Abstract

Tackling hidden hunger through soils

This report describes the work completed during a week long data study group hosted by the Alan Turing Institute. The challenge was provided by Rothamsted Research and looks at predicting soil and plant physicochemical properties from soil infrared (IR) spectra. Three datasets were explored and modelled using a combination of established and more recent data-science strategies. Due to the size, scope and variety in the datasets, multiple conclusions were drawn. Overall, our preliminary findings indicate that soil physiochemical properties were easier to model than plant physicochemical properties. Decision tree based methods were used consistently throughout the three datasets and were overall more robust than other approaches considered in our analysis. Our results are in line with the current literature; IR data can be an effective predictor of the physicochemical properties of soil and by extension, the health of the soil.

Citation information

Data Study Group team. (2020, April 29). Data Study Group Network Final Report: Rothamsted Research. Zenodo. http://doi.org/10.5281/zenodo.3775489

Additional information

Timo Breure, Cranfield University and Rothamsted Research
Chris U. Carmona, University of Oxford
Samuel Ellick, University of Bristol
Ben Evans, University of Bristol
Ali Fahmi, Queen Mary University
Stephan Haefele, Rothamsted Research
Kirsty Hassall, Rothamsted Research
Markus Loning, UCL
Diego Perez Ruiz, University of Manchester
Darya Shchepanovska, University of Bristol
Cathy Thomas, Rothamsted Research

Turing affiliated authors

Download list