application of the analytic hierarchy process for selecting and

Universities CoUnCil on Water resoUrCes JoUrnal of Contemporary Water researCh & edUCation

issUe 146, pages 50-63, deCember 2010

Application of the Analytic Hierarchy Process for Selecting and Modeling Stormwater

Best Management Practices Kevin D. Young, Tamim Younos, Randel L. Dymond,

David F. Kibler, and David H. Lee

Virginia Polytechnic Institute and State University, Blacksburg, VA

Abstract: Engineers are increasingly expected to consider several influential criteria when selecting a single best management practice (BMP) for stormwater management. These criteria include site physical characteristics; local, state, and federal pollution control ordinances; stakeholder input, and BMP implementation and long-term maintenance costs. This paper discusses the development of a software-aided approach based on the Analytic Hierarchy Process (AHP) as a decision-making tool for selecting stormwater management BMPs. Supported with input from a geographic information system, the AHP can provide an objective, mathematically-based alternative to the existing, often subjective BMP selection approaches. In this study, the developed AHP decision support software was applied in a demonstration site in the Town of Blacksburg, Virginia. First, the AHP decision support algorithm was applied to evaluate and rank BMP options for the demonstration site. Second, the EPA Storm Water Management Model (SWMM) was used to effectively model the BMPs recommended by the AHP software. Results indicate potential advantages over traditional BMP selection methods. Keywords: Stormwater management, best management practices, decision-making tool

Implementation of best management practice (BMPs) to control stormwater runoff is fundamental to most nonpoint sources pollution

control programs. A stormwater BMP is a device, practice, or method for removing, reducing, retarding, or preventing targeted stormwater runoff constitufents, pollutants, and contaminants from reaching receiving waters.

Conventional BMPs such as detention ponds are often effective for providing runoff rate attenuation but may not be as effective for water quality management purposes. Progress in developing innovative approaches to stormwater management has made a broad array of structural and non-structural BMPs available to BMP designers. Planners and engineers are increasingly expected to examine a broad range of criteria when selecting a BMP for a particular application. These criteria include physical site characteristics; local, state, and federal pollution

control ordinances; and implementation and long-term maintenance costs. Applying a large number of influential selection criteria to a wide array of available BMP options makes the selection of a single BMP for a particular application a daunting task. Consequently, stormwater management solutions too often default to the traditional “drain and detain” stormwater management approaches with little or no consideration given to innovative design strategies and BMP options.

The use of mathematically-based algorithms to assist in BMP selection is an evolving area of research, with a number of approaches recently attempted. One of these is a software-based decision support framework for developing cost comparisons on possible BMP design alternatives, assuming highly simplified conditions on the sizing and location of facilities (Lai et al. 2005). In addition, researchers have attempted to create BMP decision support tools for optimization of

50

Journal of Contemporary Water researCh & eduCationUCOWR

BMP placement and design. A number of these have been applied at the watershed scale, in combination with a rainfall-runoff-water quality process simulator, to examine the issue of local “on-site” versus sub-regional and regional scale benefits of BMP placement. Hipp et al. (2006) applied linear programming, in combination with simple runoff-pollutant loading models, to find least cost placement strategies for locating storm sewer insert filters at critical points in the City of Costa Mesa, CA storm sewer system. Zhen et al. (2004) describe use of the Scatter Search algorithm, combined with AnnAGNPS for physical watershed process simulation, to find least cost solutions for placing structural BMPs (detention basins) on a watershed basis. This work represents a departure from the use of genetic algorithms and their emphasis on randomization in selecting the component members of an optimal solution (Otero et al. 1995; Yeh and Labadie 1997). Carter et al. (2008) applied an extension of the earlier work by Zhen et al. (2004) to the Sun Valley watershed in the City and County of Los Angeles to examine least-cost combinations of porous pavement, infiltration trenches, and vegetated swales. Both distributed and regional placement strategies were evaluated.

A similar use of genetic algorithms for selecting the least-cost mix of agricultural BMPs on a 1014-ha watershed in Virginia is reported by Veith et al. (2004). The genetic algorithm optimization method was used successfully to reduce nonpoint source pollutants when compared to a “targeting” approach in which BMPs are sited uniformly across the watershed. A variation on genetic algorithms called the species conserving genetic algorithm, has been employed in tandem with the Soil Water Assessment Tool (SWAT) model to explore the best combination of BMPs from an assortment of detention and infiltration ponds, grassed waterways, and grade stabilization structures (Artita et al. 2007; Kaini et al. 2007). This work is notable because it examines not only the spatial distribution of BMP alternatives, but also employs BMPs other than structural detention systems.

The objective of this paper is to apply the Analytic Hierarchy Process (AHP) as a decision-making tool for selecting stormwater management BMPs. The described application of the AHP

algorithm is supported with input from a geographic information system (GIS). Specific objectives of this paper are:

1. Evaluation of the most common factors influencing the selection of urban BMPs,

2. Development the AHP decision support algorithm and software,

3. Application of the AHP algorithm to a demonstration site in a GIS environment, and

4. Modeling of various runoff management strategies on the demonstration site.

Methods

Factors Influencing the BMP Selection Process

When selecting a single BMP from a pool of competing options, numerous factors influence the decision making process. Most of these factors can be categorized as either physical site constraints or functional performance goals. Physical site constraints include the practice’s contributing drainage area, site soil type(s), topography, and other geologic factors. Functional goals include reducing the peak rate and volume of runoff from a developing site, and the removal of targeted pollutants from the runoff. Yet other criteria exist that are neither a functional goal nor a physical site constraint. These criteria include the practice’s aesthetic benefit (or liability), implementation and maintenance costs, and public safety issues associated with the BMP.

