PeaSat: Satellite imaging tool for predicting yields

Project summary

The UK is a major producer of vining peas, which has a small window for processing the harvest. Timing this right is critical for a successful crop but can be challenging to coordinate. ADAS, in partnership with HMC and funded by the European Space Agency, has developed a satellite imaging-based yield and harvest prediction tool to support growers and processors.

The background

PeaSat will provide a web-based yield and harvest date prediction service that enables processors to better understand potential inputs, manage their processing lines and therefore help to reduce bypass waste from vining pea crops. It uses machine learning to provide a yield prediction to growers based on crop variety and sowing date, supported by the use of satellite imagery and weather data.

The challenges

Each year, around 35,000 hectares of vining peas are planted across the East of England and Scotland. Despite significant improvements in crop planning and agronomy, there is no way to provide accurate early forecasts of yield, so growers and processors often struggle to coordinate throughout the harvesting period.

There is a small window of typically 1–2 days when the vining pea crop is fit for harvest. Following harvest, the crop must be processed and frozen within 4 hours, or 2.5 hours for the highest quality standards. If the vining pea crop cannot be harvested and processed within this time, it is ‘bypassed’ and wasted.

The solutions

ADAS worked with HMC, a pea growing cooperative, on the INNO-VEG project, which was set up to develop innovative methods for conducting research into field vegetables. It was through this project that the potential for remotely sensed data to estimate the yield of vining peas was identified.

Following this, ADAS and HMC applied for funding from the European Space Agency for a feasibility study to further investigate the potential for remote sensing. To further develop and demonstrate the project, ADAS is now working with Hutchinsons – owners of Omnia Digital Farming, a comprehensive farm management software that will provide farmers and processors with access to the PeaSat service.

The feasibility study covered 5 years’ worth of vining pea yield data provided by HMC. Cloud-free (Sentinel-2) satellite imagery was obtained over the growing season for each field, and numerous vegetation indices were calculated. Weather data were used to estimate crop development (based on temperature accumulation) through the season. Relationships among timing, variety, year, vegetation index and yield were explored to identify the best data to use for yield estimation.

A range of machine learning models was used to train and test the best available data combinations. When tested on real-world data from 2024, the model achieved an R² score of 0.71. This means it was able to explain 71% of the variation in crop yields based on the input data it was given. This is a strong result, showing the model can reliably predict how much yield to expect under different conditions. The demonstration project will further refine the model through the use of additional years and parameters to train the model and test it further. A weather-based growth model, used to predict yield for the minority of fields for which cloud-free satellite imagery is not available within the necessary time window, will also be developed.

The impact

Though still in its test phases, it is anticipated that PeaSat will reduce the amount of wasted (‘bypassed’) crop from the current 5-year national average of 8% down to 6% of the vining pea area. Across the UK, this saving is equivalent to 680 hectares growing 3060 tonnes of peas worth €1.8 million per year.

Currently, growers enter and update estimated yields and harvest dates in shared spreadsheets and are in frequent phone contact with processors. These requirements and processes will be reduced, as the service will make predictions available simultaneously to growers and processors.