AutoFloods

Overview

AutoFloods maps flooded areas from Sentinel-1 SAR (radar) satellite imagery, for a region and time period you specify – producing per-date and per-month flood-extent rasters, without requiring a background in radar remote sensing to use.

Under the hood, it detects flooded pixels with per-pixel Z-score anomaly detection: a dry-season VV/VH backscatter baseline (mean + standard deviation) is fit per tile, and wet-season scenes are flagged flooded wherever backscatter drops significantly below that baseline, with a terrain-slope mask to suppress false positives on steep ground.

To scale to large areas reliably, the workflow processes one AOI (“tile”) at a time rather than one country- or state-scale mosaic – see flood_mapper in the API reference for the per-tile pipeline this drives.

Data sources

AutoFloods reads Sentinel-1 RTC (radiometrically terrain corrected) imagery through a pluggable STACSource interface, with two implementations:

  • MPCSource – Microsoft Planetary Computer, the original backend.
  • OPERASource – NASA’s OPERA RTC-S1 product via ASF/CMR, hosted on AWS. Native 30m resolution; reads via a download-then-open pattern rather than streaming, which has proven substantially more reliable under sustained/concurrent load in this project’s own testing.

Swap between them (or add a new source) via the source= argument to flood_mapper – see Getting Started.

Installation

pip install --pre autofloods

For contributors working from a local clone:

pip install -e .

See pyproject.toml for the pinned dependency list. MPCSource needs no credentials for typical use (a subscription key is optional, raising the request rate limit); OPERASource requires a NASA Earthdata Login entry in ~/.netrc. See Getting Started for setup steps.

System dependency for OPERASource: mosaicking uses the GDAL command-line tools (gdalbuildvrt), which pip install does not provide – rasterio’s pip wheel bundles its own GDAL library internally but doesn’t expose the CLI binaries. Install them separately:

# Debian/Ubuntu, or Google Colab
apt-get install -y gdal-bin

# conda/mamba (any platform)
conda install -c conda-forge gdal

See gdal.org/download.html for other platforms. MPCSource does not need this. If gdalbuildvrt is missing, OPERASource raises a clear error pointing back here rather than a raw FileNotFoundError.

Citation

If you use this package (or part of it), please cite:

Tripathy, P., & Malladi, T. (2022). Global Flood Mapper: a novel Google Earth Engine application for rapid flood mapping using Sentinel-1 SAR. Natural Hazards, 114(2), 1341-1363. https://doi.org/10.1007/s11069-022-05428-2

Tripathy, P., Malladi, T., Balakrishnan, K., & Parath, N. (n.d.). AutoFloods: A Python package for automated flood mapping at scale using Sentinel-1 SAR [Manuscript in preparation].

Affiliation & funding

Department of Geography, University of California, Santa Barbara.

Funded by NASA FINESST (Award #80NSSC25K0392).