Table 1 presents the available selection criteria in the “BMP Selector” software. Users of the software may include as many or as few of these influential criteria as deemed necessary to assist BMP selection for a particular site or application.

The criteria deemed “essential” to the BMP selection process will almost certainly vary among a project’s stakeholders. While a site owner or developer may view annual BMP maintenance cost as the paramount selection consideration, the stormwater management designer likely views any number of unique site characteristics and/or technical performance goals as more critical to the BMP selection process. The criteria shown in Table 1 provide users of the BMP Selector software

51

UCOWR

Application of Analytic Hierachy Process

Journal of Contemporary Water researCh & eduCation

with a wide array of influential criteria to consider during the BMP selection process.

The AHP by design does not possess an objective function, penalty function, or randomization procedure. Rather, it is a procedure for the systematic evaluation and ranking of BMP alternatives, based on a wide range of criteria for selection and implementation of BMPs. The paramount benefit of the AHP as it relates to BMP selection is its ability to objectively and simultaneously consider an unlimited number of these criteria. Selective inclusion of the criteria depicting physical site characteristics enables the user to adapt the selection process such that the chosen BMP is feasible and appropriate for the site. Inclusion of the relevant performance goals contributes to achieving local, state, and federal regulatory stormwater management requirements. The performance goals for this study were decided in consultation with a select group of stakeholders who attended several focus group meetings with project investigators.

Development of the AHP Decision Support Algorithm and Software

The Analytic Hierarchy Process (AHP) was first developed by mathematician Thomas Saaty (Saaty

1980). The AHP is an algorithm capable of assisting complex decision-making problems. Perhaps the greatest strength of the AHP is that, although its foundation lies in complex matrix manipulation, it can be applied effectively without requiring the user to possess an in-depth knowledge of multi-criteria decision-making theory. “Fundamentally, the AHP works by developing priorities for alternatives and the criteria used to judge the alternatives” (Schmoldt 2001).

Fundamental to the AHP decision support algorithm is the construction of pairwise comparison matrices. Within the context of the herein described application of the AHP, these pairwise comparison matrices serve to compare and rank the various BMP alternatives in terms of each of the selection criteria identified in Table 1.

The AHP by design does not possess an objective function, penalty function, or randomization procedure. The AHP can be used for the systematic evaluation and ranking of BMP alternatives that require a wide range of criteria for selection and implementation of BMPs. The paramount benefit of the AHP as it relates to BMP selection is its potential to objectively and simultaneously consider an unlimited number of these criteria. Selective inclusion of the criteria depicting physical site

Contributing Drainage Area (CDA) Geologic FactorsCDA < 0.40 hectare (1 ac) shallow ground water depthCDA 0.40 - 2.0 hectare (1-5 ac) shallow bedrock depthCDA 2.0 - 4.0 hectare (5-10 ac) presence of hotspot runoffCDA 4.0 - 10 hectare (10-25 ac) best management practice costCDA > 10 hectare (25 ac) installation

CDA Impervious Fraction Annual Maintainence< 21 percent public safety concerns21 - 37 percent aesthetic benefit/liability38 - 66 percent peak runoff rate attenuation ability> 66 percent ability to recharge ground water

Hydrologic Soil Group (HSG)* Water Quality ImprovementHSG A removal of total suspended sediment (TSS)HSG D removal of total phosphorus (TP)Site Slopes/Topography removal of total nitrogen (TN)* HSG B and C soil classes are omitted, as they do not typically preclude the installation of any BMP.

Table 1. Available selection criteria in the BMP selector software.

Young, Younos, Dymond, Kibler, and Lee52


characteristics enables the user to adapt the selection process such that the chosen BMP is feasible and appropriate for the site. Inclusion of the relevant BMP performance goals contributes to achieving local, state, and federal regulatory stormwater management requirements. The performance goals for this study were decided in consultation with a select group of stakeholders who attended several focus group meetings with project investigators. Stakeholders for this study included county engineer, town engineer and planner, planning district commission planner, university physical infrastructure engineer, watershed roundtable representative (citizen groups), land developer, and state agency stormwater management experts. Expanded discussion on the AHP’s conceptual structure and further application to the BMP selection process is provided by Young et al. (2009). Below is an overview of the four steps for AHP application to BMP selection in chronological order.

Construction of Pairwise Comparison Matrices (Step One). Upon identifying all possible BMP alternatives (from which a single alternative is to be selected), it is necessary to identify the relevant criteria influencing the selection process. Because the previously described criteria exhibit varying units (or in some cases no units at all), mathematical evaluation of the criteria requires the operator to determine the relative “scale,” or performance, of the alternative BMPs in terms of each criterion. This task is accomplished by employing the Scale of Relative Importance (Saaty 1980). This scale and others developed since Saaty’s initial work, permits pairwise comparisons within the AHP. Saaty’s scale of relative importance is shown in Table 2.

Employing the scale of relative importance, one is able to construct BMP comparison matrices for each selection criterion. This step evaluates the performance of each BMP alternative against the other BMP alternatives in terms of the various selection criteria. These comparison matrices are of dimensions M×M, “M” being the total number of BMP alternatives considered.

