Skip to content

Use of Sentinel2 time serie to detect spruce dieback

Notifications You must be signed in to change notification settings

JoLeBelge/s2-spruce-dieback

Repository files navigation

Spruce dieback detection using Sentinel-2 imagery

Motivation

The ongoing thesis of Arthur Gilles related to the future of forest tree species Picea Abies in Wallonia (Belgium) have required to analyse the trend and extend to the bark beetle crisis that occured between 2018 and 2022. We have choosen to works with the Sentinel-2 spatial imagery with a methodology for detection of spruce dieback similar to the one developped by Dutrieux et al. (see the FORDEAD python package).

This repository serves as supplementary material for the scientific paper of Gilles A. (2023)

Arthur, G., Jonathan, L., Juliette, C. et al. Spatial and remote sensing monitoring shows the end of the bark beetle outbreak on Belgian and north-eastern France Norway spruce (Picea abies) stands. Environ Monit Assess 196, 226 (2024). https://doi.org/10.1007/s10661-024-12372-0

Documentation

There is two main documents that serve as documentation for our work:

  1. A french document that describe how is arranged the source code and how it works : /documentation/methodoAnalyseSentinel2TimeSerie2021.pdf This document was primarilly written for an internal use.
  2. The scientific paper of Gilles et al. 2024

Plus, the spruce dieback maps generated for Wallonia are available in the geoportal Forestimator. Users can either view or download such annual health maps.

Content of the repository

This repository is a compilation of three projects, all related to the use of Sentinel 2 time series, plus a certain number of documentation files either in french of in english. Each C++ projects represent one c++ application. Project managers is Qt (see .pro file)

s2 time serie (s2_ts.pro)

Detection of picea abies under stress caused by Ips Typographus using copernicus Sentinel 2 time serie analysis.

This application performs the downloading of Sentinel 2 - L2A product, the computation of complex SWIR Ratio and the comparison of this spectral indice with a theorical CRswir for this date in order to discriminate pixel that are suffering from water stress.

s2_postProcess.pro

This application performed post-processing of result from the detection of bark beetle damaged from S2 images. Post-processing is devoted to the change of coordinates systems of the results, to the comparison of annual health maps to better understand the evolution of the bark beetle outbreak, and to some spatial filtering that changes isolated pixels classified like "normal cutting" into "sanitary cutting" if they are surrounded by an area of sanitary cutting.
This project require Micmac c++ library.

s2_carteEss.pro

The goal is to compute one cloud-free mosaic of S2 tiles for every year quarter and then to use this time series of images ( 4 pseudo-dates x 12 bands) in order to train and apply a Random Forest classifier for species determination (or any other classification models). This sofware is twofold. Its first section is devoted to the training of a RF model (optimal sampling of observation balanced for each forest species classes and preparation of data to train RF with the ranger implementation of RF). The second part serves to apply the resulting model, so each pixels requires to run a RF.

Authorship and credit

The C++ code is developped by Jonathan Lisein (Liège University - Gembloux Agro-Bio Tech) which as followed the methodology of FORDEAD and re-used snippet of code from Nicolas Latte (Liège University - Gembloux Agro-Bio Tech). Arthur Gilles is involved in the documentation writting.

Our research were funded by the "Plan Quinquénal" of Wallonia, Belgium, and by the Interreg project "RegioWood II" contact: liseinjon@hotmail.com