Enhancement and Development of Oat Quantity and Quality Models Using Machine Learning for Supply Chain Stability

DESCRIPTION: Predicting oat quality (e.g., grain protein, beta glucan etc.) as well as yields before harvest can support operational efficiency within Quaker supply chain across the North America (US, Canada) and UK. Compared to other crops, there are few crop models, either process-based or statistical models, available for oat yield and none for oat quality. PepsiCo has developed an in-house proprietary model to predict oat yield and quality, specifically:

1. United Kingdom (UK) spring oat yield model (mechanistic derived model based on 5 years of UK data).
2. UK spring oat beta glucan and protein models (statistical models based on 5 years of UK data).
3. North America (NA) spring oat yield model (machine learning (ML) model based on StatsCan data).
4. NA spring oat beta glucan and protein models (machine learning models based on 4 years of NA data).

This project seeks to improve and extend these models including; 1) developing UK winter oat models for yield plus beta glucan using existing datasets, 2) Updating and enhancing the accuracy of the NA spring beta glucan and protein models using newly available data 3) develop NA variety specific quality and yield models and 4) evaluate accuracy gain against climatology for in-season prediction using operational seasonal forecasts (SCF) and sub-seasonal forecasts.

Specific Work/Deliverables to be performed:
The purpose of the research project is to improve PepsiCo's oat yield and quality predictions using Machine Learning techniques. The objectives of the projects are as follows:
i) Development of UK winter oat yield model.
ii) Refinement of PepsiCo spring oat quality models using ML techniques based on field data from Canada.
Development of variety specific spring oat quality model. Development of UK winter oat quality models.
iii) Investigate accuracy of seasonal climate forecasts and the performance of oat yield and quality models
for in-season forecasting using climatology, operational seasonal forecasts and sub-seasonal forecasts.