Estimating Regionalized Planning Costs of Green Infrastructure and Low-Impact Development Stormwater Management Practices: Updates to the US Environmental Protection Agency’s National Stormwater Calculator

Abstract

Estimating regional costs of green infrastructure (GI) and low-impact development (LID) stormwater management controls is an important issue for many municipalities and water utilities in the United States. The National Stormwater Calculator (NSWC) is a site-scale, planning-level tool. A regional cost estimation methodology was recently added to the NSWC, enabling users to estimate region-specific capital and maintenance costs of commonly used GI controls. This paper discusses the approach used to estimate and regionalize costs in the NSWC.

Introduction

The National Stormwater Calculator (NSWC) was developed by the US Environmental Protection Agency (USEPA) in response to the need for a tool that uses more advanced modeling approaches [e.g., continuous simulation of urban stormwater best management practices (BMPs) and long-term hourly historical weather] with a simple user-friendly interface. This tool was developed to evaluate hydrologic response from urban stormwater runoff using various stormwater management solutions to control urban stormwater runoff impacts for small drainage areas (≤4.9 ha or ≤12 acres) and to assist in the planning of green infrastructure (GI) and low-impact development (LID) implementation (Rossman and Bernagros 2019). The NSWC leverages computational components of USEPA’s Stormwater Management Model (SWMM; Rossman 2015), and is an open-source software application with code available on a GitHub repository (Rossman et al. 2019). The NSWC is a screening-level planning tool that can help users nationwide (including Puerto Rico and US Virgin Islands) to understand whether GI projects can provide adequate, long-term, site-specific hydrologic control of stormwater runoff given historical rainfall. The NSWC offers access to simple analyses to evaluate design and site suitability factors for achieving GI-based retention and volumetric control objectives, as well as alternative cost comparisons. GI controls included in the NSWC are downspout disconnection, rain harvesting, rain gardens, green roofs, street planters, infiltration basins, and permeable pavements. In response to stakeholder interests and concerns over regional GI construction and operation and maintenance costs, as well as capital return on investment, a regional cost estimation module was developed for the NSWC and released in April 2017. Outreach by USEPA was conducted, with a focus on the NSWC cost estimation module, at multiple professional conferences and public webinars (Berner et al., National Stormwater Calculator: Low-Impact Development Stormwater Control Cost Estimation Programming & Future Enhancements, unpublished report; Berner, Lifecycle Cost Analysis of Green Infrastructure Workshop; USEPA National Stormwater Calculator: Low-Impact Development Stormwater Control Cost Estimation Module and Future Enhancements, unpublished report). See also Berner (2017), Bernagros (2017, 2018a, b).

Several cost estimating tools that provide capital and operation and maintenance cost estimates for GI projects were reviewed to establish a baseline understanding of existing tools and provide insight for the development of the NSWC cost estimation module. These existing tools typically focused on ease of use. Features included build-up of unit costs, sensitivity to change, consistency of output, inclusion of regulatory and permitting costs, and clarity of cost calculations. Existing national cost estimating tools that were reviewed included the following: Green Values National Stormwater Management Calculator (CNT 2009); Best Management Practice (BMP) and LID Whole-Life Cycle Cost Models Version 2.0 (Water Research Foundation 2009); Best Management Practices—Rational Estimation of Actual Likely Costs of Stormwater Treatment (BMP REALCOST) (Olson et al. 2013); US Department of Transportation’s (DOT) BMP Evaluation Tool (Transportation Research Board 2014); Department of Defense’s (DoD) Stormwater Management Optimization Toolbox (SMOT) Model Selection Tool (MST); US Army LID Planning and Cost Tool (USACE 2018); and Autocase SITES (Autocase 2018). The development methodology for the NSWC cost estimation module was partly based on strengths and weaknesses of existing costing tools evaluated (Table 1).

Table 1.

