Use of Visible and Near-Infrared Reflectance Spectroscopy Models to Determine Soil Erodibility Factor (<i>K</i>) in an Ecologically Restored Watershed

• Main Authors: Qinghu Jiang, Yiyun Chen, Jialiang Hu, Feng Liu
• Format: Article
• Language: English
• Published: MDPI AG 2020-09-01
• Series: Remote Sensing
• Subjects: soil erodibility; visible and near-infrared reflectance spectroscopy; model transferability; model improvement; watershed landscapes
• Online Access: https://www.mdpi.com/2072-4292/12/18/3103

This study aimed to assess the ability of using visible and near-infrared reflectance (Vis–NIR) spectroscopy to quantify soil erodibility factor (<i>K</i>) rapidly in an ecologically restored watershed. To achieve this goal, we explored the performance and transferability of the developed spectral models in multiple land-use types: woodland, shrubland, terrace, and slope farmland (the first two types are natural land and the latter two are cultivated land). Subsequently, we developed an improved approach by combining spectral data with related topographic variables (i.e., elevation, watershed location, slope height, and normalized height) to estimate <i>K</i>. The results indicate that the calibrated spectral model using total samples could estimate <i>K</i> factor effectively (<i>R</i>²_CV<i> </i>= 0.71, <i>RMSE</i>_CV = 0.0030 Mg h Mj⁻¹ mm⁻¹, and <i>RPD</i>_CV = 1.84). When predicting <i>K</i> in the new samples, models performed well in natural land soils (<i>R</i>²_P = 0.74, <i>RPD</i>_P = 1.93) but failed in cultivated land soils (<i>R</i>²_P = 0.24, <i>RPD</i>_P = 0.99). Furthermore, the developed models showed low transferability between the natural and cultivated land datasets. The results also indicate that the combination of spectral data with topographic variables could slightly increase the accuracies of <i>K</i> estimation in total and natural land datasets but did not work for cultivated land samples. This study demonstrated that the Vis–NIR spectroscopy could be used as an effective method in predicting <i>K</i>. However, the predictability and transferability of the calibrated models were land-use type dependent. Our study also revealed that the coupling of spectrum and environmental variable is an effective improvement of <i>K</i> estimation in natural landscape region.