Determining surface magnetic susceptibility of loess-paleosol sections based on spectral features: Application to a UHD 185 hyperspectral image

Abstract

Magnetic susceptibility (MS) records of loess-paleosol sequences have been considered a measure of the degree of pedogenic activity and are considered to be excellent proxies for terrestrial climatic fluctuations. However, the MS of single (vertical) path variations occasionally represents site-specific influences rather than monsoonal changes (depending on the position of the path). Few studies have used remote sensing techniques to map the surface MS information of loess-paleosol sections. Hyperspectral techniques provide an efficient, economical and quantitative alternative. In this study, stepwise regression was used to build MS estimation models based on spectral features. Six MS models based on spectral features were established. Test datasets indicated that our models are very successful, all resulting in R2>0.92 and RMSEs ranging from 4.5736 to 6.80475. The slope change between 810nm and 880nm (b880/b810) observed in all models played an important role in MS estimation. Models 5 and 6 have higher RMSEs and relatively lower SAM values, although the R2 values are both above 0.95. The RMSEs of the first four models are similar. Therefore, the first four models were thought to be more stable and useful.UHD 185, a new generation of commercial hyperspectral imaging sensor, was used for surface MS mapping of a loess-paleosol section by model 1 and model 2. The MS map corresponded well to the loess sequences. The MS values obtained from the UHD 185 data are convincing and consistent with the measured data (R2>0.85). The trend in changing MS values is clear, suggesting that model 1 and model 2 could produce reasonable loess-paleosol section surface maps from the UHD 185 image, although there is a linear offset between the estimated MS and the measured MS. The methodology proposed here can be used to map MS on a much larger scale. Because of the limit of the spectral range, the performances of model 3 and model 4 with the image were not discussed. However, they have been shown to be successful according to the laboratory test data.