SCET (Shoreline Change Estimation Toolkit)

A pure-Python toolkit for extracting shorelines from segmented raster imagery and computing shoreline change rates. Transect casting, shoreline/transect intersection, and change-rate calculation are delegated to the OpenDSAS CLI (dsas), a fast, cross-platform reimplementation of USGS DSAS.

Installation

pip install scet-toolkit

This installs scet-toolkit's dependencies including opendsas, which provides the dsas CLI used for transect/intersection/rate calculations.

The deep-learning segmentation step (scet.segmentation, used by SCET.process_folder/process_image) is included: the DeepLabV3+ network is plain PyTorch, so there is no mmcv/mmsegmentation or GDAL to build and any Python 3.9+ works. torch is CUDA-build-sensitive: if the default wheel isn't right for your platform, install torch from https://pytorch.org first (matching your CUDA version), then pip install scet-toolkit. Inference uses the GPU when available and falls back to CPU.

Usage

from scet import SCET, Config

config = Config(checkpoint_file="path/to/checkpoint.pth")
scet = SCET(config)
scet.process_folder(input_naip_zipfiles_folder="...", output_folder="...")

Or drive the shoreline pipeline directly, given a folder of already-segmented, dated raster images (filenames ending in a 4-digit year, e.g. 1_2001.tif):

from scet.options import Options
from scet.pipeline import generate_result_from_folder

generate_result_from_folder("images/", "output/", Options())

This writes shoreline.shp, baseline.shp, transect.shp, and intersection.shp to the output folder.

Development

conda create -n my_env python==3.12
conda activate my_env
bash build.sh  # pip install -e .

Testing

pip install -e ".[test]"
pytest                 # full suite, including end-to-end pipeline tests
pytest -m "not e2e"    # unit tests only (skip anything that shells out to `dsas`)

The end-to-end tests build synthetic georeferenced rasters, run the full images -> shoreline/baseline -> dsas cast/cal pipeline, and check the output shapefiles. They need the dsas executable (installed with the opendsas dependency) on PATH and self-skip if it is missing.