The final comparison matrix is termed the criteria judgment matrix and is used to reflect the user-defined importance of the individual criteria

themselves. The criteria judgment matrix is of dimension N×N, “N” being the total number of user-chosen criteria. It is during the construction of the criteria judgment matrix that the operator is able to prioritize the criteria influencing the selection of the competing alternatives.

Entries into the BMP comparison and criteria judgment matrices are expressed in terms of the importance intensities illustrated in Table 3. For instance, consider a matrix comparing BMP alternatives “A,” “B,” and “C” in terms of criterion “One.” “By convention, the comparison of strength is always of an activity appearing in the column on the left against an activity appearing in the row on top” (Saaty 1980). An element in the matrix is equally important when compared with itself, and thus the main diagonal of all judgment matrices must be 1. Employing Table 3, consider the following scenario:• In terms of satisfying criterion “One,” BMP A

demonstrably outperforms BMP B. • In terms of satisfying criterion “One,” BMP C

weakly outperforms BMP A. • In terms of satisfying criterion “One,” BMP C

absolutely outperforms BMP B.

Intensity of Importance

Definition

1 The alternatives being compared con-tribute equally to the defined object.

3 One alternative is favored slightly over the other in terms of achieving the defined objective.

5 One alternative is favored strongly over the other in terms of achieving the defined objective.

7 One alternative is favored very strong-ly over the other in terms of achieving the defined objective.

9 The evidence favoring one alternative over the other is absolute in terms of achieving the defined objective.

2, 4, 6, 8 Intermediate values available express user-defined comparisons

Table 2. Scale of relative importance (adapted from Saaty 1980.)

53

UCOWR



Following the aforementioned convention, notice that the relative importances from Table 3 are found in row one, while their reciprocal values are found in column one. At this point, the comparison matrix of criterion “One” appears as in Table 3.

This step in the AHP is repeated until BMP comparison matrices are constructed for each influential criterion which the software developer has chosen to make available in the BMP selection process. BMP comparison matrices are “hard-coded” and their values do not change on a site-by-site basis. The rankings expressed in these matrices were derived from an extensive literature review which investigated BMP installation parameters, construction and maintenance costs, and functional performance. Pollutant removal matrices were developed from data found in the National Pollutant Removal Database (CWP 2000), the International Stormwater BMP Database (U.S. EPA, ASCE 2005), and the Technology Acceptance and Reciprocity Partnership clearinghouse on proprietary BMPs (MASTEP 2006). The user may select as many or as few of these criteria as desired for an individual BMP selection scenario.

The final task in this step is the construction of a criteria judgment matrix that weighs the importance of each selection criterion by comparing it against all other selection criteria. The BMP comparison matrices remain static, and do not change on an application basis. The evaluations of individual BMPs in terms of the various available criteria are based on literature review, data compilation, and professional judgment. In practice, the user may select as many criteria to influence a single BMP selection scenario as he or she wishes. Additionally, the chosen criteria may be weighted differently on an application-by-application basis. Therefore, in contrast to the BMP comparison matrices, the criteria judgment matrix is dynamic and may be altered by the introduction or removal of criteria or by choosing to modify the weighting of individual criteria.

Extraction of Priority Vectors (Step Two). Upon constructing the various BMP comparison matrices as well as the criteria judgment matrix, the analyst then proceeds to the next step, which is to extract the relative importance implied by each matrix.

This task is accomplished by employing matrix algebra to compute the right principal Eigenvector of each judgment matrix. Mathematically, the principal eigenvector for each matrix, when normalized, becomes the vector of priorities for that matrix (Saaty 1980). Eigenvector extraction can be facilitated through the use of proprietary mathematical software. Or, in the absence of proprietary mathematical software, a number of computationally accessible methods exist to estimate the priority vectors with acceptable accuracy. Below is the estimated Eigenvector for the matrix presented in Table 3.

{ } (1) The priority vector indicates that in terms of criterion “N,” alternative C is highly prioritized, with alternatives A and B ranking second and third, respectively. The quantified results support the previously described qualitative comparisons of alternatives A, B, and C.

Consistency Evaluation (Step Three). This step in the algorithm does not contribute directly to the goal of selecting a single BMP from the pool of competing alternatives, but rather exists to provide a logic-based consistency check on the validity of the analyst’s entries into the various comparison matrices. Execution of this step in the algorithm ensures that each matrix is within an acceptable consistency tolerance, and therefore does not inadvertently violate the comparison values intended by the analyst. The consistency evaluation within the scope of the AHP algorithm is a potentially lengthy and complex operation and detailed discussion of the procedure is beyond the scope of this paper. For details, the interested reader is referred to Young et al. (2009).

A B CA 1 7 1/3B 1/7 1 1/9C 3 9 1

Table 3. BMP comparison matrix (criterion “one”).

