A Three-Tiered Approach to Long Term Monitoring Program ... of... · ing an effective groundwater...

Bioremedia tion Journal, 8(3-4):147-165, 2004 Copyright ~ 2004 Taylor and Francis Inc. ISSN: 1040-8371 DOh 10.1080/10889860490887518

( ~ ) Taylor & Francis Taylor & Francis Group

A Three-Tiered Approach to Long Term Monitoring Program Optimization

Carolyn Nobel Parsons, 1700 Broadway, Suite 900 Denver, CO 80290, USA

John W. Anthony Mitretek Systems, 7720 E. Belleview Avenue, Suite BG6, Greenwood Village, CO 80111, USA

ABSTRACT Long term monitoring optimization (LTMO) has proved a valuable method for reducing costs, assuring proper remedial decisions are made, and streamlining data collection and management requirements over the life of a monitoring program. A three-tiered approach for LTMO has been developed that combines a qualitative evaluation with an evaluation of temporal trends in contaminant concentrations, and a spatial statistical analysis. The results of the three evaluations are combined to determine the degree to which a monitoring program addresses the monitoring program objectives, and a decision algorithm is applied to assess the optimal frequency of monitoring and spatial distribution of the components of the monitoring network. Ultimately, application of the three-tiered method can be used to identify potential modifications in sampling locations and sampling frequency that will optimally meet monitoring objectives. To date, the three-tiered approach has been applied to monitoring programs at 18 sites and has been used to identify a potential average reduction of over one-third of well sampling events per year. This paper discusses the three-tiered approach methodology, including data compi- lation and site screening, qualitative evaluation decision logic, temporal trend evaluation, and spatial statistical analysis, illustrated using the results of a case study site. Additionally, results of multiple applications of the three-tiered LTMO approach are summarized, and future work is discussed.

KEYVVORDS long term monitoring network optimization, Mann-Kendall temporal trends, geostatistics, kriging

Address correspondence to Carolyn Nobel, Parsons, 1700 Broadway, Ste 900, Denver, CO 80290, USA. E-mail: [email protected]

INTRODUCTION

Groundwater monitoring programs have two primary objectives (U.S. Environmental Protection Agency [USEPA], 1994; Gibbons, 1994; USEPA, 2004a):

147

1. Evaluate long-term temporal trends in contaminant concentrations at one or more points within or outside of the remediation zone, as a means of monitoring the performance of the remedial m ea su re (temporal objective); and

2. Evaluate the extent to which contaminant migration is occurring, particularly ifa potential exposure point for a susceptible receptor exists (spatial objective).

The relative success of any remediation system and its components (including the monitoring network) must be judged based on the degree to which it achieves the stated objectives of the system. Design- ing an effective groundwater monitoring program involves locating monitoring points and developing a site-specific strategy for groundwater sampling and analysis so as to maximize the amount of relevant information that can be obtained while minimiz- ing incremental costs. Relevant information is that information required to effectively address the temporal and spatial objectives of monitoring. The effectiveness of a monitoring program in achieving these two primary objectives can be evaluated quan- titatively using statistical techniques. In addition, there may be other important considerations associated with a particular monitoring program that are most appropriately addressed through a qualitative assessment of the program. The qualitative evaluation may consider such factors as hydrostratigra- phy, locations of potential receptor exposure points with respect to a dissolved contaminant plume, and the direction(s) and rate(s) of contaminant migration.

Parsons has developed an approach to evaluating and optimizing long-term monitoring programs at mature sites (site characterization is complete, and a long-term monitoring program is in place and has been executed for some period of time) where the costs of long-term monitoring represent a significant fraction of the costs of an environmental response action. Parsons' three-tiered approach for long term monitoring optimization (LTMO) consists of a qualitative evaluation, an evaluation of temporal trends in contaminant concentrations, and a spatial statistical analysis. After all three phases (or "tiers") of the evaluation have been completed, the results of the three tiers are combined to assess the degree to which the monitoring network addresses the primary objectives of monitoring. A decision al-

gorithm is applied to assess the optimal frequency of monitoring and the spatial distribution of the components of the monitoring network, and to develop recommendations for monitoring program optimization. "Optimization" in this paper refers to the application ofa rule-based procedure (based on professional judgment and statistical analysis) that is applied to recommend a monitoring program that is effective at meeting the two objectives of monitoring while balancing resources and qualitative con- straints, rather than being a formal mathematical optimization. This paper describes methods that have been applied by other investigators to the problem of monitoring-program optimization, presents Parsons' three-tiered approach, summarizes the re- suits of'several sites to which the three-tiered LTMO approach has been applied, and discusses future WO rk.

EXAMPLE APPLICATIONS OF METHODS FOR DESIGNING,

EVALUATING AND OPTIMIZING MONITORING PROGRAMS

Although monitoring network design has been examined extensively in the past, most previous studies have addressed one of'two problems (Reed et al., 2000):

1. application of" numerical simulation and optimization techniques for screening monitoring plans for detection monitoring at landfills and hazardous-waste sites, and

2. application of ranking methods, including geostatistics, to augment or design monitoring networks for site-characterization purposes.

Storck et al. (1995) used a simulation approach to

examine ways to design and evaluate groundwater monitoring networks for leaking disposal facilities. A Monte Carlo simulator was used to generate a large number of equally-likely realizations of" a random hydraulic conductivity field and a contaminant source location. A numerical model simulat- ing groundwater flow and contaminant transport was used to generate a contaminant plume for each realization of the hydraulic-conductivity field. The re- suits of the transport simulations then were used as input to an optimization model, which generated

148 NOBEL AND ANTHONY

optimal trade--off curves among three conflicting objectives:

1. maximize probability ofcontaminant detection, 2. minimize cost of monitoring network (i.e., mini-

mum number of monitoring wells), and 3. minimize volume of contaminated groundwater.

The model was applied to a hypothetical scenario in order to examine the sensitivity of the trade-off curves to various model parameters.

Kelly (1996) applied a numerical model ofgroundwater flow and contaminant migration together with knowledge of locations of potential contaminant sources, to determine screened-interval elevations and locations for 75 monitoring wells in 35 clusters, in a network protecting the municipal wellfield of Independence, Missouri.

Dresel and Murray (1998) used a ranking approach to assist in the design of a groundwater monitoring network at the U.S. Department of Energyas Hartford site in Washington. A geostatistical model of existing plumes was used to generate a large number of realizations of contaminant distribution in groundwater at the facility. Analysis of the realizations provided a quantitative measure of the uncertainties in contaminant concentrations, and a measure of the probability that a cutoffvalue (e.g., a regulatory concentration standard) would be exceeded at any point. A metric based on uncertainty measures and declus- tering weights was developed to rank the relative value of each monitoring well in the network design. This metric was used, together with hydrogeologic and regulatory considerations in identifying candidate locations for inclusion in, or removal from the network.

Hudak et al. (1993) applied a ranking methodology to the design of a detection-monitoring network for the Butler County Municipal Landfill in southwest Ohio. A geographic information system (GIS) was used to assign relative weights to candidate monitoring locations on the basis of distance to possible contaminant sources, location relative to probably contaminant migration pathways, and distance to a potential receptor. The G IS application was found to be relatively straightforward to implement, was capable of addressing established regulatory policy, and could be used to address several monitoring objectives.

