Moisture spectral characteristics and hyperspectral inversion of fly ash-filled reconstructed soil
Ke Xia,Shasha Xia,Qiang Shen,Bin Yang,Qiang Song,Yunfei Xu,Shiwen Zhang,Xu Zhou,Yan Zhou
DOI: https://doi.org/10.1016/j.saa.2021.119590
2021-05-15
Abstract:To explore problems with the fast estimation method of moisture content (MC) in reconstructed soil under human disturbance, this paper used a fly ash-filled reconstructed soil region as the research object and obtained experimental data by Fieldspec4 high spectrometry and the laboratory drying method. The response characteristics of MC were analyzed from the original spectral data that underwent mathematical transformation and the spectral index data, and a corresponding inversion model was established. Combined with the successive projections algorithm (SPA), the model was optimized with a better fitting effect, and the optimal inversion model was obtained. The results showed that the composition of soil and fly ash were different, resulting in obvious differences in the shape of the spectral curve, but both had large moisture absorption peaks near 1420 nm and 1920 nm. After mathematical transformation, the correlation between the spectral reflectance and MC was enhanced, in which the absolute value of the maximum correlation between the soil moisture content (SMC) was 0.839, and the absolute value of the maximum correlation between the fly ash moisture content (FMC) was 0.801. Among them, the first-order differential of multivariate scattering correction (MSC') and the first-order differential of logarithm ((lgR)') had higher fitting accuracy for FMC and SMC, respectively. The scale and sensitivity of significance variables were greatly improved based on the spectral index of two-band operation. Better FMC and SMC models were constructed based on the difference soil index (DSI) under mathematical transformation, and R2 were 0.73 and 0.87, respectively. After SPA optimization, the predictive ability of the model was further improved, in which the predictive accuracy R2 of FMC and SMC reached up to 0.87 and 0.96, respectively, and the RPD was greater than 3. This shows that the DSI model based on MSC' and (lgR)' combined with the SPA method can be used as an effective means of predicting the MC in fly ash-filled reconstructed soil. These research results provide the theoretical basis and technical support for the application of soil near-earth sensing technology and rapid estimation of the MC of reconstructed soil under human disturbance.