![Hier klicken, um den Treffer aus der Auswahl zu entfernen](images/unchecked.gif) |
Titel |
Using visible and near-infrared diffuse reflectance spectroscopy for predicting soil properties based on regression with peaks parameters as derived from continuum-removed spectra |
VerfasserIn |
Radim Vasat, Ales Klement, Ondrej Jaksik, Radka Kodešová, Ondřej Drábek, Lubos Boruvka |
Konferenz |
EGU General Assembly 2014
|
Medientyp |
Artikel
|
Sprache |
Englisch
|
Digitales Dokument |
PDF |
Erschienen |
In: GRA - Volume 16 (2014) |
Datensatznummer |
250092133
|
Publikation (Nr.) |
EGU/EGU2014-6459.pdf |
|
|
|
Zusammenfassung |
Visible and near-infrared diffuse reflectance spectroscopy (VNIR-DRS) provides a rapid and
inexpensive tool for simultaneous prediction of a variety of soil properties. Usually, some
sophisticated multivariate mathematical or statistical methods are employed in order to
extract the required information from the raw spectra measurement. For this purpose
especially the Partial least squares regression (PLSR) and Support vector machines (SVM)
are the most frequently used. These methods generally benefit from the complexity
with which the soil spectra are treated. But it is interesting that also techniques
that focus only on a single spectral feature, such as a simple linear regression with
selected continuum-removed spectra (CRS) characteristic (e.g. peak depth), can
often provide competitive results. Therefore, we decided to enhance the potential of
CRS taking into account all possible CRS peak parameters (area, width and depth)
and develop a comprehensive methodology based on multiple linear regression
approach. The eight considered soil properties were oxidizable carbon content (Cox),
exchangeable (pHex) and active soil pH (pHa), particle and bulk density, CaCO3 content,
crystalline and amorphous (Fed) and amorphous Fe (Feox) forms. In four cases
(pHa, bulk density, Fed and Feox), of which two (Fed and Feox) were predicted
reliably accurately (0.50 < R2cv < 0.80) and the other two (pHa and bulk density)
only poorly (R2cv < 0.50), we obtained slightly better results than with PLSR
and SVM. In one case (pHex) we achieved a significantly higher, although just
reliable, accuracy (R2cv = 0.601) than with PLSR and SVM (R2cv = 0.448 and
0.442, resp.). But most interestingly, in the case of particle density, the presented
approach outperformed the PLSR and SVM dramatically offering a fairly accurate
prediction (R2cv = 0.827) against two failures (R2cv = 0.034 and 0.121 for PLSR
and SVM, resp.). In last two cases (Cox and CaCO3) a slightly worse results were
achieved then with PLSR and SVM with overall fairly accurate prediction (R2cv >
0.80).
Acknowledgment: Authors acknowledge the financial support of the Ministry of
Agriculture of the Czech Republic (grant No. QJ1230319). |
|
|
|
|
|