Chieniawski et al. (1995) used a simulation approach combined with a ranking approach to examine the problem of optimizing detection monitoring at a waste facility under conditions of uncertainty. A numerical model was used, together with stochastic realizations ofcontaminant transport, to generate numerous realizations of contaminant movement for use as input to a multi objective optimization model. The optimization model was solved using a genetic algorithm, and generated trade-off curves comparing the relative cost of a particular monitoring network design with the probability that the network could detect a leak.

The studies described above dealt primarily with detection monitoring and global approaches to the design of new monitoring networks. By contrast, few investigators have formally addressed the evaluation and optimization of long-term monitoring programs at sites having extensive pre-existing monitoring networks that were installed during site characterization. The primary goal of optimization efforts at such sites is to reduce sampling costs by eliminating data redundancy to the extent practi- cable. This type of optimization usually is not intended to identify locations for new monitoring wells, and it is assumed that the existing monitoring network sufficiently characterizes the concentrations and distribution of contaminants being monitored. It also is not intended for use in optimizing detection monitoring.

Reed et al. (2000) developed and applied a simulation approach for optimizing existing monitoring programs using a numerical model of groundwater flow and contaminant transport, several statistically- based plume-interpolation techniques, and a formal mathematical optimization model based on a genetic algorithm. The optimization approach was used to identify cost-effective sampling plans that accurately quantified the total mass of dissolved contaminant in groundwater. Application of the approach to the monitoring program at Hill Air Force Base (AFB) indicated that monitoring costs could be reduced by as much as 60 percent without significant loss in ac- curacy of mass estimates. Reed and Minsker (2004) and Reed et al. (2001 and 2003) extended this work using several different mathematical optimization algorithms to address multiobjective monitoring optimization problems.

Cameron and Hunter (2002) applied a spatial and temporal optimization algorithm known as

A THREE-TIERED APPROACH TO LONG TERM MONITORING PROGRAM OPTIMIZATION 149

the Geostatistical Tern poral/Spatial (GTS) Optimiza- tion Algorithm to the evaluation and optimization of two existing monitoring programs at the Massachusetts Military Reservation (MMR), Cape Cod, Massachusetts. The GTS algorithm is intended for use in optimizing long-term monitoring (LTM) networks using geostatistical methods, and was developed to ensure that only those monitoring data sufficient and necessary to support decisions crucial to monitoring programs are collected and analyzed. The algorithm uses geostatistical methods to optimize sampling frequency and to define the network ofessential sampling locations. The algorithm incorporates a decision pathway analysis that is separated into temporal and spatial (i.e., frequency and location) components, which are used to identify temporal and spatial redundancies in existing monitoring networks. The results of the temporal analysis applied to the monitoring programs at MMR indicated that sampling frequency could be reduced at most locations by40 to 70 percent. The results ofthe spatial analysis indicated that 109 of the 536 wells included in the two monitoring programs at MMR were spatially redundant, and could be removed from the programs. More recently, Cameron and Hunter (2004) applied the GTS algorithm to monitoring programs at three other sites, and confirmed that use of this optimization approach could generate savings

ranging from 30 percent to 63 percent of monitoring costs.

PARSONS' THREE-TIERED APPROACHml N ITIAL STE PS

The initial steps in the Parsons Three-Tiered Ap- proach involve obtaining available site information and screening it to determine if the site is a potential candidate for LTMO. Table 1 presents the types of data required to conduct a three-Tiered LTMO, and the purposes for which those data are collected. Data acquisition and database processing are often the most difficult and time-consuming steps in the LTMO evaluation. Ideally, historical analytical data and site map information should be available in electronic database and GIS format, respectively.

General characteristics of a site that is suited for a LTMO include:

Site characterization has been completed, and the spatial distribution of contaminants in the subsurface is well-characterized.

A long-term monitoring program has been established.

Data that fulfill the requirements in Table 1 are available in an accessible format (e.g., electronic database).

TABLE 1 Data used to characterizes the monitoring system, contaminant plume, and surrounding area over time and space that is needed for a three-liered LTMO assessment

Data needed Purpose

Required information Description of current monitoring program

Well locations/coordinates Historical COC analyses/results Configuration of potentiometric surface:

Groundwater flow direction, velocity and gradient

Hydrogeologic conditions

Well completion intervals/hydrogeologic zone

Cleanup goals/regulatory limits Locations of potential receptors/

compliance points Useful Information

Logistical/policy considerations Site features (roads, building,

rivers, property boundaries) Water levels through time Geochemical data

Establish baseline conditions, purpose of monitoring program and rationale for monitoring wells

Determine spatial distribution of monitoring points Define concentrations of COCs a in space and time; Confirm primary COCs Evaluate direction and rate of groundwater movement and contaminant

migration

Identify geologic or other controls on occurrence and movement of groundwater and dissolved COCs

Determine depth of sample collection in groundwater system and potential zones

Establish cleanup limits and areas of concern requiring monitoring Identify areas and/or migration directions of concern

Identify regulatory/public priorities and potential for program implementation Create spatial context for monitoring program; Develop base map of site

for reporting Identify dry wells; evaluate seasonal effects Determine natural attenuation parameters

a C O C _ c o n t a m i n a n t of c o n c e r n .

15o NOBEL AND ANTHONY

A long-term monitoring program has been established.

Concentration data are available for more than 10 wells (more than 30 wells is preferable), monitoring the same contaminant plume and completed in the same water-bearing unit or monitoring zone.

Analytical results are available from at least four historical sampling events over two years or more.

If a system (e.g., site remedy) is in place, it is not expected to change dramatically during the next few years.

The regulatory environment is flexible enough to consider program modifications.

Monitoring objectives establish the purpose(s) of monitoring, but often are poorly defined. Conse- quently, the stated monitoring-program objectives should be thoroughly evaluated during the initial stages of LTMO, and modified, if necessary, to reflect actual program objectives. Additionally, site personnel, regulators, and stakeholders should be involved throughout the LTMO process, as they can provide essential information about regulatory and policy issues and other qualitative information that drive monitoring priorities that might not be available from other information sources.

PARSONS THREE-TIERED APPROACHmM ETHODOLOGY

Case Study Background

The qualitative, temporal, and spatial analysis tiers are illustrated using the results ofan optimization of a monitoring program that included a subset of the existing wells at a case study site at the former Mather Air Force Base near Sacramento, California (Mather AFB). The case study results presented in this paper do not include the entire results for the Mather AFB study, and are not intended as a stand-alone case study, but rather as example results to demonstrate the Three-Tiered methodology. Complete Three- Tiered case studies are available elsewhere (USEPA, 2004b). Various types of contaminants were intro- duced to soil and groundwater in the subsurface at Mather AFB during the course of routine Base operations that included fuel storage and delivery, fire-fighting training, equipment maintenance, waste disposal, and other industrial activities conducted

throughout the operational history of the Base. The primary contaminants ofconcern (COCs) at the case study site include carbon tetrachloride (CC14), tetra- chloroethene (PC E), and trichloroethene (TCE). The monitoring program examined in the case study included 306 wells completed in multiple hydrogeologic zones. Some of the wells are not included in the current monitoring program; others are sampled at frequencies ranging from quarterly to biennially, for a total of 643 well-sampling events per year. The site monitoring program is reviewed annuallyby the remedial project managers team; the three-tiered approach was applied in 2003 to assess the existing program and to identify potential additional opportunities for optimization.

In this paper, results from sixteen wells completed in the upper (water-table) aquifer at Mather AFB are used to illustrate the results of each phase of the three-tiered evaluations. Examination of the results obtained from this arbitrary group of wells is not intended to present the comprehensive analysis of the case study site, but rather to demonstrate the methodology and types of results produced by the analysis.

