Introduction

This document serves two purposes: provide COPA with (1) clarification of the implications of ground-water tracing results in the basin that encompasses the Lemon Lane Landfill, and (2) information about ground-water tracing techniques using fluorescent dyes.

In order to describe ground-water flow and contamination migration at the Lemon Lane Landfill site several ground-water tracing experiments were conducted by injection of tracers (fluorescent dyes and other substances) into wells in the landfill, or into sinkholes in the periphery of the landfill. Sampling of ground water and surface water in the hinterland of the landfill was then done to detect or document recovery of those tracers. McCann and Krothe (1992) have summarized the results of the initial experiments, more recent experiments have not been published, information regarding the most recently performed tracing experiments has been obtained from Mike McCann of Westinghouse/CBS Corporation. For the initial tests where fluorescent dyes were used sampling was by use of packets of activated charcoal in springs, streams, and wells. Activated charcoal can adsorb the dye, which can be removed later for analysis. In two of the early tracing experiments, lithium bromide was used as the tracer, once by itself and another time in addition to fluorescent dyes.

Two fluorescent tracers that were used in these initial experiments Fluorescent Brightening Agent 28 (C.I. FBA 28) and sodium fluorescein (uranine) (C.I. Acid Yellow 73) are also used in industrial applications and are thus common contaminants in ground-water and surface water bodies. One tracer used in these initial experiments was C.l.Direct Yellow 96. This substance is not a dye per se but a dye intermediate compound. Intermediate compounds are used to manufacture dyes, some are fluorescent and useable as tracers in the same way that fluorescent dyes can be used, DY 96 is such a compound. Both FBA 28 and DY 96 are collected on unbleached cotton pads and are viewed under an ultra-violet light to detect their characteristic blue or yellow fluorescence (Alexander and Quinlan, 1992). Their sensitivity to being adsorbed onto cotton is assumed to be high and therefore many organizations used only cotton samplers and did not analyze water samples for these dyes. In these experiments DY 96 was never detected on cotton samplers, but was reported to be present in water samples.

The results of the tracing experiments were generally similar and consistent with results of contamination (PCB) signatures, exceptions were the results of the high-flow testing where three fluorescent dyes were used, uranine (AY 73), FBA 28, and DY 96.

The locations where these dyes were recovered included locations where the other tracers had been recovered and where PCBs were documented, but also locations where dyes had not been recovered previously and where there were never any documented detections of PCBs. Also a few locations had bromide present but no PCBs (N.B. bromide was used alone in the initial low-flow test and also used in addition to dyes in the high-flow test). Sites having the highest PCB concentrations were: Illinois Central Spring, Quarry Spring, the Slaughterhouse Spring Complex, and some detections at the east and west Stoney Springs, PH Road and Snoddy A Spring.

Lemon Lane Landfill - Hydrogeological Setting

Hypothetical Flow Routes from the Landfill Area

The results of ground-water tracing experiments and contaminant data from the landfill and its hinterland are a test of this flow and discharge hypothesis. There is additional evidence that the area of the landfill might drain to the these springs, that is potentiometric data from wells, however many assumptions are made when such maps are used that do not have to be made with ground-water tracing data. These maps show the slope in the water table depicted at both high and low flow suggest that the primary flow routes generally would be to the north and south, flow to the south would be expected to be discharged at the Illinois Central Spring, and flow to the north would be expected to discharge at the Slaughterhouse Spring complex as they, collectively have the largest discharges (see McCann and Krothe,1992). Potentiometric data are not necessarily unusable because of the limiting assumptions associated with them, Quinlan and Ray (1989) show that traced flow routes often coincide with troughs in such maps, but most site investigations do not have as much data as there were available in that case, and the position of the potentiometric troughs would be much more difficult to infer.

Initial Ground-Water Tracing Experiments, Lemon Lane Landfill

Both Urban and Cresmont A Springs have small discharges and form the headwaters of Cascade Branch that drains to the northeast into Griffy Creek.

During this low-flow test it was also observed that Quarry Spring and Illinois Central Spring are the same discharges, where Illinois Spring sinks and reemerges as the discharge at Quarry Spring (based upon temperature, electrical conductivity and discharge data - this was later supported by additional ground-water tracing [Mike McCann, personal communication, 1999]).

Using flow lines inferred from the potentiometric maps (McCann and Krothe, 1992) it also can be hypothesized that the probable flow routes from the landfill are primarily north and south. So this first tracing result was essentially consistent with the ground-water hydrology inferred at the site from the potentiometric data. The three sites where Br~ was detected, but could not be inferred to have been part of slug of tracer that passed were MW-8D, and Urban and Cresmont Springs, (with the exception of MW-8D) do not have such supporting evidence that they are hydraulically connected to the landfill. Since the tracer used was Br~ it must be evaluated whether at these three sites this Br~ was the same or not the same slug of Br~ that was injected. There are several compounds that could be source of Br~ in an industrial setting, one being ethylene dibromide which is in diesel fuel and as a spray for insect infestations in trees. Well MW-8D is a monitoring well at the landfill site and is relatively close to the tracer injection locations such that it could have detected Br~ from the experiment.

If the analytical (PCB) data are evaluated for the three sites MW-8D, Urban Spring and Cresmont A Spring. Well MW-8D and Urban Spring has shown some PCBs detections but both it and Cresmont A springs are small discharges and are likely to have others sources for PCBs as well as tracers, so based upon the detections of both PCBs and Br~ it cannot be said that they are hydraulically connected to the landfill without some uncertainty. In any industrial setting low-concentrations of PCBs should be viewed cautiously as there are PCBs in used oils and hydraulic oils that often spilled accidentally. The only way to test whether a contaminant is from a certain source is to repeat a tracing experiment from that source with different tracers and get the same results, and have the contaminant there too (providing the transport characteristics would theoretically allow that to occur). It would also be highly desirable to completely delineate the boundaries of the basins that feed the various springs, although this can be reasonably estimated when base-flow discharge is known with some precision (Quinlan et al., 1 995).

