Simon Schreiner (2013)
Flächendeckende Bestimmung von Elementkonzentrationen an Bodenprofilen durch abbildende Laborspektroskopie
(Wissenschaftliche Publikation der Bachelorarbeit aus dem Fach Umweltgeowissenschaften, Schwerpunkt Fernerkundung)
Visible/near infrared (VNIR, 400 nm to 1000 nm wavelength) and shortwave infrared (SWIR, 1000 nm to 2500 nm wavelength) laboratory imaging spectroscopy with spatial resolutions of 63 μ m and 250 μ m, respectively, was used for mapping contents of C, N, Fe, Al, and Ca in soil profiles. Four soil cores of 30 cm to 50 cm length were collected at a Regosol and a Cambisol site and scanned hyperspectrally after drying. Small sam- ples (ROI; Regions of Interest) were taken from the cores and analysed chemically as references for regression analyses. Partial least-squares regression (PLSR) and multivariate adaptive regression splines (MARS) models between spectral information and elemental contents of reference samples were established for VNIR and SWIR data separately and for the combined datasets. The regression models were applied to the hyperspectral image data to create spatially explicit maps of elemental contents for the soil profiles. PLSR yielded slightly better regression accuracies than MARS, and PLSR maps were more plausible in visual inspection. The optimal spectral range for elemental mapping was inconsistent, but in most cases the addition of SWIR data to VNIR data resulted in improvements of the elemental content estimations.
