Application of the Hydrocarbon Spill Screening Model to Field Sites James W. Weaver Proceedings of the American Society of Civil Engineers Conference on Non-aqueous Phase Liquids in the Subsurface Environment: Assessment and Remediation (To appear) November 12-14, 1996 Washington, D.C. 1 Weaver ------- Application of the Hydrocarbon Spill Screening Model to Field Sites James W. Weaver1 Abstract The Hydrocarbon Spill Screening Model, HSSM, was developed for estimating the impacts of petroleum hydrocarbon contamination on subsurface water resources. The model simulates the release of the hydrocarbon at the ground surface, formation of lens in the capillary fringe, dissolution of constituents of the gasoline, and trans- port to a receptor in the aquifer. Field data from two case histories were used to de- velop input parameter sets for HSSM. In one case there were aqueous concentration data from an extensive monitoring network. In the second case the monitoring net- work was small, but the date and volume of the release could be estimated. Both of these cases have features that are well suited for testing of the model. In both cases the model was able to reproduce the trends in the data set and the concentrations to within an order of magnitude. Introduction The Hydrocarbon Spill Screening Model (HSSM) was intended as a simplified model for estimating the impacts of petroleum hydrocarbons on subsurface water resources (Weaver etal., 1994 and Charbeneau etal., 1995). In this paper the model was applied to two sites with releases from leaking underground storage tanks. The data were drawn from State Agency case files and were not intended for research purposes. The objectives of the work were to determine if HSSM could reproduce the observed contaminant distributions and to demonstrate the effect of data gaps on model results. These applications demonstrate the effects of parameter uncertainty on model results, because in each case some information for running the model was not available. 1 National Risk Management Research Laboratory, United States Environmental Protection Agency, Ada, Oklahoma 74820 1 Weaver ------- Field Data Sets Contamination from underground storage tank releases is usually characterized by contaminant concentrations in the ground water and soils. Normally, data are collected for benzene, toluene, ethylbenzene and the xylenes (BTEX), and total pe- troleum hydrocarbons (TPH). Water samples give concentrations that in effect are related through time and space by the transport equation: dc rjR— = X/ -D\/ c - q-x/c - XrjRc + J(t) (1) where // is the porosity, R is the retardation factor, c is the contaminant concentra- tion in the ground water, q is darcy velocity, D is the dispersion constant, A is a first order decay constant, J(t) is the amount of mass per unit volume of aquifer added per unit time. Each term on the right hand side of equation 1 can cause the con- centration to change. Apparent dilution along the length of contaminant plumes is characterized by the dispersion constant (term 1: y • D y c), which is assumed to depend upon the seepage velocity and inherent dispersivity of the aquifer. Water flowing along divergent streamlines can cause reduction in concentration (term 2: q ¦ y c). In equation 1 biodegradation is assumed to follow first order decay (term 3: Xr/Rc). The last term, J(t), is the source/sink term which can include losses due to hydrolysis, extraction or volatilization, for example. It may also include gains due to loading from the contaminant source. For leaks from underground storage tanks the source of contamination is the re- leased hydrocarbon liquid. Normally the volume and timing of the release is un- known. The magnitude of the source term in equation 1, however, depends directly upon the volume, rate and timing of the release. Typical data from a site that re- flect the hydrocarbon liquid are free product levels observed in wells, and concen- trations from soil samples. Free product levels can not provide a reliable estimate of the release volume, because there is not a clear relationship between the free prod- uct levels in wells and the amount of free product in the formation (Kemblowski and Chiang, 1990). Free product recovery data can obviously give a lower bound on the release volume. Core extracts could provide more reliable estimates of the product volume, but they are only taken upon installation of wells and may not be analyzed over their full length. Hydrogeologic data normally consists