5 mm/h, ocean class ≥ 95%.
  • 150×150 segments, stride 100. Keep if valid_frac≥0.995, SWH < 2 m, 8×MAD outlier fraction ≤ 0.002, σ_DD/2 ≤ 8 cm.
  • FFT2 Gaussian low-pass, k0 = 1/4 km⁻¹. Keep k=0 and l=0. Restore mean.
  • Local Gaussian remap, L = 1 km, R = 2 km, k-NN = 48.

Progress: tail -f /spray/swot/statistics/oi_250m_fft4km/logs/process.log

40×40 Unsmoothed tiles for the KaRIn noise catalog (not the 250 m OI product):

python swot/extract_unsmoothed_tiles.py --workers 8

Tiles land under /spray/swot/statistics/raw_tiles/. Per-tile statistics (double-difference noise, Expert SWH/wind join) are in /spray/swot/statistics/diagnostics/. Single-tile OI lives in swot/oi_regrid_tile.py.

Next: Tutorial 6 — PIXC or the /spray path reference.