ARCHIVED REPORTS_XR0002666 - Laserfiche WebLink

Home Browse Search

• 9 296 HYDROCARBON CONTAMINATED SOILS AND GROUNDWATER GROUNDWATER REMEDIATION LIMITS 297 observed concentrations of benzene in a recovery well If the remedial objectives • = raw data used in regression were based on some other contaminant, this method still could be applied 5500 ° raw dolo not used in regression Data from a recovery well were utilized principally because it is the most fre- ht = asymptote quently sampled well at many sites undergoing remediation Extrapolation of trends ° in recovery wells may not correctly predict the time to achieve acceptable quality ° prediction interval in all monitoring wefts, considering such factors as well placement, screened in- CL 00°° tervals, and(lie horizontal and vertical distribution of contamination in the aquifer Nonetheless, the recovery well is the focal point of the pump-and-treat technology, o and its potential limitations were evaluated statistically in this study a 900 ° The seven characteristic data sets used for method evaluation are presented in T 1 the API report 2 An eighth data set was submitted subsequent to that work and Zhas also been used for subsequent evaluation of the PC software The data records 600 Q 0 range in duration from 9 to 22 months All data sets exhibit an asymptotic region, 1 with average concentrations of benzene at the asymptote ranging from 2 to 300 - • ° t parts per billion (ppb) 300 • " • Two alternative statistical techniques were evaluated in Phase i The first tech- -------------- ----�-- ••00 % ti " roque was based on regressing the concentration, C, versus time t The second technique was based on regressing the natural logarithm of concentration, 1nC, 120 240 360 a Po 600 versus t For both techniques,an asymptotic condition is indicated if die slope(dC/dt) TIME (DAYS) is not signifit,antiy different from zero and if the 95% upper and lower confidence Figure 17 1 Linear asymptotic regression results for one of seven data sets used for bounds on the slope are both near zero The logarithmic regression is based on method evaluation the full data set, whereas the linear technique includes an objective identification of the asymptotic portion of the data set This step for the linear case was neces- sary because, with a data set of the type illustrated in Figure 17 1, a flat, straight The apparent leveling off of groundwater quality is indicative of a technologi- line can only be fit through data after the initial period of concentration decline cal limitation on the effectiveness of the in-place remedial technology It may F These two alternatives were selected because it was unclear at the initiation be infeasible to further improve groundwater quality via the technology in use ' of the study whether the apparent asymptote was actually greater than zero Several at the site mechanisms that have been documented to affect the fate and persistence of In 1988, the American Petroleum Institute(API) funded Environmental Science hydrocarbons in soil and groundwater are expected to exhibit first order kinetics 3 & Engineering, Inc , to develop and evaluate alternative statistical procedures If the overall restoration process during remedial action were a first-order removal that Lan be used to document and quantify the asymptotic condition Preliminary process (I e , contaminant is being eliminated from the system in direct propor- results of this study were published in 1989, and an API report documenting the tion to its concentration), then the logarithmic regression would closely fit the method evaluation was published in 1991 'z The second, ongoing, phase of that data A first-order process is asymptotic, but the asymptotic concentration limit study is to develop personal computer(PC)software to support wider dissemina- should be at zero At study initiation it was unclear whether the appearance of tion of the statistical procedures This chapter summarizes the previous work and a nonzero asymptote simply resulted from a small trend coupled with large data presents additional results and hypothetical applications based on the to-be-released variability The results of the method evaluation indicate that this is not the case, PC software and that a nonzero asymptote exists The alternative linear regression technique is expected to fit the data better if METHODS- PHASE 1 a valid nonzero asymptote exists This would also tend to indicate that fundamentally distinct processes control the initial rapid reduction as contrasted with the subse- quent asymptotic interval For example, the initial reduction may result from the Groundwater quality data from seven sites undergoing active remediation were rapid removal of contamination from the aquifer by the pump-and-treat technolo- selected to test alternative statistical procedures These data sets were selected gy,whereas the asymptotic period could be attributable to the slow release of con- front approximately 20 site data sets provided by API The data sets represent i taminant from soils, a process that is not accelerated by the pump-anId-treat system