SpiPy#

PyPI version Documentation Status Build and Test Python 3.9-3.14 License: MIT

📦 View Source on GitHub | 📖 Documentation | 🐛 Report Issues

SpiPy is a Python implementation of SPIRES (Snow Property Inversion From Remote Sensing), originally implemented in MATLAB (SPIRES GitHub repository).

Overview#

SPIRES retrieves snow properties (grain size, dust concentration, fractional snow-covered area) from satellite multispectral imagery by inverting reflectance spectra using lookup tables generated from Mie-scattering theory.

Key features:

  • Hybrid Python/C++ implementation for performance (3000x speedup over pure Python)

  • Support for MODIS, Sentinel-2, and Landsat data

  • SWIG bindings for optimized interpolation and optimization routines

  • NLopt-based nonlinear optimization

Installation#

Quick Install (PyPI)#

pip install spires

Note: Pre-built binary wheels are available for Linux and macOS (Python 3.9-3.14). For other platforms or to build from source, see below.

Install from Source#

Important: Use conda-forge for all dependencies. The apt version of nlopt does not include required C++ headers.

# Install build tools and nlopt (required)
conda install -c conda-forge swig gxx gcc nlopt

# Install all dependencies (recommended)
conda install -c conda-forge numpy h5py scipy xarray netCDF4 gdal geopandas matplotlib tox sphinx dask jupyterlab pyproj

Git LFS#

This repository uses Git LFS for test data. Install Git LFS before cloning:

# macOS
brew install git-lfs

# Linux
sudo apt install git-lfs

# Initialize
git lfs install

Build and Install#

# Build SWIG extensions
python3 setup.py build_ext --inplace

# Install package
pip install .

# Or install with optional dependencies
pip install ".[dev,test,docs]"

Usage#

See the examples/ folder for Jupyter notebooks with detailed use cases.

Basic usage:

import spires

# Load lookup table
interpolator = spires.LutInterpolator(
    lut_file='tests/data/lut_sentinel2b_b2to12_3um_dust.mat'
)

# Process imagery to get fractional snow-covered area
fsca = spires.get_fsca(...)

Development#

Building Wheels#

Build a wheel for the active Python interpreter:

pip install build
python -m build --wheel

Build wheels for multiple Python versions using tox:

tox -e py39,py310,py311,py312

Note: When using pyenv, wheels for Python 3.9 may incorrectly build for x86 instead of arm64 on M1 Macs. Use a conda environment to build correctly.

Building C++ Extensions Manually#

The setuptools build process handles SWIG bindings automatically. To build manually:

cd spires
make

Or specify paths explicitly:

NUMPY_INCLUDE=$(python -c "import numpy; print(numpy.get_include())")
g++ -shared -o spires_module.so spires.cpp -I$NUMPY_INCLUDE

Testing#

Run doctests:

pytest --doctest-modules

Documentation#

Install documentation dependencies:

pip install ".[docs]"

Build documentation:

cd doc/
make html

Lookup Tables and Test Data#

Lookup Tables#

Simulated Mie-scattering snow reflectance lookup tables are available on Zenodo:

DOI

  • MODIS: LUT_MODIS.mat (537 MB)

  • Sentinel-2: lut_sentinel2b_b2to12_3um_dust.mat (70 MB)

Download using the helper script:

python scripts/download_test_data.py --luts

Or download directly:

curl -L -o LUT_MODIS.mat https://zenodo.org/records/18701286/files/LUT_MODIS.mat
curl -L -o lut_sentinel2b_b2to12_3um_dust.mat https://zenodo.org/records/18701286/files/lut_sentinel2b_b2to12_3um_dust.mat

Note: The Sentinel-2 LUT is also included in the repository via Git LFS. Landsat lookup tables are planned.

