115 development and demonstration of a method to evaluate ... · to evaluate bio-sampling...

1. Introduction

The DHS Science and Technology Directorate(S&T) is tasked with researching and organizingthe scientific, engineering, and technologicalresources of the United States, leveraging theseexisting resources into technological tools to help

protect the homeland and supporting technol-ogical tools with standards. Standards developmentis critical to the deployment and performance evalua-tion of detection technologies, but the ability todetect contamination is equally dependent on theability to find and recover contamination throughsampling.

Volume 115, Number 2, March-April 2010Journal of Research of the National Institute of Standards and Technology

113

[J. Res. Natl. Inst. Stand. Technol. 115, 113-147 (2010)]

Development and Demonstration of a Methodto Evaluate Bio-Sampling Strategies UsingBuilding Simulation and Sample Planning

Software

Volume 115 Number 2 March-April 2010

W. Stuart Dols, Andrew K.Persily, and Jayne B. Morrow

National Institute of Standardsand Technology,Gaithersburg, MD 20899

and

Brett D. Matzke, Landon H.Sego, Lisa L. Nuffer, and BrentA. Pulsipher

Pacific Northwest NationalLaboratory,Richland, WA 99352

[email protected]@[email protected]@[email protected]@[email protected]

In an effort to validate and demonstrateresponse and recovery sampling approachesand technologies, the U.S. Department ofHomeland Security (DHS), along withseveral other agencies, have simulated abiothreat agent release within a facility atIdaho National Laboratory (INL) on twoseparate occasions in the fall of 2007 andthe fall of 2008. Because these eventsconstitute only two realizations of manypossible scenarios, increased understand-ing of sampling strategies can be obtainedby virtually examining a wide variety ofrelease and dispersion scenarios usingcomputer simulations. This research effortdemonstrates the use of two software tools,CONTAM, developed by the NationalInstitute of Standards and Technology(NIST), and Visual Sample Plan (VSP),developed by Pacific Northwest NationalLaboratory (PNNL). The CONTAMmodeling software was used to virtuallycontaminate a model of the INL testbuilding under various release anddissemination scenarios as well as a rangeof building design and operation para-meters. The results of these CONTAMsimulations were then used to investigatethe relevance and performance of various

sampling strategies using VSP. One of thefundamental outcomes of this project wasthe demonstration of how CONTAM andVSP can be used together to effectivelydevelop sampling plans to support thevarious stages of response to an airbornechemical, biological, radiological, ornuclear event. Following such an event(or prior to an event), incident details andthe conceptual site model could be usedto create an ensemble of CONTAMsimulations which model contaminantdispersion within a building. Thesepredictions could then be used to identifypriority area zones within the building andthen sampling designs and strategies couldbe developed based on those zones.

Key words: agent; biological; modeling;planning; response; sampling

Accepted: December 12, 2009

Available online: http://www.nist.gov/jres

Editor’s Note: This paper was originally published as NIST Technical Note 1636, Development and Demonstration of a Method to EvaluateBio-Sampling Strategies Using Building Simulation and Sample Planning Software, June 2009. All of the content from the original publicationhas been included except for the Executive Summary and the table of contents.

In the event of a bio-terror attack like the contami-nated letter incidents of 2001, a building is firstsampled to assess the extent of contamination anddetermine the appropriate decontamination method.Effective surface decontamination is critical to reduc-ing the risk of physical transport and reaerosolizationof contaminant materials. Post-decontaminationsampling is used to determine efficacy of the deconta-mination effort and to assess the ability to clear thebuilding of any remaining contamination. In order toincrease the confidence in our ability to detect potentialbiological contamination events, the performance ofsampling strategies and uncertainties associated withsampling methodologies must be characterized. Inresponse to a 2005 report written by the GovernmentAccountability Office (GAO-05-251) DHS organized aworking group, the Validated Sampling Plan WorkingGroup (VSPWG), composed of multiple federalagencies and national laboratories to ensure that theoverall process of sampling activities has been vali-dated. In response to the VSPWG efforts to developperformance characteristics for surface samplingmethodologies, agency efforts include evaluating theimpact of methodology limitations and samplinglocations on the overall confidence of detecting surfacecontamination. The generation of a sampling planbased on probabilistic methods to choose the samplinglocations provides a level of confidence in theoverall ability to detect low amounts of surfacecontamination.

Over the past two years, a facility at Idaho NationalLaboratory served as the location for a multi-agency effort to experimentally evaluate the perform-ance of sampling strategies to characterize andclear a building after a bio-contamination event. Bothtime and cost are significant constraints in large-scalefield tests such as those conducted in Idaho. Modelingprovides a means to expand our understanding of thedistribution and transport of biological agents in abuilding beyond what can be explored by the physicaland resource limitations of experimental testing. Modelparameters can be adjusted to study the impact ofvariations in physical parameters that cannot beaddressed experimentally in either a practical or eco-nomic manner.

The purpose of this project was to integrate and coor-dinate the use of CONTAM [2], an airflow and contam-inant dispersal simulation program, with the VisualSample Plan (VSP) [3] computer program to provide

individuals and agencies responding to a biologicalcontamination event with a software-based approach togenerating sample plans. The objectives of this projectwere to demonstrate the process of using CONTAMsimulations to develop zoning strategies and samplingdesigns in VSP and to explore whether there are consis-tent patterns of contaminant dispersion that would helpin pre-planning zoning strategies within the building.Bio-threat release scenarios were generated for a singleoffice building of moderate complexity using theCONTAM program (developed by and maintained atNIST). Release amounts, locations and circumstancesincluding outdoor weather, ventilation system opera-tion and other relevant building parameters wereutilized to generate nearly four hundred release scenar-ios. Results of the NIST generated bio-threat dispersionand deposition scenarios were imported into the VSPsoftware. VSP was developed at the Pacific NorthwestNational Laboratory (PNNL) to generate samplingplans that can include probabilistic sample site determi-nations and user selected judgmental sampling loca-tions. PNNL visually analyzed the simulation resultsgenerated by NIST with CONTAM, looking for pat-terns, gradients and uncontaminated areas. In order tosimplify the task of analyzing so many different scenar-ios, PNNL then performed a cluster analysis to grouptogether scenarios that resulted in similar contamina-tion patterns. Incident details, facility layout andCONTAM results were then utilized to classifythe building spaces into four zoning categories accord-ing to their likelihood of being contaminated. The zon-ing procedure utilized the framework establishedby the Environmental Protection Agency (EPA) andCenters for Disease Control (CDC) through theVSPWG [4] to classify an area as a range from definite-ly contaminated to highly unlikely to be contaminated.Sampling approaches were generated for each zoneclassification.

Sampling plans generated utilizing CONTAM andVSP may provide individuals and agencies responding toa biological contamination event with a tool forevaluating the effects of building parameters, explorezoning strategies that cover most scenarios, and estab-lishing sampling strategies for various scenarios. Thecomputer tools can provide a means of assessingsampling strategies generated under different releasescenarios prior to an actual contamination event and canaide in pre-planning and response plan generation by thefirst responder community for a biological attack.


114

A discussion of this effort, as well as a detaileddescription of the CONTAM and VSP tools, are provid-ed in Sec. 2. The INL facility and the development ofthe CONTAM simulations are discussed in Sec. 3.Detailed results which address each of the threeresearch objectives are presented in Sec. 4. Conclusionsare given in Sec.5.

2. Description of Modeling Tools

Two main software tools were employed in this proj-ect: CONTAM [2] and Visual Sampling Plan (VSP) [3].CONTAM was used to simulate a set of contaminantrelease scenarios, and VSP was used to perform the sta-tistical analysis of the simulation results. CONTAMprovided results in the form of mass of contaminantdeposited within each room of the simulated building.A third program was developed to convert theCONTAM results into a format that VSP couldimport. The following sub-sections describe eachof these tools in more detail, with specifics on howthese tools were used in this project provided inSection 3.

2.1 CONTAM

CONTAM is a computer program that is used to sim-ulate building airflow and contaminant transport on awhole-building scale. The program was developed bythe National Institute of Standards and Technology(NIST). CONTAM is referred to as a multi-zone ornodal network model, because the air volumes thatmake up a CONTAM representation of a building aretreated as a set of nodes interconnected by airflowlinks. These nodes or zones are characterized by asingle pressure, temperature and contaminant concen-tration, though pressure does vary hydrostatically with-in each zone.

For the purposes of analysis, a building is idealizedby subdividing it into a set of zones, typically includingrooms, major plenums between floors, and inter-floorshafts and chases. Zones are then interconnected by both intentional airflow paths, e.g., doors, windows,louvers and unintentional paths such as cracks thatoccur along the junctions between building partitions.Airflows occur between zones due to pressures arisingfrom various driving forces including wind, tempera-ture differences (stack effect) and mechanical systems.CONTAM allows the definition of mechanical ventila-tion systems either via a simplified representation of anair handling system or a fully detailed duct model.

