Soil zinc content estimation using GF-5 hyperspectral image with mitigation of soil moisture influence

Abstract

Hyperspectral imagery has a high potential for large-area estimation of soil heavy contents. However, soil moisture significantly influences spectral analysis accuracy, which many existing studies on soil metal estimation have overlooked. This study investigates the impact of soil moisture on the characteristic spectral range of Soil Spectrally Active Constituents (SSAC) by analyzing soil spectra under varying moisture conditions. Based on this analysis, the SSAC characteristic bands were identified and subjected to segmented Orthogonal Signal Correction (OSC)to mitigate moisture influence. Then, a stacking ensemble model was constructed based on the corrected SSAC bands. A total of 105 soil samples were collected from the Dongsheng coalfield mining area in the Inner Mongolia Autonomous Region, China, alongside Chinese Gaofen-5 (GF-5) satellite hyperspectral imagery acquired simultaneously. The results demonstrate that the segmented OSC can effectively mitigate the influence of soil moisture when moisture is 15% or less. After applying the segmented OSC, the accuracy R² of the test set is improved significantly from 0.0508 to 0.7697. Additionally, the stacking ensemble model outperformed conventional single models, demonstrating superior accuracy in estimating soil heavy metal content. The use of SSAC characteristic bands also reduced model overfitting. The estimated spatial distribution of soil zinc (Zn) content in the study area is accurate and reasonable, indicating high reliability and applicability of the proposed method. This approach provides a robust solution for precise soil metal estimation under varying moisture conditions.