PhD (M/F) Analysis of tree physiology in urban environments: modeling and machine learning using optical and radar imagery

29 Mar 2024
Job Information

Organisation/Company

CNRS

Department

Littoral, environnement, télédétection, géomatique

Research Field

Environmental science
Political sciences » Science and society

Researcher Profile

First Stage Researcher (R1)

Country

France

Application Deadline

18 Apr 2024 - 23:59 (UTC)

Type of Contract

Temporary

Job Status

Full-time

Hours Per Week

35

Offer Starting Date

2 Sep 2024

Is the job funded through the EU Research Framework Programme?

Not funded by an EU programme

Is the Job related to staff position within a Research Infrastructure?

No

Offer Description

The datasets made available to the PhD student for optical and radar simulations will be pre-processed (including radiometric and geometric image correction, urban models).

The candidate will take part in occasional experimental campaigns to acquire the data required for his thesis, as well as for the MONI-TREE project as a whole.

The PhD will take place in LETG-Rennes, with expected missions to ONERA's Palaiseau for works regarding EMPRISE model

In a context of global climate change, ongoing alterations are already causing extreme weather events such as storms, heat waves and prolonged droughts, posing major challenges for urban areas. According to the World Bank, 54% of the world's population lives in urban areas, and this figure is set to rise to 80% by 2050. Faced with these major environmental changes and their potential impacts, it is crucial to develop policies and tools aimed at strengthening the resilience of cities and adapting them to these profound transformations. It is now well established that vegetation, and trees in particular, deliver effective solutions for mitigating the impacts of climate change. Trees can help refresh cities, improve air quality and promote biodiversity. However, to play a positive role, trees must survive and maintain their health in an environment that is undergoing drastic transformations, including increased water and heat stress and resource constraints. The summers of 2022 and 2023 revealed the damaging consequences of long periods of water stress and high temperatures, resulting in the death of many trees. Nevertheless, some of them managed to survive despite these hostile environmental conditions. This surprising observation is of interest to municipal managers and it requires understanding the influence of the environment on tree condition, with a view to guiding future planting strategies in land management.

The aim of the MONI-TREE project, funded by the French National Research Agency (ANR), is to gain a better understanding of how trees react to extreme stress, detect senescence at an early stage, and create maps of tree well-being in urban environments, accompanied by warning systems. To achieve this, the project aims to collect a wide range of physiological data on trees (measured and simulated) under real (urban) and controlled (experimental greenhouse) conditions, as well as data on their environment (3D urban model, climatic conditions, etc.). Machine learning techniques will then be used to identify the various relationships between these data and produce global information maps on the physiological state of trees in the city.

Satellite remote sensing is an essential tool for monitoring vegetation on a global scale, providing essential data for natural resource management. One of the major challenges for researchers is to understand the physiological parameters sensitive to different sensor acquisition technologies.On the one hand, optical remote sensing, based on visible and near-infrared light, makes it possible to assess, among other things, spatial cover and plant species, leaf density, leaf pigments involved in photosynthetic activity (e.g. chlorophylls), water and dry matter contents that can be linked to water stress and living biomass. However, optical sensors can be hampered by cloud cover or low-light conditions affecting the signal acquired at the optical detector. In contrast, radar remote sensing, using microwaves, can penetrate clouds and operate day and night. It can supply information on the three-dimensional morphology of vegetation, including height and density, as well as humidity. However, the interpretation of this radar data has not yet been validated for the estimation of these vegetation parameters.Exploring the complementarity, or even fusion, of optical/radar data has been rarely studied for so many physiological parameters for tree vegetation, especially in urban environments, due to the complexity of 3D relief and the recurrence of mixed pixels (mixing the contribution of trees with their environment) for satellite remote sensing observations with generally low spatial resolutions (of the decametric order).
The main objective of this thesis is to establish links between the observations collected by these two remote sensing acquisition technologies (optical and radar) and tree-specific physiological parameters relating to their biomass, growth, structural and biochemical properties. To achieve this goal, we plan to design advanced modeling and deep learning algorithms that will integrate optical and radar images from the Copernicus program (Sentinel-1 and Sentinel-2), as well as in-situ measurements, in order to make accurate estimates of these physiological parameters.

The thesis work will be divided into three parts:

1) Prospective analysis of radar signal sensitivity for quantifying tree functional status

We will examine which plant parameters are most sensitive to radar signal variations. To do this, we will use ONERA's EMPRISE code (https://www.emprise-em.fr/ ) [Lebarbu et al., 2021] to create simulated image databases, including 3D urban mock-ups representing scenes with different tree physiological conditions (moisture/water stress, structure) and instrumental configurations (viewing angles and resolutions), including configurations of Sentinel-1. A sensitivity study will identify the most influential parameters, in order to select them for later estimation using hybrid inversion methods (training/testing of machine/deep learning methods on simulated databases and application on the image). We will adapt an inversion code already used in the optical domain to move towards a common optical/radar approach. An assessment of the estimation performance for each selected parameter will be carried out, together with the best estimation method.

