Introduction
The development and application of property-transfer functions is an important approach for predicting unsaturated hydraulic properties from more easily measured bulk properties. At the Idaho National Laboratory (INL), the unsaturated zone is comprised of thick basalt flow sequences interbedded with thinner sedimentary layers. Buried hazardous waste in the surficial soil is a possible source of contamination to the underlying Snake River Plain aquifer, which can be as deep as 200 m below land surface. Determining the unsaturated hydraulic properties of the sedimentary layers is one step in understanding water flow and solute transport processes through this complex unsaturated system.
Approach
This study uses multiple linear regression analysis to construct simple property-transfer functions for estimating the water retention curve for deep sediments at the INL. The regression models were developed using laboratory measurements on 109 sediment core samples collected at depths of 9 m to 175 m at two facilities within the southwestern portion of the INL, the Radioactive Waste Management Complex (RWMC) and the Vadose Zone Research Park (VZRP). These data included water retention measurements, the curve fit parameters for which are the dependent variables of the property-transfer functions, and bulk properties (such as bulk density and various representations of the particle-size distribution), which are the potential independent variables.
The Rossi-Nimmo junction model was used to represent the water retention measurements. Three parameters define this retention curve model: 1) saturated water content (qsat), 2) a scaling parameter for matric pressure (yo), and 3) a curve shape parameter (l). The bulk property data and optimized hydraulic parameter values were used to develop a separate regression model for each parameter. The predicted parameters were then used to calculate the water retention curve from saturation to oven dryness. A selection process for the independent variables, referred to as “all possible subsets regression,” was used to determine the best predictive model for each hydraulic parameter.
Preliminary results
Preliminary regression results show that textural class percentages were consistently better able to explain the hydraulic parameters than were other potential representations of the particle-size distribution. The adjusted coefficient of determination (adjusted R2) for the best models, which consisted of some linear combination of textural class percentages and bulk density, ranged between 0.2 and 0.5 when all observations were included in the regression analyses. The residuals were close to normally distributed and were fairly homoscedastic when plotted versus the predicted dependent variable values. The low adjusted R2 values may indicate that the bulk property data used in calibrating the models are not sufficient to completely predict the hydraulic parameters or may indicate significant measurement errors in the dependent or independent variables. Other bulk property data not available for calibrating the property-transfer functions, such as mineralogy, specific surface areas, or adsorption capacities, might correlate more strongly with the hydraulic parameters, and thus may be useful in future regression analyses. The property-transfer functions from this study provide a basis for development of a theoretical model that relies on physical relationships between the pore-size distribution and the bulk properties of the media and that should be more universal in its application throughout the INL and other geographic locations.
Future work
An important next step is to develop a property-transfer model to predict unsaturated hydraulic conductivity as a function of water content (K(q)). This can be accomplished by (1) parameterizing measured K(q) curves using a simple power law function, or (2) combining the Rossi-Nimmo (1994) junction model with Mualem’s (1976) capillary bundle model. Multiple linear regression formulas oversimply and may misrepresent the relationships between hydraulic and bulk properties because they are based on statistical correlations rather than physical and theoretical relationships. By applying interpretation and refinement to the regression model based on physical considerations, a more theoretical property transfer model can be developed for stronger assurance of model reliability and probably greater simplicity. This type of model will be more adaptable in its application to other sites as well.
For more information contact: UZ Flow Webmaster
Last modified: Wed Sep 17 13:47:04 PDT 2003