0.2950.0570.649



Ranking of Competing BMP Alternatives (Step Four). The final step in executing the AHP algorithm begins with construction of the BMP decision matrix. Column entries in the BMP decision matrix are simply comprised of the Eigenvectors (priority vectors) obtained from each individual BMP comparison matrix. The decision matrix is of dimensions M×N, “M” representing the number of BMP alternatives being considered, and “N” indicating the total number of influential criteria for which BMP comparison matrices were constructed. Considering three possible BMP alternatives (A, B, and C) and three selection criteria (i, j, and k), the following priority vector subscript convention can be adopted:

{ } (2)

The decision matrix would appear as follows:

[ ] (3)

To obtain the overall ranking of the alternatives, the BMP decision matrix is multiplied by the transpose (column version) of the row priority vector from the criteria judgment matrix. Considering the following subscript convention for the row priority vector of the selection criteria:

{Avei Ave

j Ave

k } (4)

The matrix multiplication operation is then formulated as follows:

[ ]{ } (5)

Executing this operation accomplishes weighting of each of the individual BMP priority vectors by the corresponding selection criteria. The overall rank of each alternative is shown as follows:

Rank of alternative A = AiAvei+ Aj

Ave

j+ AkAve

k (6)

Rank of alternative B = BiBvei+ BjBve

j+ BkBve

k (7)

Rank of alternative C = CiCvei+ CjCve

j+ CkCve

k (8)

An

Bn

Cn

Ai Aj Ak

Bi Bj Bk

Ci Cj Ck

Ai Aj Ak

Bi Bj BkCi Cj Ck

Avei

AvejAve k

Figure 1. Schematic flow chart of AHP algorithm in BMP selection.

Identify Influential Selection Criteria

Obtain or Construct Criteria-Specific BMP Pairwise Comparison Templates

Extract Priority Vector for Each BMP Comparison

Template

Enter Priority Vectors into the BMP Decision Matrix

Evaluate Relative Importance of

Each

Apply Scale of Relative Importance to Create the Criteria Judgment

Matrix

Extract the Criteria

Priority Vector

[ ] x { } → BMP Ranking Vector

55

UCOWR



Table 4 presents the various GIS layers compiled for the Town of Blacksburg, Virginia to provide input data for a case study using the BMP Selector software. The Parcel Boundaries layer is used to aggregate the various physical site parameters shown in Table 4. Attributing these physical parameters to individual parcels is useful when the proposed BMP installation will serve relatively small development projects confined to a single parcel. This enables users of the software to easily obtain the impervious fraction, soil type, and average slope for the specific parcel upon which a BMP installation is proposed. Similarly, the Sub-Watershed Boundaries layer is a set of polygons delineating the urban sub-watersheds contributing runoff to the Town’s storm sewer system. These urban watershed delineations reflect those identified as part of the Town of Blacksburg’s 2007 MS4 permit. The ability to evaluate the various physical site parameters on the sub-watershed scale is useful when the chosen BMP(s) will serve large developments spanning multiple parcels.

Application of the BMP SelectorThe first step in applying the BMP decision

support software to a site is to identify the criteria which should influence the BMP selection process. Upon selection of those criteria that are critical to the particular BMP selection scenario, the user qualitatively defines the relative importance of each individual criterion. This process is inherently subjective. However, certain guidelines do exist that improve the usefulness of results obtained from the software.

First, it is suggested that any physical site constraints be given the highest degree of influence during the BMP selection process. The rationale behind this recommendation lies in the fact that the chosen BMP must be suited to the unique physical constraints of the site upon which it is to be installed. For example, consider a site whose in situ soils are primarily HSG D. These soils will not exhibit the minimum infiltration rate required for installation of the infiltration BMPs. Consequently, if this physical site constraint was omitted or not prioritized during BMP selection, the algorithm could conceivably rank an infiltration BMP very favorably. Such a situation would require the user to manually

The alternative with the greatest rank is the most desirable, while successively lower ranks indicate less desirable alternatives. Figure 1 provides a schematic depiction of the AHP algorithm’s use as a BMP selection tool.

Results and DiscussionIn order to evaluate the utility of the AHP

decision support software, a demonstration site was chosen in the Town of Blacksburg, Virginia. Upon selection of the site, studies were performed in two steps. First, the AHP decision support algorithm was applied to evaluate and rank BMP options for installation on the demonstration site. Second, EPA Storm Water Management Model (SWMM) was used to construct a hydrologic model of the demonstration site in order to effectively model the BMPs recommended by the AHP software.

The project demonstration site is comprised of two phases. Phase I occupies approximately 10.5 hectares (26 acres), with the proposed development comprised of an outdoor shopping mall, a theater complex, restaurants, and a stormwater detention facility. Phase II comprises approximately 5.26 hectares (13 acres) on the northern portion of the site and is proposed as a “big box” retail development. It will consist of one large building and an associated parking area. Prior to its 2008 development, the site was predominately undeveloped open space.

Various BMP selection criteria for which BMP judgment matrices are available in the Virginia Tech BMP decision support software were shown in Table 1. As previously discussed, these judgment matrices are hard-coded and do not change on a site-by-site basis. However, the user may select as many or as few of these criteria as desired for an individual BMP selection scenario.

Many of the criteria influencing the selection of a BMP for a particular runoff management application are functions of a site’s physical characteristics. While some of these physical site characteristics can be adequately evaluated only through the completion of a detailed site investigation, some can be sufficiently assessed through the use of GIS software. This GIS-based desktop assessment is of particular value when stormwater management options are being evaluated at the planning stage of development.



override the BMP rankings obtained from the algorithm.

As with physical site constraints, it is generally advisable to provide any regulatory or functional stormwater management objectives with a great deal of influence during the BMP selection process. Frequently, the driving factor behind a BMP installation is to meet the stormwater management requirements imposed by the local review authority and/or the State. These regulatory requirements may include flood control in the form of peak runoff rate attenuation, providing ground water recharge, or a reduction in pollutant loads found in the runoff. If a regulatory requirement, such as peak runoff rate reduction, was omitted or not prioritized during the BMP selection process, the algorithm may rank a particular BMP very favorably when, in reality, that BMP is not capable of providing any runoff rate reduction.