Contaminant concentrations within the building zonesare affected by the location and strength of contaminantsources and removal mechanisms, e.g., material sinksand filters, as well as intra- and inter-zonal airflows.Detailed descriptions of CONTAM and multi-zonemodels in general, are available from several references[5-7] and the NISTwebsite [8].

One important aspect of CONTAM that is pertinentto this project is the contaminant element used tosimulate the deposition of the biological agent. Particledeposition is typically characterized by a depositionvelocity [9], and CONTAM provides the abilityto model contaminant removal by deposition. TheCONTAM deposition velocity model determinesthe mass of contaminant deposited on a givensurface during a simulation based on the followingequation:

Md = mass deposited [kg]ρair = density of air [kg/m3]vd = deposition velocity [m/s]AS = deposition surface area [m2]

C(t)= mass fraction of contaminant [kg/kg]t = time [s], (integration period was the

same for all simulations)

CONTAM allows the definition of as manydepo-sition surfaces in a zone as desired, usingdifferent deposition velocities for each as appropriate.However, all deposition surfaces must be placedwithin a CONTAM zone. This means that eachof the deposition surfaces will see the samezone concentration, represented by C(t) in theprevious equation, within each zone.

2.2 VSP

VSP is a software tool that employs statisticallydefensible approaches in developing sampling designsand analyzing data. It was developed by the PacificNorthwest National Laboratory to support the DataQuality Objectives (DQO) process [10]. It ensures theright type, quality, and quantity of data are gathered tosupport confident decisions and helps in evaluatingtradeoffs between increased confidence in decisionsand costs or number of samples required.

VSP allows the user to draw or import floor plans ofbuildings and generate and apply sampling designs tothose floor plans. Sampling can be applied to the floor


115

2

1

( )t

d air d stM v A C t dtρ= ∫

of a room or set of rooms, but can also be expanded toa 3D representation to include walls, ceilings, andfurniture. Sample types (swab, wipe, vacuum) and pointsof interest (HVAC system, release point, sampling zones, etc.) can be user-specified. Many statisticalsampling designs are available including random,systematic, sequential, adaptive cluster, collaborative,stratified, transect, multi-increment, combined judg-mental and random, rank set, and judgmental sampling.A number of statistical analysis modules are availableto complement the various sampling approaches.Statistical evaluations of the data are provided withdecision recommendations, and most modules providean automated report in text form.

For a particular release scenario, each room mayhave its own sampling design or be combined withother rooms to form a zone (which is different from thedefinition of zone used in CONTAM). Samplingdesigns can be generated for each room or zone, andresults can be imported and analyzed using built-instatistical tests, visual analysis, and geostatisticalmapping.

VSP has been developed with support from theU.S. Department of Energy (DOE), EPA, theDepartment of Defense, the Department of HomelandSecurity (DHS), CDC, and the United KingdomAtomic Weapons Establishment. It is freely available athttp://vsp.pnl.gov. In addition to supporting sampling within buildings, VSP has many sampling designand statistical analysis modules focusing on soils,sediments, surface water, streams, groundwater, andunexploded ordnance sites. VSP is recommended bymany regulators for defensible sampling design andstatistical analysis.

2.3 CONtoVSP

CONTAM and VSP were developed independentlyof each other. Therefore, their inputs and outputs arenot immediately compatible. A process was developedto convert the CONTAM results into a format thatVSP could import. This process involved defining across-reference file format that would map the spacedefinitions of VSP into the deposition surfaces ofthe CONTAM building representation, modifyingCONTAM results files to include information neededto facilitate the mapping, and developing a computerprogram to perform the conversion and generate theinput files for VSP. The conversion program is referredto as CONtoVSP.

Typically, the VSP representation of a building sur-face would be subdivided into a set of grid cells that

can take on numerical values, which in this case wouldbe the mass deposited within the area of the cell.CONtoVSP obtains the mass deposited on each depo-sition surface from a CONTAM contaminant summaryfile (.csm) and apportions the total mass into a set ofVSP grid cells that correspond to the deposition surfaceof CONTAM. The number of VSP cells associated witha room or sampling area is defined within VSP via themap file (.vsp). The number of cells that correspond toa CONTAM zone must also be set in the cross-refer-ence file (.crf) in order for CONtoVSP to generate aproperly formatted input file for VSP which requires avalue for each sample cell defined in the VSP map file.This import file format can be viewed with VSP via theView—Coordinates menu command.

CONtoVSP enables the mass deposited on eachCONTAM deposition surface to be distributed to theVSP grid cells either uniformly or randomly with a nor-mal distribution. If the mass is to be distributeduniformly, then each cell will contain the same amountof mass, equal to the total mass deposited on theCONTAM deposition surface divided by the number ofcells. If distributed normally, then a random set ofvalues will be generated using a normal probabilitydistribution function with a mean and standard devia-tion which are specified in the cross-reference file.CONtoVSP prevents negative values from being gener-ated and normalizes the results by the total massdeposited on each CONTAM deposition surface.

3. Contaminant Transport SimulationUsing CONTAM

This section describes the details of the modelingeffort. It includes a description of the building, theCONTAM model, and the scenarios that were modeledin this study.

3.1 Building Description

The building used for this study is PBF-632 locatedat the Idaho National Laboratory (INL) and shown inFig. 1. Floor plans for the two floors of PBF-632 areshown in Fig. 2 and Fig. 3. Each floor is approximate-ly 24.4 m × 15.2 m (80 ft × 50 ft) for a total of 372 m2

(4 000 ft2) per floor. As designed, each floor contains aconstant volume air handler located within a mechani-cal room of the floor it serves. Outdoor air is brought inthrough an intake duct on the return air side of each airhandler. Supply air ducts are located above suspendedceilings on the floor which they serve, and the space


116

above the suspended ceilings also serves as a return airplenum. Return ducts draw air out of these plenumsfrom just above each of the two mechanical equipmentrooms. Supply air is provided to all occupied spacesexcept the restrooms and janitorial closet. Two returnair grilles are located in the ceilings along the mainhallway of each floor. Dedicated exhaust fans serve the

four restrooms (two on each floor) and each of themechanical rooms. Total design supply airflows forthe 1st and 2nd floors are 1,200 L/s (2,550 cfm) and1,180 L/s (2,500 cfm) respectively. Design restroomexhaust flows are 94 L/s (200 cfm) for each restroomand 47 L/s (100 cfm) for each mechanical room. Thedesign outdoor air intake rate is unavailable.

117


Fig. 1. INL Building PBF-632.

Fig. 2. 2nd Floor Plan.

3.2 CONTAM Model

The CONTAM model of the building was based onthe building as designed. Design information wasobtained from mechanical system drawings and elec-tronic floor plans (AutoCAD format files, i.e., dwgfiles). Initially, the dwg files were imported into a 3Darchitectural Computer Aided Design (CAD) programthat was then used to convert the main building geo-metry into an IFC-compliant (Industry FoundationClass) file. This file was then converted, via an in-house IFC to CONTAM converter, to an initial versionof a CONTAM project file (.prj file) that included thebuilding zones (rooms), doors and windows. Once thebasic file was obtained it was modified to includebuilding envelope and inter-zone leakages, plenumsand a full duct system. For the purposes of this study,no leakage measurements were performed, so theenvelope and inter-zone leakage values were assumedbased on data in the literature [11, 12]. It is important tonote that the purpose of this model was not to represent

the actual building as-built, but to provide a realisticvirtual test bed within which multiple release scenariosand building configurations could easily be appliedand upon which the sampling strategies could beexercised.

Each level of the building is represented inCONTAM by a schematic of the floor plan via theCONTAM sketchpad. The model used in this studyconsists of four levels—the 1st and 2nd floors and theirrespective plenums that contain the ductwork. Thesefour schematics are shown in Fig. 4 through Fig. 7. Asseen in the figures, the building is represented by vari-ous icons which include zones, airflow paths betweenzones, ducts and the deposition surfaces. Zones aredelineated by the solid black lines. Airflow paths arerepresented by the diamond-shaped icons located alongthe walls and on the floors of all but the bottom level.The double blue lines and icons represent the duct sys-tem. Each zone has a deposition surface icon associat-ed with it. The CONTAM colors and icons are definedas follows:

118


Fig. 3. 1st Floor Plan.

The release agent was modeled as a particle havingan equivalent diameter of 1.06 μm. The depositionsurfaces were assumed to consist of only the floorarea of each zone. A constant deposition velocity of

3.78 × 10–5 m/s [13] was assumed for all depositionsurfaces within the CONTAM model. These surfaceswere subdivided into 61 cm × 61 cm (2 ft × 2 ft) gridcells within VSP.

119


Fig. 4. CONTAM representation of 1st Floor Plenum.

120


Fig. 5. CONTAM representation of 1st Floor.

Fig. 6. CONTAM representation of 2nd Floor Plenum.

3.3 Scenarios Modeled

The main benefit to using simulations to evaluatesampling strategies is the ability to perform multiplerelease scenarios under varying building operating andenvironmental conditions. CONTAM allows for releasesto be simulated from within any of the building zoneswith varying amounts and time profiles. It also allows forreleases external to the building by either targetingspecific building openings or assuming a concentrationprofile around the building. CONTAM also allows thesimulation of various building configurations includingvariations in door and window opening status, variationsin fan on/off status and ventilation flow rates, as well asvariations in weather conditions.

