Digital mapping of surface soil salinity in the central part of Khuzestan Province using supervised image classification algorithms

Document Type : Research Paper

Authors

• Mohammad Abiyat ¹
• Saeid Hojati ²
• Ahmad Landi ³
• Asim Biswas ⁴

¹ Department of Soil Science, Faculty of Agriculture, Shahid Chamran University of Ahvaz, Khuzestan, Iran

² Full Professor, Department of Soil Science, College of Agriculture, Shahid Chamran University of Ahvaz

³ Full Professor, Department of Soil Science, Faculty of Agriculture, Shahid Chamran University of Ahvaz, Khuzestan, Iran

⁴ School of Environmental Science, Ontario Agricultural College, University of Guelph, Guelph N1G2W1, Ontario, Canada

Abstract

Keywords

• Environmental Variables
• Principal Component Analysis
• Pixel-Based Method
• Object-Oriented Analysis

Main Subjects

• پیدایش و رده‌بندی خاک

Introduction

Soil salinity is one of the major causes of land degradation in arid and semi-arid regions, negatively impacting agricultural productivity. Remote sensing and satellite imagery are essential tools for monitoring soil salinity, as image classification techniques enable the accurate identification and detection of saline areas. Bavi County, located in Khuzestan Province is a key agricultural region, making soil salinity assessment crucial for farmers to adopt sustainable soil and water management practices and prevent crop yield reduction. This study aims to create a digital map of surface soil salinity using supervised classification algorithms.

Methodology

First, 350 soil samples were collected from a depth of 0–10 cm using conditioned Latin hypercube sampling. In the laboratory, electrical conductivity (EC) was measured in a 1:2 soil-to-water extract to determine soil salinity. Environmental variables were derived from remote sensing indices based on satellite imagery and topographic features extracted from a digital elevation model (DEM). Principal Component Analysis (PCA) was applied to reduce data dimensionality and enhance interpretability. Initially, 80% of the data were used for training, and 20% for model validation. To evaluate the impact of sample size on algorithm performance, an alternative split of 70% training and 30% testing was also applied. For soil salinity classification, both pixel-based supervised methods (Maximum Likelihood, Minimum Distance, Decision Tree, Random Forest, Artificial Neural Network, and Spectral Angle Mapper) and object-based approaches (K-Nearest Neighbors and Support Vector Machine) were employed. Model performance was assessed using overall accuracy and the Kappa coefficient.

Results and Discussion

Field data analysis revealed an average EC of 28.51 dS/m, indicating an extremely soil saline class. PCA results showed that the first six principal components (PC1–PC6), with eigenvalues greater than one and cumulative variance exceeding 91%, contained the most significant information for classification. Key environmental variables influencing salinity detection included the Brightness Index (BI), Vegetation Soil Salinity Index (VSSI), Difference Vegetation Index (DVI), and Green Normalized Difference Vegetation Index (GNDVI), all of which exhibited high factor loadings (0.99). After selecting the optimal components, these were compiled into a multi-band data_set for supervised classification. The results indicated that a reduction in the number of training samples decreased classification accuracy. Among the classification methods, the object-based Support Vector Machine (SVM) and pixel-based Random Forest (RF) achieved the highest accuracy in distinguishing salinity classes. In contrast, the Spectral Angle Mapper (SAM) algorithm performed the poorest in detecting soil salinity. The SVM-derived salinity map revealed that most of the study area falls under the extremely saline class (greater than 16 dS/m), highlighting the critical salinity status of the region. The widespread distribution of extreme salinity may be attributed to anthropogenic activities (e.g., irrigation with saline water) and natural factors (e.g., drought). Visual assessment indicated that severely saline soils were predominantly located in the western part of the study area.

Conclusion

This study demonstrated that integrating remote sensing data and topographic variables with advanced classification methods, particularly object-based approaches, provide an effective tool for monitoring soil salinity in arid regions. The resulting salinity maps can support soil and land management in affected areas. Implementing management strategies such as cultivating salt-tolerant crops, improving drainage systems in highly saline zones, and continuous monitoring using updated maps can contribute to soil conservation and mitigate land degradation.

Author Contributions

First author: Collecting and analyzing data, writing — original draft preparation, review, and editing. Second and Third authors: Conceptualization, writing — review and editing, methodology, guidance, research oversight, project administration. Fourth author: Advisement, writing — editing.

Data Availability Statement

Data is available on request from the authors.

Acknowledgements

The authors would like to thank the Deputy of Research Technology and Social Relationships at Shahid Chamran University of Ahvaz (SCU.AS1403.365), as well as the Iran National Science Foundation (INSF, Project No. 4038641), for their financial support of this research.

Ethical considerations

The authors avoided data fabrication, falsification, plagiarism, and misconduct.

Conflict of interest

The author declares no conflict of interest.