A new research study aims to demonstrate how the combination of digital X-ray signatures with data-driven approaches can replace the time-consuming elements of expert led mineral identification and allow for new insights into the role of soil minerals in the environment.
Soils are a complex mixture of minerals, water, air, organic matter, and home to a huge variety of organisms. Minerals are the major component of many soils, made up from a combination of primary minerals derived from the soils geological “parent materials”, along with secondary minerals such as clay and iron oxides. Soils are responsible for many functions relating to biodiversity and are vital to life on earth, but their complexity has made it notoriously challenging to systematise soil property – mineralogy relationships.
Conventional approaches for the assessment of soil mineralogy are primarily based on X-ray diffraction (XRD) measurements. Modern instrumentation and precision sample preparation methods allow for the collection of remarkably reproducible XRD patterns. In recent years, this has resulted in databases containing thousands of reproducible, geo-referenced, XRD measurements and associated soil properties. Such datasets present unique opportunities to advance the understanding of how soil minerals govern or influence soil properties, processes, and functions.
Each XRD pattern within these databases can be treated as a detailed mineral signature of a soil, from which minerals can be identified and quantified. The steps of mineral identification and quantification are notoriously time consuming, and thus not suitable for datasets of this size. If instead the XRD patterns are treated simply as digital signatures, which encode information on both soil mineral (peaks) and amorphous (background) components, then computational, data-driven, approaches such as data mining and cluster analysis become possible.
The project, supported by the Macaulay Development Trust, is led by its fellow Dr Benjamin Butler, digital mineralogist in the Institutes Environmental and Biochemical Sciences group, he explained: “We applied machine learning to a dataset of Scottish soil XRD patterns to predict and interpret a range of soil properties. The X-ray diffraction data can be considered a mineralogical signature of each soil sample; therefore, this approach gave us the opportunity to investigate relationships between soil mineralogy and soil properties using data-driven technology for the first time.
Not only was the machine learning algorithm able to predict several soil properties accurately, but the way in which it selected mineral variables from the data allowed us to evaluate specific mineral contributions to the soil properties. Since minerals are the major component of most soils, this study ultimately showed the utility of machine learning to investigate soil mineralogy – soil property relationships in greater detail. Further application will help us better understand the role of soil minerals in Scotland and beyond.”
