The application of machine learning to understand the
dynamics of soil structure
Publication Date
December 12, 2025
Creator
Abstract
Machine learning and artificial intelligence are quickly changing the world. Just a few years ago, these concepts usually provoked visions of theoretical fiction. However, now they are becoming commonplace across a vast array of industries and even in our daily lives.
Soil structure is a very important factor in soil health and has strong applications in agricultural studies and improvement. It is also a dynamic, complex, and very hard-to-measure component. Scientific advancements which improve the accessibility of soil structure quantifications are of great importance and benefit to the soil science community.
X-ray Computed Tomography (X-ray CT) is a non-destructive and non-invasive technique for observing and analysing soil structure. It has been used in soil science for several years for various tasks, leading to a large volume of data to be analysed. A lot of this data is often discounted, due to the extreme volumes produced.
Machine learning allows for high-throughput data processing, unrestricted by human processing and labour limitations. Data processing and analysis via machine learning algorithms might also notice new trends or patterns in the data that would not be noticed with human analysis. This means that machine learning algorithms could be the perfect solution to take a deeper dive into the data collected from X-ray CT, get more information out of the data, and potentially find new patterns.
Soil structure can vary greatly due to the soil composition, weather conditions (temperature, rain, or lack thereof), crops, compaction, tillage management and the biodiversity in the soil. In recent years there has been a lot of interest in zero tillage, as part of a series of measures usually deployed under the banner of conservation or regenerative agriculture and how it may be beneficial over conventional tillage. This research looks at how the tillage practices impact the soil structure and in particular the pore network, and how these properties may vary over a growing season. Different machine learning techniques are looked at and applied to demonstrate how machine learning can be used alongside X-ray CT images and soil data to improve processes and observe new information.
Ultimately, the large volume of data produced by X-ray CT make excellent training datasets for machine learning algorithms. The neural networks demonstrated were mostly successful at identifying and predicting soil type, water content and tillage treatment based on soil structure. Shortcomings could be overcome by increasing the volume and variation of training data, as well as making use of more advanced machine learning techniques such as multi-resolution networks.
Item Type
ethesis
Thesis Type
PhD
Supervisors
Subjects (LC)
Associated Schools / Departments
School of Biosciences (UK)
eprints ID
82656
UoN Repository URI
Except where otherwise noted, this item's license is described as
File(s)![Thumbnail Image]()
Name
CarolineRoy Corrected Thesis FINAL.pdf
Type
Full-text
Description
Examined
Size
9.15 MB
Format
Adobe PDF
Checksum (MD5)
3ef9057af016ee083d8e307444f08eee