Table 1 provides a list of the factors that were consid-ered in the scenarios simulated and their associatedlevels. This is by no means an exhaustive list of potentialvariables and variations, but it was considered to be themost ambitious set of scenarios allowed under theresources available for this project. This set of factorsand levels yields a total combination of 384 scenarios.

A project file generation tool was developed to create the384 test cases and batch files were used to run CONTAMand the CONtoVSP conversion tool. In this way the fullset of simulations could be performed more quickly andreliably.

The release zones are shown in orange in Fig. 5 andFig. 7 (near the bottom-right corners of the schematics)and the zones having the dedicated fans are shown indark green (top-left corners of the schematics). In thecase of the outdoor release, the contaminant is simulatedto be uniformly distributed around the building. All sim-ulations were for 24 h periods with 1 min time steps. Allreleases were modeled to occur for 1 min starting 1 h intothe simulation. When the dedicated outdoor air fan wasturned on, the duct that served the zone into which theoutdoor air was provided was closed off in order to pre-vent the agent from being re-circulated into the zone.Thus, when the release was from an interior source, thesezones would serve as a form of shelter-in-place (SIP)zones and should have very little to no contamination.This was necessary to support one of the objectives ofthis research effort, i.e., the development of zoningstrategies.

121


Fig. 7. CONTAM representation of 2nd Floor.

122

4. Analysis of Simulations and SamplingStrategies Using VSP

One of the fundamental outcomes of this project wasa demonstration of how CONTAM and VSP can beused together to effectively develop sampling plans tosupport the various stages of response to an airbornechemical, biological, radiological, or nuclear (CBRN)event. Following a CBRN event, incident details andthe conceptual site model (CSM) could be used tocreate CONTAM predictions of contaminant dispersionwithin a building. These predictions would then be usedto identify priority area zones within the building, andthen sampling designs and strategies could be devel-oped based on those zones. As discussed in Sec. 2.1 and3.2, a sizeable number of input parameters must bespecified in order to execute the CONTAM model. It is

plausible that reasonable estimates could be made forsome of the CONTAM parameters using the incidentdetails and the CSM. However, some of the CONTAMinput parameters will remain unknown or at least lesswell known. Consequently, a group (or ensemble) ofCONTAM dispersion predictions could be made byholding constant the input parameters which are knownand varying the input parameters that are unknown.Visual analysis of the ensemble of predictions couldthen be used to develop a priority zoning strategywithin the building. Sampling plans, in turn, wouldbe developed in VSP to match the sampling objectivesand characteristics of each of the zones. In short,CONTAM predictions can be used to create moreefficient sampling designs in VSP. The integratedCONTAM-VSP process is illustrated in Fig. 8.


Table 1. Factors and Levels for simulated scenarios

Factors Levels Number of Levels

Contaminant Release

Outdoor Remote

Location 1st floor Lobby 41st floor Office2nd floor Office

Low (1.0 mg)Amount Medium (500 mg) 3

High (1.0 g)

HVAC Operation

1st and 2nd floor supply fans Both On (10 % outdoor air)(and Outdoor air intake rate) Both Off (no outdoor air)

1st and 2nd floor exhaust fans On 2Off

Dedicated outdoor air fan Inactive 2Active on 1st and 2nd floor offices

Weather

Outdoor temperature 0 ºC 220 ºC

Wind speed 1 m/s 25 m/s

TOTAL Simulations 384

PNNL used VSP to analyze a large number of simu-lated scenarios generated by CONTAM. Five zoningstrategies were developed based on typical patternsobserved in the CONTAM simulation results. For eachzoning strategy, four sampling approaches were evalu-ated for the various scenarios: judgmental, purely prob-abilistic, purely probabilistic with zoning, and com-bined judgmental and random (CJR) with zoning.

Section 4.1 discusses the process of using visual andcluster analysis to explore whether there are consistentpatterns of contaminant dispersion that would help inpre-planning the zoning strategies within the building.The definitions and characteristics of the various zonesare presented in Sec. 4.2. The demonstration of how theresults from Sec. 4.1 are used to develop zoning strate-gies is presented in Sec. 4.3. Section 4.4 presents com-parisons of the performance of the various samplingapproaches across five typical scenarios identified bythe visual and cluster analysis shown in Sec. 4.1.

4.1 Identifying Patterns of Contaminant Dispersion

4.1.1 Visual Analysis

The results of the 384 CONTAM simulations wereconverted by CONtoVSP (Sec. 2.3) into the file formatused for VSP files. In this format, each floor has its ownfile containing the response values (contaminant levels)for each grid cell organized by rooms. Files were pro-vided using both the uniform and normal distributionmethods for distributing contaminant to grid cells with-in each room (Sec. 2.3). All numerical analyses useddata from the uniform distribution method. The meth-ods produced similar images when viewed in VSP, andeither set of data (uniform or normal) could be used for visual analyses. However, data from the normal distri-bution method have more texture instead of givingrooms solid colors, and this makes visual comparisonsof adjacent rooms somewhat easier. Therefore, allimages in Sec. 4 of this report are from data using

123


Fig. 8. Illustration of CONTAM-VSP integrated sample design process.

the normal distribution method. To better account forboth linear and non-linear patterns on the floors, PNNLcreated a second set of files that use base 10 logarithmsof the raw values, so that both the raw and log-scaleddata could be analyzed. This was done for both floorsin every simulation, resulting in four VSP files for eachof a scenario's sets of data (uniform or normal), the rawand log of the 1st floor values, and raw and log of the2nd floor values. Visual analysis was performed usingthe Color By Values feature of VSP [3]. This featuremaps the data values in each cell to a continuous colorscale, making it easy to visualize the location and inten-sity of contamination. A total of 3,072 (2 × 2 × 2 × 384)images were produced and analyzed visually with theobjective of identifying scenarios that:

1) Showed contamination gradients or trends onindividual floors.

2) Showed floors with little or no contamination.3) Showed other interesting patterns including

unique patterns isolated to only a handful ofsimulations, or scenarios that summarized alarger number of scenarios with similar disper-sion of contaminant.

Using VSP’s menu commands to open and view eachof the 3,072 files would clearly be a slow process. So,a “scenario viewer” was developed using a Pythonscript to automate the process.

Scenarios are called up one-at-a-time, allowing theuser to look at the 1st floor, and then the 2nd, by hittinga “Next” button. If a scenario fit one of the three cate-gories, a “Save” button allowed the scenario to bestored for future use. If both files have been viewed, ora scenario was saved, the viewer proceeds to the nextscenario. The raw and log values were analyzed inseparate runs of the viewer for both the uniform andnormally distributed sets of data.

Approximately 50 % of the scenarios fell into one ofthe three categories listed above. Examples of scenariosnot included in this 50 % include cases in which most,if not all, of the rooms were contaminated with no dis-cernible gradient, or contamination was uniform acrossall or most of the rooms.

4.1.2 Cluster Analysis

The visual analysis revealed many different patternsacross the different scenarios. The next step was to per-form a numerical analysis of the data that would help

put the results of the visual analysis into perspective. Itwas decided that a cluster analysis was well-suited forsummarizing the numerical results.

We chose to use the rooms as the individual units forthe cluster analysis. The numerical values used in theanalysis were the average contamination levels of thegrid cells in each room. This results in having the clus-ter analysis generally group together scenarios withsimilar levels of contamination in the same rooms sothat scenarios with similar dispersion patterns and sim-ilar contaminant levels in rooms will tend to grouptogether.

Results of a partial least squares (PLS) regression[14] bi-plot are shown in Figure 9 generated using TheUnscrambler software [15]. In PLS, each room has avariable stating the scenarios' levels of contamination,and these rooms are simultaneously modeled using theexperimental variables previously outlined in Table 1.Unlike traditional regression methods, all rooms aremodeled together taking into account their correlationsand the data structure of both the contamination levelsof the rooms and the experimental variables. Clustersof scenarios can be shown graphically with the regres-sion variables overlayed on the plot to show whichvariables are positively or negatively correlated with acluster. Variables on the outside of the plot and near acluster will tend to be positively correlated withthat cluster. Variables on the outside of the plot andfar away from a cluster will tend to be negativelycorrelated with that cluster. In Fig. 9, clusters near the“High Contamination” are those with the highestlevels of contamination, and likewise “LowContamination” indicates clusters with lower levels ofcontamination.

PLS was used largely because in addition to identify-ing clusters, it also attempts to identify how the combi-nations of the various CONTAM factors may influencethe way in which the scenarios are clustered. Theresults show approximately 15 distinguishable groupsof scenarios. Groups were largely determined by threeof the explanatory variables that were significant in thePLS model: release point, the amount of contaminantreleased, and whether the supply fans were on or off. Inorder to reduce the large number of CONTAMsimulations into smaller groups for analysis, a 3 × 3 × 2factorial structure was used to create 18 clusters byusing release point (1st floor office and 1st floor lobbycombined, 2nd floor, or outdoor), the amount of con-taminant released (low, medium, or high), and supplyfan status (on or off).

