Sensor-based soil monitoring, which uses infrared spectroscopic (IR)
approaches (Vis/NIRS, MIRS) and energy dispersive X-ray fluorescence
spectroscopy (XRF), can quantify soil properties that are essential to
derive and evaluate soil functions in high spatial and temporal
resolutions. However, collection and evaluation in laboratory-based and
portable sensor-based methods have not yet been optimised, making it
difficult to critically assess possibilities and limitations of these
methods.This proposal aims at optimising the IR and XRF data collection
for spectrally-active soil properties using laboratory-based and
portable methods for representative soil type complexes of selected soil
landscapes. The individual methods are optimised with regard to
measuring modes (diffuse reflectance infrared Fourier transform vs.
attenuated total reflection for MIRS, powder vs. pellets for XRF),
sample preparation (in-situ, dried, dried and sieved, dried and ground),
measuring conditions and required sample numbers. Specific depth
functions for the different soil properties will be defined in the
studied sample profiles.A systematic comparison of calibration
algorithms for the quantification of spectrally-active soil properties
by Vis/NIRS, MIRS and XRF will be performed. Linear and non-linear
calibration approaches and spectral deconvolution will be employed as
functions of different sample sizes in multiple partitions. Spectral
variable selection and spiking using existing databases will be used.
The accuracy of soil profile characterisation by portable and
laboratory-based spectroscopic methods as well as a potential
transferability of calibrated estimation models will be systematically
investigated on individual profiles of the studied soil type complexes.
The analysis of site-specific relationships between spectrally-active
and spectrally-inactive or spectrally poorly-defined soil properties
will be performed both chemometrically and by regression analyses using
the spectrally-active soil properties as predictors. The synergistic IR
and XRF data analysis using the laboratory-based and mobile approaches
will be based on low-, mid- and high-level as well as hybrid data fusion
approaches with an emphasis on outer-product analysis, low-level fusion
with subsequent automatic wavelength range selection, mid-level fusion
using the principal components and high-level fusion using the
approaches of Bates-Granger and Granger-Ramanathan.This project provides
the tools with which to optimize the data collection of sensor-based
soil monitoring, to improve the determination accuracy of differently
complex soil properties by data fusion approaches, and to evaluate the
applicability of a spatial and profile-related sensor-based monitoring
for selected soil type complexes.