Combining spectroscopy and big data analytics to monitor soil carbon dynamics in agricultural systems

Background

Managed soils are at the nexus of the global challenges of climate change, food security and water resources. Soil carbon sequestration – removing carbon from the atmosphere and storing it in soil organic matter – is considered a key global warming mitigation strategy (Paustian et al., 2016). Soil carbon is one of the main indicators of soil health and resilience, and developing better understanding of the links between land degradation and climate change is critical. Increased soil carbon levels also improve soil moisture retention and drainage, contributing to agricultural resilience to extreme weather events e.g. floods and droughts (Rawls et al., 2003). Soil carbon is a positive indicator of sustainable land management, as stated in the global indicator framework of the United Nations Sustainable Development Goals (SDG Indicator 15.3.1). In order to determine the rate and extent of carbon sequestration in soils, it is necessary to determine the carbon content in a rapid and, ideally, non-destructive manner. However, conventional methods such as Loss on Ignition and Chromate Oxidation require separate determinations for inorganic- and organic-C, are time consuming, expensive, destructive and can generate chemical wastes with associated disposal costs.

The use of infrared (IR) spectroscopy in soil science over the past two decades has revolutionised how such analyses are undertaken. IR spectroscopy is non-destructive, requires minimal sample preparation and does not involve any chemicals. Measurements only take a few seconds and multiple soil properties can be analysed in a single scan by relating the spectral information to the analyte in question, using advanced multivariate statistical procedures such as Principal Components Regression, Ridge Regression, Partial Least Squares Regression, and their extended models that create sparse solutions for more accurate predictions. The use of soil spectroscopy in estimating soil carbon content has increased exponentially in the last few years, supported by the development of large-scale soil spectral libraries (Sanderman et al., 2021).

Objectives

  1. To assess the impact of land use and agricultural management practices on the main biological, chemical and physical soil health indicators, with focus on soil carbon temporal dynamics and spatial variability, using soil spectroscopy.
  2. To identify specific relevant segments of the spectra that correlate with total soil carbon and selected soil carbon fractions in samples from arable and grassland fields under different management practices including contrasting soil tillage methods, crop rotations, agroforestry, and fertilizer and organic amendment applications.
  3. To develop novel statistical and machine learning methods in jointly enabling accurate prediction and performing variable or feature selection in challenging correlated high-dimensional spectra data.

Methodology

The student will make use of the University of Leeds 350-hectare highly instrumented experimental farm. Soil samples collected at the experimental farm will be analysed using direct chemical and biological assays, as well as soil spectroscopy in the Mid-Infrared range using Fourier Transform Infrared (FTIR) spectrometer. Reflectance spectroscopy will be linked to soil properties, using machine learning methods such as Convolutional Neural Networks, Partial Least Squares Regression, Cubist Regression Tree, Random Forest, and Support Vector Machine (Barra et al., 2021, Song et al., 2021).

Potential Impact

The combination of soil spectroscopy and machine learning techniques represents a step change in the soil analysis capabilities, providing a contribution to programmes such as DEFRA’s 25 Year Environment Plan, the Environmental Land Management (ELM) scheme, and the UK Net Zero targets.

Fit to NERC Science

This project is aligned to NERC’s ‘Terrestrial & freshwater environments’ research subject. Specifically the project aligns to the following NERC research areas: (1) Soil science – by assessing physical, chemical and biological properties and processes in soils using novel analytical methods; and (2) Biogeochemical cycles – by considering the fluxes and cycling of soil carbon within and between the biosphere and the physical environment.

Supervision

The student will work under the supervision of Dr Marcelo Galdos (Institute for Climate and Atmospheric Science, School of Earth and Environment) and Dr Arief Gusnanto (Department of Statistics, School of Mathematics) at the University of Leeds.

Requirements

The student should have a keen interest in soil processes and environmental issues with a strong background in one or more of agricultural sciences, earth sciences, soil science, environmental sciences or related discipline.  Strong analytical/statistical/fieldwork skills are desirable but not essential, as full training will be provided during the PhD.

Key references

Barra, I., Haefele, S.M., Sakrabani, R., Kebede, F. (2021). Soil spectroscopy with the use of chemometrics, machine learning and pre-processing techniques in soil diagnosis: Recent advances – A review. Trends in Analytical Chemistry 135, 116166.

Paustian, K., Lehmann, J., Ogle, S., Reay, D., Robertson, G.P., Smith, P. (2016). Climate-smart soils. Nature 532, 49-57.

Rawls, W.J., Pachepsky, Y.A., Ritchie, J.C., Sobecki, T.M., Bloodworth, H. (2003). Effect of soil organic carbon on soil water retention. Geoderma 116(1-2): 61-76.

Sanderman, J., Savage, K., Dangal, S.R.S., Duran, G., Rivard, C., Cavigelli, M.A., Gollany, H., Jin, V.L., Liebig, M.A., Omondi, E.C., Rui, Y., Stewart, C. (2021). Can agricultural management induced changes in soil organic carbon be detected using mid-infrared spectroscopy? Remote Sensing 13, 2265.

Song, Y., Shen, Z., Wu, P., Viscarra Rossel, R.A. (2021). “Wavelet geographically weighted regression for spectroscopic modelling of soil properties.” Scientific Reports 11(1): 17503.