# Towards assessing agricultural land suitability with causal machine learning

**Authors:** Georgios Giannarakis, Vasileios Sitokonstantinou, Roxanne Suzette Lorilla, Charalampos Kontoes  
**arXiv:** [2204.12956](https://arxiv.org/abs/2204.12956)  
**Date:** 2022-04-27  
**Summary page:** https://www.summarizepaper.com/en/paper/2204.12956v1/

## Key points

- Agricultural land suitability is critical for sustainable and resilient agriculture against climate change impacts
- Causal machine learning can estimate intervention impacts on an outcome of interest based on observed characteristics
- A data-driven framework leverages earth observations to assess agricultural land suitability as a geospatial impact assessment problem
- Estimated effects of agricultural practices serve as a land suitability score and guide decision-making
- The approach is formulated as a causal machine learning task and demonstrates potential for agricultural planning in a changing climate
- Crop rotation and landscape crop diversity were analyzed using double machine learning to estimate their effect on Net Primary Productivity (NPP) within the Flanders region of Belgium from 2010 to 2020
- Results show that while crop rotation had an insignificant effect, landscape crop diversity had a small negative effect on NPP with considerable effect heterogeneity in space for both practices.
- This study highlights the potential benefits of using causal machine learning to assess agricultural land suitability and guide decision-making towards more sustainable agriculture practices in the face of climate change.

## Summary

The suitability of agricultural land for specific management practices is a critical factor in ensuring sustainable and resilient agriculture against the impacts of climate change. Recent advances in causal machine learning have enabled the estimation of intervention impacts on an outcome of interest, based on observed characteristics. In this study, the authors introduce a data-driven framework that leverages earth observations to assess agricultural land suitability as a geospatial impact assessment problem. The estimated effects of agricultural practices on agroecosystems serve as a land suitability score and guide decision-making. The authors formulate this approach as a causal machine learning task and demonstrate its potential for agricultural planning in a changing climate. Specifically, they extract information about two key agricultural management practices - "crop rotation" and "landscape crop diversity" - from crop type maps, account for climate and land use data, and use double machine learning to estimate their heterogeneous effect on Net Primary Productivity (NPP) within the Flanders region of Belgium from 2010 to 2020. The results show that while the effect of crop rotation was insignificant, landscape crop diversity had a small negative effect on NPP. The authors also observe considerable effect heterogeneity in space for both practices and analyze it further. Overall, this study highlights the potential benefits of using causal machine learning to assess agricultural land suitability and guide decision-making towards more sustainable agriculture practices in the face of climate change.