124


The resulting clusters of scenarios were defined asfollows:

1) 1st floor release (lobby or office), low amount,supply fan on

2) 1st floor release (lobby or office), mediumamount, supply fan on

3) 1st floor release (lobby or office), high amount,supply fan on

4) 1st floor release (lobby or office), low amount,supply fan off

5) 1st floor release (lobby or office), mediumamount, supply fan off

6) 1st floor release (lobby or office), high amount,supply fan off

7) 2nd floor office release, low amount, supply fanon

8) 2nd floor office release, medium amount, supplyfan on

9) 2nd floor office release, high amount, supply fanon

10) 2nd floor office release, low amount, supply fanoff

11) 2nd floor office release, medium amount, supplyfan off

12) 2nd floor office release, high amount, supply fanoff

13) Outdoor release, low amount, supply fan on14) Outdoor release, medium amount, supply fan on15) Outdoor release, high amount, supply fan on16) Outdoor release, low amount, supply fan off17) Outdoor release, medium amount, supply fan off18) Outdoor release, high amount, supply fan off.

125


Fig. 9. Partial Least Squares (PIS) Bi-Plot of the 384 CONTAM simulations.

Clusters involving 2nd floor and outdoor releasepoints (7-18) each consisted of 16 CONTAM simula-tions. Clusters involving 1st floor release points con-tained 32 CONTAM simulations because the 1st flooroffice and 1st floor lobby release points were combinedsince they resulted in similar dispersion patterns andlevels throughout most of the building.

The visual analysis of the clusters also revealed pat-terns regarding the variables used for choosing theseinitial clusters. First, the supply fan appeared to be alarge factor in determining how much of the buildingwas contaminated. When the supply fan was on, all ormost of the building was contaminated. When the sup-ply fan was off, often only a smaller portion of thebuilding around the release point was contaminated.

The amount of contaminant released affected theamount of contaminant in each of the rooms, but didnot appear to have a large effect on dispersion patterns.For instance, the simulations with the same factorsother than having low, medium, or high amounts ofcontaminant tended to have the same rooms contami-nated regardless of the amount of contaminant. This isto be expected, because the contaminant has no effecton airflow patterns and deposition rates as modeled byCONTAM.

4.1.3 Visual Analysis of Clusters

Following the initial visual analysis and the clusteranalysis, a new visual analysis of the scenarios in eachindividual cluster was then performed. In the originalvisual analysis of all 384 CONTAM simulations, it waslearned that the log-scaled data was much moreintuitive for finding patterns using VSP. Therefore,the remaining analysis was performed using only thelog-scaled data. Also, in order to maintain consistencyacross scenarios during the visual analysis, the colorscales were fixed in VSP as opposed to each scenariobeing auto-scaled based on its individual range ofvalues. Images were produced of maps of the 1st

and 2nd floors for each scenario and their levels ofcontamination shown by a color scale as in theoriginal visual analysis. This time, color scaleswere kept fixed and the images were organized bycluster.

As a result of this analysis, the set of clusters werereduced even further from 18 down to 5. After examin-ing images within each cluster, it became more appar-ent that the amount of contamination had very littleinfluence on which rooms were contaminated, andtherefore the amount of contaminant would not be amajor factor when determining zoning strategies.Clusters with the same release point and supply fansetting were combined as were those with outdoorreleases. For outdoor releases, it was obvious that asingle zoning strategy was sufficient since the outdoorrelease scenario (a plume surrounding the building)always contaminated each room in the building to atleast a small extent. This resulted in the following fiveclusters, which from this point forward will be referredto as Clusters 1-5:

1) 1st Floor (Office or Lobby) Release, SupplyFan OFF (96 CONTAM scenarios)

2) 1st Floor (Office or Lobby) Release, SupplyFan ON (96 CONTAM scenarios)

3) 2nd Floor Office Release, Supply Fan OFF (48CONTAM scenarios)

4) 2nd Floor Office Release, Supply Fan ON (48CONTAM scenarios)

5) Outdoor Release (96 CONTAM scenarios).

Images from one of the scenarios in each of Clusters1-5 are shown in Fig. 10 through Fig. 19 as examples ofcontaminant deposition patterns for the scenarios ineach cluster.

126



127

Fig. 10. Cluster 1:1st Floor (Office or Lobby) Release, Supply Fan OFF—1st Floor Log scale.

Fig. 11. Cluster 1:1st Floor (Office or Lobby) Release, Supply Fan OFF—2nd Floor Log scale.


128

Fig. 12. Cluster 2:1st Floor (Office or Lobby) Release, Supply Fan ON—1st Floor Log scale.

Fig. 13. Cluster 2:1st Floor (Office or Lobby) Release, Supply Fan ON—2nd Floor Log scale.


129

Fig. 14. Cluster 3: 2nd Floor Office Release, Supply Fan Off—1st Floor Log scale.

Fig. 15. Cluster 3: 2nd Floor Office Release, Supply Fan Off—2ndFloor Log scale.


130

Fig. 16. Cluster 4: 2nd Floor Office Release, Supply Fan ON—1st Floor Log scale.

Fig. 17. Cluster 4: 2nd Floor Office Release, Supply Fan ON—2nd Floor Log scale.


131

Fig. 18. Cluster 5: Outdoor Release—1st Floor Log scale.

Fig. 19. Cluster 5: Outdoor Release—2nd Floor Log scale.

4.2 Zone Classifications

The concept of dividing the building into classes isconsistent with the Multi-Agency Radiation Surveyand Site Investigation Manual (MARSSIM) developedby EPA, Nuclear Regulatory Commission (NRC), andDOE [16]. According to the MARSSIM, during thecharacterization phase of a release event, the rooms ofthe building should be divided into one of four possiblepriority areas or zones. These zones are not to be con-fused with CONTAM zones. The following is the set ofclassification criteria [17] into which the building areaswere categorized for this study:

Class 1 Zone: Definitely Contaminated or Assumedto Be Contaminated (i.e., ExtremelyHigh Likelihood of Being Contam-inated).

Class 2 Zone: High Likelihood of Being Contam-inated.

Class 3 Zone: Low Likelihood of Being Contam-inated.

Class 4 Zone: Extremely Low Likelihood of BeingContaminated [18].

Based on the scenarios analyzed, the building used inthis case can be zoned into Class 1, 2, or 3 zones. A sep-arate zoning strategy was developed for each of the fiveclusters of scenarios, and these will be shown later. Therelatively small size of the building, both floor area andnumber of floors, as well as the tendency of the venti-lation system to mix the air throughout each floor, leadsto reduced expectancy for Class 4 zones. Therefore, allrooms would be classified as at least Class 3 zones.Class 4 zones are more likely to occur in situationswhere a release would be isolated to a relatively smallarea of a large facility, either due to the nature of therelease (i.e., location, amount or mechanism), buildingconfiguration, or system operation.

The sampling approaches for each zone class aredifferent because the sampling objectives and decisioncriteria can be based on the pre-event informationdeveloped for each part of the building through, forexample, the visual and cluster analysis carried outherein. For example, during the characterization phaseof a release event, Class 1 and Class 2 zones mightwarrant the use of a sampling approach that is gearedtowards determining the extent of contamination.While Class 3 or 4 zones can be handled with a clear-ance sampling approach where enough samples aretaken to determine with high confidence that little or nocontamination is present in the area, and decontamina-

tion may not be necessary. The recommended samplingstrategies for each phase of the response and each zoneare presented below.

4.2.1 Class 1 Zones

Class 1 zones are known or assumed to be contam-inated, such as the location of the release point of thecontaminant, or rooms that would almost certainly becontaminated (provided the release point is known).These rooms will presumably need to be decontaminat-ed. If desired, judgment samples can be taken using thesampling expert's best professional judgment, whichmay include choosing locations in rooms most likely tobe contaminated and surfaces from which an accuratesample can be most easily extracted. The red area inFig. 20 shows a Class 1 area, which in this case is theknown release point. In some cases where a very largearea is expected to be contaminated (usually in a muchlarger building), a hotspot sampling design may beapplied to delineate areas of contamination.

4.2.2 Class 2 Zones

Class 2 zones are assumed to have a high probabili-ty of being contaminated, but there is no obviousevidence that points to it conclusively being contami-nated. Initially, judgmental samples may be taken in themost likely areas of contamination. In Fig. 20 the yel-low areas are Class 2 Zones. These are rooms surround-ing the suspected release point (the red Class 1 Zone).If unacceptable levels of contamination are present,then the area can be reclassified as a Class 1 zone andcan be scheduled for decontamination. If judgmentalsamples are clean, then a more extensive statisticalsampling design can be applied to determine a courseof action for the zone. A hotspot sampling design couldbe implemented to delineate possible areas of contami-nation, or the initial assessment of the zone being Class2 may be reconsidered. A clearance design might beapplied to determine with statistical confidence that ahigh percentage of the area is free of contamination.

