Background: Soil properties such as porosity, bulk density and soil water content are subject to often rapid change due to complex interacting drivers, including physical, chemical and biological processes. A comprehensive characterization and prediction of properties in a soil profile could bring substantial benefits to agricultural security and land management decision-making. Soil profiles contain a wealth of information about the character and properties of the soil. Interpretation of a soil profile requires the recognition and identification of many of these different characteristics, each of which contribute to an overall assessment of the soil’s ability to provide ecosystem services and its response to specific drivers.
Rationale: Traditionally, the expertise required to recognise soil characteristics and integrate them into a useful description of the profile requires expertise and time. The use of Machine Learning (ML) to interpret soil profile characteristics from profile imagery has been successful in estimating specific properties such as carbon content, pH and texture. The proposed work will develop a tool through machine learning techniques and model development approaches (e.g. least absolute shrinkage and selection operator (LASSO)), to identify and predict properties and functions of soil within a profile. The project will focus on (1) the recognition of specific diagnostic soil properties, (2) integration of diagnostic properties into an overall soil profile assessment, and (3) estimation of soil characterisation and ecosystem service provision.
- Degree at undergraduate or Masters level in a relevant environmental subject
- Experience in using statistical software packages (e.g. R, Matlab, SPSS)
- Good English skills (spoken and written)
- Willingness to carry out field work
- Training in soil profile characterization
- Experience in image analysis
- Experience in programming or app development
- Experience in machine learning / artificial intelligence implementation
Supervision: The student would be based largely at the James Hutton Institute in Aberdeen, with additional time spent at University of Aberdeen and Queens University Belfast. Matt Aitkenhead (Hutton) would be the primary supervisor, responsible for day-to-day activities, soil science and ecosystem service aspects and overall project management. Quan Gan (UoA) would be responsible for supervision on image analysis and machine learning approaches. Jennifer McKinley (QUB) would be responsible for supervision on geostatistics and mathematical modelling. Clare Bond (UoA) would be responsible for supervision on app development and stakeholder engagement.
Partner organisations: The student would become a member of the British Society of Soil Science (BSSS) and through the Society would participate in training activities (e.g. soil survey, farm management) and conferences (annual BSSS conference, World Congress of Soil Science in Glasgow, August 2022). Application would be made to the BSSS Grants and Awards Committee for funding to attend additional conferences and for additional field work equipment.
The James Hutton Institute has two research farms at Glensaugh (upland grazing, Aberdeenshire) and Balruddery (arable crop rotation, near Dundee). Both of these would be used as field sites for soil sampling and imaging, providing data on a wide range of soil types and land management conditions. Similar work would be carried out at farms associated with QUB in Northern Ireland. The student would also take part in soil survey work at Hutton Institute field sites at Dava Bog (Flow Country, northern Scotland) and in the Cairngorm National Park, and at Stirling University soil survey training pits near Stirling.
At each of the field locations, data would be captured on soil image and soil profile characteristics. The locations chosen cover a range of soil properties and management options that encompass the breadth of soil conditions in the UK. Additional information on land management and soil-based ecosystem services would be captured through farmer interviews and the use of existing map resources at the Hutton Institute.
Funding and eligibility information available here.
|Profile: Quan Gan|
Institution: University of Aberdeen
Department/School: School of Geosciences
|Profile: Jennifer McKinley|
Institution: Queen's University, Belfast
Department/School: School of Natural and Built Environment
|Profile: Clare Bond|
Institution: University of Aberdeen
Department/School: School of Geosciences
Matt Aitkenhead, James Hutton Institute
 Aitkenhead, M.J., Donnelly, D., Sutherland, L., Miller, D.G., Coull, M.C., Black, H.I.J., 2015. Predicting Scottish topsoil organic matter content from colour and environmental factors. European Journal of Soil Science 66, 112-120.
 Matt Aitkenhead, David Donnelly, Malcolm Coull, Richard Gwatkin, 2016. Estimating soil fertility indicators with a mobile phone. Chapter 7 of ‘Digital Soil Morphometrics’; Book series “Progress in Soil Science” by Springer (http://www.amazon.co.uk/Digital-Soil-Morphometrics-Progress-Science-ebook/dp/B01DXI295Y).
 Aitkenhead, M.J., Coull, M.C., Gwatkin, R., Donnelly, D., 2016. Automated soil physical parameter assessment using smartphone and digital camera imagery. Journal of Imaging 2(4), 35; DOI 10.3390/jimaging2040035.
