Python GeoInformatics Lab Environment-Plus A headless Docker environment combining SAGA GIS, GRASS GIS, WhiteboxTools, and OTB with a complete Python geospatial stack for spatial analysis, remote sensing, and machine learning — with GPU acceleration support.

What is included

GIS platforms

• SAGA GIS 9.3.2
• GRASS GIS 8.4.2
• WhiteboxTools 2.4.0
• Orfeo ToolBox 9.1.1

Python stack

• Core geospatial libraries: GDAL 3.12.2, PROJ 9.7.1, GEOS 3.14.1, Shapely, Fiona, Rasterio, GeoPandas
• Scientific computing: NumPy, SciPy, Pandas, Xarray
• Terrain and hydrology: RichDEM, pysheds, Landlab, TopoToolbox, xarray-spatial
• Machine learning: PyTorch, TensorFlow, scikit-learn, scikit-image
• GPU computing: CuPy, CUDA-enabled PyTorch and TensorFlow
• Visualization: Matplotlib, Plotly, Folium, Holoviews, Seaborn
• Image processing: OpenCV, scikit-image
• Web mapping: Leafmap, Geemap, ipyleaflet, localtileserver
• Cloud and remote sensing tools: Earth Engine API, STAC tools, odc-stac, Planetary Computer
• Point clouds: laspy, PDAL
• Documentation tools: Sphinx, Jupyter Book
• Jupyter Lab

Core characteristics

• Headless operation
• Preconfigured environment
• Version-pinned and reproducible: every package is fixed to an exact version, and a build manifest is written into the image
• CLI and Python access to supported tools
• Docker and Apptainer workflows
• GPU-capable machine learning stack, subject to host driver compatibility

Getting started

Pull from Docker Hub

docker pull dockagile/pygile-plus

Build from source

git clone https://github.com/Geoinformatics-Lab/PyGILE-Plus.git
cd PyGILE-Plus/docker
docker build -t pygile-plus .

The Dockerfile is self-contained; no additional files are required.

Run

CPU

docker run -it --name pygile-plus \
  -p 8888:8888 \
  -v $(pwd)/data:/workspace/data \
  dockagile/pygile-plus

GPU

docker run -it --name pygile-plus \
  --gpus all \
  -p 8888:8888 \
  -v $(pwd)/data:/workspace/data \
  dockagile/pygile-plus

GPU use requires: - NVIDIA GPU - compatible NVIDIA driver - NVIDIA Container Toolkit on the host

Jupyter Lab starts automatically at http://localhost:8888.

Only /workspace/data is mounted in the examples above. Files written elsewhere in the container are lost when it is removed. Mount the whole /workspace directory to persist notebooks and outputs as well.

For more detailed Docker usage, see docker/README.md.

Usage inside container

conda activate pygile

Basic checks

python -c "import grass.script as g; print('GRASS OK')"
grass --version
saga_cmd --version
whitebox_tools --version
otbcli_BandMath -help

GPU verification

python -c "import torch; print(torch.cuda.is_available())"
python -c "import tensorflow as tf; print(tf.config.list_physical_devices('GPU'))"
python -c "import cupy; print(cupy.cuda.runtime.getDeviceCount())"

If GPU is not detected, verify host driver compatibility.

Reproducibility

Every package is pinned to an exact version, and source tarballs are checksum-verified during the build. A manifest recording engine versions, compiler, checksums, library linkage, and the full resolved package list is written into the image:

docker run --rm dockagile/pygile-plus \
  cat /opt/manifest/build_manifest.txt

Also in /opt/manifest/: conda_explicit.txt (exact conda package URLs and build hashes) and pip_freeze.txt.

The environment can be verified at any time:

docker run --rm dockagile/pygile-plus pygile-verify

Deterministic numerics are enabled by default (OMP_NUM_THREADS=1 and equivalents) so floating-point results are reproducible across runs. Override at run time if throughput matters more than bit-identical output.

For citation, record the image digest:

docker inspect --format='{{index .RepoDigests 0}}' \
  dockagile/pygile-plus

Known limitations

• OTB Python bindings (otbApplication) target Python 3.8 and are unavailable in this Python 3.11 environment. The OTB CLI is fully functional.
• saga_cmd returns exit code 134 on teardown despite writing correct output. When scripting SAGA, check for output files rather than return codes.
• SAGA links the system GDAL 3.0.4 while the Python stack uses GDAL 3.12.2. Raster read and write are bit-identical between the two, so values are unaffected.
• PyTorch is present as both a conda build and a CUDA wheel; the wheel takes precedence in site-packages.

Tool locations

conda_env_path = "/opt/conda/envs/pygile"
saga_cmd = "/opt/saga/bin/saga_cmd"
grass_command = "grass"
otb_bin = "/opt/otb/bin"
whitebox_tools = "/opt/conda/envs/pygile/bin/whitebox_tools"
build_manifest = "/opt/manifest/build_manifest.txt"

Applications

• Remote sensing analysis and classification
• Raster and vector processing
• Terrain and hydrological analysis across multiple engines
• Spatial statistics and geostatistics
• Machine learning on geospatial data
• Educational and research workflows
• Web mapping and interactive visualization
• Cloud-based geospatial analysis
• HPC workflows through Apptainer

System requirements

• Docker
• Recommended memory: 8 GB minimum, more for large workloads
• Recommended storage: about 25 GB or more for the image and data
• Network access for initial pull or build
• For GPU use: NVIDIA GPU and compatible host setup
• For HPC: Apptainer or Singularity with --nv support when GPU passthrough is needed

HPC and Apptainer

apptainer pull docker://dockagile/pygile-plus
apptainer run --nv pygile-plus_latest.sif

License information

All included software remains under its respective upstream license. Refer to the original project documentation for specific license terms.

Reference and citation

Awasthi, B., Ninsawat, S., Raghavan, V., & Nemoto, T. (2026). Geoinformatics-Lab/PyGILE-Plus: PyGILE-Plus v1.1.0 (v1.1.0). Zenodo. https://doi.org/10.5281/zenodo.18512474

Awasthi, B., Ninsawat, S., Raghavan, V., & Nemoto, T. (2025). PyGILE-Plus: Python GeoInformatics Lab Environment-Plus (1.0.0). Zenodo. https://doi.org/10.5281/zenodo.16146572