Quick Start¶

Two independent workflows are covered here. Pick the one that matches your data:

GCOV workflow — backscatter COGs from NISAR GCOV products (gamma0/sigma0 intensity)
GSLC workflow — power/amplitude/phase/coherence from NISAR GSLC complex SLC products

Steps 1–2 (install and credentials) are shared.

GCOV Workflow¶

Backscatter amplitude or intensity COGs from NISAR GCOV products (track 105, frames 17–18).

IMPORTANT: Ideally run on an AWS ec2 instance in us-west-2 where NISAR data reside (32GB RAM recommended for full scenes, less for subsets). Outside us-west-2 add --https to the search command. Output supports s3://my-bucket/prefix/. If full scenes are requested, caching is automatically turned on (--cache y) See full documentation.

1 -- Install¶

mamba create -n openseppo -c conda-forge openseppo aria2 matplotlib
conda activate openseppo

2 -- Credentials¶

Store your NASA Earthdata login in ~/.netrc once:

machine urs.earthdata.nasa.gov
    login <your_username>
    password <your_password>

Then pre-cache the bearer token:

seppo_earthaccess_credentials -t

3 -- Find scenes¶

Search for NISAR GCOV scenes for track 105, frames 17 and 18: (omit --https flag if on an ec2 instance in us-west-2)

seppo_nisar_search --track 105 --frame 17 18 --start_time_before 2026-01-17 \
    -o urls.txt --https

urls.txt now contains one URL per line.

4 -- Inspect available grids¶

Check which frequencies and polarizations are in the first file before converting:

seppo_nisar_gcov_convert -i urls.txt -lg

5 -- Convert to GeoTIFF¶

Convert to amplitude-scaled Cloud Optimized GeoTIFFs at 20 m resolution, clipped to a subset provided in projection coordinates crossing the two frames, with a time-series VRT stack:

seppo_nisar_gcov_convert -i urls.txt -o out/ \
    -amp -projwin 636357 3497674 655829 3480149 -tr 20 20 -v

Replace out/ with an S3 path (e.g. s3://my-bucket/NISAR/out/) if preferred.

Output:

out/
  NISAR_..._hh_AMP.tif       <- HH backscatter (uint16, amplitude, COG)
  NISAR_..._hv_AMP.tif       <- HV backscatter (uint16, amplitude, COG)
  NISAR_..._hhhv_AMP.vrt     <- per-scene snapshot VRT (both pols)
  NISAR_..._hh_AMP.vrt       <- time-series VRT stack (1 band per date)
  NISAR_..._hv_AMP.vrt        <- time-series VRT stack

Optional: ancillary layers¶

Extract mask, number of looks, and the gamma-to-sigma conversion factor. Ancillary grids bypass backscatter scaling and receive specialized downscaling (mask: priority 255>0>subswath, numberOfLooks: sum):

seppo_nisar_gcov_convert -i urls.txt -o out_anc/ \
    -projwin 636357 3497674 655829 3480149 -tr 20 20 \
    --vars mask numberOfLooks rtcGammaToSigmaFactor \
    -v

Output: _mask.tif (uint8, nodata=255), _nlooks.tif (float32), _gamma2sigma.tif (float32).

Common variants¶

# dB float32, no spatial subset
seppo_nisar_gcov_convert -i urls.txt -o out/ -dB -v

# Compact uint8 browse + dual-pol ratio, 100 m, WGS84
seppo_nisar_gcov_convert -i urls.txt -o out/ \
    -DN -dpratio --no_single_bands \
    -t_srs 4326 -tr 0.001 0.001

# Sigma0 conversion
seppo_nisar_gcov_convert -i urls.txt -o out/ -sigma0 -amp

# Rebuild VRTs from existing TIFs (auto-detects mode)
seppo_nisar_gcov_convert -o out/ -ro

# Show output summary with /vsis3/ paths for QGIS
seppo_nisar_gcov_convert -o s3://my-bucket/out/ -S -vsis3

See Examples for the full reference, or CLI Reference for all options.

Load the time-series VRT stack into Python¶

The per-track time-series VRT (one band per acquisition date) can be opened lazily with rioxarray -- no data is read until you actually index or compute:

import rioxarray

s3 = "s3://seppo1-data/NISAR/test3/NISAR_L2_PR_GCOV_005-010_105-105_A-A_017-018_4005_DHDH_A_20251117T111904_20260116T112015-EBD_A_hh_AMP.vrt"

# Open lazily -- bands = acquisition dates, nothing is read yet
da = rioxarray.open_rasterio(s3, chunks={})
print(da)  # shape, dims, CRS, dtype

# Spatial subset and compute
subset = da.isel(x=slice(100, 300), y=slice(100, 300))
subset.isel(band=0).compute().plot()

The band dimension corresponds to acquisition order. Each band's Description attribute holds the acquisition date; the companion .dates sidecar file lists them one per line in band order.