of information on re- gional hydrology and geology from published sources, well logs, and measured or estimated hydraulic conductivities. Well logs define the vertical and spatial stratig- raphy. The Hydrocarbon Spill Screening Model The Hydrocarbon Spill Screening Model (HSSM) was intended as a screening model for estimating the impacts of hydrocarbon (or light nonaqueous phase liq- uid, LNAPL) releases to the subsurface (Weaver et al., 1994, and Charbeneau et al., 2 Weaver ------- 1995). The model consists of three modules that treat transport in the vadose zone, the formation and decay of an oil lens in the capillary fringe, and transport of solu- ble constituents of the LNAPL in the aquifer to receptor locations. The model uses semi-analytical solutions of the transport equations which include many of the im- portant physical and chemical processes. It does not include all processes which may be important, and because of the usage of semi-analytical solutions, it does not account directly for heterogeneity. There are 36 physical and chemical parameters required to run HSSM. Each of these has an impact on the model results. Depending on the specific model scenario, some of the parameters are far more important than others. Of the entire suite of pa- rameter values, only a small number of the most important parameters were varied to achieve the fits to the data described below (Table 2). Features of the data sets Both of the data sets simulated below were accepted by the State Agency that was responsible for managing the site and thus met the regulatory requirements for assessment of contamination. In neither of these cases was modeling of the spill considered an essential or integral assessment activity. Table 1 lists some features of each spill. The data used in this paper came from excerpts of state agency case files (Weaver et al., 1996, and State of Utah, 1996). Each data set has unique fea- tures that led to its inclusion in this study. Hagerman Avenue has a large number of monitoring wells and Mountain Fuels has estimates of the date and volume of re- lease. For each case model results for benzene are discussed below. These results are intended to illustrate certain features of the cases and the ability of the model to duplicate them. Mountain Fuels, Salt Lake City, Utah At the Mountain Fuels site in Salt Lake City, Utah, approximately 7500 gallons of gasoline leaked from a tank that was punctured in 1979 (State of Utah, 1996). Four monitoring wells were sampled from June 1991 to December 1994 to charac- terize the resulting aquifer contamination (Figure 1). Thus these data date from 12 to 15 years after the release. Concentrations in two of the four monitoring wells were always below the detection limits. One of the remaining wells is clearly directly in the path of the contaminant plume (MW-2). The fourth well (MW-1) appears to be on the edge of the plume, both because of its geographic location and its concentra- tion history. Parameter values given in Table 2 were used in the simulation. The hydraulic conductivity and gradient were estimated from the field data. The other parame- ters listed in the table were estimates that resulted in order of magnitude matches to the observed data. At the furthest down gradient well, MW-2, the concentrations 3 Weaver ------- Item Hagerman Ave. Mountain Fuels E. Patchogue, Salt Lake City, New York Utah Release Date Unknown Known Release Volume Unknown est. 7500 gal Composition of Fuel Unknown Unknown Mass in the Ground Water Est. Unknown Cores Analyzed 30 - Monitor Wells 48 O) 4 Vertical Plume Definition Yes No Sample Rounds 3 8 W Data Points per Sample Round 210 4 Slug Tests 13 Unknown Pump Tests 1 0 26 multilevel samplers and 22 screened wells O 6 samples were taken from MW-2 number not given in case file excerpt Table 1: General features of the two data sets decline steadily. This implies that for all the simulations and the field data that the peak concentration has already passed this receptor. Table 3 lists the simulated peak concentrations, maximum mass fluxes to the aquifer, and times of their occurrence. With increasing conductivity, the benzene is released sooner at a higher maximum rate. The peak concentrations in the aquifer occur sooner, but with higher veloci- ties the peak concentrations may decline due to the effect of increasing dispersion (which is proportional to the velocity). Item Hagerman Ave. Mountain Fuels E. Patchogue, Salt Lake City New York Utah Hydraulic Conductivity 