The second set of tracing experiments was done in high-flow conditions in the summer of 1988 and used three different dyes (C.I. Acid Yellow 73, uranine; C.I. FBA 28; and C.I. DY 96) as well as more Br~ added to each injection. The results of this test were considerably different from the results of the low-flow experiment. The results also suggested, that in some cases, the DY 96 migrated further than the FBA 28 tracer and the C.I. AY 73 (uranine or sodium fluorescein) that were used at the same time. Also there were sites where two out of three of the dyes were detected and sites where only one dye was detected and other sites where all three dyes that were used were detected. This is rather strange as the fluorescence of this compound is greatly enhanced by its adsorption onto cotton and is much less fluorescent in water, if it had not been detected on cotton it should not have been detected in water. The following point should be added, even though no DY 96 was detected on cotton it was reported to be present in water samples that were archived and run about one year later (Mike McCann, Westinghouse/CBS personal communication, 1999). These results are quite puzzling as the fluorescence of DY 96 is much higher on cotton that it is in water samples.

The apparent recovery pattern of the C.I. DY 96 is radial from the landfill area. The recovery pattern of the C.I. FBA 28 is similar. The recovery pattern for uranine is slightly different in that less uranine was recovered west and southwest of the landfill area. This is problematic as this is the most conservative tracer that was used, and should have migrated the greatest distance. It is also well established that uranine dye is a much better tracing dye than is either DY 96 or FBA 28, and it is known to be among the most if not the most conservative dye there is. It should be pointed out that, when used as tracers, FBA 28 (and uranine) have many other potential sources, principally car radiators and domestic sewage, great care must accompany interpretation of results like these and design of the experiments must be quite rigorous.

One way to resolve such problems is to assume that there should be PCBs at any site where dye was recovered if it is hypothesized that contaminants migrated from the landfill to those locations. Reviewing the information from both the tracing results of both the low-flow and the high-flow tests and the contaminant results, PCBs were only detected at the following sites: Illinois Central Spring, Quarry Spring, and Stony East Spring, Stony West Spring, Snoddy Spring, PH Road Spring and Slaughterhouse Spring. It is significant that all these sites are, with regards to the site, along the steepest hydraulic gradients. It is therefore reasonable to assume that the PCBs that were detected at these locations might have the Lemon Lane Landfill as their source. To say they are all from Lemon Lane Landfill is flawed logic, because it has not been disproved that they originated at other locations within the basins that discharge through those same springs. However since Lemon Lane Landfill is such a large source it is highly probable that they originated there. The situation is actually much more complicated because there are several sources of PCBs in the vicinity of the Lemon Lane Landfill, and apart from Illinois Central, Quarry and Slaughterhouse Springs, which seem to be clearly hydraulically connected to the Lemon Lane Landfill, PCBs reported at Urban Spring and the Stoney Springs (east and west), might have their PCB sources at locations other than Lemon Lane Landfill.

Stanton and Smart (1981) showed that when flow was highest, tracer-dye recoveries were highest, velocities fastest, and tracer passage quickest. This relationship is useful for interpreting the high-flow ground-water tracing results from Lemon Lane Landfill. Two of the three dyes that were used in this experiment are commonly associated with accidental spillage, uranine (used in car radiator coolant) and FBA 28 used in laundry detergent The third dye that was used (actually a dye intermediate product) C.I. DY 96 appears to have migrated literally all over the place. However, based upon the findings of Stanton and Smart (1981) and the association~of PCB migration with suspended sediments, if the C.I. DY 96 migrated to all the locations that it appears to have done, then in the same high-flow conditions so should PCBs have migrated to the same locations. There is therefore a problem with interpretation for all those sites that have dye detections but no PCB detections, as having Lemon Lane Landfill as the only possible source for any tracer dye detected.

The detections of dyes in nearby domestic wells confirm a typical form of drainage for the epikarst where flow is locally horizontally along the bedrock surface in various directions until being captured by vertical conduits usually located in the throats of sinkholes. Dyes that migrated from the injection locations in the landfill area were recovered in such wells to the east and northwest of the landfill, even though the largest springs are located to the south and southeast. It is often the case in the epikarst that ground-water flows in directions that are not initially in a direction directly towards any particular spring that is the likely discharge location. These initial horizontal and vertical pathways are tributary to cave streams that eventually discharge at a spring or springs, the tributaries typically form a dendritic pattern, there are many such examples in the world (Courbon et al., 1986). The directions of soil-water and ground-water flow in the hinterland of closed depressions are all influenced by the local hydrology and locally steep hydraulic gradients into the closed depressions (Williams, 1983; Gunn, 1981).

Both Illinois Central and Quarry Springs discharge PCBs in considerable quantities and highest concentrations are being discharged from Illinois Central Spring during following storm events. Sporadic detections of PCBs are also documented at Urban Spring, although the orifice of this spring orifice is along a much shallower hydraulic gradient from the landfill, and its discharge is relatively small when compared to the discharge from Illinois Central Spring or Quarry Spring. Also, when compared to the elevation of the landfill, Urban Spring is at a higher elevation than the other springs (Illinois Central, Quarry, etc.). The spring is also near an automobile repair facility (many automobile oils contain PCB's), this causes a problem deciphering where the source of PCBs at Urban Spring could be, since the discharge is small the basin feeding it would be small (Quinlan et al., 1995) - so it would most likely have a local source rather than Lemon Lane Spring. From a combination of geomorphological, contaminant, PCB, hydraulic gradient, and discharge data, it can therefore be hypothesized that the most likely discharge locations for ground-water from beneath the landfill are these Illinois Central and Quarry springs.