Using openSEPPO in Python / Jupyter¶

For integrating openSEPPO into a Python script or notebook -- including programmatic search and conversion -- see the Python API / Jupyter Integration.

GSLC Workflow¶

Power, amplitude, phase, and interferometric coherence from NISAR GSLC complex SLC products.

IMPORTANT: Same AWS/us-west-2 advice applies. GSLC pixels have fixed ~5 m azimuth (Y) spacing; X spacing varies by bandwidth (77 MHz: 2.5 m, 40 MHz: 5 m, 20 MHz: 10 m, 5 MHz: 40 m). Use --square to produce square pixels in one step.

1 -- Install and credentials¶

Same as the GCOV workflow -- see steps 1–2 above.

2 -- Find GSLC scenes¶

Search for NISAR GSLC scenes (same track/frame syntax):

# Search GSLC scenes -- track 135, frame 77
# (omit --https if on an ec2 instance in us-west-2)
seppo_nisar_search --product GSLC --track 135 --frame 77 \
    --start_time_before 2026-02-01 -o gslc_urls.txt --https

gslc_urls.txt now contains one GSLC HDF5 URL per line.

3 -- Inspect available grids¶

seppo_nisar_gslc_convert -i gslc_urls.txt -lg

4 -- Convert to power COGs¶

Convert all GSLC scenes to power intensity GeoTIFFs, square pixels, clipped to an area of interest:

seppo_nisar_gslc_convert -i gslc_urls.txt -o out_gslc/ \
    -pwr --square \
    -projwin 72.39 23.21 72.82 22.81 -projwin_srs EPSG:4326 \
    -v

Output:

out_gslc/
  NISAR_..._HH_pwr.tif        <- HH power intensity (float32, COG)
  NISAR_..._HH_pwr.vrt        <- time-series VRT (1 band per date)

5 -- Extract complex SLC for coherence¶

To compute coherence you need the raw complex pixels. Extract them with -cslc:

seppo_nisar_gslc_convert -i gslc_urls.txt -o out_cslc/ \
    -cslc \
    -projwin 72.39 23.21 72.82 22.81 -projwin_srs EPSG:4326 \
    -v

Output:

out_cslc/
  NISAR_..._HH_cslc.tif      <- HH complex SLC (complex64, tiled GTiff)
  NISAR_..._HH_cslc.vrt      <- time-series VRT

6 -- Compute interferometric coherence¶

Compute sequential pairwise coherence from the complex SLC TIFs. Use a 3×9 window for more isotropic ground coverage at 20 MHz pixel spacing (10 m × 5 m):

seppo_nisar_coherence -i out_cslc/*_HH_cslc.tif -o out_coh/ \
    -window 3 9 -v

Output:

out_coh/
  NISAR_..._HH_COH_w03x09_20251119_20260118.tif   <- uint8 COG, DN=round(coh*200)
  NISAR_..._HH_COH_w03x09.vrt                     <- time-series VRT

Recover coherence values: coh = DN / 200 (nodata=255).

7 -- Coherence with crop and downscale¶

Crop to the area of interest and downscale 2× after coherence estimation:

seppo_nisar_coherence -i out_cslc/*_HH_cslc.tif -o out_coh/ \
    -window 3 9 \
    -projwin 72.39 23.21 72.82 22.81 -projwin_srs EPSG:4326 \
    -d 2 \
    -v

8 -- All pairs, reprojected to geographic¶

seppo_nisar_coherence -i out_cslc/*_HH_cslc.tif -o out_coh_geo/ \
    -window 3 9 -pairs all \
    -t_srs EPSG:4326 -tr 0.0001 \
    -v

Common GSLC variants¶

# Amplitude output (GCOV-compatible uint16)
seppo_nisar_gslc_convert -i gslc_urls.txt -o out/ -amp --square

# Wrapped phase
seppo_nisar_gslc_convert -i gslc_urls.txt -o out/ -phase

# HH only, anisotropic downscale (2x cols, 4x rows)
seppo_nisar_gslc_convert -i gslc_urls.txt -o out/ -pwr -vars HH -d 2 4

# Float32 coherence (no DN encoding)
seppo_nisar_coherence -i *.tif -o out/ -no_DN

# Write coherence directly to S3
seppo_nisar_coherence -i *.tif -o s3://my-bucket/coh/ -window 3 9

See GSLC Convert CLI Reference and Coherence CLI Reference for all options.