43, 94, 149 m/d 0.75, 1.5, 3.0 m/d Hydraulic gradient 0.0013 0.02 Water Table Fluctuation 0.1 m 0.1 m Initial Benzene Concentration 8200 mg/L 4100 mg/L Half Life 2250 days 69.3,30 days Longitudinal Dispersivity 10,15,20 m 15m Table 2: Main Adjusted Parameters for the Cases The simulated concentration distributions at MW-2 shown in Figure 2 decline 4 Weaver ------- Figure 1: Mountain Fuels, Salt Lake City, Utah uniformly. The concentrations are seen to be dependent upon the hydraulic conduc- tivity because of their dependence upon the release rate of benzene from the gasoline and advective-dispersive transport in the aquifer. The value of concentration can also be adjusted by changing the degradation rate constant as noted in the figure. Both the simulation with a Ks = 1.5 m/d and half life of 30 days and that with Ks = 0.75 m/d and half life of 69.3 days (loss rate of 0.01% per day) give a similar match to the field data. The breaks in the concentration curves for Ks = 1.5 m/d and Ks = 3.0 m/d occurring at about January 1, 1990 and January 1, 1996, respectively, are caused by the ending criterion used in HSSM (Weaver et al., 1994). Observation well MW-1 was located within the oil lens generated by HSSM. Conductivity Time Maximum Time Maximum Concentration Mass flux m/d d //g/L d kg/d 0.75 968 0.1281 731 0.0066 1.5 839 4.7276 502 0.0262 3.0 390 2.9253 306 0.0742 Table 3: Mountain Fuels Simulated Peak Concentrations, MW-2 5 Weaver ------- CT3 O ~CX3 CD CJ cz o 250 200 1 50 1 00 50 0 Ks = 0.75 m/d, h.l. = 69.3 m/d Ks = 1 .5 m/d, h.l. = 69.3 d Ks = 3.0 m/d, h.l. = 69.3 days Ks = 1 .5 m/d, h.l. = 30 days /\ Field Data MW-2 cvj co lo co r-— cn cn cn en cn cn cn cn cn cn cn cn Date Figure 2: Model results and Field data for MW-2 Mountain Fuels, Salt Lake City, Utah Therefore no contaminant concentrations were calculated for this location by the aquifer module. The concentrations in the ground water below the lens, however, give an indication of concentrations that would be observed in MW-1. Figure 3 shows a comparison of the model and the data for this well. Three values of hydraulic con- ductivity were used in the simulations: the reported average of 1.5 m/d, and half and twice this value. As the hydraulic conductivity increased, the rate of release of benzene to the aquifer increased. Thus the benzene concentrations for the higher conductivity simulations decrease more rapidly than for the low conductivity cases. The case with the lowest estimate of conductivity (0.75 m/d) falls through the scatter of the field data, but the variation in concentration observed in the monitor well can- not be matched by the model. Both MW-1 and MW-2 are best fit by the simulation with conductivity of 0.75 m/d and half life of 69.3 days. Hagerman Avenue, East Patchogue, New York The gasoline spill at East Patchogue, New York is described in detail in another paper in this proceedings (Weaver et al, 1996). In that paper, the extensive data set was used to estimate the ground water flow velocity, the volume of gasoline released and the mass of BTEX and methyl /er/-buty\ ether, MTBE, released to the aquifer. The data from the site suggest a release volume of at least 13,200 gallons, which contains 420 kg of benzene (Weaver et al., 1996). The release likely occurred as a 6 Weaver ------- Figure 3: Model results and Field data for MW-1 Mountain Fuels, Salt Lake City, Utah series of continuing leaks over several years. The tanks at the service station were re- moved in 1988, so any releases ended that year. Since some fraction of the gasoline contained MTBE, that gasoline was released after 1979 when MTBE was approved for use as an octane enhancer. The MTBE in the aquifer traveled from the source zone to its center of mass in 1994 and 1995 in 16 years or less. Using the centers- of-mass calculated by Weaver et al. (1996), the MTBE plume traveled at the rates listed in Table 4. The rates show remarkable consistency suggesting that the aver- age transport time, averaged over the duration of the contamination event, is nearly constant for distances between 1387 m and 1583 m from the suspect source. The rate would have been 0.65 m/d if the entire release occurred on December 31, 1988 and the 0.25 m/d if the release began on January 1, 1979. The ground water flow velocity influences advective transport in the aquifer and the rate of release of mass from the oil lens. These two processes must both be con- sistent with the data from the field for the simulation to be