Influential BMP selection criteria that are not categorized as a physical site constraint or a regulatory/functional objective can be given the degree of influence deemed appropriate by the individual user or stakeholder group.

Adhering to these guidelines, Table 5 qualitatively summarizes the relative importance assigned to each of the influential criteria introduced on the demonstration site.

Based on a detailed study of the site development plans, the most logical application of the software to the demonstration site was to evaluate sub-drainage basins whose areas ranged between one and five

acres. Subdivision of the demonstration site at this scale yielded a total of six sub-watersheds, not including rooftops. Within the context of this application, roofs were considered to be individual sub-watersheds. Four of the six sub-watersheds fell into the 38 to 66 percent impervious cover range. The remaining two sub-watersheds were greater than 66 percent impervious cover.

First, the software was applied on the demon-stration site’s four sub-watersheds whose impervious fraction ranged from 38 to 66 percent. Reflecting the qualitative criteria assessments shown in Table 5, Figure 2 identifies porous pave-ment as the highest ranking BMP alternative for this runoff management scenario. Given that the four sub-watersheds considered in this scenario are comprised of paved parking spaces, drive aisles, and parking lot islands, the results appear to be physically feasible in terms of installation of the BMP.

Further examination of the output reveals that the user-defined, qualitative criteria rankings were successfully expressed quantitatively. All physical site constraints and functional objectives were ranked equally with one another, and given the greatest overall influence (reference in Figure 2: Drainage Area, Impervious Fraction, Peak Mitigation Performance, TSS and TP Removal). Following these criteria, safety and nuisance liability was considered to be of moderate importance. Finally, the selection criteria depicting aesthetics and costs were given the lowest degree of influence in the BMP selection.

Spatial BMP Selection Criteria Description Data Source(s)parcel boundaries tax parcel boundaries shape file from town of Blacksburg

GISsub-watershed boundaries sub-watershed delineations shape file of sub-watersheds devel-

oped during the town of Blacksburg MS-4 permitting process

land cover identification of percentage imper-viousness within parcel and sub-wa-tershed boundaries

compilation of vector-format GIS data from various sources

soil type identification of area-weighted hy-drolic soil group within parcel and sub-watershed boundaries

NRCS Soil Survey Geography (SSURGO) database.

Site Slopes identification of average land slope within parcel and sub-watershed boundaries

town of Blacksburg LiDAR data

Table 4. GIS data compiled for the town of Blacksburg, Virginia.

57

UCOWR



Next, the BMP selector was applied on the demonstration site’s two sub-watersheds whose impervious fraction exceeded 66 percent. Reflecting the qualitative criteria assessments shown in Table 5, the AHP algorithm identified rainwater harvesting as the highest ranking BMP alternative. Given that the two sub-watersheds represented by this scenario are comprised of paved parking spaces, drive aisles, and parking lot islands, rainwater harvesting is not a feasible BMP. Further examination of the results revealed sand filters as the next highest ranking BMP alternative. Given the high imperviousness and small drainage areas comprising the two sub-watersheds of interest, the installation of stormwater sand filters appeared to be fully feasible.

Finally, the site’s proposed building rooftops were considered as individual sub-watersheds. For this application, neither CDA nor impervious fraction was included as selection criteria, but all other criteria remained as in the previous selection scenarios. Reflecting the qualitative criteria

assessments shown in Table 5, the AHP algorithm identified rainwater harvesting as the highest ranking BMP for management of runoff from the rooftops of the demonstration site.

SWMM Results

The final task entailed construction of three EPA SWMM models to evaluate two alternative runoff management strategies for the demonstration site. First, a baseline model was constructed for the demonstration site. The baseline model did not include any stormwater BMPs, and served only to depict the behavior and characteristics of stormwater runoff from the demonstration site if it were left unmanaged following project construction. Next, a centralized stormwater management model was created for the demonstration site to depict the traditional approach to stormwater management. This management approach is termed “centralized” because runoff from both phases of the site was

Criterion Category Relative Importancecontributing drainage area physical constraint highestimpervious percentage physical constraint highestaesthetic benefit/liability other moderatesafety and nuisance liability other moderately highimplementation cost other moderateannual O&M cost other moderatepeak mitigation ability functional objective highestrelative TSS removal functional objective highestrelative TP removal functional objective highest

Table 5. Qualitative assessment of influential selection criteria on demonstration site.

Figure 2. BMP selector output (watershed imperviousness 38 to 66 percent).



directed to one of three large storage facilities. These mass storage facilities were preceded by proprietary water quality control devices per the actual site design plans. Finally, a distributed, source-control stormwater management model was built for the demonstration site. The distributed model applied those BMPs suggested by the software (pervious pavement, sand filters, and rainwater harvesting devices).

In order to effectively model the BMPs recommended by the AHP software, templates were developed within the EPA SWMM modeling environment. These templates are constructed of typical SWMM elements such as storage units, nodes/treatment nodes, links, pumps, weirs, and outfalls.

