Precise Monitoring of Soil Salinity in China's Yellow River Delta Using UAV-Borne Multispectral Imagery and a Soil Salinity Retrieval Index

• Published in: Sensors (Basel, Switzerland) Switzerland), 2022-01, Vol.22 (2), p.546
• Main Authors: Yu, Xinyang, Chang, Chunyan, Song, Jiaxuan, Zhuge, Yuping, Wang, Ailing
• Format: Article
• Language: English
• Subjects: Agricultural land; Agriculture; Algorithms; Arable land; Back propagation networks; Cameras; China; Correlation analysis; Correlation coefficients; Drone aircraft; Image acquisition; Laboratories; Least squares method; Methods; Monitoring; Neural networks; optimal retrieval model; Parameter sensitivity; Precipitation; random forest; Reflectance; remote sensing; River ecology; Rivers; Root-mean-square errors; Salinity; Salt; Serotonin uptake inhibitors; Soil; soil salinity sensitive parameter; Soils, Salts in; support vector machine; Support vector machines; Sustainable development; Unmanned Aerial Devices; Unmanned aerial vehicles; Vegetation
• ISSN: 1424-8220; 1424-8220
• DOI: 10.3390/s22020546

Monitoring salinity information of salinized soil efficiently and precisely using the unmanned aerial vehicle (UAV) is critical for the rational use and sustainable development of arable land resources. The sensitive parameter and a precise retrieval method of soil salinity, however, remain unknown. This study strived to explore the sensitive parameter and construct an optimal method for retrieving soil salinity. The UAV-borne multispectral image in China's Yellow River Delta was acquired to extract band reflectance, compute vegetation indexes and soil salinity indexes. Soil samples collected from 120 different study sites were used for laboratory salt content measurements. Grey correlation analysis and Pearson correlation coefficient methods were employed to screen sensitive band reflectance and indexes. A new soil salinity retrieval index (SSRI) was then proposed based on the screened sensitive reflectance. The Partial Least Squares Regression (PLSR), Multivariable Linear Regression (MLR), Back Propagation Neural Network (BPNN), Support Vector Machine (SVM), and Random Forest (RF) methods were employed to construct retrieval models based on the sensitive indexes. The results found that green, red, and near-infrared (NIR) bands were sensitive to soil salinity, which can be used to build SSRI. The SSRI-based RF method was the optimal method for accurately retrieving the soil salinity. Its modeling determination coefficient ( ) and Root Mean Square Error ( ) were 0.724 and 1.764, respectively; and the validation , , and Residual Predictive Deviation ( ) were 0.745, 1.879, and 2.211.