Rasaei et al. (2025) Can environmental clustering reveal soil profile patterns? A depth-based approach at field scale

Identification

• Journal: CATENA
• Year: 2025
• Date: 2025-11-30
• Authors: Zahra Rasaei, Fereydoon Sarmadian, Azam Jafari, Trevan Flynn
• DOI: 10.1016/j.catena.2025.109695

Research Groups

• Soil Science Department, Faculty of Agricultural, University College of Agriculture and Natural Resources, University of Tehran, Karaj, Iran
• Soil Science Department, Faculty of Agriculture, Shahid Bahonar University of Kerman, Kerman, Iran
• Department of Agronomy, University of Fort Hare, Alice, South Africa
• Department of Research, Add1Tecnologies, Richmond, USA

Short Summary

This study numerically clustered soil profiles using multiple algorithms and a "pedogenon"-inspired methodology at a field scale to create meaningful soil management units. The Mahalanobis distance hierarchical clustering (HM) algorithm performed best, providing clear separation of soil properties with depth and aligning well with expert understanding.

Objective

• To numerically cluster soil profiles using multiple algorithms, following a “pedogenon”-inspired methodology applied at the field scale, to simplify soil landscapes into meaningful management units and reduce subjectivity in soil groupings.

Study Configuration

• Spatial Scale: A 5.8 x 10⁸ square meter (58,000-hectare) low-relief landscape in an arid to semi-arid region in Iran.
• Temporal Scale: Contemporary analysis of existing soil profile data.

Methodology and Data

• Models used: Hierarchical clustering using Euclidean distance (HC), Mahalanobis distance (HM), k-means (KM), Partitioning Around Medoids (PAM), fuzzy c-means (FCM), and biclustering (BiC).
• Data sources: Environmental soil-forming factors and ten physio-chemical soil properties at harmonized depth intervals, derived from original field observation data (Sarmadian, Original data).

Main Results

• The Mahalanobis distance hierarchical clustering (HM) algorithm achieved the lowest Akaike information criterion (AIC), indicating the best fit for clustering soil profiles.
• HM was followed by FCM, KM, HC, PAM, and BiC in terms of performance (from best to worst AIC).
• HM clusters aligned well with the conceptual understanding of soil scientists (USDA Soil Taxonomy) and provided a clearer separation in soil properties with depth compared to a legacy Soil Taxonomy subgroup map.
• The clustering algorithms that were easiest to interpret were also the most accurate, enhancing communication across disciplines.
• The biclustering (BiC) model, despite weaker performance, showed moderate alignment with conventional algorithms, particularly k-means (KM) cluster patterns.

Contributions

• First-time evaluation of a biclustering (BiC) algorithm for soil classification.
• Demonstrated that numerical clustering, especially using the HM algorithm, can provide more objective, reproducible, and interpretable soil classifications than traditional expert-based methods.
• Applied a "pedogenon"-inspired methodology at the field scale for numerical soil classification, contributing to the development of precise and reproducible soil classification methods.
• Enhanced communication across scientific disciplines, policymakers, and farmers through accurate and interpretable soil classifications.

Funding

Not explicitly mentioned in the provided text.

Citation

@article{Rasaei2025Can,
  author = {Rasaei, Zahra and Sarmadian, Fereydoon and Jafari, Azam and Flynn, Trevan},
  title = {Can environmental clustering reveal soil profile patterns? A depth-based approach at field scale},
  journal = {CATENA},
  year = {2025},
  doi = {10.1016/j.catena.2025.109695},
  url = {https://doi.org/10.1016/j.catena.2025.109695}
}