It should emphasized that the results of the initial tracing experiments performed at the site suggested that the discharge locations for the site were principally Illinois Central Spring and Quarry Springs (McCann and Krothe, 1992), the high-flow results suggested dye had migrated to many surface and ground-water locations far beyond those springs.

There would be no logical reason why ground water would flow to these other distal locations given the low-gradient, long, and thus inefficient Towpaths, rather than use the more efficient pathways to Illinois Central Spring and Quarry Spring, where most of the PCBs are being discharged. The third most efficient pathway based upon hydraulic gradient would be to the Slaughterhouse Spring Complex, also a site where PCBs have been detected. It would be illogical to suggest that merely in high flow that dyes would be transported to sites, but never PCBs.

In the most recent tracing investigations at Lemon Lane, direct analysis of dye in water samples has been done, and in some cases continuous flow fluorometry was done. Generally summarized the results of these experiments show flow to Illinois Central Spring and Quarry Spring to the south and the Slaughterhouse spring complex to the north, the sites where DY 96, uranine, or FBA 28 were detected in the 1988 high-flow test, were not monitored during some of these tests.

Based upon the tracing results it is logical to conclude that the springs where both dyes and PCBs were detected are the discharge locations for Lemon Lane Landfill. The detections of dyes elsewhere but without PCBs in most cases suggests that the source of those dyes must be different from the source of the PCBs.

Ground-Water Tracing Using Fluorescent Dyes

Dyes used for tracing are often the same dyes used in many industrial applications (e.g., see Green, 1990) and this often results in the accidental spillage into the environment the same dyes that are used for tracing making correct identification of tracing dyes imperative. In karst terranes ground-water is easily contaminated this way and in the same way tracing experiments can be contaminated too. Quinlan (1990) described the use of a tracer as a label that was added to a surface- or ground-water flow system that could be recovered and identified elsewhere. The ideal example of a tracing experiment would be the use of a dye as a tracer that was not present in the ground water as a result of contamination. Unfortunately, in reality in many aquifers dyes are present from accidental spillage. Since they can be detected at such a low concentrations, even very minute quantities of dyes, not being used as tracers, but that are accidentally spilled, are detected at many springs, streams, and wells. This makes identification of recovery of the dye being used as a tracer a little more challenging.

The last point is the most important, because it is not merely the recovery of the same kind of dye that is tracing, but recovery of the same dye that was injected, and being cognizant of that. This small but vital final point depends upon not only the expertise and experience of the organization doing the tracing experiment, but the way each individual test is designed, the assumptions that are made about the experiment, and the statistical reliability of the data collected and correct and logical interpretation of those results.

The more rigorous the design of a tracing experiment the more likely that the results can be interpreted logically, because there will be less chances for systematic errors. However, no amount of experience and expertise are going to help when experimental design is unsound or illogical and is based upon an inadequate hypothesis that is essentially untestable. Correct identification of recovery of those dyes that were used for tracing is absolutely essential. Laboratory techniques used for analysis have to be appropriate and robust also. No amount of checking and rechecking (e.g., what is referred to as QA/QC) of results of infrequently-collected individual samples can rectify the fundamental error being made where a certain dye is identified as being present in a sample, but is not (or more appropriately cannot) be identified as not being the same slug of dye that was injected.

What is becoming a trend by some tracing organizations (not including Cambrian Ground Water Co.) is use of a technique where grab samples of charcoal-elutant or water data are treated to statistical curve-fitting techniques. Even though based upon average velocities in karst (Worthington, et al., 1999) an inadequate sampling frequency is employed, the reliability of the analysis of each individual sample is taken to be an increase in the reliability of the data as a whole. Whatever the results and goodness of fit, a single data point it is still a single data point and the statistical probability of observing whatever is sought is practically zero, or at best a 50-50 toss up between being dead right or dead wrong.

In many cases this technique is applied to the analysis of a single sample containing multiple dyes - the resulting statistical inadequacy being a multiple of the single data point problem. Techniques such as this are claimed to be "state of the art" despite the fact that they are typically based upon a single observation with an accompanying probability of being a random and false indication of near 100%. Another claim is made that up to 7 dyes can be used at the same time, and a curve fitting program can be used to separate them when they are recovered together (Tucker and Crawford, 1999). Recent work using the software that was used by Tucker and Crawford (1999) shows that because of reactions of the dyes used and other statistical problems curve fitting techniques do not help results involving multiple dyes. It must be stressed very highly that because charcoal methods often rely on a single data point, they do not obey the principle of the Law of Large Numbers. This is simply where enough samples have been collected to have statistical reliability in the first place to support a claim that a dye that was injected was recovered - the whole point of tracing ground-water. The claim by Tucker and Crawford (1999) that seven dyes can be analyzed in a single sample and quantified using curve fitting techniques is misleading, it is note simple step from a having a sample containing multiple dyes to actually determining which dyes are present, the principle problem being chemical reactions causing subtle changes in the fluorescence (Cambrian Ground Water Co. and others, research in progress). Again it has to be stressed that analysis of a single sample and carries with it the astronomical uncertainty associated with a single observation.

