Groundbreaking advancements
The recent presentation at the IEEE WCCI 2024 by the RIL of Soil Health introduced groundbreaking advancements in soil health monitoring using hyperspectral spaceborne technology.
This study focused on analyzing key soil indicators, such as soil organic carbon, clay content, and pH levels, in the region of Central Macedonia, Greece. In the scope of our ScaleAgData project, these are some preliminary insights which tested the use of hyperspectral spaceborne images in this project's pilot area.
Towards regular and detailed soil health monitoring
In light of the new EU soil monitoring law, which emphasizes sustainable soil management and the protection of soil resources, the RIL's work is particularly significant. This legislation mandates regular and detailed soil health assessments to ensure long-term soil fertility and environmental health. Hyperspectral spaceborne monitoring represents a major leap forward in achieving these goals, offering enhanced precision and coverage compared to traditional multispectral data.
Hyperspectral imaging provides a far more detailed spectrum of data, capturing a wide range of wavelengths that allow for more accurate identification and analysis of soil components. This advancement enables researchers to detect subtle variations in soil properties that multispectral data might miss. It allows for continuous monitoring over large areas, providing invaluable data to support sustainable land management practices. This technology will help meet the requirements set forth by the EU soil monitoring law, ensuring more informed and effective soil conservation strategies.
In this paper (not yet published by IEEE, but will be available soon here) , the novel approach utilized Differential Evolution (DE) with multi-objective optimization techniques to build fuzzy rule-based systems with enhanced predictive performance and interpretability. It was tested on both the 2015 LUCAS topsoil dataset on laboratory spectra and PRISMA hyperspectral images, demonstrated significant improvements over conventional methods. By optimizing the number of rules and selecting relevant input features, the methodology achieved a balance between accuracy and interpretability, proving to be as effective as the classical Random Forest algorithm.
Next steps
The RIL of Soil Health plans to collaborate with the Flanders Research Institute for Agriculture, Fisheries and Food (ILVO) in Belgium to test federated learning approaches.
In addition, data from Hyperfield-1 of Kuva Space will be tested once this becomes operational.
This collaboration aims to further enhance the robustness and scalability of soil health monitoring systems, leveraging diverse datasets and advanced machine learning techniques to support agriculture and environmental stewardship across Europe.
More information?
Nikolaos Tsakiridis
tsakirin [at] auth.gr