In a groundbreaking gathering in late January in Wageningen, experts and researchers from around the world celebrated the launch of the first-ever AgML workshop aimed at advancing the use of machine learning for agricultural modelling. This is the beginning of an important international collaboration.
Agricultural models play a crucial role in understanding the effects of climate change on the global food system and improving resilience. Machine learning (ML) methods emerge in advancing scientific research, with their potential to learn complex, nonlinear relationships from high-dimensional data, and thus significantly improve the practice of agricultural modelling.
Understanding benefits and challenges
Over the past year, the Agricultural Model Intercomparison and Improvement Project (AgMIP) has made coordinated efforts to better understand the benefits and challenges of ML methods in agricultural modelling. This is how AgML has emerged as a research initiative of AgMIP: A transdisciplinary community of researchers aiming to better understand the benefits - and pitfalls - of ML methods in agricultural modelling tasks.
The goals of AgML include sharing knowledge, promoting best practices for using ML tools in agricultural modelling, quantifying ML performance in crop modelling, and developing new ML methods adapted to the unique challenges of the agricultural sector.
Discussions and hackaton
The workshop, led by Ioannis Athanasiadis, professor of Artificial Intelligence and Data Science at WUR, and Lily Belle Sweet, PhD candidate at UFZ, brought together representatives from FAO, EU JRC, CGIAR, NASA and several universities to discuss new ML paradigms and methodologies, as well as the need for rigorous evaluation of models.
In addition to discussions, participants participated in a hackathon to build their own ML models, which will be analysed in AgML's model comparison experiments. These experiments aim to provide reliable estimates of ML architectures and methodologies capability to be used for agricultural modelling applications.
Multidisciplinary expertise
AgML also focuses on building open datasets for assessing ML models in predicting climate impacts on agriculture and subnational yield forecasts for different crops and regions worldwide.
By bringing together expertise from different disciplines, AgML is committed to developing AI benchmarks for reproducible, comparable and interpretable modelling of agricultural and food systems.
This is the beginning of an important international collaboration facilitated by the AgMIP network, and its open to all interested parties including students, academics and the private sector.