Evaluation of existing costing tools: strengths and weaknesses

Name of tool	Description	Strengths	Weaknesses
National Green Values Stormwater Management Calculator	Web-based tool, most similar in terms of “ease of use” to what was desired for the NSWC	• National tool for grey and GI practices: construction, annual maintenance, and lifecycle costs •Provides cobenefits	• Static costs (2009) • Single sources for costs data • Costs data primarily from the Chicago area
LID Whole-Life Cycle Cost Models Version 2.0	Set of spreadsheet tools to evaluate whole life costs of stormwater BMPs	• National tool: lifecycle costs (capital and maintenance), utilizes RSMeans data	• Static costs (2008)
BMP REALCOST Tool	Spreadsheet tool that provides BMP performance (volume and pollutant load reduction) and cost estimates (construction and annual maintenance) for the Denver, CO, region. Uses design standards from the Urban Drainage and Flood Control District	• National tool: construction and maintenance costs • Utilizes national costs from Engineering News- Record • Accounts for the land value costs (Denver area)	• Static costs (2008) • Costs do not account for regional differences; regional costs only for Denver area
DoD’s Stormwater Management Optimization Toolbox (SMOT): Model Selection Tool (MST)	SMOT incorporated a Model Selection Tool (MST) for selecting the best model approach (continuous simulation or simple design storm approach) based on key factors that influence the necessary modeling sophistication to meet Energy Independence and Security Act (EISA) 438 volumetric compliance guidance for the minimum cost possible. When continuous simulation is recommended, the most cost-effective combination of BMPs that might be installed are determined using USEPA’s System for Urban Stormwater Treatment and Analysis IntegratioN (SUSTAIN) model and the BMP Decision Support System (BMPDSS) tool (USACE 2019). SMOT is planned to be a publicly available tool (H. Howard, personal communication, 2019). SUSTAIN provides construction unit costs for individual components of each BMP.a BMPDSS, developed in 2005, provides BMP costs based on excavation volume and land costb	• Optimizes selection of GI practices, at the least cost to meet EISA 438 requirements • National tool for DoD facilities: construction and land costs • Utilizes continuous simulation modeling	• Static costs (2005 and 2009) • Only applies to DoD facilities • Does not provide maintenance costs
US Army LID Planning and Cost Tool	Provides planning-level construction and maintenance costs divided into direct and indirect costs. Direct costs include materials, labor, and equipment while indirect costs include subcontractor and prime contractor overhead, productivity, profit, and contractor bond. The cost data used in the Army LID Planning and Cost Tool are from the Micro-Computer Aided Cost Estimating Software System and Area Cost Factors (installation location) derived from RSMeans in 2016 (USACE 2018)	• National tool for DoD facilities: construction and maintenance (direct and indirect costs) • Costs derived from RSMeans	• Static costs (2016) • Only applies to DoD facilities
US DOT’s BMP Evaluation Tool	A national stormwater BMP evaluation spreadsheet tool for transportation projects	• National tool for transportation projects: lifecycle, construction, and maintenance costs • Costs based on annual pollutant load removal	• Static costs (2014) • Only applies to transportation projects
Autocase SITES	A national proprietary site planning costing tool that provides triple bottom line cost analysis results for GI projects	• National tool: construction, operation, and maintenance costs; including cobenefits	• Limited documentation and many undisclosed calculations • Proprietary software, limits public use

^aData from Shoemaker (2009).

^bData from Tetra Tech (2010).

Several characteristics separate the NSWC from other existing cost estimating tools. The NSWC provides annually updated regional cost adjustments that are calculated by automatically comparing year-to-year changes in US Department of Labor’s (US DOL) Bureau of Labor Statistics (BLS) online cost data. The NSWC also addresses cost uncertainty by reporting cost estimates as a range rather than a single value. Spatial cost variations due to materials and labor cost differences across the United States are accounted for in the NSWC by automatically calculating and assigning an appropriate default regional cost-adjustment factor based on the user’s location, and by allowing users to select or provide a different factor if desired.

A major drawback of most existing GI cost estimating tools is that cost data can quickly become outdated, which limits their utility. The NSWC costing module provides annually updated national and regional planning-level estimated capital investment cost ranges, as well as average annual operation and maintenance costs of GI stormwater management controls for small (≤4.9 ha, ≤12 acres) GI projects in the contiguous United States, Hawaii, and Alaska. The NSWC provides users cost estimates as a range (low and high) of the potential construction and annual operation and maintenance costs of project(s).

