Quantifying moisture and roughness with Support Vector Machines improves spectroscopic soil organic carbon prediction

Abstract

The challenges of Vis-NIR spectroscopy are permanent soil surface variations of moisture and roughness. Both disturbance factors reduce the prediction accuracy of soil organic carbon (SOC) significantly. For improved SOC prediction, both disturbance effects have to be determined from Vis-NIR spectra, which is especially challenging for roughness. Thus, an approach for roughness quantification under varying moisture and its impact on SOC assessment using Support Vector Machines is presented here.

Hyperspectral selection based algorithm for plant classification

“…Machine learning is considered as one of the effective analysis tools. However, the main obstacles are finding the relevant information to the problem and dealing with the overfitting issues [12][13][14]. The former can be alleviated by using feature selection techniques.…”

Understanding the influencing factors and mechanisms (land use changes and check dams) controlling changes in the soil organic carbon of typical loess watersheds in China

“…Conventional SOC measurement methods require considerable time and effort and therefore have relatively high operating budgets. Numerous studies have been carried out to obtain accurate estimations of SOC distributions by using easily measurable information, which includes intrinsic and extrinsic factors (Rial et al, 2016; Romer et al, 2014; Thompson & Kolka, 2005). For example, Liu et al (2012) used soil characteristics (bulk density, soil pH, and clay contents) to estimate the distribution of SOC, while Devine et al (2020) used vegetation and topographic data.…”