Estimating soil profile salinity under vegetation cover based on UAV multi-source remote sensing

Abstract

Abstract Soil salinization is the most prevalent form of land degradation in arid, semi-arid, and coastal regions of China, posing significant challenges to local crop yield, economic development, and environmental sustainability. However, limited research exists on estimating soil salinity at different depths under vegetation cover. This study employed field-controlled soil experiments to collect multi-source remote sensing data on soil salt content (SSC) at varying depths beneath barley growth. Three types of feature variables were derived from the images and filtered using the boosting decision tree (BDT) method. In addition, four machine learning algorithms coupled with seven variable combination groups were applied to establish comprehensively soil salinity estimation models. The performances of estimation model for different crop coverage ratios and soil depth were then evaluated. The results showed that the gaussian process regression (GPR) model, based on the whole variable group for depths of 0 ~ 10 cm and 30 ~ 40 cm, outperformed other models, achieving validation R2 values of 0.774 and 0.705, with RMSE values are 0.185% and 0.31%, respectively. For depths of 10 ~ 20 cm and 20 ~ 30 cm, the random forest (RF) models, incorporating spectral index and texture data, demonstrated superior accuracy with R2 values of 0.666 and 0.714. The study confirms that SSC can be quantitatively estimated at various depths using the machine learning model based on multi-source remote sensing, providing a valuable approach for monitoring soil salinization.

Keywords

• Soil salt content
• Feature selection
• Machine learning models
• Soil depth
• Barley