4.2.3 Class 3 Zones

Class 3 zones are assumed to have a fairly low prob-ability of being contaminated. In Fig. 20, the blue areasare Class 3 Zones. On the 1st floor these rooms are notdirectly adjacent to the room in which the releaseoccurred and tended to be “upstream” from the sourcein terms of airflow direction for the cases simulated. Onthe 2nd floor, the Class 3 rooms show that most of the


132

2nd floor has a high probability of not being contami-nated. The sampling objective for Class 3 zones is totake enough samples to either confidently conclude thatthe zone is not contaminated above the acceptable con-tamination level (ACL) or detect contamination if itexists. It is recommended that enough samples be takento make a confidence statement about the zone. Thisstatement is referred to as an X %/Y % upper percentiletolerance statement [19, 20] and is stated in the form“there is X % confidence that at least Y % of thesurface area is uncontaminated (or below the ACL)”.

Two sampling strategies are recommended for sam-pling Class 3 zones: a purely probabilistic samplingapproach and a combined judgmental and random(probabilistic) approach (CJR). In a purely probabilis-tic sampling approach, random samples are taken, allsamples are expected to have an equal likelihood ofbeing contaminated, and no prior assumption is madeabout the probability of the zone being contaminated.In the CJR approach [21], randomly located (proba-bilistic) samples are used to augment judgmental sam-ples in order to make the desired X %/Y % confidencestatement. In the CJR method, judgmental samples canbe assumed to have a greater likelihood of containingdetectable contamination than probabilistic samples.Also, before any sampling takes place, a Bayesian priordistribution is determined by the a priori probabilitythat a judgmental sample will contain detectable con-tamination. This is helpful for situations in whichexperts feel there is some evidence that a zone may beclean, but they wish to take samples to achieve a highlevel of confidence. The CJR method, when appropri-ate, can be advantageous because, it requires fewertotal samples than a purely probabilistic approach.

4.2.4 Class 4 Zones

Class 4 zones have an extremely low likelihood ofbeing contaminated. These are zones for which there issufficient evidence to conclude that no contaminantcould have entered them. In general, sampling is notrequired for these zones—however, clearance samplingmay be required by stakeholders, in which case a CJRdesign would be recommended. If there is any non-negligible chance that an area could be contaminated,the zone should not be classified as Class 4.

4.3 Developing Zoning Strategies

The scenarios within each cluster tend to have simi-lar patterns of dispersion. A visual analysis was con-ducted on each of the five clusters to determine con-tamination patterns on each floor and to determinewhich rooms tended to be contaminated either all of thetime, most of the time, rarely, or never. These resultswere then used to classify rooms as Class 1, Class 2,Class 3, or Class 4 zones. The following sectionsdescribe the zoning process and strategy for each of thefive clusters.

4.3.1 Cluster 1 Zoning: 1st Floor Releases (Lobbyand Office) with Supply Fans OFF

The zoning strategy for Cluster 1, (all scenarios witha 1st floor release with Supply fans OFF) is shownin Fig. 20. The strategy for Cluster 1 as illustratedin Fig. 20 reflects a 1st floor office release, but thecluster also contains 1st floor lobby release scenarios.These release points were combined into the samecluster because the spread of contamination tendedto be very similar for the two different release points.Therefore, these scenarios were zoned the samewith the exception that the 1st floor office andlobby were zoned as either Class 1 or Class 2depending on which was the release point.

1st floor releases with Supply fans OFF resulted inthe point of release always being contaminated(the red room in Figure 20). Therefore, the releaseroom is always labeled Class 1. The surroundingrooms, hallway, and mechanical room and bathroomswere usually contaminated and are labeled Class 2zones (yellow rooms). Offices on the opposite side ofthe floor (left end of building on floor plans) wereoccasionally contaminated and are Class 3 zones (bluerooms).

The 2nd floor was usually contaminated in the areaabove the mechanical room and bathrooms of the 1stfloor. These are labeled Class 2 zones (yellow rooms inFig. 20). The remaining rooms on the 2nd floor wereoccasionally contaminated and are labeled Class 3zones (blue rooms).


133

4.3.2 Cluster 2 Zoning: 1st Floor Releases (Lobbyand Office) with Supply Fans ON

Figure 21 shows the zoning strategy for Cluster 2 (allscenarios with a 1st floor release and Supply Fans ON).On the 1st floor, the release point (office), a neighbor-ing room, and the hallways were always contaminatedand are Class 1 zones (red rooms in Fig. 1). The rest ofthe floor, with the exception of one corner office, wasusually contaminated and classified as Class 2 zones(yellow rooms). The previously mentioned corneroffice was occasionally contaminated and classified asa Class 3 zone (blue room).

The entire 2nd floor was usually contaminated withthe exception of one corner office (which correspondsto the Class 3 office on the 1” floor), and these roomswere classified as Class 2 zones (yellow rooms inFig. 21). The previously mentioned corner office wasclassified as a Class 3 zone (blue room). These twocorner offices on the 1St and 2nd floors labeled asClass 3 zones were the so-called SIP zones mentionedin Sec. 3.3.

4.3.3 Cluster 3 Zoning: 2nd Floor Office ReleasesWith Supply Fans OFF

Figure 22 shows the zoning strategy for Cluster 3 (allscenarios with a 2nd floor office release and SupplyFans OFF). On the 2nd floor, the release location and aneighboring office were always contaminated and arecategorized as Class 1 zones (red rooms in Fig. 22).The other offices around the release point and on theright half of the floor were usually contaminated, andare categorized as Class 2 zones (yellow rooms). Theremaining rooms on the other half of the building wereonly occasionally contaminated and categorized asClass 3 zones (blue rooms).

The entire 1st floor was usually not contaminated,and therefore the entire floor is categorized as Class 3(the entire floor is blue in Fig. 22).

4.3.4 Cluster 4 Zoning: 2nd Floor Office ReleasesWith Supply Fans ON

Figure 23 shows the zoning strategy for Cluster 4 (allscenarios with a 2nd floor office release and Supply Fans ON). On the 2nd floor, the release location and aneighboring office were always contaminated and arecategorized as Class 1 zones (red rooms in Fig. 23). Most of the other areas on the 2nd floor were usuallycontaminated and are categorized as Class 2 zones(yellow rooms). Two corner offices near the SIP roomwere occasionally contaminated and are categorized asClass 3 zones (blue rooms).

Most of the offices and areas on the 1st floor were usu-ally contaminated and are categorized as Class 2 zones(yellow areas in Fig. 23). Four offices in one corner ofthe floor near the SIP room were occasionally contami-nated and categorized as Class 3 zones (blue rooms).

4.3.5 Cluster 5 Zoning: Outdoor Releases

Figure 24 shows the zoning strategy for Cluster 5 (allscenarios with an outdoor release). All rooms of boththe 1st and 2nd floor were always contaminated. Thiswas most likely due to the type of outdoor release sim-ulated in this exercise, which was a plume passing overthe building, and presumably contaminant making itinto enough areas of the building that it spread to allparts either by ventilation system mixing or airflowsinduced by wind and stack effect. Since there is norelease point within the building, all areas of the build-ing were classified as Class 2 zones indicating contam-ination is highly likely.


134

Fig. 20. Cluster 1 Zoning strategy of 1st Floor and 2nd Floor for scenarios with 1st Floor Office Release and Supply Fans OFF(Red = Class 1, Yellow = Class 2, Blue = Class 3).


135

Fig. 21. Cluster 2 Zoning strategy of 1st Floor and 2nd Floors for scenarios with 1st Floor Office Release and Supply Fans ON(Red = Class 1, Yellow = Class 2, Blue = Class 3).

Fig. 22. Cluster 3 Zoning strategy of 1st Floor and 2nd Floor for scenarios with 2nd Floor Office Release and Supply Fans OFF(Red = Class 1, Yellow = Class 2, Blue = Class 3).

Fig. 23. Cluster 4 Zoning strategy of 1st and 2nd Floor for scenarios with 2nd Floor Office Release and Supply Fans ON(Red = Class 1, Yellow = Class 2, Blue = Class 3).

4.4 Comparing Sampling Strategies

Building airflow and contaminant transport model-ing can be used to compare different sampling strate-gies (as a supplement to experimental efforts). Thesingle zone (uniform concentration) assumption of themulti-zone model resulted in entire rooms being con-taminated. Therefore, hotspot designs are not applica-ble to these circumstances. Hotspot sampling designsare more suited to scenarios that result in portions ofrooms being contaminated. While hotspot scenarioswere not addressed in this study, CONTAM doesprovide the ability to simulate more detailed contami-nant distribution on a sub-room scale, e.g., using one-dimensional convection/diffusion and computationalfluid dynamics (CFD) within selected zones. Examplesof the types of sampling strategies that could bedeployed under the uniform concentration assumptionof CONTAM include the following:

1. Purely Judgmental—The sampling team usesonly their professional opinion to choose sam-pling locations.

2. Purely Probabilistic—Sample locations arechosen randomly or using a randomly placedgrid.

3. Purely Probabilistic given a Zoning Approach—Depending on the sampling objective, eitherhotspot sampling is performed, or a number ofsamples are taken to make an X %/Y % confi-dence statement.