The cost module provides construction estimates based on three primary site considerations that have been shown to have larger effects on projects costs. The first consideration is whether the project is a redevelopment or new development. This category recognizes the potential for additional costs associated with retrofitting or redeveloping an existing site, which often poses different challenges for implementation compared with new development. The second consideration is a combination of site complexity and other relevant constraints—factors that can substantially change design and construction costs. Applying a rating of poor, moderate, or excellent, these factors consist of site postconstruction soil infiltration rate, project topographic slope, existing utilities, physical obstructions, required onsite infrastructure (e.g., retaining walls or similar infrastructure), site access, haul distance, necessary dewatering, custom media blends, and the need to address geotechnical or groundwater concerns. The third consideration is whether there is a requirement or need for construction of formal pretreatment, as a separate unit or as a part of the GI controls, which may be recommended or required depending on the GI control type, location, drainage area, or local regulations. The cost estimate scenarios can be used to assist with selecting GI control types, gain a basic understanding of the change in costs with project site complexity or other site constraints, evaluate select design features, and gain an appreciation for predicted annual operation and maintenance costs per GI control type. While many other costs—such as land acquisition, permit fees, or contingencies—are recognized as important, these considerations are often site-specific and difficult to account for in a national or regional cost estimation tool.

The cost estimates are applicable to stormwater projects nationwide that range in size from <1 to 12 acres. Cost estimates are reported as a national value or can be regionalized to one of 17 geographic locations based on population centers determined by US DOL’s BLS online databases. These include the Producer Price Index (PPI) and the Consumer Price Index (CPI).

Regression models were developed to determine the most influential (predictive) variables from the PPI and CPI that estimate GI construction and annual maintenance costs and were correlated to published commercially available cost indexes (CACI). CACI are developed and sold by third-party vendors that track pricing on a national or regional basis often for a predetermined basket of goods in various markets across the country.

Several municipalities across the US require or recommend the use of the NSWC for estimating planning-level costs and stormwater runoff reduction benefits of GI projects. The Northeast Ohio Regional Sewer District (NEORSD) GI Grant Program requires the use of the NSWC for estimating the volumetric runoff reduction benefits of GI controls in grant applications (NEORSD 2017). The National Fish and Wildlife Foundation (NFWF) encourages the use of the NSWC for applications for its Southeast Michigan Resilience Fund and the Long Island Futures Fund to support the planning, design, and implementation of GI projects to mitigate flooding hazard impacts near coastal areas (NFWF 2018, 2019). The NSWC was used in a tribal climate capacity-building workshop in 2017 in Michigan to assess suitable GI controls for costs and runoff reduction benefits in support of funding for implementation projects (University of Michigan 2019).

This paper summarizes the development and function of the NSWC cost estimation module, including the costing methodology. The primary focus is on the cost regionalization process and results of internal and external validation efforts. The paper also discusses potential tool limitations.

NSWC Cost Estimation Methodology

The NSWC cost estimation module produces estimates of probable construction and maintenance costs. It then applies a cost regionalization factor based on the location of the project. Construction costs vary widely across the nation, and the regionalization factor accounts for some of that regional variability. While the focus of this effort is on the cost regionalization approach for the NSWC cost estimation module, a brief discussion of the cost estimation approach is included to provide context.

Cost Estimation

Inputs to the cost estimation module are included in the “LID Controls” tab of the NSWC’s desktop version 2.0.0.1 and consist of site suitability factors and the user’s geographical region (Fig. 1). These inputs are used to select appropriate regression equations for calculating cost ranges. The NSWC cost estimation module produces graphical and tabular estimates of probable capital costs for LID implementation (Figs. 2 and 3).

Fig. 1.

NSWC’s LID Controls screen showing cost module inputs at the lower left-hand corner. (Image © 2016 Microsoft Corporation Pictometry Bird’s Eye © 2016 Pictometry International Corp.)

Fig. 2.

Graphical results screen showing output of the NSWC cost module results.

Fig. 3.

Tabular results screen showing output of the NSWC cost module results.

The NSWC’s cost estimation approach is based on data from literature unit costs, constructed project unit costs, and commercially available unit costs for each supported GI control type. Unit quantities were estimated based on known properties of GI controls. The influential design variables and construction complexity variables include soils hydrologic group (A, B, C, or D), slope (flat, moderate, or steep), site suitability (poor, moderate, or excellent), and whether pretreatment occurs (yes, no). To estimate quantities, apply unit costs, and develop cost curves, a spreadsheet macro was created to generate GI controls of incremental sizes. The spreadsheet macro estimated capital and maintenance costs, plotted estimated costs, and determined linear regression equations for best-fit trendlines for each GI control. The regression equations were used in place of a unit cost database, enabling quick cost computational runtimes without the need for database lookups and repeated cost build-ups computations.