Direct analysis of water samples is similar to sample collection methods used in other tracing experiments that are well-published in the mainstream scientific literature. When using a sampling method such as frequently-collected water samples much information is gained that subsequently can be used for interpretation, and since data are collected continuously or semi-continuously variation can be better evaluated. Results from direct water-sample analysis provides better and more reliable information than when interpreting the results from charcoal methods (Smart et al, 1986).

These authors also point out that quantitative methods using water samples are far better than qualitative methods (using charcoal) because more information means that more intelligent interpretations can be made and subsequent corroborating tests can gain more information rapidly, in a far more logical sense. Field et al., (in review) also show that analysis of dye recovery curves from water-sample analysis allows intelligent inferences to made about many parameters of the flow path, and this is a useful advantage of direct analysis of water samples and quantitative methods.

As mentioned, charcoal based qualitative methods usually make one measurement per week, water-sample methods use typically one measurement per hour or 168 measurements per week. Collection of water samples by continuous flow means one measurement per minute can be made which means 10,080 measurements per week can be made. Often a measurement is made every 5 seconds and averages taken every minute and stored in memory, this is equivalent to 120,960 measurements per week. Worthington et al., (in press) shows that traced velocities from sinking streams to springs are all faster than 0.001 m/s (about 100 m/day) and can be as fast as almost 2 m/s (Worthington and Davies [unpublished data]), with a geometric mean velocity of 0.022 m/s [1.9 km/day]. Average velocities are such that most slugs of dye pass through springs in one or two weeks justifying the - at least - hourly sampling rate used in water-sample collection. Using such velocities and quantitative tracing data it can be shown that many such slugs of tracer are only sampled once or twice using charcoal methods (typically weekly sampling). One sample per week also cannot hope to show the details of the nature of the slug of dye that passes a particular monitoring point in a few days. Many pass in a few hours or days at very low concentrations (Figure 1).

Weekly is the about the normal frequency used for collection and replacement of charcoal packets (Alexander and Quinlan, 1990). The problem however is that the packets cannot be exchanged in the stream at too rapid a rate as there would be insufficient time for the dye adsorption reaction to take place, so sampling frequency cannot be increased by very much. A period of one week being the appropriate time for exposing the charcoal to dye has largely been assumed but is not based upon many scientific measurements, and the chemistry of adsorption is different for different dyes and in some cases the adsorption is not strong enough to retain the dye for long and in different waters adsorption might be variable and somewhat unpredictable (Fisher et al., in press) .

Even though dye can be collected by adsorption onto activated charcoal that is placed in the spring, stream or well, it must be removed in the laboratory for analysis. The combined technique is referred to as the charcoal elutantmethod and by its nature is a much more complicated technique than direct analysis of water samples, for example the spectrum obtained in elutant is significantly different to that obtained in water for the same dye. The eluant is a solvent such as isopropyl alcohol in an alkaline medium, which after treating the charcoal for a specified amount of time, usually one hour, is decanted into a test tube and then analyzed in a scanning spectrofluorophotometer.

Unfortunately none of the various methodologies for elusion (i.e., the preparation step prior to analysis) has ever been formerly published in peer-reviewed, independently-refereed analytical chemistry journals, so are based only upon individuals professional opinions, and some experiments written up in theses (Smart, 1972), but this study only involves one dye (C.I. Acid Red 388, rhodamine WT). Fisher, et al., (in press) is a paper that does explore elusion but only for a few dyes. This does not mean that the slightly different elusion techniques would be inherently problematic as most charcoal-based methods are generally similar, and most yield results, but the quality of those results is inferior to results from frequently-collected water samples. When dyes are recovered at low concentrations on charcoal where, when typically less than an absolute minimum number of samples are collected, uncertainty can be astronomical. Once each charcoal sample is elated, the analysis on a spectrofluorophotometer is similar to that of water samples.

It should also be pointed out that adsorption by charcoal is not selective to dyes and many other fluorescent compounds in the water are adsorbed, mostly naturally occurring from the decay of leaf and wood matter and substances leached from the soil. This accumulation of many compounds makes the part of the spectrum where many dyes fluoresce congested and causes additional problems with interpreting spectra. The dyes end up in a very complicated mixture consisting of all the fluorescent compounds that were adsorbed onto the charcoal. When such spectra are obtained, particularly when multiple dyes have been recovered, a mathematical technique can be used to decompose the spectra into the individual peaks that are present. Although this is a much-used technique when using charcoal packets and analysis of the elutant, it should not be forgotten that the result is based upon one measurement at one location and carries very little statistical reliability. Also, the deconvolution technique is a curve fit to the whole data and does not necessarily carry any more reliability than the raw spectrum. The whole aspect of spectral analysis should be viewed in the context that nothing in science is magic and the assumptions that have to be inevitably made might be inappropriate, with a lowering of reliability rather than an increase.

We should point out very clearly that if enough samples are collected (they can be with automatic water sampling machines and continuous-flow fluorometry), there are no such problems with collection and analysis of water samples, even at very low concentrations, where Figure 1 shows the recovery of two dyes with very similar spectral properties, with both recoveries being less than 1 part per billion at their maximum concentrations on the recovery curve. This example also clearly shows the sensitivity of water-sample analysis for tracing ground water using fluorescent dyes.