Qualitative Evaluation

The first phase of the three-tiered evaluation involves examining the current groundwater monitoring program qualitatively, to identify potential opportunities for LTMO based on factors such as hy- drostratigraphy, locations of potential receptors with respect to the dissolved plume, and the direction(s) and rate(s) of contaminant migration. An effective groundwater monitoring program will provide information regarding contaminant plume migration and changes in chemical concentrations through time at appropriate locations, enabling decision-makers to verify that contaminants are not endangering potential receptors, and that remediation is occurring at rates sufficient to achieve remedial action objectives within a reasonable time frame. The design of the monitoring program should therefore include con- sideration of existing receptor exposure pathways, as well as exposure pathways arising from potential future use of the groundwater.

Performance monitoring wells located upgradient, within, and immediatelydowngradient from a plume provide a means of evaluating the effectiveness of a groundwater remedy relative to performance criteria.


Long-term monitoring of these wells also provides information about migration of the plume and temporal trends in chemical concentrations. Groundwa- ter monitoring wells located downgradient from the leading edge of a plume ("sentry wells') are used to evaluate possible changes in the extent of the plume and, if warranted, to trigger a contingency response action if contaminants are detected.

Primary factors to consider when qualitatively evaluating a groundwater monitoring program for potential optimization opportunities include at a minimum:

Aquifer heterogeneity, Types of contaminants, Distance to potential receptor exposure points, Groundwater seepage velocity, Potential surface-water impacts, and The effects of the remediation system.

These factors will influence the locations and spacing of monitoring points and the sampling frequency. Typically, the greater the seepage velocity and the shorter the distance to receptor exposure points, the m o r e frequently groundwater sampling should be conducted.

The qualitative evaluation considers multiple factors in developing recommendations for continua- tion or cessation ofgroundwater monitoring at each well. In some cases, a recommendation is made to continue monitoring a particular well, but at a reduced frequency. Typical factors considered in developing recommendations to retain a well in, or remove a well from, a LTM program are summarized in Table 2. Typical factors considered in developing recommendations for monitoring frequency are

summarized in Table 3. Each well in the monitoring network is evaluated based on the logic presented in Tables 2 and 3, and a recommendation is made to remove the well from the monitoring program, or to retain well while continuing to sample at a specified frequency. (A recommendation to discontinue monitoring at a particular well based on the information reviewed does not necessarily constitute a recommendation to physically abandon the well. A change in site conditions might warrant resumption of monitoring at some time in the future at wells that are not currently recommended for continued sampling.) Table 4 presents the qualitative results and rationale for a subset of wells at the case study site.

Temporal Evaluation One of the most important purposes of long-

term monitoring is to confirm that the contaminant plume is behaving as predicted. Temporal data (chemical concentrations measured at different points in time) can be examined graphically, or using statistical tests, to evaluate dissolved-contaminant plume stability. If removal of chemical mass is occurring in the subsurface as a consequence ofattenu- ation processes or operation ofa remediation system, mass removal will be apparent as a decrease in chemical concentrations through time at a particular sampling location, as a decrease in chemical concentrations with increasing distance from chemical source areas, and/or as a change in the suite of chemicals through time or with increasing migration distance.

Temporal chemical-concentration data can be evaluated by plotting contaminant concentrations through time for individual monitoring wells, or by plotting contaminant concentrations versus

TABLE 2 Decision logic used in the qualitative evaluation phase to determine whether to retain or exclude a well from a monitoring network

Reasons for retaining a welt in monitoring network Reasons for removing a well from monitoring network

Well is needed to further characterize the site or monitor changes in contaminant concentrations through time

Well is important for defining the lateral or vertical extent of contaminants.

Well is needed to monitor water quality at compliance point or receptor exposure point (e.g., water supply well)

Well is important for defining background water quality

Well provides spatially redundant information with a neighboring well (e.g., same constituents, and/or short distance between wells)

Well has been dry for more than 2 years a

Contaminant concentrations are consistently below laboratory detection limits or cleanup goals

Well is completed in same water-bearing zone as nearby well(s)

aPerlochc water-level monitoring slqoulct be performect in c~ry wells to confirm that the upper Dounctary of the saturated zone remains below tlqe well ~reen. If the well becomes re-wettect, tlqen ~ts inclusion m tlqe monitoring procjram slqoulc~ De evaluatect. A well that has been dry for more than two years shoulct be reDlacec~ w~th a new well screened at a c~eeDer interval ~f grounctwater monitor ing at that location ~s deemed to be requ~recL


TABLE 3 Decision logic used in the qualitative evaluation phase to determine whether to modify the monitoring frequency at a well

Reasons for increasing sampling frequency Reasons for decreasing sampling frequency

Groundwater velocity is high Change in contaminant concentration would

significantly alter a decision or course of action Well is necessary to monitor source area or operating

remedial system Cannot predict whether concentrations will change

significantly over time

Groundwater velocity is low Change in contaminant concentration would not significantly alter

a decision or course of action Well is distal from source area and remedial system

Concentrations are not expected to change significantly over time. or contaminant levels have been below groundwater cleanup objectives for some prescribed period of time

downgradient distance from the contaminant source for several wells along a particular groundwater flowpath, over several monitoring events. Plotting tern poral concentration data is recommended for any analysis of plume stability (Wiedemeier and Haas, 2000); however, visual identification of trends in plotted data may be a subjective process, particularly if(as is likely) the concentration data do not exhibit a uni- form trend, but are variable through time, as illustrated in Figure 1.

The possibility of arriving at incorrect conclusions regarding plume stability on the basis of visual ex-

amination of temporal concentration data can be reduced by examining temporal trends in chemical concentrations using various statistical procedures, including regression analyses and the Mann-Kendall test for trends. The Mann-Kendall nonparametric test (Gibbons, 1994) is well-suited for evaluation of environmental data because the sample size can be small (as few as four data points), no assumptions are made regarding the underlying statistical distribution ofthe data, and the test can be adapted to account for seasonal variations in the data. The Mann-Kendall test involved calculating a trend statistic S at each

TABLE 4 Illustration of qualitative evaluation results that shows the pre-optimization (i.e., current) sampling frequency, qualitative analysis recommendation (exclude, retain, recommended frequency) and the rationale for the recommendation for each well

Qualitative Current analysis recommendation

sampling Weft ID frequency Exclude Retain Frequency Rationale

MAFB-006 Not sampled ,/ MAFB-033 Semi-annual , / Semi-annual MAFB-037 Annual , / Annual MAFB-047 Annual , / Annual MAFB-048 Annual , / Annual MAFB-085 Semi-annual , / Semi-annual MAFB-086 Quarterly ,/ Semi-annual MAFB-087 Quarterly ,/ Semi-annual

MAFB-088 Semi-annual , /

MAFB-089 Annual , / Annual MAFB-090 Semi-annual , / Semi-annual MAFB-091 Not sampled ,/

MAFB-092 Annual , / Annual MAFB-093 Not sampled ,/

MAFB-095 Quarterly ,/ Quarterly

MAFB-096 Annual , / Annual

Upgradient well. no historical COO detections. Monitors TCE hot spot a. Monitors water quality near southeastern plume boundary. Monitors water quality downgradient of hot spots. Monitors water quality near western plume boundary. Montiors water quality downgradient. Monitors water quality downgradient of extraction wells. Defines extent of hot spot. monitors water near MBS

EW6ABu. Reduce frequency due to COCs consistently under MCLs.

