sweets

End-to-end InSAR workflow that turns a single AOI + date range into unwrapped interferograms, a displacement timeseries and a velocity raster.

What sweets gives you

• Cloud-optimized data downloads
  • Sentinel-1 burst-subsetting via burst2safe
  • OPERA CSLC-S1 and NISAR GSLC AOI-based subsetting
• Geocoding
• DEM + water-mask prep
• Geometry stitching
• Phase-linking, unwrapping, timeseries network inversion, and velocity estimation via dolphin

Interchangeable input sources, accessible through the same Workflow object:

--source	What it is	Tools used
safe (default)	Sentinel-1 Level-1 bursts that intersect your AOI via burst2safe. Geocoded SLCs (GSLCs) created by COMPASS.	burst2safe, s1-reader, COMPASS
opera-cslc	Pre-made OPERA L2 CSLC-S1 HDF5s from ASF DAAC, plus their matching CSLC-STATIC layers for geometry. Faster than safe when OPERA has produced for your AOI (CONUS, see OPERA product suite for coverage) if you are fine with their processing parameters (e.g., 5 meter x 10 meter UTM posting).	opera-utils
nisar-gslc	NISAR L2 GSLC products geocoded in UTM.	opera-utils

An optional --do-tropo flag adds a post-dolphin OPERA L4 TROPO-ZENITH correction step for safe and opera-cslc (not yet wired for NISAR).

Install

We recommend pixi for managing local environments.

git clone https://github.com/isce-framework/sweets.git && cd sweets
pixi install
pixi shell

That drops you into an env where sweets, dolphin and compass are all on the PATH and sweets is installed in editable mode.

GPU mode (Linux + CUDA 12+ only). There's a gpu pixi environment that swaps the CPU isce3 build for isce3-cuda, which accelerates COMPASS geocoding, cross-multiplication and resampling on the GPU. dolphin itself runs phase-linking on the GPU through JAX+CUDA regardless of environment.

pixi shell -e gpu        # activate once
sweets run config.yaml   # now uses isce3-cuda

macOS can't install this environment (conda-forge doesn't ship osx-arm64 builds of isce3-cuda); pixi shell without -e gpu is the right default for every non-CUDA machine.

A conda-forge feedstock exists for sweets. It currently tracks the pre-v0.2 API; we'll bump it to the new release once the fork dependencies (s1-reader, COMPASS, opera-utils, dolphin, spurt) are merged upstream and a v0.2 release is cut.

Usage

Command line

sweets config --help
sweets run --help

To configure a workflow the minimum inputs are an AOI (--bbox or --wkt), a date range (--start / --end), and — for the Sentinel-1 path — the relative orbit (--track). --out-dir is where raw downloads land; --work-dir holds the workflow outputs.

Raw Sentinel-1 bursts (default):

sweets config \
  --bbox=-102.96 31.22 -101.91 31.56 \
  --start 2021-06-05 --end 2021-06-22 \
  --track 78 \
  --swaths IW2 \
  --out-dir ./data \
  --work-dir ./pecos_demo \
  --output pecos_demo/sweets_config.yaml

sweets run pecos_demo/sweets_config.yaml

Pre-made OPERA CSLCs (faster for CONUS, no COMPASS needed):

sweets config \
  --bbox=-102.96 31.22 -101.91 31.56 \
  --start 2021-06-05 --end 2021-06-22 \
  --source opera-cslc \
  --out-dir ./data \
  --work-dir ./pecos_opera \
  --output pecos_opera/sweets_config.yaml

sweets run pecos_opera/sweets_config.yaml

With the optional tropospheric correction step:

sweets config --source opera-cslc --do-tropo ... --output pecos_opera/sweets_config.yaml
sweets run pecos_opera/sweets_config.yaml

NISAR GSLCs:

sweets config \
  --bbox=-121.10 36.55 -120.95 36.70 \
  --start 2025-10-01 --end 2025-12-15 \
  --source nisar-gslc \
  --track 42 --frame 70 \
  --out-dir ./data \
  --work-dir ./salinas_nisar \
  --output salinas_nisar/nisar.yaml

sweets run salinas_nisar/nisar.yaml

Starting at a later step

sweets run accepts --starting-step N so you can skip earlier stages if the outputs are already on disk:

sweets run config.yaml --starting-step 2  # skip download, run geocode + dolphin
sweets run config.yaml --starting-step 3  # just (re-)run dolphin

Configuration from Python

You can also build a Workflow directly:

from sweets.core import Workflow
from sweets.download import BurstSearch

w = Workflow(
    bbox=(-102.96, 31.22, -101.91, 31.56),
    search=BurstSearch(
        track=78,
        start="2021-06-05",
        end="2021-06-22",
        swaths=["IW2"],
        out_dir="data",
    ),
    work_dir="pecos_demo",
)
w.to_yaml("pecos_demo/sweets_config.yaml")
w.run()

The same pattern works for OperaCslcSearch and NisarGslcSearch — pick whichever variant matches your data source. Workflow.search is a discriminated union keyed on a kind field, so a YAML file with search: {kind: opera-cslc, ...} round-trips correctly.

License

This software is licensed under your choice of BSD-3-Clause or Apache-2.0 licenses. See the accompanying LICENSE file for further details.

SPDX-License-Identifier: BSD-3-Clause OR Apache-2.0