Journal of Environmental Biology

Spatial variability of soil organic carbon in Tamil Nadu: Integrating remote sensing and machine learning approaches

B.K.R. Bobbiti¹, S. Maragatham¹*, R. Santhi¹, D. Balachander², D. Vijayalakshmi³, V. Davamani⁴, V. Duraisamy⁵ and M. Gopalakshinan⁴     

¹Department of Soil Science & Agricultural Chemistry, Tamil Nadu Agricultural University, Coimbatore, India-641 003

²Department of Agricultural Microbiology, Tamil Nadu Agricultural University, Coimbatore-641 003, India

³Department of Crop Physiology, Tamil Nadu Agricultural University, Coimbatore, India-641 003

⁴Directorate of Natural Resource Management, Tamil Nadu Agricultural University, Coimbatore-641 003, India

⁵Division of Soil Resource Studies, ICAR-NBSS & LUP, Nagpur-440 033, India

Received: 10 June 2024                   Revised: 31 August 2024                   Accepted: 03 October 2024

*Corresponding Author Email : maragatham.s@tnau.ac.in                           *ORCiD: https://orcid.org/0000-0002-5787-2390

Abstract

Aim: The experiment was carried out to estimate the geographical distribution of soil organic carbon (SOC) in Tamil Nadu, India, using advanced machine learning models.

Methodology: High-resolution remote sensing data from Landsat 8, climate indices from TerraClimate, and terrain indices from the Shuttle Radar Topography Mission (SRTM) were integrated to construct reliable SOC estimation models. The models employed in this study included Random Forest (RF), Quantile Neural Network (QNN), Cubist, and Bootstrapped Random Forest (BRF). These models were selected for their ability to capture complex relationships between SOC and environmental covariates.

Results: The QNN and Cubist models displayed superior accuracy with mean absolute error (MAE) values of 0.48 and 0.47, respectively. Variogram analysis indicated moderate spatial dependence across all models, with the Cubist model exhibiting the highest partial sill and sill values.

Interpretation: These findings demonstrate the effectiveness of combining machine learning with remote sensing for SOC estimation. The results suggest that these models can be valuable tools in improving the accuracy of SOC distribution maps, which are essential for sustainable crop production. By identifying potential SOC sequestration areas, this research contributes to the development of targeted soil management strategies that prioritize soil health and support sustainable agricultural practices.

Key words: Carbon sequestration, Digital soil mapping, Estimation models, Environmental covariates, Land management

Copyright © 2025 Triveni Enterprises. All rights reserved. No part of the Journal can be reproduced in any form without prior permission. Responsibility regarding the authenticity of the data, and the acceptability of the conclusions enforced or derived, rest completely with the author(s).