Redundant with well MAFB-206. which has higher TCE concentration.

Defines extent of plume. Monitors potential increase in TCE above MCL. Spatially redundant. MAFB-092 is sampled annually.

COCs not detected historically. Defines plume extent to west. Available in reserve for future sampling if MB-3 becomes

operational. Monitors hot spot and effects of extraction system near

MBS EW 1A (declining PCE concentrations). Reduce sampling frequency due to COCs consistently

under MCLs. Hydraulically upgradient of plume.

a"Hot sDots" are deflnec~ to be areas where contaminant concentrations exceed the cleanup level by a factor of 10 or more.


Increasing Trend

Confidence Factor High

Decreasing Trend No Trend

Confidence Factor Low

>. Variation High Variation Low

FIGURE 1 Conceptual representation of trends, confidence factors, and levels of variation to illustrate potential concentrations measured over time.

well using Equation 1.

S = ~ sign(Xj - Xk) k l j k+l

(2)

where

sign(X i - Xk) = 1 if (X i -- Xk) > 0 sign(X i -- Xk) = 0 if (X i -- Xk) = 0 sign(Xj - Xk) = -- 1 if (Xj -- Xk) < 0 Xi = concentration at time i n = total number of observations

A positive Mann-Kendall trend statistic, S, indicates an increasing trend, a negative S indicates a decreasing trend and a S of zero indicates neither an increasing nor decrease trend. For example, assume samples from a well during four consecutive monitoring events that measure 50, 30, and 10 and 10 respectively. This corresponds to an S value o f - 5 , which indicates a decreasing trend. A test is applied

to assess if the observed trend is statistically significant or could potentially be attributed to random variability. A statistically-significant negative trend or a positive trend provides statistical confirmation of temporal trends that may have been identified visu- ally from plotted data, and the significance of the test depends on the sample size. An exact test is applied for small sample sizes (less than 10) and an approxi- mate method is used for sample sizes greater than 10. Both tests are described in detail in Gilbert (1987). In this analysis a confidence level of 95% was used to evaluate whether a temporal trend is exhibited by contaminant concentrations detected through time in samples from an individual well. Note that the re- suits of a Mann-Kendall test may be misleading if the temporal data from an individual monitoring location are serially correlated (Gibbons, 1994). Usually, however, serial correlation of temporal data is of concern only if sampling is conducted more frequently than once per quarter (E1-Shaarawi and Niculescu, 1992).


The relative value of information obtained from periodic monitoring at a particular monitoring well can be evaluated by considering the location of the well with respect to the dissolved contaminant plume and potential receptor exposure points, and the presence or absence of temporal trends in contaminant concentrations in samples collected from the well. The degree to which the amount and quality of information that can be obtained at a particular monitoring point serve the two primary (i.e., temporal and spatial) objectives of monitoring must be considered in this evaluation. For example, the continued non-detection of a target contaminant in groundwater at a particular monitoring location provides no information about temporal trends in contaminant concentrations at that location, or about the extent to which contaminant migration is occurring, un- less the monitoring location lies along a groundwater flowpath between a contaminant source and a potential receptor exposure point (e.g., downgradient of a known contaminant plume). Therefore, a monitoring well having a history of contaminant concentrations below detection limits may be providing little or no useful information, depending on its location.

A trend of increasing contaminant concentrations in groundwater at a location between a contaminant source and a potential receptor exposure point may represent information critical in evaluating whether contaminants are migrating to the exposure point, thereby completing an exposure pathway. Identifi- cation of a trend of decreasing contaminant concentrations at the same location may be useful in

evaluating decreases in the areal extent of dissolved contaminants, but does not represent information that is critical to the protection of a potential receptor. Similarly, a trend of decreasing contaminant concentrations in groundwater near a contaminant source may represent important information regarding the progress ofremediation near, and downgradient from the source. By contrast, the absence of a statistically significant (as defined by the Mann-Kendall test at a 95% confidence level) temporal trend in contaminant concentrations at a particular location within or downgradient from a plume indicates that virtually no additional information can be obtained by frequent monitoring of groundwater at that location, in that the results of continued monitoring through time are likely to fall within the his- toric range of concentrations that have already been detected (Figure 2). Continued monitoring at locations where no temporal trend in contaminant concentrations is present serves merely to confirm the results of previous monitoring activities at that location.

The temporal trends and relative location of wells can be examined to determine if a well should be retained, excluded, or continue in the program with reduced sampling. Figure 3 presents a flowchart demonstrating the methodology for utilizing trend results to draw these conclusions. Trend results are obtained for multiple chemicals and recommendations based on the most conservative result (e.g., recommend retaining a well if any of the trend analyses for individual COGs result in a recommendation to retain the well). Table 5 presents the temporal trend

OC ~ • I Historical Results

I Likely Future . . . . . . . . . . . . . . I___ Results

t - o U

Time

FIGURE 2 Conceptual representation of continued monitoring at a location where no temporal trend in concentration is present. Where no trend exists, future results can be predicted to be within the range of historical results.


Yes ~ -SExclude/Reduce> Yes -~,... Frequency

- - Nc - / Retain ~'~ No Y

~ ~ Y e s ]d/~Exclude/Reduce "x ~ Y e s . Y e s >~x,,,. Frequency J

No ~ Retain >

!o ~ Y e s < Retain >

" ~ NC _S Exclude/Reduce"~ "k Frequency J

Yes

No

/ Retain

\ roquonc J

Retain >

_f Exclude/Reduce ~'~ -~.... Frequency J

*If increasing trend occurs in a source area that is undergoing remediation, the well should be retained.

FIGURE 3 Temporal evaluation flow chart that shows how a well is recommended for retention, exclusion or change in monitoring frequency based on the well's calculated Mann-Kendall statistical trend result, location in the plume, and recent sampling results.

results and resulting LTMO recommendations for a subset of the case study wells. These recommendations to retain or exclude/reduce sampling are based solely on the value of the temporal information. The

temporal evaluation results are ultimately combined with the qualitative and spatial evaluation results in the summary evaluation to develop final recommendations for each well.


TABLE 5 Illustration of temporal evaluation results that shows tile Mann-Kendall trend results for tile case study contaminants of concern, along with the recommendation to exclude/reduce or retain tile associated well in tile monitoring network under evaluation, and the associated rationale

Exclude/ Well ID CCL4 PCE TCE reduce Retain Rationale

MAFB-006 ND ND ND , /

MAFB-033 No trend Decreasing No trend , /

MAFB-037 ND ND No trend , / MAFB-047 ND Decreasing Decreasing , /

MAFB-048 ND No trend Increasing MAFB-085 ND <PQL ND MAFB-086 ND Decreasing Decreasing , / MAFB-087 ND No trend No trend , /

MAFB-088 No trend No trend No trend , /

MAFB-08g No trend ND No trend , /

MAFB-090 <PQL No trend Decreasing , /

MAFB-091 <4meas <4meas <4meas Not analyzed MAFB-0g2 ND ND ND , / MAFB-0g3 ND ND <PQL , /

MAFB-095 No trend Decreasing <PQL MAFB-0g6 No trend <PQL <PQL , /

, / , /

, /

No COCs a detected in monitoring history (1/96-10/99) upgradient.

Decreasing PCE < MCL; no-trend downgradient of source area.

ND or no-trend cross-gradient (TCE < 2 fAg/L). Decreasing trends in plume downgradient.

