![Hier klicken, um den Treffer aus der Auswahl zu entfernen](images/unchecked.gif) |
Titel |
Digital mapping of topsoil organic carbon in Grand-Duchy of Luxembourg using a compilation of legacy soil data |
VerfasserIn |
Antoine Stevens, Simone Marx, Bas van Wesemael |
Konferenz |
EGU General Assembly 2014
|
Medientyp |
Artikel
|
Sprache |
Englisch
|
Digitales Dokument |
PDF |
Erschienen |
In: GRA - Volume 16 (2014) |
Datensatznummer |
250095895
|
Publikation (Nr.) |
EGU/EGU2014-11371.pdf |
|
|
|
Zusammenfassung |
Providing localized predictions (with uncertainties) of soil properties is needed to assist soil
surveyors and land managers, and inform/assist the political debate with quantified estimates
of the status and change of the soil resource. Such maps can be produced with data
originating either from purpose-built soil monitoring networks (SMN) or from previous soil
measurements exercises such as soil testing for farmers by commercial/institutional soil
laboratories. Although SMN’s are likely to provide better and less biased estimates of soil
attributes because of their optimal sampling strategy, SMN’s are costly to establish, maintain
and re-visit. Data gathered from other sources, on the other hand, are often more
numerous which might favor greater accuracy in a geostatistical context. Another
advantage is that these data are often acquired continuously so that derived-maps can
be rapidly and easily improved or updated with upcoming data. In this study, we
produce a map of the topsoil Organic Carbon (OC) content of croplands, grasslands,
vineyards and forest land of the Grand-Duchy of Luxembourg using more than 2000
samples analyzed for OC (by dry combustion) in 2012-2013 by an accredited soil
laboratory in Luxembourg. To model OC content, this study relies on a set of spatial
covariates with a resolution of 90 m, including elevation and its derivatives, land
cover, soil texture, climate and livestock intensity. To avoid problems related to
co-linearity in the independent variables, the covariates were transformed using Principal
Component (PC) analysis, retaining PC components explaining more than 99% of the
variation. Different prediction models were developed for each of the four land cover
classes using either Generalized Additive Models (GAM) or Random Forest Kriging.
For cropland soils, the model is characterized by a R2 = 0.75 and RMSE = 4 g C
kg-1 and the variables used in the model are geographical coordinates and PC
components related to elevation and aspect. Models developed for the other land
covers were less successful (R2 < 0.3) with a large part of OC variation occurring at
small scales. Maps show overall a clear south-north gradient in OC in Luxembourg. |
|
|
|
|
|