Quantitative derivation of key soil parameters on the basis of hyperspectral remote sensing data

Abstract

The Subtropical Thicket Biome (Eastern Cape, South Africa) with its dense shrub vegetation in pristine conditions stores unusually large amounts of carbon for a semi-arid region. Therefore it is a region of extraordinary importance [7]. Currently the region undergoes fundamental changes caused by heavy grazing of goats which is accompanied by a serious depletion of carbon stocks above and below ground. The key soil parameters clay, iron and organic carbon can be used to assess land degradation. In this context these parameters are quantified using hyperspectral remote sensing data and multiple linear regression techniques. First, the correlation of spectral features with measured soil contents is modelled using unsupervised partial-least-squares regression analysis and second a feature-based approach is tested.