4. Combined Judgmental/Random (CJR) withZoning—Both judgmental and probabilisticsamples can be taken to make X %/Y % confi-dence statements. Judgment samples potentiallycarry more weight, and a Bayesian prior isapplied that factors in existing, but not

sufficient, evidence that the area may be clean.Random samples are also taken to supplementjudgmental samples and to establish a sufficientnumber of samples to achieve the X %/Y %confidence statement.

The following example is presented to show howdifferent sampling strategies can be compared. In thisexample, the four sampling strategies mentioned aboveare applied to Cluster 1. VSP was used to generate theprobabilistic sample locations. Because entire roomswere contaminated in the methodology used, locationsof judgment samples were not critical for this exerciseand do not affect the results. Therefore, judgmentalsample locations were selected by PNNL based looselyon locations from previous live tests [22]. However,during an actual contaminant release, judgment samplelocations would be chosen by contamination samplingexperts depending on the amount of information knownabout the release [4]. If judgment sample locationswere to influence the results, such as scenarios whererooms were only partially contaminated, the assistanceof a sampling expert should be used to choose judg-ment sample locations.

The zoning strategy shown previously in Fig. 20 isdesigned for scenarios from Cluster 1 (1st FloorReleases with Supply Fan OFF). First, the four sam-pling strategies for this case will be applied by showingsample locations, summarizing the strategies in a table,and overlaying the sampling strategy onto the groundtruth data based on the specific scenario shown inFig. 10 and Fig. 11. The strategies will be compared toshow how they differ with respect to the conclusionsthat can be drawn by applying them. Summaries andconclusions for the remaining four clusters will be pro-vided in table form with respect to the zoning strategiesapplied on each floor.


136

Fig. 24. Cluster 5 Zoning strategy for 1st and 2nd Floor for scenarios with Outdoor Release (Yellow = Class 2).

Sample sizes for the purely judgmental and purelyprobabilistic approaches are the same for Clusters 2-5as in Cluster 1. For the two strategies that includezoning, samples sizes within the zones are also thesame in Clusters 2-5 as in Cluster 1, but it should benoted that some zoning strategies have fewer zonesresulting in fewer samples. For instance, the 1st floor ofCluster 3 consists entirely of a Class 3 zone, so theconclusions are made without taking the additionalsamples needed for a separate Class 2 zone. Addition-ally, Cluster 5 has only Class 2 zones on both floors.Therefore, Cluster 5 scenarios do not need the addition-al samples needed for separate Class 1 and Class 3zones.

It is important to note that the following comparisonsof the four sampling strategies are based on single real-izations of various sampling plans which have beenapplied to the data generated by CONTAM simulationswhich in this case are being taken as the ground truth.Consequently, the conclusions which can be drawnfrom a particular sampling design are relevant only for1) the particular set of judgmental and/or probabilisticsample locations selected in that design and 2) theparticular results generated by CONTAM for thatsimulation. For this reason, a rigorous statistical evalu-ation of the four sampling strategies would only bepossible with thousands of realizations of the varioussampling designs to verify the precision of their statis-tical confidence statements. Nonetheless, the applica-tion of the four sampling strategies to the five clusterscenarios below provides a valuable illustration of thestrengths and weaknesses of each of the samplingstrategies. A detailed report of a rigorous statistical val-idation of the sampling designs for buildings in VSP iscurrently in preparation (as part of the DHS fundedChemical Operation Technology Demonstration).

4.4.1 Cluster 1—1st Floor Release With SupplyFan OFF

Table 2 shows the four sampling methods and num-ber of samples required for the Cluster 1 scenarios, i.e.,1st floor releases with the supply fans OFF. Samplelocations for the purely judgmental approach are shownin Fig. 25 and Fig. 26. Sample locations for the purelyprobabilistic approach are shown in Fig. 27 andFig. 28. Sample locations for the purely probabilisticwith zoning approach are shown in Fig. 29 and Fig. 30.Sample locations for the CJR with zoning approach areshown in Fig. 31 and Fig. 32.

The purely judgmental approach proves useful inidentifying some rooms as being contaminated. Basedon the ground truth overlay (Fig. 10 and Fig. 11), manyrooms were sampled once or twice with no samplesexceeding the ACL level. This is not enough samples tostate with a high degree of statistical confidence thatthe room is clean. Additional sampling would berequired to confidently identify these rooms as clean.This shows that judgmental sampling is effective atidentifying rooms as contaminated when all grid cellsare contaminated.

The purely probabilistic approach also detected con-taminated areas. However, because the two purelyprobabilistic sampling designs were applied to theentire 1st floor and the entire 2nd floor, the possiblecleanliness of smaller areas on those floors could not bereadily quantified with X %/Y % confidence state-ments. This is the case in the evaluation of the simulat-ed scenario. If the samples are considered on an indi-vidual basis, many of the same conclusions drawn fromthe purely judgmental sampling approach can be made,i.e., some rooms can be identified as being contaminat-ed. The purely probabilistic approach has the potentialadvantage of making X %/Y % confidence statementsabout an entire floor if none of the samples are abovethe ACL.

The purely probabilistic with zoning approach has apotential economic advantage over non-zoningapproaches since the floors are sub-divided. This canlead to cost savings associated with eliminatingportions of a floor as candidates for decontamination.When this sampling strategy is applied to the simulatedscenario, the result is that the Class 3 zones on the 1stand 2nd floor are deemed 95 % confident that at least95 % of the grid cells are below the ACL. This is morethan half of the total surface area which could reducedecontamination costs if the building configuration anddecontamination methods were amenable to suchsavings. It is important to note a 95 %/95 % was usedin this exercise because of the tendency for rooms to beeither entirely clean or entirely contaminated. In othertypes of release scenarios where there is the potentialfor very small portions of rooms to be contaminated, itis recommended that a stronger confidence statementbe used such as 95 %/98 % or 95 %/99 %.

The CJR with zoning approach takes advantage ofprior information to effectively make conclusions aboutsome portions of the building while requiring fewersamples. The OR method yields similar conclusions tothe purely probabilistic approach, but has the added


137


138

Table 2. Comparison of sampling strategies for Cluster 1(1st Floor Release with Supply Fans ON). Comparison based on scenario shownin Fig. 10 and Fig. 11

Conclusion if no samplesSampling #Judg. #Prob.

StrategyFloor Zone

Samples SamplesTotal above the Acceptable Comparison to Simulated Scenariao

Contamination Level (ACL)

1 n/a 30 0 30No contamination found, Several rooms contaminated, othersbut cannot state statistical possibly clean but no statistical

Purely confidence confidenceJudgmental No contamination found, Several rooms contaminated, others

2 n/a 24 0 24 but cannot state statistical possibly clean but no statisticalconfidence confidence

Total 54 0 54

1 n/a 0 58 5895 % confident that 95 % of Some rooms contaminated

Purely the floor is below the ACLProbabilistic 95 % confident that 95 % of Some rooms contaminated

2 n/a 0 58 58 the floor is below the ACLTotal 0 116 116

1 Class 1 6 0 6No contamination found, The room is contaminatedbut cannot state statisticalconfidence

1 Class 2 0 58 5895 % confident that 95 % of The rooms sampled arethe Zone is below the ACL contaminated, and become class 1

zones.Purely 95 % confident that 95 % of No contamination found. We areProbabilistic 1 Class 3 0 58 58 the floor is below the ACL 95 % confident that at least 95 % ofwith Zoning the zone is clean or below the ACL

2 Class 2 0 58 5895 % confident that 95 % of The rooms sampled arethe floor is below the ACL contaminated, and become Class 1

zones.2 Class 3 0 58 58 95 % confident that 95 % of No contamination found. We are

the floor is below the ACL 95 % confident that at least 95 % ofthe zone is clean or below the ACL

Total 6 232 238

1 Class 1 6 0 6No contamination found, The room is contaminatedbut cannot state statisticalconfidence

1 Class 2* 16 24 4095 % confident that 95 % of The rooms sampled arethe floor is below the ACL contaminated, and become class 1

zones.CJR with 95 % confident that 95 % of No contamination found. We areZoning 1 Class 3* 8 23 31 the floor is below the ACL 95 % confident that at least 95 % of

the zone is clean or below the ACL95 % confident that 95 % of The rooms sampled are

2 Class 2* 7 39 46 the floor is below the ACL contaminated, and become class 1zones.

95 % confident that 95 % of No contamination found. We are2 Class 3* 17 5 22 the floor is below the ACL 95 % confident that at least 95 % of

the zone is clean or below the ACLTotal 54 91 145

* Parameter inputs for CJR method: 50 % probability that judgment sample location contains detectable contamination and are twice as likely tocontain detectable contamination as random sample locations.

advantages provided by both Bayesian prior and judg-mental sampling. The Bayesian prior accounts for theprobability that contamination is not present, which canbe a reasonable assumption for Class 3 zones. Thisaccounts for the sampling experts’ ability to identifylikely areas of contamination, and results in much fewer

samples as shown in Table 2, i.e., 145 for CJR withzoning versus 238 for purely probabilistic with zoning.It should be noted that the Bayesian prior used in thisexercise was very conservative, and a less conservative(lower) Bayesian prior would result in fewer randomsamples being required.


