 |
| Titel |
A stochastic-geometric model of soil variation in Pleistocene patterned ground |
| VerfasserIn |
Murray Lark, Eef Meerschman, Marc Van Meirvenne |
| Konferenz |
EGU General Assembly 2013
|
| Medientyp |
Artikel
|
| Sprache |
Englisch
|
| Digitales Dokument |
PDF |
| Erschienen |
In: GRA - Volume 15 (2013) |
| Datensatznummer |
250073607
|
|
|
|
|
|
| Zusammenfassung |
| In this paper we examine the spatial variability of soil in parent material with complex spatial
structure which arises from complex non-linear geomorphic processes. We show that this
variability can be better-modelled by a stochastic–geometric model than by a standard
Gaussian random field. The benefits of the new model are seen in the reproduction of features
of the target variable which influence processes like water movement and pollutant
dispersal.
Complex non-linear processes in the soil give rise to properties with non-Gaussian
distributions. Even under a transformation to approximate marginal normality, such variables
may have a more complex spatial structure than the Gaussian random field model of
geostatistics can accommodate. In particular the extent to which extreme values of the
variable are connected in spatially coherent regions may be misrepresented. As a result, for
example, geostatistical simulation generally fails to reproduce the pathways for preferential
flow in an environment where coarse infill of former fluvial channels or coarse alluvium of
braided streams creates pathways for rapid movement of water. Multiple point geostatistics
has been developed to deal with this problem. Multiple point methods proceed by sampling
from a set of training images which can be assumed to reproduce the non-Gaussian
behaviour of the target variable. The challenge is to identify appropriate sources of such
images.
In this paper we consider a mode of soil variation in which the soil varies continuously,
exhibiting short-range lateral trends induced by local effects of the factors of soil formation
which vary across the region of interest in an unpredictable way. The trends in soil variation
are therefore only apparent locally, and the soil variation at regional scale appears random.
We propose a stochastic–geometric model for this mode of soil variation called the
Continuous Local Trend (CLT) model. We consider a case study of soil formed in relict
patterned ground with pronounced lateral textural variations arising from the presence of
infilled ice-wedges of Pleistocene origin. We show how knowledge of the pedogenetic
processes in this environment, along with some simple descriptive statistics, can be used to
select and fit a CLT model for the apparent electrical conductivity (ECa) of the soil. We use
the model to simulate realizations of the CLT process, and compare these with
realizations of a fitted Gaussian random field. We show how statistics that summarize the
spatial coherence of regions with small values of ECa, which are expected to have
coarse texture and so larger saturated hydraulic conductivity, are better reproduced
by the CLT model than by the Gaussian random field. This suggests that the CLT
model could be used to generate an unlimited supply of training images to allow
multiple point geostatistical simulation or prediction of this or similar variables. |
|
|
|
|
|
|