Title: Plant Functional Trait Retrieval from Spectra Using Deep Learning Advances

Duration: 15.06.2021 – Present

Research Area: Remote Sensing in Geo- and Ecosystem Research

Human activities are rapidly impacting global biodiversity, leading to changes in ecosystem functions and services. Monitoring plant functional diversity is therefore crucial to better understand and track these rapid changes. In this context, this PhD project seeks to contribute to large-scale plant trait retrieval, offering a cost-effective alternative and taking a step forward in creating efficient predictive models of multiple vegetation properties by combining advanced Deep Learning (DL) techniques with the unprecedented hyperspectral remote sensing (RS) data.

Aims

The project aims to develop efficient methods for mapping multiple plant traits using hyperspectral RS. Employing Convolutional Neural Networks (CNNs) and advances in DL, the research seeks to overcome data limitations and to simultaneously predict multiple traits across different vegetation types and sensors. The goal is to demonstrate the potential of DL in creating transferable predictive models for different ecosystem types.

Problem

The project addresses the challenge of translating hyperspectral reflectance into multiple vegetation properties (plant traits), overcoming data scarcity and heterogeneity.

Practical example

• Multi-trait models (Proof of concept): https://gitlab.com/eya95/multi-traitretrieval
• Inference on hyperspectral imagery (e.g EnMAP satellite imagery)

Technology

• TensorFlow, python
• Hyperspectral imagery

Publications

• Cherif, Eya, et al. “From spectra to plant functional traits: Transferable multi-trait models from heterogeneous and sparse data.” Remote Sensing of Environment 292 (2023): 113580.

Team

Lead

• Eya Cherif

Team Members

• Teja Kattenborn
• Hannes Feilhauer

Partner

• Remote sensing center for Earth system Research (RSC4Earth)