Scottish seed potatoes play a key role in the UK’s £4–5 billion potato industry. Scotland’s climate and rigorous seed certification scheme are central to limiting the spread of virus in seed potato crops, but changing environmental conditions, shifting pest dynamics, and fewer aphicide options have made managing viruses like Potato Virus Y and Potato Leaf Roll Virus more difficult. In response, this project reviewed existing virus forecasting models used in Scotland and explored how machine learning could support more accurate, up-to-date predictions.

The study built on previous analysis of data from the Scottish Seed Potato Classification Scheme and sought to understand stakeholder needs, assess the strengths and limitations of current models, develop improved national and local forecasts, and identify key risk factors influencing virus spread. Stakeholders indicated a preference for simple, clearly defined alerts and locally relevant predictions that protect grower confidentiality.

Current linear models used by SASA were found to be generally sound, though limited in their ability to capture more complex relationships in the data. New national machine learning models achieved perfect classification of above- or below-average virus levels, while local models reached over 80% accuracy. Analysis of the models highlighted the importance of factors such as prior virus levels, crop health and quality, and proximity to infection sources.

A desktop tool was developed to allow SASA to retrain models annually using the latest data, ensuring forecasts remain relevant. Replacing existing models with the new ones is recommended, along with regular updates. Forecasts can inform better virus management and support current industry guidance. Further improvements could involve incorporating more detailed local data and developing personalised forecasts for individual crops.