COCs < MCLs b, limited temporal information.

TCE < 1 fAg/L; increasing cross-gradient. Downgradient senty well. Decreasing downgradient. ND or no trend ,,~:: MCL within plume; limited

temporal information. No trend in plume; limited temporal

information. No trend/ND in plume; limited temporal

information. Decreasing/no trend downgradient of source

area.

ND cross-gradient of plume. All COCs ND or <PQL in monitoring history

(1/96-10/00) upgradient. Decreasing PCE in source area. All COCs < 1 #g/L in monitoring history

upgradient.

ND - Constituent has not been detected during history of monitoring at indicated well. No Trend - No statistically significant temporal trenc~ in concentrations. Increasing _ Statistically s~gn~flcant ~ncreasmg trencJ ~n concentrations (95% conflc~ence level). Decreasing _ Statistically s~gn~flcant c~ecreas~ng trencJ ~n concentrations (95% confidence level). <PQL - Concentrations cons~stenVy below I~ract~cal c~uant~tat~on hm~t. <4meas _ Fewer than four analytical results; Mann-Kendall trencJ not analyzed. aCOC - Contaminant of Concern. bMCL - Maximum Contaminant Level.

SPATIAL STATISTICAL EVALUATION

Spatial statistical techniques can be applied to the design and evaluation of groundwater monitoring programs to assess the quality of information generated during monitoring, and to evaluate the spatial distribution of all components of a monitoring network. In the third tier of the three-tiered LTMO approach, a spatial statistical analysis utilizing kriging error prediction is applied to determine the relative amount of spatial information contributed by each monitoring well. Geosta.listics, or the Theory of Regionalized Variables (Clark, 1987; Rock, 1988; American Society of Civil Engineers Task Commit- tee on Geostatistical Techniques in Hydrology, 1990a and 1990b), is concerned with variables having values

dependent on location, and which are continuous in space, but which vary in a manner too complex for simple mathematical description. Geostatistics is based on the premise that the differences in values of a spatial variable depend only on the distances between sampling locations, and the relative orien- tations of sampling locations that is, the values of a variable measured at two locations that are spatially "close together" will be more similar than values of that variable measured at two locations that are "far apart."

Ideally, application of geostatistical methods to the results of a groundwater monitoring program could be used to estimate COG concentrations at every point within the dissolved contaminant plume, and also could be used to generate estimates of the "error," or uncertainty, associated with each


estimated concentration value. Thus, the monitoring program could be optimized by using available information to identify those areas having the greatest uncertainty associated with the estimated plume extent and configuration. Conversely, sampling points could be successively eliminated from simulations, and the resulting uncertainty examined, to evaluate if significant loss of information (represented by increasing error or uncertainty in estimated chemical concentrations) occurs as the number of sam- piing locations is reduced. Repeated application of geostatistical estimating techniques, using tentatively identified sampling locations, then could be used to generate a sampling program that would provide an acceptable level of uncertainty regarding the distribution of COCs with the minimum possible number of samples collected. Furthermore, application of geostatistical methods can provide unbiased rep- resentations of the distribution of COCs at different locations in the subsurface, enabling the extent of COCs to be evaluated more precisely.

Fundamental to geostatistics is the concept of semivariance, which is a measure of the spatial depen- dence between sample variables (e.g., chemical concentrations) in a specified direction. Semivariance is defined for a constant spacing between samples (b)

by:

1 y(h) = ~ ~ [ g ( x ) - a ( x -q'- h)] 2 (1)

Where:

},(h) = semivariance calculated for all samples at a distance h from each other;

g(x) = value of the variable in sample at location x; g(x + h) = value of the variable in sample at a dis-

tance h from sample at location x; n = number of samples in which the variable has

been determined.

Semivariograms (plots o f / (h ) versus b) are a means of depicting graphically the range of distances over which, and the degree to which, sample values at a given point are related to sample values at adjacent, or nearby, points, and conversely, indicate how close together sample points must be for a value determined at one point to be useful in predicting unknown values at other points. An idealized semivariogram is shown in Figure 4. For b = 0, for example, a sample is being compared with itself, so normally/(0) = 0 (the semivariance at a spacing of zero, is zero), except where a so-called nugget effect is

> o

L )

>

3500

3000

2500

2000

1500

1000

500

0

m

Nugget t

Ranee IP

• m

Spherical model

Nugget: 1,500 mg 2/L z Range: 1,000 ft

Sill: 1,500 nag 21L 2

I I I I I

0 500 1000 1500 2000 2500 3000 3500 Distance (ft)

FIGURE 4 Idealized semivariogram model. A semivariogram is a function of the distance and direction separating two locations, used to quantify autocorrelation. The variogram is defined as the variance of the difference between two variables at two locations. The range is a parameter of a covariance or semivariogram model that represents a distance beyond which there is little or no autocorrelation among variables. The sill is the parameter that represents a value that the semivariogram tends to when distances get very large. The nugget represents independent error, measurement error and/or microscale variation at spatial scales that are too fine to detect.


present, which implies that sample values are highly variable at distances less than the sampling interval. As the distance between samples increases, sample values become less and less closely related, and the semivariance, therefore, increases, until a sill is even- tually reached, where y(b) equals the overall variance (i.e., the variance around the average value). The sill is reached at a sample spacing called the range of influence, beyond which sample values are not related. Only values between points at spacings less than the range of influence can be predicted; but within that distance, the semivariogram provides the proper weightings, which apply to sample values separated by different distances.

Semivariograms are a means o f depicting graphically the range of distances over which, and the degree to which, sample values at a given point are related to sample values at adjacent, or nearby, points, and conversely, indicate how close together sample points must be for a value determined at one point to be useful in predicting unknown values at other points. When a semivariogram is calculated for a variable over an area (e.g., concentrations of a C O C at every point at a site undergoing LTMO) an irregu- lar spread ofpoin ts across the semivariogram plot is the usual result (Rock, 1988). One of the most subjective tasks of the geostatistical analysis is to iden- tiff] a continuous, theoretical semivariogram model that most closely follows the real data. Although this procedure can be automated, using least-squares methods, fitting a theoretical model to calculated semivariance points usuallyis accomplished by trial- and-error, rather than by a formal statistical procedure (Clark, 1987; Rock, 1988).

Before a semivariogram can be developed for a given site, appropriate input data must be selected. In the three-tiered approach, an indicator chemical is selected as a surrogate to represent the presence of multiple COCs at the site. In the case study example, a "Total Weighted C O C " value that incorporated all o f the site COCs was used as the indicator chemical for the geostatistical analysis-at each point the concentrations o f PCE, TCE, and CCL4 were divided by their respective cleanup-level concentrations, and the values were summed to create the surrogate raw input data that was used in the analysis, as shown in Equation 3.

COCi Total Weighted COCi = ~ COCMcL

all COGs (3)

Where

i = monitoring point ofinterest

COCi = measurement o fcon taminan t of

concern at monitoring point i

COCMcL = C O C cleanup level (e.g.,

TCEMcL = 5/ag/L)

At the case study site, development of this Total Weighted C O C indicator surrogate values provided C O C concentration data over a larger spatial area than would have been the case for any individual C O C ; and the weighting factor accounted for differences in cleanup levels among COCs (e.g., a CCL4 concentration of 1/ ,g/L, [cleanup level = 0.5/,g/L] is not equivalent to a TCE concentration of 1/ag/L [cleanup level = 5 ~g/L]). At other sites where a single C O C has the greatest range in concentrations and/or the most extensive spatial distribution it can be used alone as the indicator chemical for the geostatistical analysis. The raw indicator concentrations values are typically transformed for use in the semivariogram development.

