Détection des dégâts causés par les scolytes du pin

fordead | a python package for vegetation anomalies detection from SENTINEL-2 images

fordead | a python package for vegetation anomalies detection from SENTINEL-2 images



The fordead package, developed by the Changes and Health of Temperate Forests SEC for the detection of vegetation anomalies from SENTINEL-2 time series, provides monitoring tools to address the bark beetle health crisis on spruce trees in France. It includes several tools that make use of SENTINEL-2 satellite data easier, and allow potential anomaly detection in other contexts.

The proposed method takes advantage of complete SENTINEL-2 time series, from the launch of the first satellite in 2015. It detects anomalies at the pixel level in order to analyse archive data or to carry out continuous monitoring. The detections are then updated for each new SENTINEL-2 acquisition.

Dieback detection

The detection of dieback is done in five or six steps.

  1. The calculation of vegetation indices and masks for each SENTINEL-2 date
  2. The training by modeling the vegetation index pixel by pixel from the first dates
  3. Detecting anomalies by comparing the vegetation index predicted by the model with the actual vegetation index
  4. Creating a forest mask, which defines the areas of interest
  5. The export of results as shapefiles allowing to visualize the results with the desired time step

N.B. It is possible to correct the vegetation index using a correction factor calculated from the median of the vegetation index of the large-scale stands of interest, in which case the mask creation step must be performed before the model training step.

All the documentation and user guides for these steps are available in the documentation website.

Here is an example of the result of these steps on a small area :

Period of detection

Confidence class

Visualisation tools

The package also contains built-in visualisation tools. The first one plots the time series of the vegetation index for a particular pixel, along with the associated model, the anomaly detection threshold and the associated detection.

Healthy pixel

Attacked pixel


Jean-Baptiste Féret