Test Data#

Full-resolution test imagery for validation is available on Zenodo:

DOI

  • Sentinel-2 reflectance: sentinel_r.nc (1.4 GB, 921×1347 pixels)

  • Background reflectance: sentinel_r0.nc (705 MB)

Small subsets suitable for CI/testing are included in the repository via Git LFS. See tests/data/README.md for details.

Performance#

The C++ optimizations provide significant speedups over pure Python:

Interpolation: 3000x faster (1.07 ms → 309 ns)

  • Pure Python RegularGridInterpolator: 1.07 ms

  • Vectorized Python: 143 μs

  • SWIG C++ (vectorized): 5.58 μs

  • SWIG C++ (index lookup): 309 ns

Spectrum Difference: 1000x faster (1.1 ms → 1 μs)

  • Pure Python: 1.1 ms

  • With optimized interpolator: 3.8 μs

  • C++ implementation: 1 μs

Full Optimization: 3000x faster (165 ms → 43 μs)

  • Scipy optimization: 165 ms

  • With optimized interpolator: 4.94 ms

  • With C++ spectrum difference: 3.5 ms

  • NLopt in C++: 43 μs

Known Issues#

  • SLSQP solver doesn’t work in the C++ implementation; using COBYLA instead

  • SWIG interpolator and scipy’s RegularGridInterpolator behave differently when coordinates aren’t linspace

  • COBYLA in scipy can’t set rhobeg per dimension individually, requiring problem scaling

Roadmap#

  • [ ] Optimize inversion for single location over multiple timesteps (keep R_0 constant)

  • [ ] Support xarray inputs for interpolator and spectra

  • [ ] Add Landsat lookup tables

  • [ ] Improve cloud masking workflows

License#

See LICENSE file for details.

Citation#

If you use this software, please cite the algorithm paper, software implementation, and any datasets you use:

Algorithm:

@article{bair2021spires,
  title={Snow Property Inversion From Remote Sensing (SPIReS): A Generalized Multispectral Unmixing Approach With Examples From MODIS and Landsat 8 OLI},
  author={Bair, E. H. and Stillinger, T. and Dozier, J.},
  journal={IEEE Transactions on Geoscience and Remote Sensing},
  volume={59},
  number={9},
  pages={7270--7284},
  year={2021},
  doi={10.1109/TGRS.2020.3040328}
}

Software:

@software{bair2026spipy,
  title={SpiPy: Python implementation of SPIRES snow property inversion},
  author={Bair, Edward H. and Griessbaum, Niklas},
  year={2026},
  url={https://github.com/NiklasPhabian/SpiPy},
  version={0.2.2},
  doi={10.5281/zenodo.XXXXXXX},
  note={DOI will be updated after Zenodo release. See CITATION.cff for full metadata}
}

Lookup Tables (if used):

@dataset{bair2026spires_luts,
  author       = {Bair, Edward and Dozier, Jeff},
  title        = {{SPIRES} Snow Reflectance Lookup Tables},
  year         = 2026,
  publisher    = {Zenodo},
  doi          = {10.5281/zenodo.18701286},
  url          = {https://doi.org/10.5281/zenodo.18701286}
}

Test Data (if used):

@dataset{griessbaum2026sentinel2_testdata,
  author       = {Griessbaum, Niklas},
  title        = {Sentinel-2 reflectance data for testing the {SpiPy} implementation of the {SPIRES} algorithm},
  year         = 2026,
  publisher    = {Zenodo},
  doi          = {10.5281/zenodo.18704072},
  url          = {https://doi.org/10.5281/zenodo.18704072}
}

Alternatively, see CITATION.cff or use GitHub’s “Cite this repository” feature.

Funding#

Development of this software was supported by:

Contract: W913E523C0002 Program: “Climate and natural hazards, snow-covered and mountain environment sensing research” Sponsor: Broad Agency Announcement Program, Cold Regions Research and Engineering Laboratory Monitored by: U.S. Army Engineer Research and Development Center, Hanover, NH 03755

Distribution Statement: Approved for public release; distribution is unlimited.