Because the kriging evaluation examines a two- dimensional spatial "snapshot" of the C O C concentrations, the most recently collected analytical data available are used in the kriging evaluation. Fur- thermore, different kriging analyses must be conducted for each different hydrogeologic zone. In the case study example, different kriging evaluations were conducted for each of three monitoring zones. Wells that were not sampled during the most recent sampling event, or that are completed at a depth significantly different from the other wells in the same monitoring zone, were not included in the geostatistical evaluation. In addition, C O C concentration data from extraction wells are not appropriate for use in a kriging analysis because they represent C O C concentrations averaged over the volume within the well's capture zone, and thus are not point-specific, nor temporally discrete; therefore, concentration data from extraction wells were not included in the evaluation.

Once the appropriate wells and indicator contaminant had been selected, the commercially available geostatistical software package Geostatistical Analyst TM (an extension to the ArcView ~ geographic information system [GIS] software package; Envi- ronmental Systems Research Institute, Inc. [ESRI], 2001) was used to develop a semivariogram model


depicting the spatial variation of the contaminant concentrations in groundwater. Considerable scatter of the data was apparent during fitting of the models to the Total Weighted COC indicator concentration data. Several data transformations (including a log- normal transformation) were applied in an attempt to obtain a representative semivariogram model. Ul- timately, the concentration data were transformed to "rank statistics," in which, for example, the 55 wells in the upper monitoring zone were ranked from 1 (lowest indicator concentration) to 55 (highest indicator concentration) on the basis of the indicator concentration most recently detected at each well. Tie values were assigned the median rank of the set of ranked values, for example, if five wells had nondetected concentrations, they each would be ranked "three," the median of the set of ranks-one, two, three, four, five. Transformations of this type can be less sensitive to outliers, skewed distributions, or clustered data than semivariograms based on raw concentration values, and thus may enable recognition and description of the underlying spatial structure of the data in cases where ordinary data are too "noisy" (Henley, 1981). The Weighted Total COC rank statistics then were used to develop semivariograms that most reason- ably modeled the spatial distribution of the data in the three zones. Anisotropy also was incorporated into the models to adjust for the directional influence of groundwater movement from northeast to southwest.

After the semivariogram models were developed, they were used in a kriging system as implemented by

2

,] ,

the Geostatistical Analyst TM software package (ESRI, 2001) to develop two-dimensional kriging realizations (estimates of the spatial distribution of Total Weighted COC indicator concentrations in groundwater in the case study area), and to calculate the associated kriging prediction standard errors. The median kriging standard deviation was obtained from the standard errors calculated using the entire monitoring network for each zone (e.g., the 55 wells screened in Zone WT/A). Next, each of the wells was sequen- tially removed from the network, and for each re- suiting monitoring network configuration, a kriging realization was completed using the Total COC indicator concentration rankings of the remaining 54 wells. The "missing-well" monitoring network realizations were used to calculate prediction standard errors, and the median kriging standard deviations were obtained for each "missing-well" realization and compared with the median kriging standard deviation for the "base-case" realization (obtained using the complete 55-well monitoring network), as a means of evaluating the amount of information loss (as indicated by increases in kriging error) resulting from the use of fewer monitoring points.

Figure 5 illustrates the spatial-evaluation procedure by showing kriging prediction standard-error maps for three kriging realizations. Each map shows the predicted standard error associated with a realization generated using a particular group of wells, including a "base case" realization (generated using Total Weighted COC indicator concentration data for all wells) and two different "missing well"

A) Basecase

rc~/17B

: :

MW13C

B) More Relative Information C) Less Relative Information

FIGURE 5 Illustration of the impact of missing wells on predicted standard error. Greater spatial uncertainty is represented by darker shading. (A) Basecase scenario in which all wells are included in the predicted standard error map. In scenario (B), well MW17B was removed from the monitoring network, and results in relatively greater spatial uncertainty (i.e., more dark shading) than scenario (C) in which well MW13C is removed from the network. Thus, Well MW17B contributes more relative spatial information to the network based on predicted standard error results.


realizations. Lighter colors represent areas within which spatial uncertainty is relatively lower, and darker colors represent areas within which spatial uncertainty is relatively greater; regions in the immedi- ate vicinity ofwells (i.e., data points)have the lowest associated uncertainty. Map A on Figure 5 shows the predicted standard error map For the "base-case" realization in which all wells are included. Map B shows the realization in which well MW13C was removed from the monitoring network, and Map C shows the realization in which well MC17B was removed. In- terpretation of Figure 5 indicates that when a well is removed from the network, the predicted standard error in the vicinity of the missing well increases (as indicated by a darkening of the shading in the vicinity of that well). Ifa "removed" (missing) well is in an area with several other wells nearby (e.g., well MW13C; Map B on Figure 5), the predicted standard error may not increase as much as ira well (e.g., MW17B; Map C)is removed From an area with fewer surrounding wells.

Based on the kriging evaluation, each well was assigned a "test statistic" describing the relative value of spatial information obtained from the well, calculated From the ratio of the median "missing well" kriging error to the median "base case" error. If removal of a particular well from the monitoring network caused very little change in the resulting median kriging standard deviation, the test statistic was equal to 1.0, and that well was regarded as contribut- ing only a limited amount of information to the LTM program. Likewise, if removal of" a well from the monitoring network produced greater increases in the kriging standard deviation (greater than 1 percent), this was regarded as an indication that the well contributes a relatively greater amount of information, and is relatively more important to the monitoring network. At the conclusion of" the kriging realizations for wells in the WT/A zone, each well was ranked From 1 (providing the least information) to 55 (the total number of wells included in the analysis of the WT/A zone) (providing the most information), based on the amount of information (as measured by changes in median kriging standard deviation) the well contributed toward describing the spatial distribution of'the concentrations of the Total Weighted COC indicator. Wells providing the least amount ofinformation represent possible candidates for removal from the monitoring program. The wells that contribute the most amount of relative informa-

TABLE 6 Illustration of spatial evaluation results that shows the kriging test statistic calculated during the geostatistical evaluation, the associated relative ranking of the wells being examined, along with the recommendation to exclude or retain the wells based on their relative value of spatial information. High ranking wells are recommended for retention and low ranking wells are recommended for exclusion. This illustration shows a subset of the 55 wells considered in the analysis for the upper zone

Kriging Test Kriging Well 113 a Statistic b ranking c Exclude Retain

MAFB-033 0.99997 8 , / MAFB-037 1.01295 53 MAFB-047 0.99985 3 , / MAFB-048 1.00687 49 MAFB-085 1.01038 51 MAFB-086 1.00301 43 MAF B-087 1.01130 52 MAFB-088 0.99999 12.5" , / MAFB-090 1.00034 33 MAFB-092 1.00304 44 MAFB-095 0.99997 8 , / MAFB-096 0.98347 1 , /

m d

m d

m d

, /

V , /

, /

aWells not sampled [Jurlng the most recent sampling event were e×- cludec~ from the analysis. bRatlo of the median "missing wel l " predicted standard error to median "base case" error. r l - least relative amount of information; 55 - most relative amount of information (55 wells includecJ in the geostatlStlcal zone analysis.) dWell in the " intermediate" range ana received no recommendation. e l ie values receive the mec~lan ranking of the set.

tion are candidates for continued monitoring. Those wells that result in a measurable change of less than one percent are considered to provide an "intermediate" amount ofinformation, and optimization recommendations were not generated for these wells on the basis of the spatial evaluation. The geostatistical analysis provides in relative value rankings only for those wells included in the semivariogram data set recommendations were not generated For wells that were not sampled in the most recent sampling event, on the basis of the spatial evaluation. An example of the geostatistical analysis results for select wells from the case study evaluation is presented in Table 6.

