Related Experiment Video
Updated: Jul 2, 2026
Two-Dimensional Visualization and Quantification of Labile, Inorganic Plant Nutrients and Contaminants in Soil
Published on: September 1, 2020
Soil heavy metal prediction using GF-5 hyperspectral image and environmental covariates with data augmentation
Youxin Sun1, Xia Zhang2, Yaqiong Zhang3
1State Key Laboratory of Remote Sensing and Digital Earth, Aerospace Information Research Institute, Chinese Academy of Sciences, Beijing 100101, China; University of Chinese Academy of Sciences, No.3 Datun Road, Chaoyang District, Beijing 100101, China.
This study improves soil heavy metal (SHM) prediction using hyperspectral data by integrating environmental factors and data augmentation. The novel approach enhances accuracy for lead (Pb), zinc (Zn), and nickel (Ni) monitoring.
Area of Science:
- Environmental Science
- Remote Sensing
- Geochemistry
Background:
- Hyperspectral imaging is effective for monitoring soil heavy metals (SHM) over large areas.
- Model accuracy is often limited by environmental covariates and small sample sizes.
- Accurate SHM prediction is crucial for environmental monitoring and risk assessment.
Purpose of the Study:
- To develop an accurate SHM prediction method integrating environmental covariates and data augmentation.
- To enhance the reliability and accuracy of predicting lead (Pb), zinc (Zn), and nickel (Ni) concentrations.
- To provide a framework for SHM prediction and pollution risk assessment with limited samples.
Main Methods:
- Geographic detector and Shapley additive explanations identified key environmental covariates.
- A spectral smoothing-constrained Wasserstein generative adversarial network with gradient penalty (SS-WGAN-GP) augmented the dataset.
- TabNet model was used for SHM prediction and spatial mapping.
Main Results:
- Environmental covariates significantly improved SHM prediction accuracy.
- SS-WGAN-GP data augmentation further enhanced model performance, achieving R² values of 0.70 (Pb), 0.70 (Zn), and 0.72 (Ni).
- Spatial mapping revealed elevated SHM concentrations near mining areas and major roads.
Conclusions:
- The proposed method effectively integrates environmental covariates and data augmentation for accurate SHM prediction.
- The SS-WGAN-GP is a reliable tool for generating realistic hyperspectral data samples.
- This study offers a viable solution for SHM monitoring and pollution assessment, especially with limited soil samples.
