Skip to content

Installation

Requirements

  • Python 3.10+
  • CUDA-capable GPU recommended for training (CPU works but is much slower)
  • ~2 GB disk space for dependencies

Clone and Set Up

git clone https://github.com/birdnet-team/geomodel.git
cd geomodel
python3 -m venv .venv
source .venv/bin/activate

System Dependencies (Linux/Ubuntu)

Geospatial Python packages require native libraries for GDAL, PROJ, and GEOS:

sudo apt update
sudo apt install -y build-essential python3-dev gdal-bin libgdal-dev \
    libproj-dev proj-data proj-bin libgeos-dev libspatialindex-dev

python3 -m pip install --upgrade pip setuptools wheel

Python Dependencies

pip install -r requirements.txt

Troubleshooting

If pip install fails for a geospatial package, install it individually to see the full error: 

pip install pyproj shapely fiona rtree pyarrow
Manylinux wheels provide pre-built binaries for most packages on common Python versions.

Google Earth Engine

Earth Engine access is only needed for Stage 1 (environmental data sampling). If you already have a combined parquet file, you can skip this.

  1. Sign up at earthengine.google.com
  2. Authenticate: earthengine authenticate
  3. The provided initialize_ee() helper handles authentication in scripts

Optional Dependencies

These are only needed for specific features:

# Hyperparameter autotune (--autotune)
pip install optuna

# Model export (convert.py)
pip install onnx onnxruntime onnxscript onnxconverter-common

# TFLite / TF SavedModel export
pip install tensorflow onnx2tf

Verifying the Installation

After installing, you can verify everything works:

python -c "import torch; print(f'PyTorch {torch.__version__}, CUDA: {torch.cuda.is_available()}')"
python -c "import geopandas; print(f'GeoPandas {geopandas.__version__}')"
python -c "import h3; print(f'H3 {h3.__version__}')"