The problems with charcoal techniques are just being understood (Fisher et al., in press). Their research has shown that activated charcoal adsorbs certain tracing dyes by forming relatively weak chemical bonds with other dyes forming much stronger bonds. The problem is that previously charcoal has been claimed or at least assumed to accumulate small concentrations of dye in water thus effectively lowering the detection limit (it has been claimed by up to 400 fold) - but there are no data that have been published in peer-reviewed journals to support this. Fisher et al. (in press) shown that the systematics of charcoal chemistry and its application to dye adsorption and elusion are far from simple and assumptions that were previously made are far from appropriate. In fact these experiments show that for one particular dye a much smaller percentage is actually adsorbed on the charcoal than was previously thought. For certain other dyes, a larger percentage of dye is adsorbed on the charcoal, but is much harder to remove (elate) for analysis. What is potentially equally troublesome to the charcoal method is that fluorescent compounds that closely resemble (are almost indistinguishable from) some dyes can be found present on charcoal before any dye has contacted the charcoal in any way. These compounds are probably dyes, or are mixtures of fluorescent compounds that appear as close relatives to the dyes used for tracing.

Data showing such effects and comparing actual dye peaks with peaks on activated charcoal not having contacted dye are included with this technical memorandum. There are peaks that look like dyes on raw charcoal, but they seem to be removed by water flowing through the charcoal, although there is no systematic study that has ever been done to show exactly what is happening, and it is not known whether the compounds are removed or a merely masked by other fluorescence, and might return on charcoal as the background of other fluorescent compounds varies. It is also true that when any form of ground-water tracing is done, that unless a proper hydrogeological description of the flow system is hypothesized prior to the tracing commencing, it might be impossible to determine whether the dyes being recovered have not originated at numerous other sources in the ground-water basin being studied - a ground-water basin that also includes the site in question. It cannot be emphasized enough that when it comes to fluorescence, a dye looks the same fluorometrically on analytical instruments that are set up similarly, and there is no magic button to press to show where that dye originated. In fact no sophisticated analytical method on any instrument can tell you where the dye originated, unless there are enough data collected to provide the probability that it is not chance, which literally means that the sophistication of the analysis is far less important than the design of the experiment.

No amount of experience, professional opinion, or computer software capable of mathematically analyzing the spectra can tell the difference between fluorescein dye from a leaking radiator and fluorescein that was injected in a well or sinkhole, or a hydraulic oil colored with a rhodamine dye that has leaked from a car's transmission or power steering system and one that was injected into a well or sinkhole for ground-water tracing, spectrally they are exactly the same. The only way to tell the difference is by designing the experiment such that the difference can be shown to be so when all the results are evaluated. Because (typically) when using charcoal only one sample a week is evaluated, based upon the documented velocity of ground water in karst terranes Worthington et al., 1994), the statistical reliability of analysis of one sample is practically nought. This is true regardless of what instrumentation is used or what mathematical routines are used on the data, the probability of observing an event in one observation is practically zero. Collection of one sample per week in systems where flow is so much more rapid ignores many protocols of sampling statistics and carry a low probability of correctly describing the event being sought.

Experiments done at numerous sites where charcoal or cotton detectors were initially used have often yielded, sometimes significantly different results when the tracing experiments were repeated, and frequently-collected water samples are analyzed and those results interpreted. Often the latter type of results are the only results that conform with hydrogeology and contaminant data. When several repeated tests are done using direct analysis of water samples and those results interpreted this is often the case. What is most obvious is that when repeated tests are done using water samples the results often are supported by other data (geology, geomorphology and surface hydrology), which, in the same area was not the case with the charcoal results. Independent tests using other injections and recoveries usually put this beyond doubt and often show the false positive results from the charcoal might have been predicted if interpretations were done using all other information.

The only way to rigorously design any tracing experiment is to design it such that the results follow the basic principles of sound scientific practice. This involves developing a hypothesis using all existing information available - and testing the system to try to cause that hypothesis to fail, and, being able to repeat whatever results are obtained and to show that those results were not the result of a random, accidental event. Unless these scientific practices are performed it cannot be argued that the experimental design is scientific and therefore that there is not some flaw in the logic of that design and subsequent interpretation of the results. The results of poorly designed experiments are inherently specious. AISO7 the results of poorly-designed, illogical, unscientific experiments cannot be logically interpreted and are thus inherently statistically biased and because of their inherent unreliability, essentially useless.

There are other tests that can be done to develop robust experimental design in any project. The simplest is that if contaminants that are characteristic of contamination at the site are not found in a spring or well where dyes are recovered, and the relative transport characteristics are reasonably similar, then it should be inferred that the source of the dye that was detected is not likely to be the same as the source of the contaminants. This is especially true with regards to PCBs, because karstic velocities in conduits are rapid enough (Worthington et al., [in press]) to transport PCBs in suspension, and PCBs should be transported in a somewhat similar fashion to dyes. Most dyes that are commonly used as tracers are conservative, or put another way, are transported through the system with minimal adsorption, so it would be quite unlikely that such contaminants would be transported along a pathway where tracers dyes were not. If contaminants are detected at locations and dyes used as tracers not recovered, then it should be inferred that the contaminants came from a source that was different from the source of the dye. Particularly with charcoal-packet data, because of the infrequent data collection that is done, such inconsistencies should be further investigated. When the charcoal technique is used often only a single water sample collected at the same time as the charcoal packet. If charcoal is eluted and contains dye, then so should the water samples contain that dye, and if the dye originated from the site, so should there be contaminants at that site, providing that there is some comparability between the transport characteristics of the dyes and the contaminants. This logic is often missed.

It is often the case that when a well is being sampled, because of the logistical problems of regularly sampling many wells, relatively infrequent water samples are collected. This does not necessarily compromise the data because wells often do not intersect preferential flow zones or macro fissures (Benson and La Fountain, 1984), and are as a consequence less sensitive to detecting the passage of tracers or contaminants. Neither the contaminants or the tracers apparently arrive and pass through the well at the same velocity as they would at a discharge location where all the greatest concentration of flow passes. If the experiment is designed properly then the passage of the tracer at some location in the monitoring system that is being sampled at the appropriate frequency and can thus explain the passage of lower concentration of tracers at other locations, again, if the experiment is properly designed it can yield results that can be logically explained and interpreted. In the execution of science in ground-water investigations there are few if any short cuts to take without encountering obstacles.

