Growers need reliable production forecasts of their crops, eg lettuces, in order to be able to meet the demands of retailers both in terms of volume (eg numbers of lettuces) and timing (demand for salads increases in the summer). The opportunity for using AI in those scenarios comes up in multiple ways, writes Matthew Smith, Chief Product Officer at fellow Agri-Tech Centre, Agrimetrics.
Overall, the problem boils down to how to modify how crops are being grown now in order to maximise the profitability when they are sold in future. The ideal scenario is to be able to maximise the supply of sellable produce while minimising waste. AI can address this in multiple ways:
- By providing forecasts of how much supply there will be in future
- By providing more accurate and cost-effective estimations of the current status of the crops and
- By making recommendations of crop management actions that could be taken in light of the forecasted supply and how that matches sales strategy and forecast.
Reliable information
Practically speaking, this results in a producer receiving information that tells them the current status of their crops, forecasts how much will be produced and suggests courses of action that could be optimise the impact of their decisions.
AI is used in all of these scenarios because it is the key computational method for enabling the various data streams to be brought together and processed in such a way that the different forms of output can be created. It allows users to take into account all the relevant data streams and provides the computational flexibility required to allow us to develop algorithms that can be validated as accurate and reliable.
Key challenges
A key challenge – and mistake – for such projects is failing to account of the long-term maintenance and use of the algorithms, post any initial proof of concept or value. AI algorithms need maintenance, they need to be improved, re-trained and updated. Growers rarely have the skills to do this inhouse, but neglegting this key aspect will mean the reliability, safety and overall trust in the algorithm will deteriorate over time.
By far the hardest part developing an AI algorithm for growers is ensuring that relevant specific factors are accounted for in training the AI algorithm.
Crop growth varies by myriad genotypic, environmental and management factors. It is simply not possible to account for all these factors, not least because there is inadequate data to allow identification of their effects. A careful balancing act is needed to identify the factors that can be accounted for (e.g. Crop varieties) and that are worth accounting for.
Transferable insights
These insights are transferable across agriculture. A major issue is the initial cost of setting up any crop forecasting system for a specific crop, yet a lot of what is required is common across different crops, which can be incorporated across applications.
This means it should be possible – and highly valuable – to develop crop forecasting capabilities to serve the needs of multiple growers with a common pool of services. Hence the transfer of both the insights and the methods, across agriculture, is a great route to improving the profitability and sustainability of primary production.
AI in agriculture
What I have described is just one of many opportunities to derive value from the application of AI in agriculture and as discussed at the Oxford Farming Conference and in my paper Getting value from artificial intelligence in agriculture in which I review the landscape of opportunity and impact for the next 10 years. As I state in that paper, the more immediate requirements will be to improve the quantity and quality of precision information about what is happening on the farm by improving what is being detected and measured.
Improved data and understanding lead to increasingly accurate predictions, enabling more optimal decisions about how to manage farm systems and stimulating the development of decision support and recommender systems. Artificial intelligence will also be needed to enable organisations to harness the value of information distributed throughout supply chains, including farm data.
Empowering farmers
AI clearly has a lot to offer agriculture, but we must also be aware that there are also likely to be negative impacts, such as disruption to the roles and skills needed from farm workers. We need to consider the social and ethical impacts of AI each time a new capability is introduced.
In order to benefit from AI maximally, the collection of data – and the weaving of it into decision support systems – has to become part of farming. For that to happen, we need to ensure farmers retain ownership of the data used for such purposes and continue to benefit from its application.
This is one of the longer-term goals of the Data Marketplace operated by my organisation Agrimetrics (one of the UK AgriTech centres) and we are currently working on how to enable farmers to be empowered by their data for use in such scenarios in a way that is entirely under their control.