To illustrate how the semiautomated cost curve generation approach works, we describe our generation of cost curves for rain gardens (Fig. 4). A spreadsheet macro was created to size rain gardens based on several design variables, including allowable ponding depth [varied from 10.2 cm (4 in.) to 30.5 cm (12 in.)], and soil media thickness [varied from 15.2 cm (6 in.) to 91.4 cm (36 in.)]. The macro was used to vary the size of the rain garden from 0.9 m² (10 ft²) to 27,000 + m² (300,000 + ft²) in over 70 increments, while recalculating the cost after each size change based on unit cost build-ups applied to estimated quantities of its components. This was repeated for changes in design variables to generate tables of costs that were then used to create cost curves (Fig. 4).

Fig. 4.

Example of the rain garden cost curves in the NSWC.

For each supported GI control, six cost curves representing high and low ranges for simple, typical, and complex design scenarios were developed. The definitions of “simple,” “typical,” and “complex” are described further in the NSWC User’s Guide (Rossman and Bernagros 2019). Each design scenario represents a combination of assigned design and influential variable values. There is one set of cost curves for each of the seven GI controls included in the NSWC. The determination of simple, typical, or complex is based on several factors, including whether the project is new or a redevelopment, site suitability, topography, soil type, and whether pretreatment is included.

Cost Regionalization Methodology

Cost regionalization accounts for variations in construction costs among major metropolitan areas. The cost regionalization methodology is based on the computation of a regional cost adjustment factor using the geographical location of a project. The factor is then applied to estimated construction costs to account for the variability of costs across the nation. Cost regionalization factors are calculated using publicly available data from the BLS via its open application programming interface (API). BLS data are used to generate regression equations based on correlations to published CACI such as RSMeans cost indices (Gordian 2015) and Engineering News-Record (ENR 2015). Although regionalization of costs was a significant goal in developing the cost module of the NSWC, maintaining the relevance of cost estimates over time without the need for tool updates was also a desired outcome. The cost module was therefore developed to access the BLS API directly to ensure that estimates are always based on the latest available BLS data, thereby meeting both objectives.

Regression models were developed to determine the most influential (predictive) PPI and CPI variables that estimate GI construction and annual maintenance costs. The PPI measures the average change over time in the sale prices received by domestic producers for their output (BLS 2019b). PPI data are assembled from over 25,000 establishments providing approximately 100,000 price quotations per month for individual products specified through disaggregation. Industries and products undergo systematic resampling as needed to account for changing market conditions. PPI is an indicator that reflects price change from the perspective of the seller. PPI data are available on a national scale only and not by city or region.

The CPI is a measure of the average change over time in prices paid by urban consumers for a predetermined “market basket” of consumer goods and services (BLS 2019a). The CPI represents price change from the purchaser’s perspective. It includes 200 categories organized into eight major groups, including food and beverages, housing, apparel, transportation, medical care, recreation, education and communication, and other goods and services. CPI data are collected and reported on a regional basis, and also averaged to represent a national index.

The regional cost methodology, data, and quality assurance documents have been published on data.gov (Bernagros 2020).

A summary of the seven-step methodology used to develop regionalization factors from PPI and CPI data is described below and depicted in Fig. 5.

Fig. 5.

Regionalization approach process flow diagram.

Step 1—National PPI data review:

The goal of this step was to identify the set of goods and services included in the producer-based PPI that were likely to predict changes in GI construction or operation and maintenance costs. All PPI categories from the BLS database were assessed for inclusion. Engineering judgment was used to identify an initial shortlist of PPI categories likely to be predictive of GI control costs because of their use in or relation to construction of GI controls. Relevant PPI categories identified included concrete storm sewer pipe, asphalt paving mixture, engineering services, and construction sand and gravel (Table 2).

Table 2.

Potential BLS PPI variables (national)

Cost variable	Starting year
All other plastic pipe, including storm drain	2012
Asphalt and tar paving mixture	2005
Asphalt paving mixtures and block manufacturing	1976
Burial vaults and boxes, precast concrete	1978
Concrete contractors, nonresidential building	2008
Concrete culvert pipe	2000
Concrete pavers	1981
Concrete storm sewer pipe	2000
Construction sand and gravel	1982
Diesel fuel	1985
Dump trucking	2015
Engineering services	1997
Graders, rollers, and compactors	2004
Non-building-related engineering projects	1997
Nursery, garden, and farm supply services	2003
Plastic water pipe	1988
Ready-mix concrete manufacturing	1965
Real estate brokerage, nonresidential	2010
Regular gasoline	1977
Tractor shovel loaders (skid steer, etc.)	1976
Transportation engineering products	2010
Wood chips	1981

