Cloud-based soil salinity mapping using remote sensing and machine learning approaches

doi:10.1016/j.jafrearsci.2026.105993

Cloud-based soil salinity mapping using remote sensing and machine learning approaches

2026-01-01

0

PRE

AI

S

S. Raja Ranganathan *

N

Nada Alzaben

A

Abdulaziz G. Alghamdi

H

Hatim Dafaalla

DOI：10.1016/j.jafrearsci.2026.105993

Original

Original request for help

Save

Abstract

En 中文

Soil salinity is a prevalent environmental issue, especially in coastal environments, which exacerbate the conditions favouring salt accumulation in the root zone (saline intrusion, poor drainage, and evapotranspiration). The study focused to create a robust, cloud-based method for mapping soil salinity using remote sensing data and more sophisticated machine learning approaches. Remote imagery from Sentinel-2 was used to generate spectral indices, while topographic variables were derived from DEM products within Google Earth Engine. Two ensemble learning models were created: Random Forest (RF) and Light Gradient Boosting Machine (LightGBM), to determine soil electrical conductivity (EC) based on the environmental variables above. The two models were evaluated with calibration, validation and cross-validation tests. The RF produced the best results with R2 values of 0.81, 0.68, 0.59 and RMSE of 14.8, 18.45 and 20.15 dS m-1 for the data splits. The LightGBM produced R2 values of 0.73, 0.65, 0.56 and RMSE values of 18.9, 18.75 and 20.25 dS m- 1, reflecting lower predictive accuracy compared to other models, yet still within an acceptable range. The relative performance index of quantification (RPIQ) confirmed the improved predictability of RF. The results confirm the feasibility of cloud-computing capabilities and satellite-based environmental characteristics with machine learning to assess and monitor soil salinity in dynamic coastal environments. The method also provides land managers and policy makers with a scalable option to assess salinity risk to assist with sustainable agricultural practices in regions affected by salt.

Keywords:

Google earth engine

Electrical conductivity

Vegetation indices

Ensemble learning

Random forest

LightGBM

Cloud-based soil salinity mapping using remote sensing and machine learning approaches

Abstract

Journal

Organization