Digging deeper: How AI is helping us understand the soil beneath our feet
Machine learning is helping scientists reveal key drivers of soil carbon storage, improving climate models and supporting better management of forests, farmland and natural ecosystems.
Soils may appear unremarkable, yet they are one of the planet’s most important climate regulators. They store more carbon than the atmosphere and all vegetation combined, and play a central role in global carbon cycling. Accurately predicting how much carbon they hold and how this might change in the future remains a major scientific challenge.
Recent research led by Dr Shoji Hashimoto (Forestry and Forest Products Research Institute, Japan) shows how machine learning is helping scientists to better understand soil carbon dynamics. Earth system models often struggle to reproduce real-world soil carbon measurements, leaving considerable uncertainty in climate projections. Machine learning can address this by analysing large global datasets to identify the factors that matter most.
The study found that while many models give significant weight to plant productivity, field data indicate that other influences are equally or more important. These include temperature, soil nitrogen, mineral content and land cover. By systematically comparing models with measured soil carbon stocks, researchers can detect missing processes and improve the performance of future models.
This work has implications beyond scientific modelling. A clearer picture of soil carbon storage supports more informed management of forests, farmland and natural ecosystems, strengthening strategies for climate change mitigation. As artificial intelligence methods continue to advance, they are becoming valuable tools for building models that reflect the full complexity of nature.
Read the full HoliSoils publication
Reference
Hashimoto, S., Bruni, E., Ťupek, B., Yamashita, N., Toriyama, J., Mori, T., Imaya, A., Guenet, B., Ito, A., & Lehtonen, A. (2025). Mining global soil carbon datasets: Can modern machine learning uncover the missing pieces of process-based models? Environmental Research Letters, 20(10), 101003. https://doi.org/10.1088/1748-9326/adfe83
This summary was prepared in collaboration with Dr Shoji Hashimoto (Forestry and Forest Products Research Institute, Japan). The original article is openly accessible in Environmental Research Letters.