Soil survey will involve identification and characterization of soil profiles, using diagnostic horizon criteria from the FAO-WRB World Soil Classification. In the field, this will include horizon delineation and colour estimation, sample acquisition and preparation, and hand texturing. Soil profile characterization using smartphone technology will be carried out in the later stages of the project, including the development of a protocol for preparation and imaging of a soil profile. Participation in the International Soil Judging Contest will include traditional and smartphone-based soil profile classification.
Soil pH and carbon measurement (Walkley-Black) will be carried out in the laboratory, along with bulk density measurement and analysis for cation exchange and other soil chemical components. The student will be assisted in this by the James Hutton Institute’s soil analytical laboratory team.
Geostatistical approaches to link soil properties to environmental characteristics will be carried out using GIS software packages (ArcGIS/QGIS). Statistical analysis and machine learning will be accomplished using R. App development will be carried out using libraries in R, including shiny. Literature review of existing methods for digital soil analysis will be carried out in the early stages of the project to identify knowledge gaps in the literature and identify key statistical and geostatistical analytical approaches.
Expected Training Provision
Through the British Society of Soil Science, the student would receive training in soil survey field work and in conference presentation. At the James Hutton Institute, they would attend training courses in GIS (ArcGIS, QGIS), statistical programming (R) and paper writing. At University of Aberdeen, training would be received in machine learning approaches in R, in image analysis methods and in app development.
The research carried out by the student will produce new understanding of the interactions between environmental and soil characteristics. It will also provide new ways of measuring soil properties directly in the field, enabling researchers to carry out rapid soil characterization at low cost and with greatly reduced time/effort. Additionally, information on land management practices at the field sites will provide insights into the impacts of management on specific soil-based properties, processes, functions and ecosystem services.
Impact on land management
The outcomes of the project will include an app for automated soil characterization and assessment, providing land managers with a tool for rapid assessment of soil health and identification of management challenges such as compaction, nitrogen runoff, erosion risk and pH limitations to cropping. The app will be freely available and applicable to a wide range of agricultural types including grassland, arable cropping and heathland.
Impact on environmental policy
Linkages between soil properties and ecosystem services will inform future policy development on soil management and agri-environment schemes. The project will provide better understanding of how specific soil management practices can influence ecosystem services under different environmental conditions. This will provide a mechanism for the development of targeted land management advice that incorporates site-specific conditions.
The student would be based largely at the James Hutton Institute in Aberdeen, with additional time spent at University of Aberdeen and Queens University Belfast. At the Hutton Institute, the student will be provided with desk space, computing equipment and other necessary resources for the project. The Institute has a strong cohort of postgraduate students, with approximately 100 PhD students in the Postgraduate School and several members of staff responsible for administrative activities, pastoral care and training. The Institute organizes an annual postgraduate conference at which students share ideas and discuss their projects.
Matt Aitkenhead (Hutton) would be the primary supervisor, responsible for day-to-day activities, soil science and ecosystem service aspects and overall project management. He will provide the student with regular contact meetings and support during field work, accompanying them during field training and coordinating administrative and computing requirements. As Geoinformatics Group leader, Dr Aitkenhead has extensive experience in geostatistics and environmental app development and has published extensively in relevant topics.
Quan Gan (UoA) would be responsible for supervision on image analysis and machine learning approaches. This supervision would be carried during meetings and activities at the University of Aberdeen, where the student would also have access to the University’s extensive computing, library and administrative resources. Dr Gan would supervise development of image analysis and machine learning approaches through the use of R, building on existing methods to develop a suite of approaches appropriate for soil profile imaging and characterization analysis. Meetings at UoA would be monthly, with additional participation in regular meetings with the full supervisory team.
Jennifer McKinley (QUB) would be responsible for supervision on geostatistics and geomathematical modelling. The student would spend time at QUB in Belfast with access to necessary facilities and would engage in field trips to University-run research sites.
Clare Bond (UoA) would be responsible for supervision on app development and stakeholder engagement.
Year 1: Development of soil profile library and associated soil profile feature and characteristics dataset. Development of framework for image analysis and characterisation. Development and training in machine learning systems.
Year 2: Development and testing of machine learning system for soil profile characterisation and recognition. Translation of soil property characterisation into ecosystem service description.
Year 3: Apply the Machine Learning system to identity soil characteristics from profile imagery and interpretation of soil characteristics.
Non-CASE partner: James Hutton Institute