RiverDetectWood

This Python script allows to detect river dead wood using IGN BD Ortho data. Firstly, it calculates indexes to be added to orthophotographs. Secondly it applies the random forest model on the orthophotographs. Finally, it output a shapefile layer with dead wood polygons and their metrics. This tool only works on Windows or Linux.

This tool is under CC BY-NC 4.0.

Clone the repository

Firstly you need to clone the repository :

git clone https://github.com/g-grimmer/RiverDetectWood.git
cd RiverDetectWood

Installation

Prerequisites

• Python 3.9
• GDAL
• Rasterio
• Scipy
• Joblib
• Geopandas
• Tqdm
• Orfeo Toolbox (OTB) version 9.x

Install Python Packages

We recommand to use Anaconda.

conda create -n myenv python=3.9
conda activate myenv
conda install -c conda-forge numpy gdal rasterio shapely geopandas scipy scikit-learn joblib tqdm

Install Orfeo Toolbox (OTB)

Install OTB from website

Firstly you have to install OTB from https://www.orfeo-toolbox.org/download/

Execute otbenv.bat

Here you must execute the file otbenv.bat (or otbenv.profile on linux):

On Windows

C:\path\to\otb\bin\otbenv.bat

On Linux

On linux, you can follow the steps describe on OTB official website. Basically, you just need to source the environment profile before launching our script on the same terminal.

source /Path/To/OTB_install/otbenv.profile

Potential installation error

If you encounter this error:

  File "pyproj\\_crs.pyx", line 2378, in pyproj._crs._CRS.__init__
pyproj.exceptions.CRSError: Invalid projection: EPSG:2154: (Internal Proj Error: proj_create: no database context specified)

Reinstall pyproj using pip:

pip install --force-reinstall pyproj

Usage

Download model and test files

The RiverDetectWood tool has been tested on orthophotographs from three river basins in France: the Buech, Loire, and Doubs. These test datasets provide diverse conditions to validate the model's performance on dead wood detection in varying river environments.

RiverDetectTool application

python rdw_tool.py input_tiff_path model_path keep_all_outputs

In this function :

• input_tiff_path is the orthophotograph path
• model_pathis the random forest model file path
• keep_all_outputs, 1 to keep all outputs, 0 to delete them

Citation

If you use this repository in your research, please cite the corresponding paper:

@article{grimmer2025automatic,
  title={Automatic detection of in-stream river wood from random forest machine learning and exogenous indices using very high-resolution aerial imagery},
  author={Grimmer, Gauthier and Wenger, Romain and Forestier, Germain and Chardon, Valentin},
  journal={Environmental Modelling \& Software},
  volume={190},
  pages={106460},
  year={2025},
  publisher={Elsevier}
}

and/or

@article{grimmer2025riverdetectwood,
  title={RiverDetectWood: A tool for automatic classification and quantification of river wood in river systems using aerial imagery},
  author={Grimmer, Gauthier and Wenger, Romain and Chardon, Valentin},
  journal={SoftwareX},
  volume={29},
  pages={102042},
  year={2025},
  publisher={Elsevier}
}