Zhao et al. (2026) Network analysis and GMM-informed machine learning for scale-dependent soil quality assessment in a Mollisol watershed

Identification

• Journal: CATENA
• Year: 2026
• Date: 2026-09-22
• Authors: Lixiang Zhao, Linfang Tan, Xin Chen, Jie Wang, Yantun Song, Zhonglu Guo, Yujie Wei, Chongfa Cai
• DOI: 10.1016/j.catena.2026.110605

Research Groups

College of Resources and Environment, Huazhong Agricultural University, Wuhan, Hubei 430070, China
State Key Laboratory of Efficient Utilization of Arable Land in China/Key Laboratory of Arable Land Quality Monitoring and Evaluation, Ministry of Agriculture and Rural Affairs, Institute of Agricultural Resources and Regional Planning, Chinese Academy of Agricultural Sciences (CAAS), Beijing 100081, China
Guangdong Provincial Key Laboratory for Green Agricultural Production and Intelligent Equipment, School of Environmental Science and Engineering, Guangdong University of Petrochemical Technology, Maoming 525000, China

Short Summary

This study integrated network analysis with Gaussian mixture model-informed machine learning to address the challenges in soil quality assessment in rolling hilly Mollisol watersheds. The framework identified scale-dependent functional hub indicators and generated a soil quality index that accurately predicted spatial heterogeneity.

Objective

• Investigate the application of network analysis for selecting function-oriented minimum data sets (MDS) in soil quality assessment.
• Develop a machine learning model that incorporates Gaussian mixture model-derived spatial priors to predict soil quality at high resolution.

Study Configuration

• Spatial Scale: Watershed and sub-watershed scales (29.17 km2 and 1 km2, respectively)
• Temporal Scale: Not specified

Methodology and Data

• Models used: Gaussian mixture model-informed extreme gradient boosting (XGBoost)
• Data sources: Topsoil samples (63 at watershed scale and 30 at sub-watershed scale), soil physical and chemical indicators (13)

Main Results

• Network analysis identified hub indicators that varied by spatial scale: capillary porosity and soil organic matter at the watershed scale, and sand, available potassium, available phosphorus, and bulk density at the sub-watershed scale.
• The Gaussian mixture model-informed XGBoost model achieved high predictive performance (R2 = 0.47) for the total data set under linear scoring.

Contributions

• This study provided a framework that integrates network-based indicator selection with environmentally informed machine learning, enabling scale-sensitive and interpretable soil quality assessment in erosion-prone agricultural watersheds.
• The results highlighted the importance of topographic factors (62.29–78.89% model importance) and elevation as the dominant predictor.

Funding

• Not specified

Citation

@article{Zhao2026Network,
  author = {Zhao, Lixiang and Tan, Linfang and Chen, Xin and Wang, Jie and Song, Yantun and Guo, Zhonglu and Wei, Yujie and Cai, Chongfa},
  title = {Network analysis and GMM-informed machine learning for scale-dependent soil quality assessment in a Mollisol watershed},
  journal = {CATENA},
  year = {2026},
  doi = {10.1016/j.catena.2026.110605},
  url = {https://doi.org/10.1016/j.catena.2026.110605}
}