140-Forest loss monitoring using multi-frequency radar and optical data

  • Doctorat
  • Temps plein
  • Moins de 2 ans
  • Master, DESS, DEA, Bac+5
  • Data Sciences


The world’s forests have undergone substantial changes in the last decades. In the tropics, Vancutsem et al. (2021, Science Advances) estimated that 17% of moist forests have disappeared between 1990 and 2019 through deforestation and forest degradation. These changes contribute greatly to biodiversity loss through habitat destruction, soil erosion, terrestrial water cycle disturbances and anthropogenic CO2 emissions. Effective tools are thus urgently needed to survey forest disturbances.

Several forest disturbances detection systems from remote sensing have already been developed, mainly based on satellite optical images. Although radar images have great potential in tropical areas as electromagnetic waves are partly insensitive to clouds, and although radar Sentinel-1 images time series are now available to all at the global scale, few forest disturbances detection systems based on radar data have been developed. For notable exception, Marie Ballère (Ph.D. student funded by CNES until December 2021) improved the method priorly developed at CESBIO by Bouvet et al. (2018, Remote Sensing) and successfully applied it to French Guiana (Ballère et al., 2021, Remote Sensing of Environment). The system is now installed on the CNES high performing cluster and will allow to provide forest disturbances maps in the tropics (https://www.spaceclimateobservatory.org/fr/tropisco-amazonie).

The work that is proposed here aims at improving the Sentinel-1 based system developed by CESBIO/CNES, described above, this way:

1) The system shall be improved by using a diversity of images acquired from existing and future satellites missions, in addition to Sentinel-1 data. This task will require the evaluation of the valuable data to be selected and used in the frame of this work, in addition to taking in hand and processing a large range of various data.

First, optical data from the openly available Sentinel-2 mission and from Planet mosaics shall be considered. Indeed, optical data would allow to improve the spatial and temporal accuracies of the detection method, as showed in one of the few studies on the radar-optical complementarity for forest disturbances monitoring (Hirschmugl et al., 2020, Remote Sensing). A variety of optical based forest detection methods, well developed so far and available in the literature, shall be tested.

Moreover, L-band radar data shall be used as well to pave the way with the use of multi-frequency radar data for forest disturbances monitoring. This is important because of the future L-band radar satellite missions such as NiSAR and Rose-L, and also to prepare the launch of the P-band radar BIOMASS mission. For this purpose, L-band SAOCOM and ALOS-4 time series data over Brazil, made available at CESBIO by partners, can be used. Because of the scarcity of the developed methods based on L-band data, new methods shall be developed.
Finally, some research is still needed to improve the detection method with Sentinel-1 data. In particular, the potential of coherent radar data will be investigated to derive new detection indicators, such as interferometric coherences or polarimetric indices (entropy, anisotropy etc.).

2) To ingest the new optical and radar data described above, a fusion method shall be selected, either to ingest backscatter raw data as inputs, or to ingest forest disturbances detected as inputs, to produce the best detection system possible. To do so, numerous methods, such as Bayesian approaches, are available. However, the potential of deep learning-based approaches will be explored, since the use of various data as inputs is easy, and also because thousands of forest disturbances in situ data have been gathered over French Guiana in the frame of Marie Ballère’s Ph.D.

3) An important improvement of the existing detection method would be to detect very small objects, i.e., smaller than 0.2 hectare. This is important for forest degradation monitoring such as selective logging (which is a hot topic right now), and also for tree mortality monitoring, which is also of interest for the ecology community. This topic will be supported by partnerships and collaborations within projects starting now, such as the ALT ANR project (leaded by Jérôme Chave, EDB). This topic represents a technical challenge as no speckle filter should be used to preserve the spatial resolution of the Sentinel-1 data.

For more Information, contact Directeur de thèse : Laurent.Ferro-Famil@univ-rennes1.fr

about the topics and the co-financial partner (found by the lab !). Then, prepare a resumé, a recent transcript and a reference letter from your M2 supervisor/ engineering school director and you will be ready to apply online !


The student shall be autonomous, rigorous, friendly and open-minded. He is not afraid of failures. He will travel at least one time to French Guiana. He knows Python programing and has at least a small experience in the use of remote sensing data. He has a Master's degree primarily in the domain of the science of the universe, spatial technology, telecommunications, environment.

Message from PhD Team

CNES will inform about the status of your application in mid-June. More details on CNES website : https://cnes.fr/en/web/CNES-en/10685-st-doctoral-grants.php