The following example illustrates a typical problem with charcoal-elusion data. The time-concentration curve in Figure 2 shows the apparent recovery of uranine C.I. Acid Yellow 73 (sodium fluorescein the same dye as used in the initial tracing at Lemon Lane) - at a spring that was possibly connected by a prominent fault to a landfill area in northeast Tennessee. When evaluated with other tracing results done at the same time, this dye was apparently being recovered several kilometers further away that any other dye, and in the opposite direction to the direction that every other dye used for tracing was being recovered. The fault was clearly also a possible conduit so, one year later (to try and confirm that this had really happened) a different dye C.I. Acid Red 87 - Eosin was injected into exactly the same sinkhole. Although the repeat injection was done almost in the same week of the following year and inferred to be during similar hydrological conditions, the second dye (Eosin) was never recovered at the same spring. This second test suggested that all the uranine detected after the first injection was from sources other than the injection location. Subsequently numerous other tracing tests using different dyes were completed that showed that there was almost certainly a ground-water divide between the landfill site and the spring, therefore it was extremely unlikely (if not impossible) that the initial recovery of uranine was as a result of injection at the landfill. Figure 2 shows the time-concentration curve with data for the common tracing dye uranine (sodium fluorescein) as apparently recovered in the actual case described above. No other data supported a connection between the site and the spring. It could not be shown that the dye at the spring was the dye that was injected into the sinkhole.

AISO7 the velocities obtained in the subsequent tests suggested that the apparent recovery of uranine was at a far too rapid velocity as compared to the tracer velocities in the subsequent test results. As in any scientific endeavour there is not 100 percent certainty that the first recovery was not from the landfill injection, but based on trying to repeat the result it is extremely unlikely. Careful examination of the area around the spring found several empty containers of antifreeze solution that could have been the source of the initial dye that was apparently recovered. The dye in the antifreeze is also C.I. Acid Yellow 73, uranine, and when analyzed on a scanning spectrofluorophotometer yields an identical emission peak for uranine.This example shows clearly that even when there is strong geological evidence that a site is hydrologically connected to a spring, and even when tracing is done that supports that hypothesis, testing it independently is the only way to be sure that the interpretation of those results is correct.

The use of scanning spectrofluorophotometers for analysis of water and elutant samples will help resolve whether the fluorescence in a sample if caused by a particular dye or not, but will never help determine whether the dye being analyzed is the same aliquot of dye that was injected. A problem exists where in the eyes of many tracing practitioners and regulators more sophisticated analysis is directly related to more reliable results. But bad information is actually worse than no information, it helps no one. Not knowing if information is good or bad is as potentially catastrophic as having no information when important decisions are involved. Assuming that information is good because a modern instrument is used must be weighed against the results of the 1877 tests done in the Danube, where no instrument was used, but an equally reliable result was obtained - additionally because three, not one tracer was used simultaneously (the dye fluorescein, uranine, C.I. Acid Yellow 73), sodium chloride (common salt) and shale oil were used simultaneously).

A serious problem exists with the desire to model a site's hydrogeology on a computer. The assumptions that have to be believed in the model can be easily shown to be inappropriate, and this probably applies to all aquifers (Cherry, 1994, Konikow and Bredehoeft, 1995) and certainly applies to karst and fractured-rock aquifers. Papers that are pertinent to a discussion of modeling such as Quinlan et al., ( 1996) show clearly that numerical modeling of ground-water flow on computers cannot describe velocity or transport of contaminants (ironically these are the two parameters that are desired from the modelled output)- but there are many who still use such models to describe such parameters. These models are rarely tested with empirical data (such as tracing), and many modelers act insulted at even the prospect of having to test their results. What is most mind boggling is that often when the models are tested and shown to be inadequate, the test data are questioned and not the model! In one classic example, a model that showed hydraulic heads such that the water table must be above the ground surface (i.e., a surface water body), the person responsible for the model was asked - where was the lake that was inferred in the model, at that location? The reply was that whatever was inferred was irrelevant because the modelled data was a close fit to the head data used, and the hydraulic data were modelled "correctly." It is unfortunate that almost all modelled data end up being less damaging to an industrial client's case that tracing data, and this has led to many such "experts" working for consulting companies representing industrial clients saying "I never want to testify again without a computer model."

The Cherry (1995) reference above is proofthat when the assumptions that are made about the parameters used in such models are tested with data - they fail, and these assumptions are tenuous if not inappropriate at best.

In Oak Ridge recently, hydrological testing was done near the boundary of the Y-12 nuclear weapons plant (operated by a contractor of the U. S. Department of Energy). Computer models and other hydrological data were used to "predict" that the contaminant plume would migrate, but only in about ten years to a spring that was known, about 1 kilometer from the plant boundary. During this testing the State of Tennessee sampled this spring continuously for dye (DOE and its contractors did not, their conceptual model did not describe that the flow would reach there in less than 10 years, so to them it was a non-issue). The tracing dye had been injected into a monitoring well and another down gradient monitoring well about 150 meters away was pumped continuously; both wells are over 1 kilometer from the spring. The dye that was injected was recovered in the down gradient and pumped well, so a connection between the well where dye was injected and the pumped well was proven (their declared goal in the test). Forty-eight hours after the pump was switched offthe dye emerged from the spring 1 kilometer awayÄwhere a set of fingerprint contaminants had before. The point should be made that although the test that was conducted was successful, the point that the contaminants were discharging from the spring was ignored in the test. Since it was known to be discharging contaminants one wonders why not? Although this example suggests they were blatantly unethical it is probably more appropriately explained by realizing that there is a conviction amongst many ground-water professionals that aquifers really have certain assumed qualities and that testing them with the idea that conditions are different is futile.