139

Fig. 26. Purely Judgmental sample locations on 2nd Floor.

Fig. 25. Purely Judgmental sample locations on 1st Floor.


140

Fig. 27. Purely Probabilistic sample locations on 1st Floor.

Fig. 28. Purely Probabilistic sample locations on 2nd Floor.


141

Fig. 30. Purely Probabilistic with Zoning sample locations on 2nd Floor.

Fig. 29. Purely Probabilistic with Zoning sample locations on 1st Floor.(Blue = random, ink = judgment).


142

Fig. 31. CJR with Zoning sample locations on 1st Floor.(Blue = random, Pink = judgment).

Fig. 32. CJR with Zoning sample locations on 2ndFloor.(Blue = random, Pink = judgment).

4.4.2 Cluster 2—1st Floor Release With SupplyFan ON

Conclusions from comparing the simulated scenarioto the different sampling approaches for Cluster 2 areshown in Table 3. Because most rooms were contami-nated, judgment sampling worked well at detectingcontaminated areas. Purely probabilistic sampling alsoidentified contaminated areas. However, because thepurely probabilistic sampling design was applied to theentire 1st floor, the possible cleanliness of smaller areason that floor could not be readily quantified with anX %/Y % confidence statement. Both of the samplingstrategies using a zoning approach were able to catego-rize the Class 3 zones as uncontaminated with 95 %confidence that at least 95 % of the area was below theACL, but the CJR with zoning approach required fewersamples.

4.4.3 Cluster 3—2nd Floor Release With SupplyFan OFF

Conclusions from comparing the simulated scenarioto the different sampling approaches for Cluster 3 areshown in Table 4. Judgment sampling worked well atdetecting contaminated areas but was not able to clearthe first floor as being uncontaminated due to insuffi-cient sample size. Purely probabilistic sampling identi-fied contaminated areas on the 2nd floor resulting in

95 % confidence that at least 95 % of the 1st floor was below the ACL. Both of the sampling strategies using azoning approach were able to categorize the Class 3zones (all of the 1st floor and part of the 2nd floor) asuncontaminated with 95 % confidence that at least95 % of the area was below the ACL, but the CJR withzoning approach required fewer samples.

4.4.4 Cluster 4—2nd Floor Release with SupplyFan ON

Conclusions from comparing the simulated scenarioto the different sampling approaches for Cluster 4 areshown in Table 5. Because most rooms were contami-nated, judgment sampling worked well at detectingcontaminated areas. Purely probabilistic sampling alsoidentified contaminated areas. However, because thetwo purely probabilistic sampling designs were appliedto the entire 1st floor and the entire 2nd floor, the pos-sible cleanliness of smaller areas on those floors couldnot be readily quantified with X %/Y % confidencestatements. Both of the sampling strategies using a zon-ing approach were able to categorize the Class 3 zoneon the 2nd floor as uncontaminated with 95 % confi-dence that at least 95 % of the area was below the ACL,but the CJR approach required fewer samples. Both ofthe strategies with zoning led to the reclassification ofthe 1st floor Class 3 zone as a Class 1 zone, becauseone of the rooms contained detectable contamination.


143

Table 3. Comparison of sampling strategies for Cluster 2 (1st Floor Release with Supply Fans ON). Comparison based on scenario shown inFig. 12 and Fig. 13

Sampling Approach Floor Class 1 Zones Class 2 Zones Class 3 Zones

Purely All but one room contaminated. For this room, there is not a high level of confidence thatJudgmental 1 the room is clean due to insufficient sample size.

ANDPurely

2All but two rooms contaminated. For these rooms, there is not a high level of confidence

Probabilistic that the room is clean due to insufficient sample size.Purely 1 Conclude the zone isProbabilistic contaminated Conclude with 95 % confidence that atwith Zoning least 95 % of the zone does not contain

2 N/A detectable contaminationConclude the zone is

1 Conclude the zone is contaminated Conclude with 95 % confidence that atCJR with contaminated least 95 % of the zone does not containZoning detectable contamination, but with

2 N/A fewer samples than purelyprobabilistic.

4.4.5 Cluster 5—Outdoor Release

Conclusions from comparing the simulated scenarioto the different sampling approaches for Cluster 5 are

shown in Table 6. Because all of the rooms were con-taminated on both floors, all of the methods concludeboth floors are entirely contaminated.


144

Table 4. Comparison of sampling strategies for Cluster 3 (2nd Floor Release with Supply Fans OFF). Comparison based on scenario shown inFig. 14 and Fig. 15


1 No contamination was found, but a statistical confidence statement cannot be made duePurely insufficient sample size and non-random sampling.Judgmental Contamination was found in slightly more than half of the rooms. For the

2 uncontaminated rooms, there is not a high level of confidence that the rooms are cleandue to insufficient sample size.

1 No contamination was found. There is 95 % confidence that at least 95 % of the floorPurely does not contain detectable contamination.Probabilistic Contamination was found in slightly more than half of the rooms, thus it is concluded the

2 floor contains detectable contamination. For several uncontaminated rooms, there is nothigh level of confidence that the rooms are clean due to insufficient sample size.

Purely Conclude with 95 % confidence that atProbabilistic 1 N/A least 95 % of the floor does not containwith Zoning detectable contamination.

2 Conclude with 95 % confidence that atConclude the zone is contaminated least 95 % of the zone does not contain

detectable contamination.Conclude with 95 % confidence that at

1 N/A least 95 % of the floor does not containdetectable contamination, but with fewer

CJR with samples than purely probabilistic.Zoning Conclude with 95 % confidence that at

2 Conclude the zone is contaminated least 95 % of the zone does not containdetectable contamination, but with fewersamples than purely probabilistic.

Table 5. Comparison of sampling strategies for Cluster 4 (2nd Floor Release with Supply Fans ON). Comparison based on scenario shown inFig. 16 and Fig. 17

Purely 1 Most rooms are contaminated. For three rooms not contaminated, there is not a highJudgmental 2 level of confidence that the room is clean due to insufficient sample size.

Purely 1 Contamination was found in most, and thus it is concluded the floor contains detectableProbabilistic contamination. For three uncontaminated rooms, there is not a high level of confidence

2 that the rooms are clean due to insufficient sample size.Low levels of contamination were found in

Purely 1 N/A one room in the zone, and therefore the zoneProbabilistic is reclassified as Class 1.with Zoning Conclude the zone Conclude with 95 % confidence that at least

AND Conclude the zoneis contaminated 95 % of the zone does not contain detectable

CJR with 2 is contaminated contamination, but with fewer samples thanZoning purely probabilistic.


4.4.6 Summary of Sampling Strategy Comparisons

The scenarios examined have rooms which wereeither entirely clean or entirely contaminated. The num-ber of samples for each method was sufficient enoughso that each room was sampled at least once. Thisresults in all sampling strategies being very effective atidentifying the contaminated areas for these particularscenarios. Where the strategies differ is in the state-ments that can be made about areas where no contami-nation was found, and in the number of samplesneeded to make conclusions.

One key difference between the sampling strategiesis their ability to make X %/Y % statements about areaswhere no contamination was found. The scenariosexamined for Clusters 1 through 4 had rooms on at leastone floor that were not contaminated. For each of thesescenarios, the two strategies which used zoning wereable to identify parts of the building where it was con-cluded with high confidence (95 %) that most (95 %) ofthese areas were uncontaminated, and for three of thefour cases were able to identify such areas on bothfloors. The purely probabilistic strategy was able makean X %/Y % statement about the 1st floor of thescenario examined for Cluster 3. The purely judgmen-tal strategy could not make an acceptable X %/Y %statement about any areas. In general, the methodsusing zoning were most effective at making X %/Y %statements and identifying areas which did not requiredecontamination. The CJR strategy with zoningrequired fewer samples while making the same conclu-sions as the purely probabilistic with zoning strategy.

Scenarios where an entire floor was contaminatedwere identified by all sampling strategies with the pure-ly judgmental strategy requiring the fewest samples.

The results of this exercise support the expectedstrengths of the different methods. Purely judgmentalsampling requires fewer samples for confirming con-tamination in areas known or highly likely to be con-taminated. If areas can be classified as Class 3, apply-ing a probabilistic or CJR sampling method allows thesampling team to make an X %/Y % about these areasif no contamination is found, and decontamination maynot be required. When sampling experts have sufficientprior information about the release event, the CJRmethod allows X %/Y % statements to be made withfewer samples.