The first BMP template configuration depicts those practices which discharge their treated water quality volume to the same outlet point as runoff that bypasses the BMP. These practices include the entire pond/basin class, constructed stormwater wetlands, stormwater filtering systems, vegetated buffers and swales, and bioretention basins. Within this template configuration, a pump is utilized to convey a portion of the inflow to a treatment node. The pump rate can be adjusted to reflect the actual rate at which the water quality volume passes through the BMP. Similarly, the treatment node can be adjusted to reflect the pollutant removal efficiency of the BMP type that is being modeled. After treatment, the water quality volume is then

discharged from the practice, again using a pump rate that is reflective of the BMP’s actual physical processes. Runoff that is not treated by the BMP is simply routed through the practice using storage, elevation, and discharge parameters representative of the actual BMP type. This SWMM template configuration is shown in Figure 3.

The second BMP template configuration depicts those practices which infiltrate or otherwise completely remove a portion of the overall surface runoff volume. Within this configuration, infiltrated runoff is not reintroduced to the downstream conveyance system after its removal. These practices include the entire infiltration class of BMPs, including porous pavement. Within this template configuration, a pump is utilized to convey a portion of the inflow to an outlet/discharge node. The pump rate can be adjusted to reflect the actual rate at which the water quality volume exfiltrates from the BMP. Runoff that is not treated by the BMP is simply routed through the practice and introduced to a downstream receiving channel or other conveyance.

The final BMP template configuration is applied exclusively to depict rainwater harvesting systems. Within this configuration, the storage unit can be sized such that it holds exactly the volume available in the cistern or storage tank being modeled. Then, a weir is placed at the elevation corresponding to the maximum storage volume, and excess runoff is instantly bypassed to a downstream receiving

Figure 3. U.S. EPA SWMM 5.0 BMP template, “Treatment configuration.”

59

UCOWR



Treatment Configuration

Weir/Outlet

Treatment Node

BasinStorage UnitNode

OverflowPumpOutlet of BMP

point. The volume of runoff contained in the storage unit can be held indefinitely, or drawn down at a rate reflective of the consumption rate for the actual BMP installation.

In an effort to compare performance of the centralized and distributed stormwater management models, various performance metrics were identified. These performance metrics include peak runoff rate reduction and the removal of total suspended sediment (TSS) and total phosphorus (TP). These performance metrics were evaluated at the downstream outfall of the demonstration site. At this location, the centralized and distributed stormwater management alternatives were compared against the baseline (no runoff management) model. Within the three models, pollutant buildup and wash off was established as a function of proposed land cover. Identical buildup and wash off functions were applied to each of the three models.

As with the pollutant buildup and wash off functions, the same rainfall event was applied to each SWMM model. The so called “first flush” or water quality volume is generally derived from a short-lived storm with an excess rainfall depth of at least 12 millimeters or about one half-inch (Lawson et al. 2009). For comparison modeling of the alternative stormwater management strategies, a storm event was chosen based on its ability to produce runoff from all sub-watersheds on the demonstration site without introducing excessive inundation and flooding of the BMPs. In order to evaluate the runoff rate reduction performance of the competing stormwater management strategies, a storm event with a 24-hour rainfall depth of 32 millimeters (1.25 inches), applied over a one hour period, was chosen for comparison of the centralized and distributed models. The

chosen rainfall depth was arbitrary except that it consistently met the aforementioned goals. Table 6 summarizes the results of the three runoff models, observed at the most downstream outlet of the demonstration site.

Compared to the base model, the centralized management approach yields 13.0 percent greater runoff rate reduction than does the distributed model. This result is not entirely unexpected, due to the large storage volumes afforded by the three centralized mass-storage facilities proposed in the actual site engineering plans. This observation implies that, in order to achieve peak rate reduction comparable to a centralized stormwater management approach, a distributed approach may require a supplementary impoundment BMP that provides additional detention and controlled release. However, the storage volume required of this supplemental facility will be substantially smaller than if it were implemented as the lone runoff management BMP, as in the case of the centralized approach.

Table 6 reveals that the distributed stormwater management approach greatly outperforms the conventional, centralized approach in terms of removing suspended sediment from surface runoff. While both runoff management approaches remove a significant portion of TSS from the demonstration site’s runoff, the distributed model achieves a nearly 40 percent greater removal efficiency. Similarly, the distributed stormwater management approach greatly outperforms the conventional, centralized approach in terms of removing phosphorus from surface runoff. Though both runoff management approaches remove a significant portion of TP from the demonstration site’s runoff, the distributed model achieves 38 percent greater removal efficiency.

Model Peak Flow Rate(cubic - m/sec)

(cfs)

Reduction(percent)

TSS Load(kg)

Reduction(percent)

TP Load(kg)

Reduction(percent)

Base 1.31 (46.2) -- 115.5 -- 52.9 --Centralized 0.20 (7.0) 85 81.6 29 36.5 31Distributed 0.37 (13.0) 72 35.5 69 16.3 69

Table 6. U.S. EPA Storm Water Management Model results (observed at outlet of overall site).



Conclusion

A seemingly limitless number of possible criteria may influence the choice of a BMP for a particular site or runoff management objective. As more and more criteria are introduced to the selection process, the task of objectively selecting a BMP becomes increasingly complex. This paper discussed the development of a software-aided approach to BMP selection. Supported with input from a GIS, the BMP Selector software provides an objective, mathematically-based alternative to the inherently subjective approaches currently used to choose BMPs for management of stormwater runoff.

In an effort to evaluate the utility of the BMP selection software, a demonstration site was chosen in the Town of Blacksburg, Virginia. The selection software was utilized to identify BMP alternatives for a distributed stormwater management approach on the demonstration site. Next, SWMM models were built to provide a comparative analysis of those BMPs recommended by the software against a traditional, detention-based stormwater management approach. In terms of pollutant removal, the distributed model significantly outperformed the detention-based stormwater management strategy for both pollutants considered in the study. Additionally, the distributed stormwater management approach provided significant runoff rate reduction when compared to a baseline model; however, it was slightly less than that provided by the detention-based management approach.