Step 2—National CPI data review:

The aim of Step 2 was to identify the set of goods and services included in the CPI that are likely to influence GI construction costs. As consumer-based goods and services, the CPI has fewer items that represent GI construction costs; however, CPI data are reported for cities, metropolitan areas, and regions, and therefore can be used to develop regional cost variations. Those items that were representative of or related to GI construction were placed on the initial CPI variables shortlist. The CPI goods and services that were identified with GI construction and operation and maintenance costs are shown in Table 3.

Table 3.

Potential CPI variables

Cost variable	Starting year
Purchasing power of the consumer dollar	1980
Durables	1980
Fuels and utilities	1980
Services	1980
Utilities and public transportation	1980
Transportation	1980
Transportation commodities less motor fuel	2009
Fuel oil and other fuels	1980
Energy services	1980
Water and sewerage maintenance	1980
Motor fuel	1980
Gasoline (all types)	1980

Step 3—National PPI data correlated to published CACI:

The goal of this step was to determine which PPI categories identified in Step 1 were predictive of GI control cost variability based on regression analyses. This step developed a linear regression model using generalized least squares to determine the combination of PPI variables that best fit and predict variations in published CACI and GI construction cost data. Using the linear least squares fitting method (including a least-squares fitting polynomial), all the potentially influential variables identified in Step 1 were analyzed. Note that the least squares method seeks to minimize the sum of the squares of the residuals. Several transformations of these data, including exponential, logarithmic, and power transformations, were also investigated. R-squared values, p-values, β-value, and residual and line fit plots were examined to assess the goodness of fit of the predicted data and to test for multicollinearity and heteroscedasticity. Models that did not meet the R-squared, p-value criteria (α < 0.10) or pass multicollinearity and heteroscedasticity tests were eliminated. The analyses were performed for datasets with complete data from 1986 to 2014. Ready-mix concrete, tractor shovels, and diesel fuel were found to be the most predictive of published CACI data. The results of the regression between ready-mix concrete, diesel fuel, tractor shovel loaders, and published CACI are shown in Tables 4 and 5.

Table 4.

Regression statistics and results for published CACI versus national PPI variables of ready-mix concrete manufacturing, diesel fuel, and tractor shovel loaders

Comparison statistics	Value
Multiple R	0.997121
R square	0.99425
Adjusted R square	0.99356
F-statistic	F1,29 = 1,440.9, p < 0.0001

Table 5.

Coefficients and p-values for the regression between published CACI and national PPI variables of ready-mix concrete manufacturing, diesel fuel, and tractor shovel loaders

Variable	Coefficients	p-value
Intercept	−14.4466	0.00037
Ready-mix concrete manufacturing	0.46474	3.99 × 10−7
Diesel fuel	0.05761	0.00016
Tractor shovel loaders	0.353826	4.76 × 10−5

Step 4—National CPI data with National PPI data correlated with published CACI:

The goal of this step was to develop a regression model that was predictive of published CACI values using the national PPI and regional CPI data. The regression model used the variables from the shortlist of potential national consumer-based CPI factors identified in Step 2 and the national producer-based PPI variables from Step 3 which were regressed with the published CACI values. This step was completed to iteratively select the most appropriate combination of variables from these indices (Table 5). Using the same process as in Step 3, R-squared, p-values, β-value, residual, and line fit plots were evaluated to determine the ability of each combination of CPI and PPI variables to represent published CACI data.

The final regionalization regression model is shown in Eq. (1)

$$\begin{gathered} {\text{cost}\text{index}}_{\text{year}n}=-19.4+\left(0.113\times ready-mixconcret{e}_{\text{year}n}\right) \\ +\left(0.325\times tractorshovelloade{r}_{\text{year}n}\right) \\ +\left(0.097\times energ{y}_{\text{year}n}\right) \\ +\left(0.398\times fuelsandutilitie{s}_{\text{year}n}\right) \end{gathered}$$ (1)

Model statistics are shown in Tables 6 and 7. Note that the ready-mix concrete and tractor shovel loader variables are from the PPI, while the energy and fuels and utilities are from the CPI. The reference year for the model is 2014. Due to changes in macroeconomic conditions, computed regional adjustment factors can increase, decrease, or stay the same from year to year. The y-intercept of the model, while negative in this instance, is dependent on the reference year chosen. It accounts for some of the overlap inherent in the variables, and the equation should always produce a positive value.

