![Hier klicken, um den Treffer aus der Auswahl zu entfernen](images/unchecked.gif) |
Titel |
A Study of the Groundwater Level Spatial Variability in the Messara Valley of Crete |
VerfasserIn |
E. A. Varouchakis, D. T. Hristopulos, G. P. Karatzas |
Konferenz |
EGU General Assembly 2009
|
Medientyp |
Artikel
|
Sprache |
Englisch
|
Digitales Dokument |
PDF |
Erschienen |
In: GRA - Volume 11 (2009) |
Datensatznummer |
250027407
|
|
|
|
Zusammenfassung |
The island of Crete (Greece) has a dry sub-humid climate and marginal groundwater
resources, which are extensively used for agricultural activities and human consumption.
The Messara valley is located in the south of the Heraklion prefecture, it covers
an area of 398 km2, and it is the largest and most productive valley of the island.
Over-exploitation during the past thirty (30) years has led to a dramatic decrease of thirty
five (35) meters in the groundwater level. Possible future climatic changes in the
Mediterranean region, potential desertification, population increase, and extensive
agricultural activity generate concern over the sustainability of the water resources
of the area. The accurate estimation of the water table depth is important for an
integrated groundwater resource management plan. This study focuses on the Mires
basin of the Messara valley for reasons of hydro-geological data availability and
geological homogeneity. The research goal is to model and map the spatial variability of
the basin’s groundwater level accurately. The data used in this study consist of
seventy (70) piezometric head measurements for the hydrological year 2001-2002.
These are unevenly distributed and mostly concentrated along a temporary river
that crosses the basin. The range of piezometric heads varies from an extreme low
value of 9.4 meters above sea level (masl) to 62 masl, for the wet period of the year
(October to April). An initial goal of the study is to develop spatial models for
the accurate generation of static maps of groundwater level. At a second stage,
these maps should extend the models to dynamic (space-time) situations for the
prediction of future water levels. Preliminary data analysis shows that the piezometric
head variations are not normally distributed. Several methods including Box-Cox
transformation and a modified version of it, transgaussian Kriging, and Gaussian
anamorphosis have been used to obtain a spatial model for the piezometric head. A trend
model was constructed that accounted for the distance of the wells from the river
bed. The spatial dependence of the fluctuations was studied by fitting isotropic
and anisotropic empirical variograms with classical models, the Matérn model
and the Spartan variogram family (Hristopulos, 2003; Hristopoulos and Elogne,
2007).
The most accurate results, mean absolute prediction error of 4.57 masl, were obtained
using the modified Box-Cox transform of the original data. The exponential and the isotropic
Spartan variograms provided the best fits to the experimental variogram. Using Ordinary
Kriging with either variogram function gave a mean absolute estimation error of 4.57 masl
based on leave-one-out cross validation. The bias error of the predictions was calculated
equal to -0.38 masl and the correlation coefficient of the predictions with respect of the
original data equal to 0.8. The estimates located on the borders of the study domain presented
a higher prediction error that varies from 8 to 14 masl due to the limited number of neighbor
data. The maximum estimation error, observed at the extreme low value calculation, was 23
masl.
The method of locally weighted regression (LWR), (NIST/SEMATECH 2009) was also
investigated as an alternative approach for spatial modeling. The trend calculated from a
second order LWR method showed a remarkable fit to the original data marked by a mean
absolute estimation error of 4.4 masl. The bias prediction error was calculated equal to -0.16
masl and the correlation coefficient between predicted and original data equal to
0.88 masl. Higher estimation errors were found at the same locations and vary
within the same range. The extreme low value calculation error has improved to 21
masl.
Plans for future research include the incorporation of spatial anisotropy in the kriging
algorithm, the investigation of kernel functions other than the tricube in LWR, as well as
the use of locally adapted bandwidth values. Furthermore, pumping rates for fifty
eight (58) of the seventy (70) wells are available display a correlation coefficient of
-0.6 with the respective ground water levels. A Digital Elevation Model (DEM) of
the area will provide additional information about the unsampled locations of the
basin. The pumping rates and the DEM will be used as secondary information in
a co-kriging approach, leading to more accurate estimation of the basin’s water
table.
NIST/SEMATECH e-Handbook of Statitical Methods, http://www.itl.nist.gov/div898/handbook/,
12/01/09.
D.T. Hristopulos, “Spartan Gibbs random field models for geostatistical applications,”
SIAM J. Scient. Comput., vol. 24, no. 6, pp. 2125–2162, 2003
D.T. Hristopulos and S. Elogne, “Analytic properties and covariance functions for a new
class of generalized Gibbs random fields,” IEEE TRANSACTIONS ON INFORMATION
THEORY, vol. 53, no 12, pp. 4667-4679, 2007 |
|
|
|
|
|