Parcourir les annonces Parcourir les étudiants Publier une annonce Ajouter un CV

Dynamic forest carbon maps from high-resolution Sentinel 1 satellite data

Date de début
Date de fin

Dynamic forest carbon maps from high-resolution Sentinel 1 satellite data.
PhD proposal
University of Copenhagen, Kayrros (a Paris based EO startup), the Laboratoire des Sciences du Climat et
de l’Environnement (LSCE) and INRAE Bordeaux are looking for a PhD candidate on a joint project.
The advent of high-resolution radar data from Sentinel 1 (S1) at weekly time scale for all weather
condition and at 20 m resolution opens the opportunity to derive maps of above ground biomass. The
challenge is to convert the backscatter signal from S1 into biomass using coincident data on density and
height, which can be obtained independently from LIDAR satellite data (ICESAT and GEDI). The
proposed PhD subject will focus on producing maps of biomass and biomass change at annual scale by
combining S1 and LIDAR data with forest inventory plots from national forest inventory data. A semiempirical
model of above ground biomass will be developed based on currently used models for the
Global CCI ESA Biomass products, but calibrated using GEDI height measurements and national
inventory plot-scale data available for France and Germany. Methods will use an optimization of the
model parameters through machine learning algorithms. Direct training and validation of machine
learning tools upon height and forest inventory data will be also attempted.
Overall aim
Apply high spatial and temporal resolution Sentinel_1 satellite data in combination with LIDAR data sets
to produce annual forest and carbon stock maps for selected regions in Europe (proposed: France,
Denmark and Germany; if successful, other regions will be covered)
Specific aims and working steps
• Select a forest area in Europe with available plot data.
• Develop a semi empirical model to relate S1 backscatter parameters and LIDAR-deririved height to
above ground biomass on annual time scale for forested areas at high spatial resolution.
• Evaluate results against very high resolution datasets.
• Programming skills, preferably in Python.
• Basic understanding of satellite images, spatial analyses, statistics.
• Knowledge on machine learning, preferably deep learning.
Martin Brandt,
Philippe Ciais,


3 mars 2020