Modeling the spatial variation of calcium carbonate equivalent to depth using machine learning techniques

Area of Science:

• Soil Science
• Digital Soil Mapping
• Machine Learning Applications in Geosciences

Background:

• Inorganic carbon, primarily calcium carbonate equivalent (CCE), is a major soil carbon pool in arid and semiarid regions.
• Quantifying the variability of inorganic carbon is crucial for understanding soil properties and agricultural management in these environments.
• The Chahardowli Plain in western Iran, a representative arid region, lacks detailed inorganic carbon variability data.

Purpose of the Study:

• To model and map the spatial distribution of calcium carbonate equivalent (CCE) in the Chahardowli Plain using digital soil mapping (DSM) techniques.
• To identify key environmental predictors influencing CCE variability at different soil depths.
• To evaluate the performance of machine learning models (Random Forest and Decision Tree) for CCE prediction.

Main Methods:

• Soil samples were collected from 30 profiles (145 samples) across various depths (0-100 cm) using conditional Latin hypercube sampling.
• Calcium carbonate equivalent (CCE) was measured following GloalSoilMap.net protocols.
• Random Forest (RF) and Decision Tree (DT) models were employed to establish relationships between CCE and environmental predictors, including remote sensing and terrestrial variables.

Main Results:

• The mean CCE content significantly increased with soil depth, ranging from 3.5% (0-5 cm) to 63.8% (30-60 cm).
• Both remote sensing and terrestrial variables were important predictors, with variable importance shifting with depth.
• Channel Network Base Level (CNBL) and Difference Vegetation Index (DVI) were the most significant predictors, each accounting for 21.1% importance. Vertical Distance to Channel Networks (VDCN) was also critical in areas influenced by river activity.

Conclusions:

• Machine learning, particularly the RF model, effectively predicts and maps soil inorganic carbon (CCE) distribution.
• Topographic variables (CNBL, VDCN) and vegetation indices (DVI) are key drivers of CCE variability, especially in river-influenced zones.
• Understanding CCE distribution is vital for sustainable agricultural practices in arid regions, particularly regarding potential nutrient deficiencies caused by high carbonate levels.