floodsr 0.2.0

Flood grid super-resolution CLI

floodsr

AI-powered super-resolution for flood hazard rasters that turns coarse water grids into sharp, decision-ready inundation maps. Fuse low-res flood grids with high-res DEMs to infer fine-scale flood patterns. Built on PyTorch and GDAL, floodsr is designed for geospatial practitioners who need to enhance flood hazard data for better risk assessment and planning.

• Documentation: https://floodsr.readthedocs.io/en/latest/
• Contribute: https://github.com/cefect/floodsr/blob/master/CONTRIBUTING.md

Implemented models (see floodsr/models.json):

• ResUNet_16x_DEM: 16x DEM-conditioned ResUNet
• CostGrow_Terrain: built-in terrain-penalty growth model from Bryant et al. (2024)

Installation

see documentation for details.

basic install

pipx install floodsr

extended install

for handling rasters too large for memory, floodsr requires GDAL backends. the CostGrow terrain-penalty path also requires PCRaster.

# advanced install for VRT workflows
conda create -n floodsr-gdal -c conda-forge python=3.12 gdal pcraster -y
conda activate floodsr-gdal
python -m pip install floodsr

Use

See the Getting Started guide for a quick intro to using the CLI.

List available model versions:

Fetch a model by version into the default cache:

floodsr models fetch ResUNet_16x_DEM 

Enhance a low-resolution flood hazard raster to high resolution (tohr), fetching the DEM from the HRDEM Mosaic data source.

floodsr tohr --in lowres032.tif --fetch-hrdem 

Enhance with a local DEM file:

floodsr tohr --in lowres032.tif --dem hires002_dem.tif  

Doctor diagnostics:

floodsr doctor

For more details, see the User Guide.