Estimation of soil organic carbon by combining hyperspectral and radar remote sensing to reduce coupling effects of soil surface moisture and roughness

• DOI: 10.1016/j.geoderma.2024.116874
• Publication Date: 2024-04-01T09:14:24Z
• Authors: Ranzhe Jiang; Yuanyuan Sui; Xin Zhang; Nan Lin; Xingming Zheng; Bingze Li; Lei Zhang; Xiaokai Li; Haiye Yu
• Keywords: Soil carbon

Soil organic carbon (SOC) is important in the global cycle. Accurate estimation of SOC content cultivated land a prerequisite for evaluating sequestration potential and quality soils. However, existing prediction studies based on hyperspectral remote sensing neglect spectral response physical properties surface soil, leading to inadequate model generalization. With exponential growth data, development pixel-level soil correction methods multi-source data has become an interesting challenging topic. This method aims minimize effect spectra, thus addressing poor spatiotemporal transferability models due uncertain variations properties. In this study, strategy constructed using satellite image (HSI) synthetic aperture radar (SAR) images through multi-order polynomial regression convolutional neural networks. considers variables such as moisture (SM) root mean square height (RMSH) roughness. The were established 80 samples collected from Site 1. Afterward, performance both verified remaining 25 1 50 2. results showed that: 1) SM RMSH pixel spectrum can be significantly reduced after correcting HSI strategy. correlation coefficients between corrected ground-based increase by over 60 % compared with those original spectrum. 2) improves accuracy mapping capability content, highest RP2 0.743 RMSEP 3.455 g/kg at 3) Compared HSI-based model, network 2 5.082 5.454 g/kg, increased 0.390 0.409, respectively. 4) When predicting raw HIS, contribute more than bias, having larger bias RMSH. findings study emphasize influence research SAR data.