Table 6.

Coefficients and p-values for the selected cost model for national PPI and CPI values

Variable	Coefficients	p-value
Intercept	−19.4284	0.00037
Ready-mix concrete manufacturing	0.113389	0.05448
Tractor shovel loaders	0.325493	1.13 × 10−5
Energy	0.096662	0.13369
Fuels and utilities	0.398318	0.02372

Table 7.

Statistics for the selected cost model

Comparison statistics	Value
Multiple R	0.998145
R square	0.996294
Adjusted R square	0.995676
F-statistic	F1,29 = 1,612.9, p < 0.0001

Step 5—Monthly data check:

The inputs used in Steps 1–4 were annual average data. The goal of Step 5 was to confirm that the model developed in Steps 1–4 also represented monthly values. Using monthly data, the model was found to be within ±4% of the annual model and within ±6% of the published historic CACI. For a planning-level tool, differences observed were considered within reasonable limits. This outcome confirmed that use of annual data in the development of the regression model resulted in a model that has reasonable representativeness on a monthly scale. It should be noted that a monthly scale was not an intended temporal resolution for regional cost adjustment, and Step 5 was simply a validation step.

Step 6—Develop regional adjustment factors:

The construction costs in the NSWC were developed using an aggregation of individual construction line item unit costs selected from construction sites across the country. The goal of this step was to develop a single factor for each city or region to account for relative regional differences in construction and operation and maintenance costs. The selected regional BLS CPI data for all the major urban areas (regions) available were obtained to develop the regional cost adjustment factors. For each of the regions, region-specific CPI values for the current year were plugged into the model to generate the regional cost adjustment factors. As a calibration step, this model was compared with historical (1986–2014) CACI values (Fig. 6). The model was found to be within ±4% for most regions. The regional cost adjustment factors for each of the regions were expressed as fractions of a national cost adjustment factor with the national value set to 1.00. The intent was to allow the regional cost adjustment factors to be applied as a cost multiplier. A regional cost adjustment factor <1 indicated that the modeled construction costs in that region would be lower than the national average, and >1 indicated higher than the national average. The results were compared with other relative regional cost of living and construction cost data, and were found to be reasonable. For example, Honolulu and San Francisco produced the highest regional cost indices, while Houston had the lowest regional cost multiplier. This relationship is similar to other generalized cost of living calculators and regional price parity tools—e.g., the Council for Community and Economic Research, Bureau of Economic Analysis (BEA) Cost of Living Index—although these tools are not customized to stormwater infrastructure cost variables and are often related to other goods and services (Council for Community and Economic Research 2018; BEA 2019). Fig. 7 shows a chart of modeled cost regionalization factors for US Urban Areas (as defined by BLS) for 2000–2015.

Fig. 6.

Comparison of historical CACI and modeled data.

Fig. 7.

Model-predicted cost multipliers of US urban regions.

Step 7—Validation:

The final step in the methodology was to validate the model by comparing model estimates with case study data. The goal of this step was to verify that the cost estimation and regionalization approach produced estimates that were comparable with construction and operation and maintenance costs for relevant GI projects. Five regional case studies were identified for use in validation based on location and availability of robust cost data for constructed, representative stormwater management practices. The five studies were located at Dillwyn, Virginia, Chesterland, Ohio, Mission, Kansas, and Portland, Oregon (two sites). Maintenance costs were not available for the case study sites. Regional validation was completed by applying the NSWC to each site and comparing the regionally adjusted cost estimates’ ranges to actual construction costs for each case study. A summary of the comparisons of the actual and estimated construction costs for these case study sites is shown in Table 8.

Table 8.

Case study sites actual and estimated costs comparisons (in 2014 dollars)

Case study site and location	GI controls	Actual construction cost	NSWC estimated construction costs using regional cost index	NSWC estimated construction costs with national cost index
Metcalf Avenue Permeable pavement, Mission, Kansas	Disconnection, rain harvesting, rain gardens, permeable pavement	$1,138,000	Used national cost index of 1	$526,900–$704,800
Buckingham Elementary School, Dillwyn, Virginia	Disconnection, rain harvesting, rain gardens, permeable pavement	$268,662	Used national cost index of 1	$203,800–$270,800
Multnomah Building green roof, Portland, Oregon	Green roof	$231,333a	$370,200–$495,900	$345,981–$463,458
Ramona Apartments EcoRoof, Portland, Oregon	Green roof	$288,000b	$278,800–$374,000	$260,560–$349,532
Chester Township Hall parking lot, Chesterland, Ohio	Permeable pavement	$75,000	$118,100–$141,900	$128,369–$154,239

^aData from City of Portland, Oregon, Bureau of Environmental Services (BES) (2004).