Combined Evaluation Summary

Wells evaluated in the LTMO analysis are examined using qualitative hydrogeologic and extraction- system information, temporal statistical techniques, and spatial statistics. As each tier of'the evaluation is performed, monitoring points that provide relatively greater amounts of information regarding the occurrence and distribution oFCOCs in groundwater are identified, and distinguished From those monitoring


points that provide relatively lesser amounts of information. Once the three evaluations have been completed, the results are combined to generate a refined monitoring program that potentially could provide information sufficient to address the primary objectives of monitoring, at reduced cost. Monitoring wells not retained in the refined monitoring network could be removed from the monitoring program with relatively little loss of information. In general, the temporal and spatial evaluations are used to support and refine the qualitative evaluation; this allows professional judgment and site-specific knowledge drive the recommendations, but allows for valida- tion and consistency based on the objective statistical results.

The results of the evaluations were combined and summarized in accordance with the following decision logic, presented in Figure 6:

1. Each well retained in the monitoring network on the basis of the qualitative hydrogeologic evaluation was recommended to be retained in the refined monitoring program.

2. Those wells recommended for removal from the monitoring program on the basis of all three eval-

.

.

uations, or on the basis of the qualitative and temporal evaluations (with no recommendation resulting from the spatial evaluation) should be removed from the monitoring program. If a well was recommended for removal based on the qualitative evaluation and was recommended for retention based on the temporal or spatial evaluation, the final recommendation was based on a case-by-case review of well information. If a well was recommended for retention based on the qualitative evaluation and was recommended for removal based on the temporal and spatial evaluation, the recommended sampling frequency was based on a case-by-case review of well information.

Table 7 presents the results of the three evaluations and the final summary recommendation for the select case study wells. Often, the three evaluation tiers have consistent recommendations (i.e., all recommend retention or all recommend exclusion). In some cases, however, the qualitative evaluation recommendation differs from the temporal and/or spatial recommendation and is reviewed on a case-by-case basis, as shown in Criterion three and

No

No

'Exctude Well from Future~ Sampling j

v • ,

No

o ,,oo g (Case-By-Case) j

Yes f~Case.By.Case Review)

FIGURE 6 Flowchart that demonstrates how final monitoring network optimization recommendations are made based on the results of the qualitative, temporal and spatial evaluation results.


TABLE 7 Illustration of Three-Tiered summary evaluation results that shows the pre-optimized monitoring frequency, results obtained from the qualitative, teml~ral and spatial evaluation for each well. The three tier results determine the final recommendation to exclude, retain or change the sampling frequency of each well in the monitoring network

Qualitative Temporal Spatial Final Current evaluation e v a l u a t i o n e v a l u a t i o n recommendation

sampling Weft ID frequency E R E/Red R E R E R Frequency

MAFB-006 Not sampled , / , / NA , / - - MAFB-033 Semi-annual , / , / , / , / Annual MAFB-037 Annual , / , / , / , / Biennial MAFB-047 Annual , / , / , / , / Biennial MAFB-048 Annual , / , / , / , / Annual MAFB-085 Semi-annual , / , / , / , / Semi-annual MAFB-086 Quarterly , / , / NR , / Semi-annual MAFB-087 Quarterly , / , / , / , / Semi-annual MAFB-088 Semi-annual , / , / , / , / - - MAFB-089 Annual , / , / NA , / Annual MAFB-090 Semi-annual , / , / - - - - , / Annual MAFB-0gl Not sampled , / NA NA , / - - MAFB-092 Annual , / , / - - - - , / Biennial MAFB-093 Not sampled , / , / NA , / - - MAFB-095 Quarterly , / , / , / , / Quarterly MAFB-096 Annual , / , / , / , / Annual

E - ExrJude. R - Retain. E/Red - Excluc~e or Reduce. NA - Well not analyzed for g~ven analys~s. NR - No recommenctat~on for sDat~al analys~s; Intermediate rankea well.

four above. For example, a well could be recommended for exclusion in the qualitative and spatial evaluations because it is determined to be redundant with other wells but recommended for retention in the temporal evaluation due to increasing concentrations (Criterion three); in this situation, if it was determined that the other wells in the area are also mea- suring the increasing concentrations, this well would likely be recommended for exclusion from the monitoring network based on the summary evaluation. On the other hand, if none of the surrounding wells were demonstrating the increasing trend in chemical concentration over time then this well would be retained in the monitoring network.

Additionally, ira well is recommended for retention based on professional judgments, but is not shown to have significant temporal or spatial evaluation (Criterion four) that well would be a likely candidate for a reduction in monitoring frequency. For example, continued annual sampling of cross-gradient well Mather AFB-037 was recommended as a result of the qualitative evaluation. However, the results of the temporal evaluation indicated that no temporal trend was present in the trace-level TCE concentrations detected in this well after several years of monitoring, and there have been no historical detections

ofeither PCE or CC14 in groundwater samples from this well. The recommended sampling frequency for this well therefore was reduced from annual to biennial based on all of the available information, which indicates that the results ofcontinued monitoring of this well through time are likely to fall within the his- toric range o£ concentrations that already have been detected. Therefore, the value of information obtained from this well on a more-frequent basis is relatively low. Furthermore, in several cases the results of the temporal and/or spatial evaluation support the qualitative recommendation to reduce sampling frequency. The final recommendations also incorpo- rate the comments and requirements of regulatory- agency personnel and other stakeholders who reviewed the analysis.

At the conclusion of the LTMO analysis at the case-study site, recommendations were generated to add three formerly sampled wells not currently in the monitoring program, and to exclude eighteen wells from the program; recommendations also were generated regarding modifications to the sampling schedule frequencies. The recommendation to add three wells to the monitoring program illustrates the fact that the Three-Tiered LTMO evaluates existing monitoring network wells for both potential


exclusion (for those wells currently sampled) and for re-inclusion (for those wells in the area of interest that are not currently sampled). The added wells were recommended for inclusion based on their historical data. This refined monitoring program would result in an average of 559.5 well-sampling events per year (where 0.5 events/year represents a well sampled biannually), as compared with 643 well-sampling events per year that occur in the current monitoring program. Implementing these recommendations for optimizing the LTM monitoring program at the case-study site would reduce the number of well- sampling events per year by approximately 13 percent. Although the monitoring program examined in this case study, at 306 wells and 643 well-sampling events per year, is among the largest that has been evaluated to date using the three-tiered approach, the evaluation produced a relatively lower reduction in recommended well-sampling events than have evaluations at other sites. This may be a result of the ongoing internal optimization efforts conducted annually at Mather AFB, and may also reflect input from regulatory personnel and other stakeholders who preferred more conservative sampling frequencies (more frequent sampling than recommended based on the results of the three-tiered evaluation).

