Agrimetrics to present at prestigious European Space Agency event

Crop species can be reliably predicted earlier in the season than previously possible

Agrimetrics is using machine learning and SAR radar imagery to identify which crop species are growing within a field boundary, within a season. The goal is to make an accurate identification much earlier in the year than is currently possible. These findings will be presented at the ESA’s EO Phiweek in September.

An abstract co-authored by Agrimetrics’s Data Science Team, detailing how SAR radar data can be used to identify crop species, has been selected for presentation at the European Space Agency’s (ESA) Earth Observation (EO) Phiweek.

In the abstract, they explain how a deep learning algorithm was used to map the field boundaries of over 1.45 million UK fields. Earth Observation Data was then mapped to these boundaries, producing a series of SAR images for each field.

Using data from 2016/17, the algorithm was trained to recognise crop species depicted in the SAR imagery. Early predictions are complicated by the visual similarity of young crops, however, the algorithm predicted even early crops with 65% accuracy. By the end of the growing season, this had risen to 80%. Due to the nature of machine learning and plans to incorporate non-Earth Observation data, the scope and accuracy of predictions will increase further in the coming months.

In the mid-term, the goal is to make highly accurate predictions much earlier in the season than is currently possible. This has the potential to significantly improve in-season decision making. Organisations can build a picture of the cropping landscape for that season, without having to rely on growers self-reporting – this will be a valuable resource for researchers, commodity, and supply chain planners.

In the future, when combined with other metrics (e.g. weather data and soil composition – both available in the Agrimetrics Field Explorer API – and yield) it will be possible to uncover new and deeper insights. For example, buyers will be able to forecast not just yield, but also the quality of crops, e.g. the amount of wheat meeting milling standard.

It can also help identify and rectify the causes of the ‘yield gap’ – the phenomena where actual yields are just half of potential yields. This has important ramifications for global food security and the productivity of the UK’s agri-food system.

EO Phiweek is an annual event hosted by the ESA, with the aim of connecting multi-disciplinary communities – from innovators and start-ups to data scientists and EO researchers – to review the latest developments in Earth Observation Science. It takes place on 9 – 13 September at the ESA-ESRIN, Rome.

‍