2) Further characterization of trees using optical signals

Using the same 3D models of the scene as the previous radar studies, and common electromagnetic parameters for each element present in the urban scene, we will generate simulated optical image bases using the DART code [Zhen et al. 2023]. As with radar, a sensitivity study will be carried out, followed by an inversion estimation step. These will be largely based on work in progress in collaboration with LETG and ONERA, who have already obtained preliminary results with an urban modeling hypothesis based on the definition of Local Climate Zones (LCZ) [Stewart & Oke, 2012] for isolated and aligned trees. In particular, this work has shown that only LAI (leaf area index) and leaf chlorophyll content are accessible with Sentinel-2 data at 10m spatial resolution for certain spectral vegetation indices and geometric configurations [Le Saint et al., 2023]. In the following, the first inversion results show encouraging performance, and remain to be further investigated. We will assess whether these results can be reproduced in other contexts, including joint optical/radar modeling and inversion methodology.

3) Optical/radar data fusion

Optical and radar image databases will be merged, and estimation methods will be jointly researched for all physiological parameters accessible in optics and radar. Innovative deep learning and transfer methods will be investigated. The estimation performance of vegetation parameters obtained by radar alone (step 1), optics alone (step 2), and the combination of the two in this step will be evaluated. In particular, two variables accessible from both optical and radar perspectives, leaf moisture content and LAI, will be examined. An analysis will be carried out to determine how the contribution of optics or radar can be used to refine their estimation.

References :

Elise Colin et Laetitia Thirion-Lefevre (2023): Modéliser la rétrodiffusion radar par les forêts : une première étape pour inverser, dans Yajing Yan (dir.), Inversion et Assimilation de données de données de télédétection, Wiley,, 2023, p. 237-269. , DOI : 10.51926/ISTE.9142.ch7.
Di Martino, T., Guinvarc'h, R., Thirion-Lefevre, L., & Colin E. (2021, July). Convolutional Autoencoder for Unsupervised Representation Learning of PolSAR Time-Series. In 2021 IEEE International Geoscience and Remote Sensing Symposium IGARSS (pp. 3506-3509). IEEE.
Lebarbu, C., Ceotto, E., Cochin, C., Jouade, A., Everaere, E., & Trouve, N. (2021, March). Complete radar simulation chain: Application to maritime patrol surveillance using SAR/ISAR modes. In EUSAR 2021; 13th European Conference on Synthetic Aperture Radar (pp. 1-6). VDE.
Zhen, Z., Benromdhane, N., Kallel, A., Wang, Y., Regaieg, O., Boitard, P., ... & Gastellu-Etchegorry, J. P. (2023, May). DART: a 3D radiative transfer model for urban studies. In 2023 Joint Urban Remote Sensing Event (JURSE) (pp. 1-4). IEEE.
I. D. Stewart et T. R. Oke, « Local Climate Zones for Urban Temperature Studies », Bulletin of the American Meteorological Society, vol. 93, no 12, p. 1879 1900, déc. 2012, doi: 10.1175/BAMS-D-11-00019.1.
Le Saint T., Adeline K., Lefebvre S., Nabucet J. & Hubert-Moy L. (2023) Analyse de sensibilité du satellite Sentinel-2 pour la caractérisation des arbres en milieu urbain, TEMU (télédétection en milieu urbain), Montpellier, France (oral found on: https://www.theia-land.fr/urbain/2023-urbain/ ).

Requirements

Research Field

Environmental science

Education Level

Master Degree or equivalent

Research Field

Political sciences

Education Level

Master Degree or equivalent

Languages

FRENCH

Level

Basic

Research Field

Environmental science

Years of Research Experience

None

Research Field

Political sciences » Science and society

Years of Research Experience

None

Additional Information

Website for additional job details

https://emploi.cnrs.fr/Offres/Doctorant/UMR6554-THOCOR-001/Default.aspx

Work Location(s)

Number of offers available

1

Company/Institute

Littoral, environnement, télédétection, géomatique

Country

France

City

RENNES

Geofield

Where to apply

Website

https://emploi.cnrs.fr/Candidat/Offre/UMR6554-THOCOR-001/Candidater.aspx

Contact

City

RENNES

Website

https://letg.cnrs.fr/

STATUS: EXPIRED