This leads to a discussion about the ethics of deliberately designing a tracing experiment where the test is designed to only show what supports a predetermined conceptual model, rather than test the system for what it is. The problem is whether the professional involved believes in being an advocate or a diagnostician. The problem with testing a firmly-believed conceptual model is that the design of the test can cause support for the inappropriate model thus causing any alternative models to be rejected because there is "overwhelming support" for a certain model. This ends up being used as a tool against any criticism of the model and is probably the most insidious way that science (and it should not be referred to as that) can be misused. The same can be said for the design of such tests that do not encompass a study of what describes the flow system as a whole rather the flow system on a particular site. This is the classic situation of not wishing to know what happens to a plume after it has left a site, a rather typical situation for most sites. Fortunately tracing done properly will show whatever is happening and also whatever might not be happening. This method of tracing will also protect all parties more than expose any party. Usually the attorney of a company can see a situation of the lesser of two evils, where it is better to know too much than too little, of course, attorney jokes apart, there is also the opposite - where a lawyer will only approve doing tracing to show what is happening inside a particular fence. Professionals who indulge in this type of tracing where they deliberately design an experiment to show something that is embedded in a particular conceptual model and is based upon a set of"convenient" assumptions, are acting unethically as scientists, whatever they claim was asked of them by their clients.

Recommendations

The following figures show examples of fluorescence spectra from analysis of dye samples, charcoal elutant, and other fluorescent compounds. N.B. Although it seems that some of these examples might suggest tracing ground water using fluorescent dyes is rather difficult careful scrutiny of the spectra can differentiate between the various compounds and using appropriate scientific techniques can result in highly reliable tracing of ground water using dyes.

1-1 Emission spectrum of standard sample of the tracing dye uranine (C.I. AY 73 - sodium fluorescein), 4.75 parts per billion in water showing an emission peak with its center at about 518 nm.

1-2 Emission spectrum of standard sample of the tracing dye uranine (C.I. AY 73 - sodium fluorescein), 1 part per billion in water showing an emission peak with its center at about 513 nm. * These two spectra show that the position of the peak (in nm) is prone to shifting slightly in different waters/samples, even though it is the same substance.

1-3 Standard sample of the tracing dye uranine (C.I. AY 73 - sodium fluorescein), 1 part per billion in eluant (5% NH4OH in 70% isopropyl alcohol) showing an emission peak at about 521 nm.

1-4 Sample of charcoal subject to same preparation as typical charcoal sample - however, this charcoal has not been near any dye, it is fresh from a can. Note the peak at 516 nm - if the position of this peak shifts it could easily be mistaken for uranine.

1-5 Background sample (concentration of 13.6 parts per billion) from a charcoal packet placed in a well near an active landfill in Tennessee. Note the very strong peak at about 522 nm. Compare this to example 1-3. Sample 1-3is a laboratory standard prepared from C.I. AY 73 uranine tracing dye, 1-5is the same dye naturally discharging from a leaking landfill, in a sample collected before any tracing had begun at this facility. Instrument settings were the same for both analyses.

1-6 Another background sample from another well at the same site - note the peak here is at 519.6 nm, and is probably the same dye - uranine, but the peak has shifted slightly.

1-7 Another sampling location at the same site, uranine dye at this location also.

1-8 Unwashed charcoal refrigerated and then eluted for 1 hour (i.e., standard preparation). Note peak at about 516 nm. This is thought to be same dye as the tracing dye uranine that gets on the charcoal in the charcoal's manufacturing process.

1-9 Raw washed charcoal (washed in packet) in strong jet of water - note peak at 513 nm. This peak is reported to go away after " adequate washing," It obviously does not in every case.

1-10 Standard sample of the dye C.I. AR 52 (sulphorhodamine B) 10 part per billion in water used for tracing ground water. This dye or another dye with identical fluorescence is used to colour automatic transmission fluid (see 1-11 and 1-12).

1-11 Sample of Dextron II ATE from my Land Rover power steering box diluted in water and analyzed, note peak at 582.8 nm - compare this with 1-10 - its probably the same dye, or C.I. Basic Violet 10 (rhodamine B) - which has exactly the same fluorescence spectrum as the dye C.l. AR 388 (rhodamine WT) as used by CBS for their tracing.

1-12 Standard sample of C.I. AR 388 (rhodamine WT) tracing dye, 0.5 parts per billion in water, note peak at 578.4, compare with 1-lO, remember peaks move about and C.I. AR 388 fluoresces at a higher wavelength in a more concentrated solution (see 1-13), and on the tails of recovery curves.

1-13 Standard sample of C.I. AR 388 (rhodamine WT) tracing dye, 1.0 parts per billion in water, note peak at 578.8 (i.e., at a higher wavelength) compare with 1-12, remember peaks move about and C.I. AR 388 fluoresces at a higher wavelength in a more concentrated solution (see 1-13). This example shows this even in a very low concentration standard. 1-14 Spectrum of new style Prestone antifreeze solution, showing clearly the presence of two dyes used for tracing (uranine, C.I. AY 73, and C.I. AR 388, rhodamine WT ). These are the same dyes as used by CBS for their tracing experiments - they are the most popular and commonly-used tracing dyes.

