EGU General Assembly 2023
© Author(s) 2023. This work is distributed under
the Creative Commons Attribution 4.0 License.

Exploring the ‘Individual Treatment Effects’ (ITE) of Vegetation with Causal Inference on Soil Organic Carbon Prediction in Germany

Nafiseh Kakhani1, Thomas Gläßle2, Ruhollah Taghizadeh-Mehrjardi1,3,4, Ndiye Michael Kebonye1,4, and Thomas Scholten1,3,4
Nafiseh Kakhani et al.
  • 1Department of Geosciences, University of Tübingen, Tübingen, Germany (
  • 2Department of Computer Science, University of Tübingen, Tübingen, Germany
  • 3CRC 1070 Resource Cultures, University of Tübingen, Tübingen, Germany
  • 4DFG Cluster of Excellence “Machine Learning”, University of Tübingen, Tübingen, Germany

Carbon is an essential element and contributor to healthy soil conditions as well as ecological soil function and productivity. Additionally, carbon is a component of all plants and animals on the planet and is a necessary component of life. Natural vegetation serves as a significant but highly dynamic carbon sink. When vegetation is removed quicker than it can regenerate, for example by harvesting crops or timber, soil carbon is depleted. Thus, understanding the environmental effects and dynamics of loss of vegetation is a crucial prerequisite to turning our natural resource management from a carbon emitter to a carbon sink to avoid that and achieve sustainability. At the same time, the spatial distribution of soil organic carbon is also highly heterogeneous, with variations in climate, other soil characteristics, and land use/land cover affecting how our ecosystem reacts to the loss of vegetation. Thus, to effectively improve green metrics and contribute to the creation of future policies, it is required to conduct research on the changes in vegetation and their effect on soil organic carbon and provide regionally appropriate management advice. Here, in this research, our goal is to examine the "individual treatment effects" (ITE), which are a personalized or individualized effect estimation of one variable on the output, and utilize causal inference to address them.  Using the LUCAS dataset, we explore the heterogeneous treatment effect of percent tree coverage (PTC), as a parameter of the density of trees on the ground, on the soil organic carbon content in Germany. We do this by leveraging some parameters, such as climate data, land use/land cover information, and other information from the soil. We thus offer a data-driven viewpoint for focusing on sustainable behaviors and effectively increasing soil organic carbon content levels.

How to cite: Kakhani, N., Gläßle, T., Taghizadeh-Mehrjardi, R., Kebonye, N. M., and Scholten, T.: Exploring the ‘Individual Treatment Effects’ (ITE) of Vegetation with Causal Inference on Soil Organic Carbon Prediction in Germany, EGU General Assembly 2023, Vienna, Austria, 24–28 Apr 2023, EGU23-1083,, 2023.