Machine learning for agricultural modelling

AgML

Tell Me More

About AgML

AgML is the AgMIP transdisciplinary community of agricultural and machine learning modellers

AgML aspires to

  • identify key research gaps and opportunities at the intersection of agricultural modelling and machine learning research,
  • support enhanced collaboration and engagement between experts in these disciplines, and
  • conduct and publish protocol-based studies to establish best practices for robust machine learning use in agricultural modelling.

Close collaboration with other AgMIP activities (i.e. the Global Gridded Crop Model Intercomparison, GGCMI) will facilitate the creation of agricultural model datasets for use in cutting-edge ML research.

Interested to join us? Join the AgML mailing list!

Our perspective

Transdisciplinary coordination is essential for advancing agricultural modelling with machine learning.

The last decades have shown the many opportunities for agricultural modeling using machine learning. From field-level to global scales, machine learning has allowed better yield estimates or estimates for under-studied crops, among many other uses. The decades have also shown, however, that many pitfalls exist and must be addressed: overfitting, lack of causal insight, interpretability issues. To avoid them, we need to establish benchmarks, evaluation criteria and best practices. And for that, a transdisciplinary community is required, integrating crop modelers and plant scientists and machine learning researchers.

We detail our perspective in a new article. Read the full article here.

With the support of

Activities

AgML team is organized around model intercomparison activities.
Currently we have launched the first three tasks, and we are open to more!

Future climate impacts
on yields

By measuring the skill of machine learning models in emulating existing process-based crop models under climate change scenarios, we can evaluate and intercompare the ability of data-driven approaches to generalise outside of the training distribution.
More on the Future Crop Challenge, our benchmark dataset to evaluate ML models that estimate agricultural impacts of climate change.
Team leader: Lily-Belle Sweet (Helmholtz Centre for Environmental Research - UFZ)

Regional yield forecasting

Sub-national yield forecasting is often approached differently in terms both of available predictors and evaluation strategies. In this task, we aim to harmonize and intercompare machine learning models for forecasting crop yields in different environments and for different crops.
More about CYbench, the AgML benchmark dataset for subnational crop yield forecasting to intercompare ML models.
Team leader: Michiel Kallenberg (Wageningen University and Research, Artificial Intelligence)

Good practices for yield modelling

When using machine learning for yield modeling, many studies fail to follow what are considered good practices. We mean to build upon our group’s knowledge as well as the literature dedicated to these issues and discuss the full pipeline of a machine learning model, from its conception to its use. We also aim to contextualise these practices in different use cases.
Team leader: Monique P. Oliveira (Embrapa Agricultura Digital)

New tasks

Interested to propose or organize a new task?
Join us!

AgML Workshops

Upcoming workshop: Leipzig, November 3-5, 2025

After the first AgML workshop, hosted in Wageningen, the Netherlands, we are proud to announce our second workshop! On November, we will meet again, this time in Leipzig, to discuss our activities, hybrid modelling and much more.

Tell Me More

Join Us

Interested in AgML activities?