appropriate. Parameter values for the simulation are listed in Table 2. In separate simulations the hydraulic conductivity was taken as the average and the average plus or minus one standard deviation of values determined from slug tests performed at the site. The water ta- ble fluctuation was determined from wells near in the source zone. The other para- meters listed in Table 2 were taken as reasonable estimates that resulted in order of magnitude matches to the observed data. The critical parameters for simulating the Hagerman Avenue site were the rate 7 Weaver ------- Sample Date Distance Estimated Velocities (m/d) Round Release Date Scenarios Late Early days since rate days since rate m Dec 31, 1988 m/d Jan 1, 1979 m/d 1 Dec 16, 1994 1387 2176.25 0.64 5828.75 0.24 2 April 16, 1995 1557 2297.5 0.68 5950.0 0.26 3 Oct 17, 1995 1583 2481.5 0.64 6134.0 0.26 Table 4: Rates of Movement of the MTBE Plume and duration of the release, ground water flow velocity, degradation rate constant, and water table fluctuation. The ground water velocity was selected to be 0.40 m/d which is within the range given in Table 4. The release was assumed to occur con- tinuously from January 1, 1979 to December 31, 1988 at a rate that resulted in 50 m3 (13200 gallons) of gasoline in the aquifer. This release scenario was used be- cause the true rates and timings of releases are unknown and because it was found necessary to generate relatively low concentrations in the ground water to match the monitor well data. From sample round one and two water level data near the source, the water table fluctuation was approximately 0.33 ft (0.10 m). Figure 4 shows a comparison of the vertically averaged measured concentration in MW-13, MW-12 and MW-1,4 with the HSSM model results (see Weaver et al., 1996, Figure 1). The highest concentrations were simulated at the up gradient monitoring well (MW-13). The observed concentrations at this well (open triangles on Figure 4) decline over the sampling period indicating that the peak concentration has passed this well. MW-12 shows the highest observed concentrations (open squares) that decline with time, in contrast to the model result that indicates relatively constant concentration over this period. Being down gradient of MW-13 the concentrations would be expected to decrease if advective-dispersive transport is occurring in a uni- form aquifer. Figure 3 of Weaver et al. (1996) shows that the benzene distribution in the aquifer is unsmooth, suggesting that there is preferential flow through certain re- gions. The observed concentration in MW-1,4 increased from sample rounds one to two (open circles). This behavior is not matched by the model which indicates that the peak concentration at this receptor has yet to arrive. This result is consistent with the field data (Weaver et al., 1996, Figure 3) which show that the benzene plume is beginning to pass MW-1,4 during the sampling interval. Despite variability, each of the model results was within an order of magnitude of the field data. Figure 5 shows the effect of hydraulic conductivity variation on model results at MW-13. As expected, the benzene arrives sooner and at higher concentration as the ground water flow velocity increases. In sample round one the field data fall near 8 Weaver ------- CD O ~ccS 1500 1 000 A ~ O Model Result MW-1 3 Model Result MW-1 2 Model Result MW-1 ,4 Field Data MW-1 3 Field Data MW-1 2 Field Data MW-1 ,4 ~ CD CJ a o O 500 LO CO CD Date Figure 4: Model results and field data Hagerman Avenue, East Patchogue, New York the average curve (Ks = 94 m/d) while later data fall near the low conductivity result (Ks = 43 m/d). These results suggest that with a steeper front one curve may fit data from all three sample rounds. Figure 6 shows the effect of varying dispersivity on the results for MW-13. Decreasing dispersivity over the range shown here sharpens the front somewhat, but does not force an exact match to the data. The parameter values could be further adjusted to try to match the field data at this well, but other wells would likely remained unmatched as shown in Figure 4. Conclusions Uncertainty exists in model parameters for both data sets. In a general sense HSSM was able to reproduce the trends in the monitoring wells. These trends are largely related to the hydraulics of the system, which apparently are relatively well matched by the model. At the Mountain Fuels site, the concentrations in MW-2 decline continually, which matches the model result that indicates