The selection of stormwater management BMPs on a site is an inherently subjective process susceptible to personal experience, bias, and judgment. For many years, the detention-based approach has been the predominate means by which stormwater management is addressed on land development projects. The widespread acceptance of this strategy results in a reluctance by many engineers to look beyond detention facilities for stormwater management needs. Unfortunately, impoundment facilities are often not the best runoff management option in terms of water quality improvement and cost.

The AHP algorithm has potentially significant benefit when utilized as a decision-support tool

in the BMP selection process. First, by providing a mathematically-based, objective means by which to evaluate alternative BMPs human bias can be minimized during the selection process. This advantage is particularly noteworthy when considering a functional BMP performance objective such as the removal of pollutants from runoff. Within the framework of the BMP Selector, BMPs are ranked by their ability to achieve threshold removal values for various nutrients and suspended sediment. These rankings are derived from industry databases and are hard-coded into the software and do not change on a site-by-site basis. Users can simply identify the pollutants of interest on a site, and then apply the algorithm to objectively identify which BMPs are capable of attaining the desired level of removal.

The software also affords its user the option to consider a wide spectrum of site-specific, physical characteristics when selecting a BMP. This, again, contributes to minimizing subjectivity during the selection process. Following an extensive literature review of state stormwater management manuals and other publications, BMP ranking matrices were developed for physical site characteristics such as drainage area, land cover, soil type, slope, and others. As with those matrices that rank BMP pollutant removal performance, these matrices are static within the software. The inclusion of these types of criteria in the BMP selection process helps to objectively ensure that the chosen BMP is feasible and well suited for the site upon which it is to be installed.

An added benefit of the AHP is the ability to alter the degree to which each included criterion influences the overall selection process. Unlike the BMP comparison matrices, the criteria comparison matrix is dynamic and can be altered on a site-by-site basis through the inclusion or removal of criteria, or by altering the degree to which each individual criterion influences the selection process. This permits the user to assign a very high degree of influence to essential performance objectives such as pollutant removal and/or peak runoff rate reduction, while simultaneously providing less overall influence to non-essential factors such as aesthetics. This intrinsic characteristic of the AHP algorithm is ideal in the field of BMP selection because it allows for customization as sites and

61

UCOWR



runoff management objectives change, yet retains a high degree of objectivity throughout the process.

In terms of BMP selection, the AHP is not without its limitations. Save for a trial and error approach, there is no readily available means by which to apply the algorithm to evaluate combinations of BMPs. In many localities, when infiltration is not feasible, two (or more) BMPs placed in series are required. As the software presently exists, the only means by which to select multiple BMPs for a given site is to examine the site as individual sub-watersheds. While this may be a justifiable approach for many scenarios, it yields a group of BMPs functioning independently throughout the site of interest. This approach would likely not satisfy the requirements of a locality requiring multiple BMPs installed in series.

Finally, it must be reiterated that the AHP provides a computationally accessible means of solving complex multi-criteria decision-making problems in which the influential criteria may be numerous and/or exhibiting varying units. However, one must be cautious when employing the AHP in engineering applications. Particular caution is issued when the final priority vector yields values very close to each other. When this is the case, the user should consider introducing additional influential criteria, which may yield greater discrimination among the competing alternatives. These shortcomings, however, are not unique to the AHP algorithm. Multi-criteria decision-making algorithms should be viewed as one tool in the decision-making process. When the AHP process is applied to BMP selection, the algorithm attempts to satisfy all selection criteria while adhering to the respective weights assigned to each criterion by the user. This attempt to simultaneously satisfy potentially conflicting criteria may yield results that do not fully satisfy each criterion individually. Therefore, the BMP rankings attained by employing the AHP must be critically scrutinized.

AcknowledgementsThis project was supported by the USEPA Grant No. 83340501-0 awarded to Virginia Tech. Mr. Stuart Lehman was the USEPA project officer. Daniel Phipps (graduate research assistant) assisted with data preparation and geospatial analysis. Project stakeholders

listed below provided technical information and valuable input: Mr. Ronald Bonnema (Montgomery County, Virginia), Mr. Kevin Byrd and Mr. David Rundgren (New River Planning District Commission), Mr. David Dent (Virginia Tech Infrastructure Office), Mr. Charles Dietz and Mr. Wil Orndorff (Virginia Department of Conservation and Recreation), Mr. Michael Harvey (New River Watershed Roundtable), Ms. Meredith Jones (Developer and Tom’s Creek Investors, L.C.), Mr. Wayne Nelson (Town of Christiansburg, Virginia), and Mr. Mathew Stolte (Town of Blacksburg, Virginia).

Author Bios and Contact Information Kevin Young is a Research Associate in the Via Department of Civil & Environmental Engineering at Virginia Tech. His areas of research and teaching interest include sustainable land development, optimization of BMP selection and design, hydrology, and watershed management. He can be contacted at Department of Civil & Environmental Engineering, Virginia Tech, 200 Patton Hall, Blacksburg, VA or by email at: [email protected].

