HR: 0800h
AN: H21B-1347 [Abstracts]
TI: Characterizing Soil Hydraulic Parameter Heterogeneity Using Cokriging and Artificial Neural Network: A
Framework of Integrating Hard and Soft Data
AU: * Ye, M
EM: Ming.Ye@dri.edu
AF: Division of Hydrologic Sciences, Desert Research Institute, 755 E. Flamingo Road, Las Vegas, NV 89119
United States
AU: Schaap, M G
EM: mschaap@ucr.edu
AF: University of California, Riverside, CA 92507
United States
AU: Khaleel, R
EM: Raziuddin_Khaleel@rl.gov
AF: Fluor Government Group, P. O. Box 1050, Richland, WA 99352
United States
AU: Zhu, J
EM: Jianting.Zhu@dri.edu
AF: Division of Hydrologic Sciences, Desert Research Institute, 755 E. Flamingo Road, Las Vegas, NV 89119
United States
AB:
Characterization of the heterogeneity of hydraulic parameters that control transport processes in the vadose zone is always
difficult due to prohibitive investments involved with direct parameter measurements (so-called `hard' data). `Soft' data
such as moisture content (θ) and results derived from geophysical measurements, however, carry significant information
about media heterogeneity and should be included in site characterization, where possible. We developed a method to
incorporate both `hard' and `soft' data using cokriging and artificial neural network (ANN) analyses to generate 3D spatially
correlated structures of hydraulic parameters. This method was applied to a field injection experiment carried out in 2000
at the `Sisson and Lu' injection site at the U.S. Department of Energy's Hanford Site, WA. Available data included limited
measurements of soil hydraulic parameters (i.e., saturated hydraulic conductivity and van Genuchten parameters, totaling 70
datasets) and soil characterization data (bulk density and percentages of gravel, coarse and fine sand, silt, and clay). A 3D
initial θ field reflecting the geologic layering was available at 32 observation wells (1344 locations). We used
variograms and cross-variograms to investigate the spatial correlation and cross-correlation of the initial θ
measurements and soil characterization data variables. We used ANN-based pedotransfer functions to map soil characterization
data to hydraulic parameters. Using initial θ as a secondary variable, we used a cokriging scheme to estimate 3D
heterogeneous fields of the primary variables, the soil characterization data and, subsequently, 3D fields of the hydraulic
parameters with the pedotransfer functions. These hydraulic parameter fields were then used to simulate the field injection
experiment. The spatial moments of the measured and simulated θ were compared to evaluate the robustness of the
developed method. The θ profiles at observation wells were investigated to explore the effect of soil hydraulic
parameter heterogeneity on the movement and distribution of moisture due to injected fluid.
DE: 1847 Modeling
DE: 1866 Soil moisture
DE: 1875 Vadose zone
SC: Hydrology [H]
MN: Fall Meeting 2005