THREE-TIERED LTMO APPLICATIONS AND FUTURE WORK

As of April 2004, the Three-Tiered LTMO approach has been applied to evaluate monitoring programs at eighteen sites. The monitoring programs evaluated with the three-tiered method have ranged in size from programs in which 10 wells are sampled, to programs in which more than 300 wells are sampled. COGs addressed by these programs primarily have included fuel constituents and volatile organic compounds. Application of the three-tiered approach has produced reductions in well-sampling events ranging from 13 percent to 83 percent per year. On average at all of the sites analyzed, application of the three-tiered approach has resulted in reductions ofover one-third of the sampling events per year. The results are highly dependent on the site conditions. Typically, application of the approach at sites where monitoring is conducted relatively frequently (e.g., quarterly) and/or to programs that have not been evaluated in several years resulted in more signifi-

cant optimization opportunities. Although there are fewer opportunities for program optimization at sites with a smaller number ofwells, significant optimization opportunities can still be identified on a relative scale (i.e., potentially-significant reductions in sampling events per year). Additionally, although the fo- cus of three-tiered LTMO applications has been primarily on optimizing existing LTM programs, it also has been applied successfully to identify optimal location(s) at which to add new wells to a program, and also has been used to optimize monitoring programs for soil-vapor extraction systems.

The three-tiered LTMO method is an evolving approach that is being continuously refined with additional enhancements, review, and exposure. The most significant future enhancement to the approach will be automation ofthe spatial evaluation. This will enable multiple chemicals and time periods to be incorporated into the spatial evaluation, if desired, in place ofthe "indicator COC single time-period snapshot" currently utilized.

The three-tiered LTMO approach also is being profiled with other LTMO methods in several documents currently in preparation. During the past several years the US EPA's Office ofSuperfund Remedi- ation and Technology Innovation and the Air Force Center for Environmental Excellence (AFCEE) have been conducting a demonstration project illustrating the application of the three-tiered approach and the Monitoring and Remediation Optimization System (MAROS) to three case-study sites. MAROS is public domain software that was developed for AFCEE by Groundwater Services, Inc., and incorporates optimization procedures and decision logic that were developed in the AFCEE Long-Term Monitoring Op- timization guide (AFCEE, 2003). A summary report (USEPA, 2004b), which discusses the results of application of the two approaches to the evaluation and optimization of groundwater monitoring programs at the three sites, and examines the overall results obtained using the two LTMO methods, will be available in mid-2004. The three-tiered LTMO approach also is one of several method included in a Roadmap for LTMO, currentlyin preparation by the USEPA and the U.S. Army Corps of Engineers. The Roadmap is intended to assist managers, regulators, scientists and engineers tasked with reviewing monitoring programs in determining whether optimization is appropriate for their monitoring program, and describes available optimization methods that may


be appropriate for application to particular monitoring program s.

REFERENCES Air Force Center for Environmental Excellence, Inc. (AFCEE).

2003. Monitoring And Remediation Optimization Sys- tem (MAROS) User's Guide, Version 2.0. http:t/vwvw. gsi-net.com/software/Maros.htm Brooks Air Force Base, Texas. November.

American Society of Civil Engineers (ASCE) Task Committee on Geostatistical Techniques in Hydrology 1990a. Review of Geostatistics in Geohydrology--I. Basic concepts. Journal of Hydraulic Engineering 116(5):612-632.

ASCE Task Committee on Geostatistical Techniques in Hydrol- ogy 1990b. Review of Geostatistics in Geohydrology II. Applications. Journal of Hydraulic Engineering 1 •6(6):633- 658.

Cameron, K., and R Hunter. 2002. Optimization of LlM Net- works Using GTS. Statistical Approaches to Spatial and Temporal Redundancy. Online document available on the Worldwide Web at http://www.afcee.brooks.afmiVedrl)o/ GfSOptPaper.pdf

Cameron, K., and P. Hunter. 2004. Optimizing LTM Net- works with GTS: Three New Case Studies, In Accelerating Site Closeout, improving Performance, and Reducing Costs Through Optimization--Proceedings of the Federal Reme- diation Technologies Roundtable Optimization Conference. June 15 17. Dallas, Texas.

Cieniawski, S. E., J. W. Eheart, and S. Ranjithan. 1995. Using genetic algorithms to solve a multiobjective groundwater monitoring problem Water Resources Research 31(2): 399- 409.

C lark, I. 1987. Practical Geostatistics. Elsevier Al~plied Science, Inc.: London.

Dresel, E. R, and C. Murray. 1998. Groundwater monitoring network design using stochastic simulation. Geological So- ciety of America Abstracts with Programs 30(7): 181.

EI-Shaarawi, A. H., and S. R Niculescu. 1992. On Kendall's tau as a test of trend in time. Environmetrics 3:385-412.

Environmental Systems Research Institute, Inc. (ESRI). 2001. ArcGIS Geostatistical Analyst Extension to ArcGIS 8 Soft- ware. Redlands, CA.

Gibbons, R. D. 1994. Statistical Methods for Groundwater Monitoring. John Wiley & Sons, Inc.: New York.

Henley, S. 1981. Nonparametric GeostatisUcs. Applied Science Publishers, Inc./Halsted Press: London.

Hudak, R F., H. A. Loaiciga, and F. A. Schoolmaster. 1993. Application of geographic information systems to groundwater monitoring network design. Water Resources Bulletin 29(3): 383-390.

Kelly, B. R 1996. Design of a Monitoring We//Network for the City of independence, Missouri, Well Field Using Sim- ulated Ground-Water Flow Paths and Travel Times. US Ge- ological Survey Water-Resources Investigation Report 96- 4264.

Reed, R M., and B. S. Minsker. 2004. Striking the balance: Long term groundwater monitoring design for multiple, conflicting objectives. Journal of Water Resources and Planning Management 130(2): 140-149.

Reed, R M., B. S. Minsker, and D. E. Goldberg. 2001. A multiobjective approach to cost effective long-term groundwater monitoring using an elitist nondominated sorted genetic algorithm with historical data. Journal of Hydroinformatics 3:71-89.

Reed, R M., B. S. Minsker, and D. E. Goldberg. 2003. Simpli- fying multiobjective optimization I1: An automated design methodology for the nondominated sorted genetic algorithm Water Resources Research 39(1): 1196.

Reed, R M., B. S. Minsker, and A. J. Valocchi. 2000. Cost-effective long-term groundwater monitoring design using a genetic algorithm and global mass interpolation. Water Resources Research 36(12):3731- 3141.

Rock, N. M. S. 1988. Numerical Geology: Springer-Verlag: New York.

U.S. Environmental Protection Agency (USEPA). 1994. Meth- ods for Monitoring Pump-and-Treat Performance. Office of Research and Development. EPN600tR-94/123.

USEPA. 2004a. Guidance for Monitoring at Hazardous Waste Sites Framework for Monitoring Plan Development and implementation. USEPA Office of Solid Waste and Emer- gency Response. OSWER Directive No. 9355.4-28. January.

USEPA. 2004b. Demonstration of lwo Long-lerm Ground- water Monitoring Optimization Methods Report with Ap- pendices. USEPA Office of Solid Waste and Emergency Re- sponse. 542-R-04-001B. July.

Wiedemeier, f. H., and R E. Haas. 2000. Designing Monitoring Programs to Effectively Evaluate the Performance of Natural Attenuation. Air Force Center for Environmental Excellence (AFCE E). August.


A Three-Tiered Approach to Long Term Monitoring Program ... of... · ing an effective groundwater...

Documents

Transcript of A Three-Tiered Approach to Long Term Monitoring Program ... of... · ing an effective groundwater...