Remote sensing and machine learning for the field-scale prediction of maturity and yield in vining pea (Pisum sativum L.)

Processors and growers of vining peas face a significant annual challenge in generating timely and accurate predictions of crop harvest dates and yields. Vining peas are harvested whilst still immature, and so are highly sensitive to environmental conditions. High temperatures in particular expedite maturity quickly past the optimal range of tenderness, often resulting in a short harvest window of just one or two days.

The aim of this PhD project is to improve UK vining pea production through the development of prediction models capable of estimating vining pea harvest dates and yields further in advance than is currently achievable. This will improve planning capability and efficiency and reduce the requirement for tenderometer sampling in the approach to harvest. Research will occur as an extension of the KTP project currently underway between PGRO and the University of Nottingham, with the aim of developing a service for use by growers and processors.

This PhD project has a novel focus on the combined use of remote sensing and machine learning for the development of robust, flexible, empirical prediction models. There are many physiological cues which indicate the growth stage and yield potential of a crop, many of which can be identified remotely through acquiring field-scale canopy reflectance data like NDVI and NDRE. This project will utilise the European Space Agency’s Sentinel-2 satellite imagery in combination with associated meteorological data, forming a set of remotely-sensed ‘ground truth’ data that is crucial for any crop research. Machine learning models are key for recognising the complex relationships between crops, the environment, and management strategies which influence crop growth and development. They allow complex, non-linear, and often hidden relationships to be represented based on extensive inputs of historic industry data. Utilising regular inputs of canopy reflectance and climate data, the models will be capable of producing up-to-date predictions throughout the vining season, improving future vining pea harvests.

Sponsors:

Student:

Leah Howells (University of Nottingham, Sutton Bonington, Loughborough, LE12 5RD)

Start Date: April 2021

Duration of study: 3 years