that the peak con- centration already passed the well. At Hagerman Avenue the concentration peak had not yet reached MW-1,4 as noted on Figure 3 of Weaver et al. (1996) which was re- flected in the simulation results. For both cases the data were collected more than ten years after the release may have occurred so that the measured concentrations are below a few hundred micrograms per liter. The higher concentration data which could provide a better test of the model are not available. As noted at Hagerman 9 Weaver ------- 1 000 CD =s_ d o ~cd CD o cz o o 800 600 400 200 a kb = 43 m/d ¦ Ks = 96 m/d • Ks = 1 49 m/d A Field Data MW-1 3 *—¦— i ¦ Tlrfc CO CT) cn Date Figure 5: Effects of hydraulic conductivity variation at MW-13, Hagerman Avenue, East Patchogue, New York 1 000 CD o "cd CD o dZ o O 800 600 400 200 = 1 0 m ocL = 1 5 m ocL = 20 m Field Data MW-1 3 Date Figure 6: Effects of dispersivity variation at MW-13, Hagerman Avenue, East Patchogue, New York 10 Weaver ------- Avenue, heterogeneity plays an important role in determining contaminant concen- trations. Since HSSM cannot include heterogeneity, a possible use for the model in some situations is to use its capability for generating a mass input function for the aquifer, but then to simulate the aquifer with a numerical solute transport model. Relatively close approximations of the concentrations could be obtained for wells at these sites. In each case it was necessary to include degradation of benzene to achieve an order of magnitude estimate of concentration. This reflects the common occurrence of benzene degradation and the power of the decay constant in reducing concentrations. Where there was unsmooth variation in concentration as at Hager- man Avenue and MW-1 of Mountain Fuels, the model did not capture the fluctuation and is not capable of doing so. Hagerman Avenue was simulated with relatively low values of dispersivity (Gelhar et al., 1992) and degradation, while Mountain Fuels was simulated with relatively high values of both parameters. These suggest, tenta- tively, that dispersion and degradation are less important at Hagerman Avenue than Mountain Fuels. The examples presented in this paper showed that parameter values could be se- lected from field data sets so that HSSM matched the data to within an order-of- magnitude for Hagerman Avenue and Mountain Fuels. The model results may or may not be predictive of future conditions at the sites because of parameter uncer- tainty, model assumptions, and that the values used were fitting parameters. The two sites selected for this study are unique as they have unusual amounts of data (Hager- man Avenue) or estimates of the time and volume of the release (Mountain Fuels). These cases represent unusual opportunities for testing the HSSM against field data. The cases, also, illustrate that for many fuel spills, site data is a limiting factor in test- ing or applying models. Acknowledgement The information in this document has been funded wholly or in part by the United States Environmental Protection Agency. It has been subjected to Agency review and approved for publication. Mention of trade names or commercial products does not constitute endorsement or recommendation for use. The author thanks Joseph Haas, New York State Department of Environmental Conservation for providing the E. Patchogue data set; and Robin Jenkins, Utah Department of Environmental Qual- ity for providing the Salt Lake City, Utah data set. References [1] R. J. Charbeneau, J. W. Weaver, B. K. Lien, 1995, The Hydrocarbon Spill Screening Model (HSSM) Volume 2: Theoretical Background and Source Codes, US EPA, EPA/600/R-94/039b. 11 Weaver ------- [2] L. W. Gelhar, C. Welty, K. R. Rehfeldt, 1992, A critical review of data on field- scale dispersion in aquifers Water Resources Research, 28(7), 1955-1974. [3] M. W. Kemblowski and C. Y. Chiang, 1990, Hydrocarbon thickness fluctua- tions in monitoring wells, Ground Water, 28(2), 244-252. [4] State of Utah, 1996, Unpublished case file. [5] J. W. Weaver, R. J. Charbeneau, J. D. Tauxe, B. K. Lien and J. B. Provost, 1994, The Hydrocarbon Spill Screening Model (HSSM) Volume 1: User's Guide, US EPA, EPA/600/R-94/039a. [6] J. W. Weaver, J. E. Haas, J. T. Wilson, 1996 Analysis of the Gasoline Spill at East Patchogue, New York, Proceedings of the Conference on Non-aqueous Phase Liquids in the Subsurface Environment: Assessment and Remediation, American Society of Civil Engineers, November 14-16, Washington, D.C. 12 Weaver ------- |