References Cited

Alexander, E.C., Jr., and Quinlan, J.F., 1992, Practical Tracing of Groundwater with Emphasis on Karst Terranes, Short Course Manual, Geological Society of America (Annual Meeting Cincinnati, Ohio).

Benson, R. C. and La Fountain, L.J., 1984, Evaluation of subsidence or collapse potential due to subsurface cavities, In, Sinkholes: their geology, engineering, and environmental impact, B.F. Beck, (ed.) Balkema, Boston, p. 201-215.

Bogli, A., l99O, Karst Hydrology and Physical Speleology, Springer-Verlag, 284 p.

Courbon, P., Chabert, C., Bosted, P., and Lindsley, K., 1989, Atlas of the Great Caves of the World, Cave Books, St. Louis, 369 p.

Crawford, N. C., and Tucker, R., 1999, Non-linear curve-fitting analysis as a tool for identifying and quantifying multiple fluorescence tracer dyes present in samples analyzed with a spectrofluorophotometer and collected as part of a dye tracer study of groundwater flow. Hydrogeology and Engineering Geology of Sinkholes and Karst in, Beck, Pettit and Herring, (Eds) Balkema, Rotterdam, p. 307-312.

Field, M.S., Davies, G.J., and Pinske, P., (in review) Solute transport parameter estimation using a two-region nonequilibrium model, (submitted to Journal of Contaminant Hydrology).

Fisher, A, Alexander, S. C., and Alexander, E.C., Jr., (in press) Evaluation of eluant solutions used in ground water dye tracing. (submitted to - Journal of Cave and Karst Studies, Bulletin of the National Speleological Society).

Gunn, J., 1981, Hydrological processes in karst depressions, Zeitschrift fur Geomorphology, v. 25, no. 3, p. 313-331.

McCann, M.R., 1999, CBS Corporation, personal communication.

McCann, M.R., and Krothe, N.C., 1992, Development of a monitoring program at a Superfund Site in a karst terrane near Bloomington, Indiana, Hydrogeology, Ecology, Monitoring and Management of Ground Water in Karst Terranes Conference (3rd, Nashville) Proceedings, National Ground Water Association, Dublin, Ohio, p 349-372. Quinlan, J. F., and Ewers, R.O., 1989, Subsurface drainage in the Mammoth Cave area, in, White, W. B. and White, E.L., (Eds), Karst Hydrology, Cponcepts from the Mammoth Cave Area, Van Nostrand Reinhold, New York, p. 65-104.

Quinlan, J.F., 199O, Special problems of ground-water monitoring in karst terranes, Ground Water and vadose Zone Monitoring, ASTM STP 1053, D.M. Nielsen and A.I. Johnson, (Eds), American Society for Testing and Materials, Philadelphia, p. 275 -304.

Quinlan, J.F., and Ray, J.A.,1995, Normalized base-flow discharge of ground-water basins: A useful parameter for estimating recharge area of springs and recognizing drainage anomalies in karst terranes, in, B.F. Beck, Ed., Karst Geohazards, Rotterdam, Balkema, p. 149-164.

Smart, P.L., Atkinson, T. C., Laidlaw, I.M.S., Newson, M.D., and Trudgill, S.T., 1986, Comparison of of the results of quantitative tracer tests for determination of karst conduit networks: an example from the Traligill basin, Scotland. Earth Surface Processes and Landforms, v. 11, p. 249-261. Stanton, W. I., and Smart, P.L., 1981, Repeated dye traces of underground streams in the Mendip Hills, Somerset, Proceedings ofthe University of Bristol Speleological Society, v. 16, no. 1., p. 47-58.

Williams, P.W., 1983, The role ofthe subcutaneous zone in karst hydrology, Journal of Hydrology, v. 61, p. 45-67.

Worthington, S.R.H., Davies, G. J., and Ford, D. C., 1999 (in press), Quantification of matrix, fracture and channel contributions to storage and flow in a Paleozoic Carbonate aquifer, (in press, published in a book by Balkema, Rotterdam).

Worthington, S.R.H., 1991, Karst hydrogeology of the Canadian Rocky Mountains, Ph.D Thesis, School of Geography and Geology, McMaster Unversity, Hamilton, Ontario, LS8 4K1, 227 p

Ford, D. C., and Williams, P. W., "Karst Geomorphology and Hydrology," Unwin Hyman (London), 1989, 601 p.

Green, F., 199O, The Sigma-Aldrich Handbook of Stains, Dyes and Indicators, Aldrich Chemical Company, Milwaukee, Wisconsin, 776 p.

Quinlan, J.F., Davies, G.J., Jones, S.W., and Huntoon, P.W., 1996, The applicability of numerical models to adequately characterize ground-water flow in karstic and other triple-porosity aquifers. Subsurface Fluid-Flow (Ground-Water) Modeling, ASTM STP 1288, J.D.Ritchey and J.O. Rumbaugh, American Society for Testing and Materials, Philadelphia, PA. p.114-133. Prof. E. Calvin Alexander, Jr., and Scott Alexander, Department of Geology and Geophysics, University of Minnesota, personal communication, 1998-1999

Konikow, L. F., and Bredehoeft, J. D., 1995, Ground-water models cannot be validated, Advanced Water Research v. 15, p. 75-83. Cambrian Ground Water Co. (previously unpublished data included in this manuscript)

Cheny, J.A., 1995, Field-scale experimental studies of contaminant behaviour in ground water, Special Lecture. Geological Society of America Abstracts with Programs, Southeastern Section, 103 p.

Stephen R. H. Worthington and Gareth J. Davies, [unpublished data]