A field study was conducted to determine the applicability of multivariate
geostatistical methods to the problem of estimating and simulating pesticide
concentrations in groundwater from measured concentrations of nitrate and pesticide,
when pesticide is undersampled. Prior to this study, no published attempt had been made
to apply multivariate geostatistics to groundwater contamination.
The study was divided into two complementary aspects of geostatistics: estimation
and simulation. The use of kriging and cokriging to estimate nitrate and the herbicide
dimethyl tetrachloroterepthalate (DCPA) contaminant densities is described in Chapter I.
Measured concentrations of nitrate and the DCPA were obtained for 42 wells in a shallow
unconfined alluvial and basin-fill aquifer in a 16.5 kmĀ² agricultural area in eastern
Oregon. The correlation coefficient between log(nitrate) and log(DCPA) was 0.74.
Isotropic, spherical models were fitted to experimental direct- and cross-semivariograms
with correlation ranges and sliding neighborhoods of 4 km. The relative gain for
estimates obtained by cokriging ranged from 14 to 34%. Additional sample locations
were selected for nitrate and DCPA using the fictitious point method. A simple economic
analysis demonstrated that additional nitrate samples would be more beneficial in reducing
estimation variances than additional DCPA samples, unless the costs of nitrate and DCPA
analysis were identical.
These estimates are by definition, the Best Linear Unbiased Estimates (i.e., the
estimates with minimized estimation variance), however the requirement of minimized
variance smoothes the variability of contaminant values. The application of conditional
simulations to groundwater contamination is described in Chapter 11. Conditional
simulation allows the degree of fluctuation of nitrate and DCPA between sample points to
be assesed. With knowledge of both the 'best' estimates and the of the variability
between sample points, nitrate and DCPA groundwater contamination in the study area
can be characterized
Based on the semivariogram models found in Chapter I, univariate and
multivariate conditional simulations of nitrate and DCPA were generated using the turning
bands method and the kriging or cokriging system. Kriging was used to condition the
univariate simulations, while cokriging was used to cross-correlate and condition the
multivariate simulations. The mean of 25 conditional and coconditional simulations at 8
different locations in the study area were generated and compared to kriging and
cokriging estimates and 95% confidence intervals.
Both conditional and coconditional simulation of the DCPA and nitrate
contaminant densities showed large variations when values in different simulations were
compared. The fluctuation in values demonstrate the uncertainties in the contaminant
distributions when sample sizes are small. As a result of this unkown component,
simulated values vary widely. Coconditional simulation displayed the cross-correlation
imposed by using the cokriging system to condition the simulations. After 25
simulations, the mean remained unstable indicating that more simulations would be
required to enable comparisons with kriging and cokriging estimates. / Graduation date: 1989
Identifer | oai:union.ndltd.org:ORGSU/oai:ir.library.oregonstate.edu:1957/26873 |
Date | 23 May 1988 |
Creators | Smyth, Jeffrey D. |
Contributors | Istok, Jonathan D. |
Source Sets | Oregon State University |
Language | en_US |
Detected Language | English |
Type | Thesis/Dissertation |
Page generated in 0.0084 seconds