5. Conclusions

CONTAM and VSP have been shown to be usefulfor evaluating multiple sampling strategies as appliedto buildings under a wide range of contaminant release,building configuration and operating scenarios.CONTAM can be used to perform virtual experimentsto supplement actual experiments. Its strength lies in itsability to perform multiple simulations of a wide rangeof scenarios in relatively little time when compared toactual testing. VSP provides the capability to exploreand employ a wide range of sampling strategies and hasfunctionality specifically designed for the sampling ofbuildings. In combination, VSP and CONTAM providea powerful set of tools with which investigators can

145


Table 6. Comparison of sampling strategies for Cluster 5 (Outdoor Release). Comparison based on scenario shown in Fig. 18 and Fig. 19


PurelyJudgmental 1

AND Conclude the entire floor is contaminated. All samples were contaminated.PurelyProbabilistic 2

PurelyProbabilistic 1with Zoning Conclude the entire floor is

AND N/A contaminated. All samples were N/ACJR with 2 contaminated.

Zoning

study potential building-specific sampling strategies orobtain general knowledge related to building sampledesigns.

One of the fundamental outcomes of this project wasthe demonstration of how CONTAM and VSP can beused together to effectively develop sampling plans tosupport the various stages of response to a CBRNevent. Following such an event, incident details and theconceptual site model could be used to create CON-TAM predictions of contaminant dispersion within abuilding. These predictions would then be used to iden-tify priority area zones within the building and thensampling designs and strategies could be developedbased on those zones.

Furthermore, CONTAM and VSP could potentiallybe used in preplanning exercises for a CBRN event.CONTAM and VSP can be used to explore a priori howone might best delineate priority sampling zones with-in a building in a manner applicable to a wide range ofplausible scenarios. Viewing a large number of scenar-ios can allow the classification of an event based on areduced set of parameters. This could allow samplingexperts to initially choose better sampling approachesfor different portions of the building, lead to clearinguncontaminated areas more quickly, provide a moreaccurate means of determining the number of samplesrequired to meet specific sampling goals, and evenimprove targeting of decontamination efforts.

CONTAM and VSP can be used to evaluate theeffectiveness of various sampling plans given varyingground truth scenarios. It was demonstrated that apply-ing separate sampling designs within each zone canlead to some areas of the building being cleared whenotherwise they would have been decontaminated orrequired further sampling. Together these tools can pro-vide insight into the potential trade-offs between thenumber of samples, sampling locations, and associateduncertainty statements of the various sample designsthat could be generated for a given set of scenarios.

CONTAM and VSP can be used to identify groundtruth scenarios that yield similar contamination pat-terns, and thus could use the same sampling plan.Cluster analysis was used to show that contaminant dis-persion patterns were similar across multiple scenarios,and therefore the same sampling plan could be used forsuch cases. This revealed the potential to reduce thenumber of sampling approaches or zoning schemesrequired to successfully cover a range of possible con-tamination scenarios.

This study was based on the uniform concentrationassumption of CONTAM and empirically determineddeposition rates from other studies. Consequently,

hotspot scenarios were not addressed and variation indeposition due to other factors were not accounted for,e.g., surface roughness and orientation, airflow, tem-perature and humidity. CONTAM does provide theability to simulate more detailed contaminant distribu-tion on a sub-room scale, e.g., using one-dimensionalconvection/diffusion and computational fluid dynamics(CFD) within selected zones. This is another possiblearea to follow-up to utilize and improve upon the sim-ulation of transport phenomenon within modeling toolsin order to address a broader range of scenarios and tofurther evaluate the combined effectiveness of integrat-ing sample planning tools.

The CONTAM-VSP tool could be used to leveragethe integration of VSP with the Building RestorationOperations Optimization Model (BROOM) [23] devel-oped by Sandia National Laboratories which is anotherDHS investment. An integrated CONTAM-VSP-BROOM tool would provide users with start-to-finishsample design and collection capability. CONTAMwould provide simulated predictions to guide zoningstrategies in order to create optimal sampling designs inVSP. These sampling designs would then be passed toBROOM to facilitate data acquisition and management.Data analysis of laboratory results could then be con-ducted by BROOM and VSP. Such a tool, when prop-erly used, could reduce the time and cost of responseand restoration by facilitating the rapid design andexecution of validated, statistically defensible samplingplans.Acknowledgements

This report was developed with the support ofDr. Bert Coursey, Director of the Office of Standards,Department of Homeland Security and the NIST Officeof Law Enforcement Standards. The authors would liketo express their appreciation to Dr. Barbara Jones forher role in supporting the initiation of this project andto acknowledge the support of the members of theValidated Sampling Plan Working Group led byDr. Randy Long of the Department of HomelandSecurity.

6. References

[1] Evaluation Report September 2007: Indoor Field Evaluation ofSample Collection Methods and Strategies at Idaho NationalLaboratory. Department of Homeland Security and JointProgram Executive Office for Chemical and BiologicalDefense: Washington, D.C. (2008).

[2] G. N. Walton and W. S. Dols, CONTAM 2.4 User Guide andProgram Documentation, 7251, National Institute of Standardsand Technology: Gaithersburg, (2005).


146

[3] B. D. Matzke, et al. Visual Sample Plan Version 5.0 User’sGuide, PNNL-16939. Pacific Northwest National Laboratory:Richland, WA. nt, 37: p. 1083-1090.

[4] Validated Sampling Strategy. Environmental ProtectionAgency: Washington, D. C., EPA (not yet published).

[5] D. M. Lorenzetti, Computational Aspects of Nodal MultizoneAirflow Systems. Building and Environment 37, p. 1083-1090(2002).

[6] J. W. Axley, Indoor Air Quality Modeling: Phase II Report,NBSIR 87-3661. National Institute of Standards andTechnology Gaithersburg, USA, (1987).

[7] W. S. Dols and A. K. Persily, A Software Tool for AnalyzingBuilding Airflow and Airborne CBR Agent Concentrations, in2008 International Joint Topical Conference and Expo onEmergency Management & Robotics for HazardousEnvironments. American Nuclear Society: Albuquerque, NM,(2008).

[8] W. S. Dols, NIST Multizone Modeling Website. 2001 October6, 2008 [cited December 16, 2008]; Available from:www.bfrl.nist.gov/IAQanalysis.

[9] W. W. Nazaroff, A. J. Gadgil, and C. J. Weschler, Critique of theUse of Deposition Velocity in Modeling Indoor Air Quality,STP 1205, N. L. Nagda, Editor. American Society of Testingand Materials: Philadelphia, (1993).

[10] Guidance on Systematic Planning Using the Data QualityObjectives Process (EPA QA/G-4), EPA/240/B-06/001.Environmental Protection Agency: Washington, D.C., EPA.(2006).

[11] A. K. Persily and E. M. Ivy, Input Data for Multizone Airflowand IAQ Analysis, NISTIR 6585. National Institute ofStandards and Technology: Gaithersburg, (2001).

[12] J. H. Klote and J. A. Milke, Principles of Smoke Management.Atlanta: ASHRAE, (2002).

[13] J. T. Firrantello, et al, Effects of HVAC System and BuildingCharacteristics on Exposure of Occupants to Short-DurationPoint Source Aerosol Releases. ASCE Journal of Architectural Engineering, 13(2): p. 84-94 (2007).

[14] K. Esbensen, S. Schonkopf, and T. Midtgaard,. MultivariateAnalysis in Practice. Trondheim, Norway: CAMO, (2007)

[15] The Unscrambler computer program. Version 9.7., CAMO.Oslo, Norway, (2007)

[16] Multi-Agency Radiation Survey and Site Investigation Manual(MARSSIM), NUREG-1575, Rev. 1, EPA 402-R-97-016,Rev.1, DOE/EH-0624, Rev. 1. Environmental ProtectionAgency: Washington, D.C., EPA. (2000).

[17] B. A. Pulsipher, L. H. Sego, and L. L. Nuffer, ValidatedSampling Strategy, PNNL-17711. Pacific Northwest NationalLaboratory: Richland, WA, (2008).

[18] P. Emanuel, K. Niyogi, and J. W. Roos, eds, Sampling forBiological Agents in the Environment. ASM Press, (2008).

[19] W. M. Bowen and C. A. Bennett, Statistical Methods forNuclear Material Management, (1988).

[20] E. G. Schilling, Acceptance Sampling in Quality Control. NewYork: CRC Press, (1982).

[21] L. H. Sego, et al, An Environmental Sampling Model forCombining Judgment and Randomly Placed Samples, PNNL-16636. Pacific Northwest National Laboratory: Richland, WA,(2007).

[22] B. G. Amidan, et al., Experimental Design for the INL SampleCollection Operational Test, PNNL-17129. Pacific NorthwestNational Laboratory: Richland, WA, (2007).

[23] Recovering from Bio-Attacks, Sandia Technology 7(2), (2005).

About the authors: Stuart Dols is a mechanical engi-neer and Andrew Persily is a mechanical engineer andgroup leader in the Building Environment Division ofthe NIST Building and Fire Research Laboratory.Jayne Morrow is an environmental engineer in theBiochemical Science Division of the NIST ChemicalScience and Technology Laboratory. Brett Matzke,Landon Sego, and Brent Pulsipher are statisticians inthe Statistics and Sensor Analytics group at PNNL.Lisa Nuffer is a computer scientist in the Data Intensiveand Scientific Computing group at PNNL.


147

115 development and demonstration of a method to evaluate ... · to evaluate bio-sampling...

Documents