Tamim Younos is a Research Professor in the Department of Geography and Associate Director of the Water Resources Research Center at Virginia Tech. His areas of research and teaching interest include environmental hydrology, watershed assessment and monitoring, stormwater management, and water-energy nexus. He is the editor of a book entitled “Total Maximum Daily Load: Opportunities and Challenges (PennWell 2005). Dr. Younos is a past president of the Universities Council on Water Resources. He can be contacted at the Cabel Brand Center, 2502 Plymouth Street, Blacksburg, VA 24060 or by email at: [email protected].

Randy Dymond is an Associate Professor of Civil and Environmental Engineering and Founding Director of the Center for Geospatial Information Technology (CGIT) at Virginia Tech. His areas of interest include geospatial application in water resources management and web-enabled decision support systems for watershed management. Recent research projects include Low Impact Development, Hazard Identification and Risk Assessment (HIRA). He can be contacted by email at: [email protected].

David Kibler is Emeritus Professor of Civil and Environmental Engineering at Virginia Tech. Dr. Kibler’s areas of interest include flood control, stormwater management, and computational methods in hydrology and hydraulics. In 1996, he was selected



Massachusetts Stormwater Treatment Evaluation Program (MASTEP). 2006. Stormwater Technologies Clearinghouse. Available at: http://www.mastep.net.

Otero, J., J. Labadie, D. Haunert , and M. Daron. 1995. Optimization of managed runoff to the St. Lucie estuary. Proceedings from the First International Conference on Water Resources Engineering, ASCE. New York. 1506 – 1510.

Saaty, T.L. 1980. The Analytic Hierarchy Process. New York, New York: McGraw-Hill International.

Schmoldt, D. L., J. Kangas, G. Mendoza, and M. Pesonen. 2001. The Analytic Heirarchy Process in Natural Resource and Environmental Decision Making. Dordrecht, The Netherlands: Kluwer Academic Publishers.

U.S. Environmental Protection Agency (U. S. EPA) and the American Society of Civil Engineers (ASCE). 2005. International Stormwater Best Management Practices (BMP) Database. Available at: http://www.bmpdatabase.org.

Veith, T. L., M. L Wolfe, and C. D. Heatwole. 2004. Cost-effective BMP placement: optimization versus targeting. Transactions of the American Society of Agricultural Engineers 47(5): 1585-1594.

Yeh, C – H., and J. W. Labadie. 1997. Multi-objective watershed-level planning of stormwater detention systems. ASCE Journal of Water Resources Planning & Management 123(6): 336-343.

Young, D. K., T. Younos, R. L. Dymond and D. Kibler. 2009. Virginia’s Stormwater Impact Evaluation: Developing and Optimization Tool for Improved Site Development, Selection and Placement of Stormwater BMPs. VWRRC Special Report No. SR44-2009. Virginia Water Resources Research Center, Virginia Tech, Blacksburg, VA. Available at: http://www.vwrrc.vt.edu/special_reports.html#2009

Zhen, Xiao-Yue, S. L. Yu, and Jen-Yang Lin. 2004. Optimal location and sizing of stormwater basins at watershed scale. ASCE Journal Water Resources Planning & Management 130(4): 339.

as one of three recipients of the ASCE Wesley J. Horner Award for the paper “Sizing Stormwater Detention Basins for Pollutant Removal.” He is a coauthor of the ASCE Manual of Practice on Design & Construction of Stormwater Drainage Systems published in 1995.He can be reached at Department of Civil & Environmental Engineering, Virginia Tech, 200 Patton Hall, Blacksburg, VA or by email at [email protected].

David Lee is a Project Engineer with Withers & Ravenel in Cary, North Carolina. David holds a Bachelor’s and Master’s Degree from Virginia Tech. At the time of this study, David was a Graduate Research Associate in the Department of Civil and Environmental Engineering at Virginia Tech. He can be reached at Withers & Ravenel, 111 MacKenan Dr., Cary, NC 27511 or by email at [email protected].

ReferencesArtita, K. S., P. Kaini, and J. W. Nicklow. 2007.

Generating alternative watershed-scale BMP designs with evolutionary algorithms Proceedings from 2007 World Environmental and Water Resources Congress, ASCE. Tampa, FL.

Carter, S., X. Zhen, A. Parker, and C. Cutter. 2008. An innovative stormwater BMP decision support system for quantifying and optimizing load reductions and costs in Los Angeles. Proceedings from 2008 World Environmental and Water Resources Congress, ASCE. Ahupua’a/Honolulu, HI.

Center for Watershed Protection. 2000. National Pollutant Removal Performance Database for Stormwater Treatment Practices 2nd Edition. Ellicott City, MD.

Hipp, J. A., O. Ogunseitan, R. Lejano, and C.S. Smith. 2006. Optimization of stormwater filtration at the urban/watershed interface. Environmental Science & Technology 40(15): 4794-4801.

Kaini, P., K. S. Artita, and J. W. Nicklow. 2007. Designing BMPs at a watershed scale using SWAT and a genetic algorithm. Proceedings from 2007 World Environmental and Water Resources Congress, ASCE. Tampa, FL.

Lai, Fu-hsiung, L. Shoemaker, M. Cheng. 2005. Decision Support Framework for Placement of BMPs in Urban Watersheds. U.S. EPA Science Forum.

Lawson, S., A. LaBranch-Tucker, H. Oho-Wack, R. Hill, B. Sojka, E. Crawford, D. Crawford, and C. Brand. 2009. Virginia Rainwater Harvesting Manual, Cabell Brand Center, Salem, VA.

63

UCOWR



application of the analytic hierarchy process for selecting and

Documents