The use of visible and near-infrared reflectance measurements for identifying the source of suspended sediment in rivers and comparison with geochemical fingerprinting

Abstract

Visible and near-infrared (Vis-NIR) reflectance measurements may be an alternative technique to identify suspended sediment sources in streams of headwater catchments. In this study, we examined if Vis-NIR reflectance measurements are capable of estimating sediment source contributions to sediment yield and compared this technique with a more conventional (i.e. geochemical) technique. Two headwater catchments in Ethiopia, Unta (2,052 ha) and Desera (1,657 ha), were analysed with the same techniques in order to find similarities and differences in the results obtained. The first technique used Vis-NIR spectral analysis as a fingerprint, using a partial least squares regression model. The second technique was a quantitative composite fingerprinting technique using geochemical analysis of source materials and suspended sediment samples. As a comparison, the partial least squares model was also used on the geochemical data. In August and September 2009, 30 soil samples of three different land uses (landslides, croplands, and grazing lands) and 21 suspended sediment samples at the catchment outlet were collected. Source samples were sieved to <63 μm. Geochemical analyses consisted of total element concentrations, percentage carbon, percentage nitrogen, and atom percentage 15N and δ13C. Reflectance measurements were taken on dried source samples with a spectrometer. Neither technique was able to predict the contributions of the three land use types; they could only distinguish between landslide and topsoil material. The agreement between the results of both techniques was significant for the Unta catchment (R 2 = 0.80) but not for the Desera catchment (R 2 = 0.39). The uncertainty of the technique using Vis-NIR reflectance measurements was slightly higher than with the geochemical approach. Both techniques revealed that topsoil erosion played an important role during storm runoff discharges. Using the partial least squares model for the geochemical data revealed that uncertainty can differ greatly when using other statistical techniques. The quantitative composite fingerprinting technique using spectral signatures from both source and suspended sediment samples was able to quantify the contribution of two source materials (landslides and topsoil). It provided a faster and more cost effective alternative to the conventional geochemical procedure.