Solute Transport in Unsaturated Soil: Experimental Design, Parameter Estimation, and Model Discrimination

The objectives of this study were to: (i) examine the efficacy of two sampling techniques for characterizing solute transport under steady-state water flow, (ii) study the variation in transport model parameters with increasing depth of solute leaching, and (iii) perform model discrimination to examine the transport process operative within a field plot. Bromide, NO?, and Clwere applied sequentially to a plot instrumented with two sets of 12 solution samplers located at depths of 0.25 and 0.65 m. At the conclusion of the experiment we destructively sampled the entire 2.0 by 2.0 m plot to a depth of 2.0 m. Mass recovery by the solution samplers ranged from 63 to 83% for the three tracers, and recovery by soil excavation ranged from 96 to 105%. The mean solute velocity estimated with the solution sampler data was significantly less than that determined by soil excavation. Mean solute velocity determined from soil excavation implied an effective transport volume equal to 0.828, (where 8, is volumetric water content) for the three tracers. Solution samplers and soil excavation provided similar measures of vertical dispersion. Both sampling methods revealed a scale-dependent dispersion process in which the dispersivity increased linearly with mean residence time. The depth profiles for all three solutes were accurately described with a stochastic convective lognormal transfer function model (CLT) using the applied mass and two constant parameters (estimated from simultaneous fitting to the depth profiles). EN V I R O N M E N T A L L Y HARMFUL anthropogenic chemicals frequently enter natural ecosystems, either by accident or by accepted management practices. Increased public awareness and concern has led to an expanded regulatory effort aimed at providing accurate assessments of the environmental fate of these compounds under a wide variety of management and climatic conditions. To this end, numerous environmental fate and transport models have been developed (van Genuchten and Shouse, 1989). One principal limiting factor to model development and discrimination is the lack of experimental research that examines transport mechanisms of chemicals through unsaturated field soils (Dagan, 1986; Jury and Fliihler, 1992). As noted by Gelhar et al. (1992), considerable experimental evidence exists that supports the theory of a scale-dependent dispersion process in aquifers (Gelhar and Axness, 1983; Dagan, 1984, 1987; Sposito and Barry, 1987). This is typically shown by fitting the V (mean solute velocity) and D (dispersion coefficient) CDE parameters to tracer data sampled sequentially in time and then plotting the dispersivity, a (the ratio D/V), against time of sampling (and/or mean travel T.R. Ellsworth, Dep. of Natural Resources and Environmental Sciences, Univ. of Illinois, 1102 S. Goodwin Ave., Urbana, IL 61801; P.J. Shouse, J.A. Jobes, and J. Fargerlund, USDA-ARS, U.S. Salinity Lab., 450 Big Springs Rd., Riverside, CA 92507; and T.H. Skaggs, Dep. of Environmental Engineering, Centre for Water Research, Univ. of Western Australia, Nedlands, WA 6907, Australia. Received 27 Jan. 1995. *Corresponding author ([email protected]). Published in Soil Sci. Soc. Am. J. 60:397-407 (1996). Skaggs, J. A. Jobes, and J. Fargerlund distance), with a generally observed to increase with either of these. Theoretically, for unsaturated soil-water flow systems, scale-dependent dispersion would apply when the dominant mechanism of effective dispersion is the variation in local pore-water velocity arising from variability in the hydraulic conductivity of the soil (Russo, 1991; Russo and Dagan, 1991; Beven et al., 1993). No clear consensus exists, however, for scaledependent dispersion in solute transport experiments through unsaturated field soils (Porro et al, 1993; Bevan et al, 1993; Jury and Fliihler, 1992). The dispersion coefficient, D, has been found to vary with depth in many ways: linear increase, nonlinear increase, constant, decrease, and erratic fluctuations. In a laboratory experiment, Kahn and Jury (1990) measured the outflow concentrations of tracers from repacked and undisturbed soil columns of various lengths, flow rates, and column diameters. They found that for the repacked columns, D was constant. For undisturbed columns, D was constant for only the lowest flow rate, and at the higher flow rates D increased with column length. One of two large-scale (6-m-deep and 0.95m-diam.) column studies of solute transport through a homogeneous soil in New Mexico found that D tended to increase to a depth of 4.0 m (Wierenga and van Genuchten, 1989). However, D had an erratic relationship with depth in the second column study (Porro et al., 1993). Jaynes and Rice (1993) monitored solute transport in a heterogenous profile with solution samplers at multiple depths on a 37-m2 field plot under both drip and ponded irrigation water applications. Although quite erratic, a slight decrease in dispersivity with depth was observed under both water application methods. In a layered soil, Porro et al. (1993) showed that D was independent of depth as long as the layer thickness was small relative to the observation scale. At the field scale, Butters and Jury (1989) found that D increased to a depth of 13 m, except for a 40% decrease between 3.0 and 4.5 m. They attributed the temporary decrease to the influence of an increase in silt content between depths of 3 .O and 4.7 m. A series of studies of solute transport on a loamy sand soil (Hamlen and Kachanoski, 1992; van Wesenbeeck, 1993) in Canada used solution samplers along transects to monitor steadyflow solute transport at different fluxes. They found that the transect-scale travel time variance increased approximately as the square of the mean travel time, at least to a depth of 0.4 m, indicating that D increased linearly with travel time. In contrast, a constant D was found by Roth et al. Abbreviations: CLT, convective lognormal transfer function; Cl. confidence interval; pdf, probability density function; CDE, convection-dispersion equation; BTC, breakthrough curve; NAW, net applied water; MLE. maximum likelihood estimates; SSQ, least squares regression; MM, method of moments; RMSE, root mean square error.