Document Type : Research Paper

Authors

• Saeed Saadat
• leila esmaeelnejad
• Hamed Rezaei
• Rasoul Mirkhani
• َAmirreza Esfandyar

Soil and Water Research Institute, Karaj, Iran

Abstract

This study was conducted to model and estimate the spatial distribution of soil organic carbon (SOC) in the agricultural lands of Alborz province by integrating Landsat 8 satellite data with the Random Forest machine learning algorithm. The model was based on field data from 257 soil samples collected via a regular grid method, with their SOC content measured using the Walkley-Black method. Predictor variables included visible, near-infrared (NIR), and short-wave infrared (SWIR) spectral bands, along with relevant vegetation indices. The model's performance evaluation confirmed its high accuracy, with a coefficient of determination (R2) of 0.83 and a normalized root mean square error (NRMSE) of 12.5%. The resulting map showed an average SOC of 0.23% for the region, with the highest values concentrated in the northern and central parts of the province. Additionally, a trend analysis over the last decade showed no statistically significant changes in the mean SOC. The findings demonstrate the significant potential of combining machine learning algorithms with remote sensing data for accurate and efficient mapping of soil organic carbon at a regional scale. This approach can serve as an effective tool for sustainable soil resource management and greenhouse gas emission reduction strategies.

Keywords

• Digital Soil Mapping
• Random Forest
• Landsat 8
• Spectral Indices
• Geographic Information System (GIS)