|
Titel |
Risk Assessment of Sediment Pollution Using Geostatistical Simulations |
VerfasserIn |
J. Golay, M. Kanevski |
Konferenz |
EGU General Assembly 2012
|
Medientyp |
Artikel
|
Sprache |
Englisch
|
Digitales Dokument |
PDF |
Erschienen |
In: GRA - Volume 14 (2012) |
Datensatznummer |
250064188
|
|
|
|
Zusammenfassung |
Environmental monitoring networks (EMN) discreetly measure the intensities of continuous
phenomena (e.g. pollution, temperature, etc.). Spatial prediction models, like kriging, are then
used for modeling. But, they give rise to smooth representations of phenomena which leads
to overestimations or underestimations of extreme values. Moreover, they do not
reproduce the spatial variability of the original data and the corresponding uncertainties.
When dealing with risk assessment, this is unacceptable, since extreme values must
be retrieved and probabilities of exceeding given thresholds must be computed
[Kanevski et al., 2009]. In order to overcome these obstacles, geostatistics provides
another approach: conditional stochastic simulations. Here, the basic idea is to
generate multiple estimates of variable values (e.g. pollution concentration) at every
location of interest which are calculated as stochastic realizations of an unknown
random function (see, for example, [Kanevski, 2008], where both theoretical concepts
and real data case studies are presented in detail). Many algorithms implement
this approach. The most widely used in spatial modeling are sequential Gaussian
simulations/cosimulations, sequential indicator simulations/cosimulations and direct
simulations.
In the present study, several algorithms of geostatistical conditional simulations were
applied on real data collected from Lake Geneva. The main objectives were to compare their
effectiveness in reproducing global statistics (histograms, variograms) and the way they
characterize the variability and uncertainty of the contamination patterns. The dataset is
composed of 200 measurements of the contamination of the lake sediments by heavy
metals (i.e. Cadmium, Mercury, Zinc, Copper, Titanium and Chromium). The results
obtained show some differences highlighting that risk assessment can be influenced by
the algorithm it relies on. Moreover, hybrid models based on machine learning
algorithms and geostatistical simulations were applied to study the phenomena of spatial
nonstationarity.
References
Kanevski, M., Pozdnoukhov, A. and Timonin, V. (2009). Machine Learning for
Spatial Environmental Data: Theory, Applications and Software. Lausanne: EPFL
Press.
Kanevski, M. (Editor) (2008). Advanced Mapping of Environmental Data: Geostatistics,
Machine Learning and Bayesian Maximum Entropy. London / Hoboken: iSTE / Wiley. |
|
|
|
|
|