Up to 40% of the world’s crops are currently lost to pests. There is an urgent need for evidence to help prioritise research into minimising these losses, writes Gaby Oliver, Project Assistant on the Global Burden of Crop Loss (GBCL) project.
Indeed, the requirement is acute when you consider the world’s population is expected to reach 10 billion by 2050.
But can artificial intelligence (AI) really help solve global crop loss which puts the food security of the planet at risk?
There is currently $25.8 billion spent annually on agricultural research, but is this being spent efficiently to deal with the actual problems in crop production?
Automating global crop loss data
The GBCL initiative aims to provide authoritative evidence of the impacts, causes, and risk factors of crop loss and use evidence-based estimates of crop loss to inform direct funding, policy, and research.
It is difficult to make informed decisions on crop loss mitigation without having current data readily available. However, data on the causes of losses are often outdated and sporadic. Moreover, lack of data is not the only problem.
Manual extraction of information available in a variety of scientific papers and reports is highly time consuming and error prone.
If this process could be done automatically, it would greatly improve the agricultural research and decision making, as a better understanding of the major causes of crop loss would help to streamline research into areas where it is most required.
To work towards this goal, the GBCL initiative partnered with Rothamsted Research with funding support from the Alan Turing Institute.
Two main aims of the crop loss data project
The overarching aims of the project were to create a pipeline to extract evidence of crop loss from recent scientific literature, and to provide semantic querying access and knowledge navigation capabilities over extracted GBCL estimates.
Developing an automatic abstract classifier
As the first step, an abstract classifier was developed, to identify relevant manuscripts automatically.
The subject area experts within the GBCL team conducted string searches on CAB Abstracts and Web of Science to create datasets of papers.
The datasets were merged and cleaned to reduce the unrelated literature, and eventually resulted in a dataset with a total of 1,382 abstracts, 230 of which were deemed relevant because they contained useful information about the pest impact.
The Rothamsted team used these outputs to create an automated abstract classifier. This classifier achieved a good balance between precision and recall when tested.
Extracting crop loss evidence
Once the relevant literature could be identified, the next step was to extract the relevant information from the full texts. Project partners at Rothamsted used supervised learning with manually labelled sentences supplied by the subject domain experts at CABI to develop an automated sentence classifier.
These labelled sentences were then also used by the project partners to create the knowledge graphs that could be used for semantic querying. These knowledge graphs were built using KnetMiner, which allows scientists to search large databases to find links between genes, traits, and other information types.
Potential for AI to provide crop pest information
This project lays the foundation for an automated pipeline for extraction of information on pest impacts from literature and is a very promising development for the application of AI in agricultural research.
This project is exciting when you consider the potential future applications of this work. AI could be used to create a standardised database of compiled evidence of crop loss, which could be accessed by everyone to further the research into agricultural developments.
To have this information easily retrievable from one place would allow for many advancements. The next steps for this development would include integrating papers which are published in other languages to remove any potential bias and increase inclusivity of data.
This work was presented at the AI UK Showcase, hosted by the Alan Turing Institute, by Chris Baker, the Project Principal Investigator, from the Rothamsted team.
Main image: There is an urgent need for evidence to help prioritise research into minimising crop losses which have an impact on global food security (Credit: Pixabay).
About the Global Burden of Crop Loss initiative
The Global Burden of Crop Loss initiative is modelled after the Global Burden of Disease initiative in human health, which has transformed health policy and research, over the last 25 years through better use of data.
It aims to have a similar impact in agriculture, providing evidence to enable the global plant health community to generate actionable information and lead to a dramatic reduction in crop loss, resulting in increased food security and trade.