Causal Machine Learning for Sustainable Agriculture

Vasileios Sitokonstantinou, Emiliano Diaz, Jordi Cerda-Bautista, Maria Piles, Ioannis N. Athanasiadis, Ilias Tsoumas, Gustau Camps-Valls

Keywords: machine learning, causal inference, sustainable agriculture, food security, evidence-based decision-making, robust predictions

TL;DR: We propose that causal machine learning can guide sustainable agriculture by uncovering mechanisms, estimating effects and improving predictive models for research, policy and practice.

Abstract: Agricultural systems involve biological, physical, social and economic dimensions. Current farming practices often fail to balance these and contribute to soil degradation, biodiversity loss and greenhouse gas emissions. Understanding and managing these complex systems requires data-driven approaches that can monitor and anticipate agricultural outcomes. Machine learning (ML) has been used to predict outcomes such as crop yields, soil health or water use from observational data. However, ML often cannot explain how these outcomes are affected by human or natural interventions, nor can it reliably generalize across regions and management contexts. Causal machine learning (causal ML) combines causal inference with ML to address these limitations. It enables estimation of causal effects for agricultural questions and improves prediction by focusing on stable and generalizable features. In this perspective, we introduce methods for causal inference from observational data and approaches that embed causal knowledge into predictive models. We then outline applications in research, policy and practice and conclude with key challenges and future directions for sustainable agriculture.

Submission Number: 33