Student information
MSc thesis topic: Scaling SIF: from point observations to the entire field using UAV-based SIF and multispectral data
Sun-induced chlorophyll fluorescence (SIF) is an important measure of plant stress that can be obtained using remote sensing. However, it requires field spectroscopy using hyperspectral sensors that are expensive to obtain. Most affordable sensors are point-based, i.e. they can only measure a single footprint several centimetres across at a time, making it infeasible to obtain SIF measurements for the entire field. In contrast, a multispectral or even a simple RGB camera is affordable and can quickly assess the entire field, but it lacks spectral detail to retrieve SIF directly. In addition, physical effects that lead to SIF emission are known and can be modelled using radiative transfer models and machine learning, but it requires correct measurement or estimation of a number of model parameters.
The objective of this topic is to find out the most accurate means of scaling SIF across the whole field using machine learning on auxiliary data, which is typically multispectral raster imagery, radiative transfer model inversion, or a combination of these techniques. This can include spatial interpolation (regression kriging) and/or advanced regression algorithms, including deep learning (e.g. U-Net) and/or physical modelling using radiative transfer models such as SCOPE.
Relevance to research/projects at GRS or other groups
The research topic is part on an ongoing field of research into measuring, modelling and scaling SIF observations within the GRS group, led by prof. Lammert Kooistra.
Objectives and Research questions
The objective of this thesis topic is to scale SIF using plant traits estimated from multispectral data with physical modelling, machine learning and spatial interpolation. Research questions may include:
- How do different methods for scaling SIF from point observations to the entire field compare with one another?
- Which proxy features and/or input parameters are the most important for SIF estimation accuracy?
Requirements
- Required: Geoscripting, Machine Learning
- Optional: Spatial Modelling and Statistics, Deep Learning
Literature and information
- Sahoo RN, Gakhar S, Rejith RG, Verrelst J, Ranjan R, Kondraju T, Meena MC, Mukherjee J, Daas A, Kumar S, et al. Optimizing the Retrieval of Wheat Crop Traits from UAV-Borne Hyperspectral Image with Radiative Transfer Modelling Using Gaussian Process Regression. Remote Sensing. 2023; 15(23):5496.
- Tomislav Hengl, Gerard B.M. Heuvelink, David G. Rossiter, About regression-kriging: From equations to case studies, Computers & Geosciences, Volume 33, Issue 10, 2007, Pages 1301-1315, ISSN 0098-3004.
- Katja Berger, Jochem Verrelst, Jean-Baptiste Féret, Tobias Hank, Matthias Wocher, Wolfram Mauser, Gustau Camps-Valls, Retrieval of aboveground crop nitrogen content with a hybrid machine learning method, International Journal of Applied Earth Observation and Geoinformation, Volume 92, 2020, 102174, ISSN 1569-8432.
Expected reading list before starting the thesis research
- Gerjon Schoonderwoerd, Modelling Sun-Induced Fluorescence using Broadband Vegetation Indices, 2019.
- Wang et al (2022). Comparison of a UAV- and an airborne-based system to acquire far-red sun-induced chlorophyll fluorescence measurements over structurally different crops.
Theme(s): Sensing & measuring, Integrated Land Monitoring