^bData from City of Portland, Oregon, Bureau of Environmental Services (BES) (2011).

Results and Discussion

Two of the five cost estimates were within the range estimated by the NSWC cost module. There were variations between the actual construction costs and the estimated cost ranges in some cases. For example: (1) the upper range estimate for the Mission, Kansas, GI construction project was 62% of the actual cost; (2) the lower range estimate for the Multnomah green roof in Portland, Oregon, was 160% of the actual cost; and (3) the lower range estimate for the Chesterland, Ohio, GI project was 157% of the actual cost. The Mission, Kansas, project included an underground detention component. Because the NSWC does not support underground detention, cisterns were used as a surrogate and likely led to underestimated underground detention costs. The Amy Joslin Memorial Eco-Roof installed in the Multnomah Building was a retrofit project conducted at a site that was classified as complex. In general, green roof costs are highly variable and challenging to predict. Two factors are believed to have affected accurate cost estimation of the Multnomah green roof. First, the complex classification implies that, because the green roof was a retrofit, there was a high likelihood of poor or challenging site suitability (e.g., the roof might require additional engineering support to accommodate the green roof load). If calculated as a less complex project based on the site suitability criteria for green roofs (low, moderate, and high), the predicted range would have included the actual green roof costs. Second, the existing load-bearing capacity and structural roof materials, including a five-ply roofing system, needed no alterations to accommodate the retrofit, leading to cost savings that are not always realized with typical green roof construction (BES 2004). The Chesterland, Ohio, project was a demonstration project funded by an Ohio Environmental Protection Agency grant that may not have funded all of the construction costs, such as volunteer involvement or cost sharing not documented (Chester Township Town Hall 2012). This demonstrates how some knowledge of projects may be necessary in the cost estimation process.

The regional cost adjustment factor improved the estimated range of construction costs for the Ramona Apartments EcoRoof in Portland, Oregon, and the Chesterland, Ohio, projects compared with the national adjustment factor. The estimated cost range for the Chesterland, Ohio, project was lower with the use of the regional adjustment factor, resulting in better prediction of the actual construction costs. For the Multnomah Building green roof in Portland, Oregon, the regional cost adjustment factor estimated a range of construction costs greater than the actual construction costs. In this case, the national cost adjustment factor was a better predictor of actual costs. The remaining case study sites used the national cost adjustment factor because those projects were not located near urban centers for which regional cost adjustment factors were available.

As another point of validation, a recent publication documented GI maintenance costs from across the US and compared the data to existing GI cost tools (Clary and Piza 2017). Multiple municipalities surveyed (including Austin, Texas, Plymouth, Minnesota, Portland, Oregon, St. Paul, Minnesota, Charlotte, North Carolina, Fort Collins, Colorado, Lakewood, Colorado, Newark, Delaware, Overland Park, Kansas, Kansas City, Kansas, Charlottesville, Virginia, and Seattle, Washington) provided information about their reported GI maintenance costs for bioretention practices. The publication showed that the maintenance costs for the case studies were within the high and low maintenance cost ranges estimated by the NSWC cost module (Fig. 8). The publication gathered data on maintenance only and did not include estimated construction costs.

Fig. 8.

Comparison of bioretention average annual maintenance cost survey to USEPA and Urban Drainage Flood Control District (UDFCD) cost estimating tools. (Reprinted from Clary and Piza 2017, © ASCE.)

Limitations of the Regionalization Approach

The regionalization approach used in the NSWC cost module represents an incremental advancement in stormwater cost estimation tools. However, there are several limitations in the approach as detailed below.

Periodic Evaluation of Regression Representativeness

Because the NSWC cost module is based on a regression model of select variables, the model itself requires periodic assessments to determine whether the identified variables reliably predict GI control costs. Model coefficients or representative BLS variable updates may be required if the tool becomes less predictive over time. The update frequency is dependent upon the rate of change of macroeconomic conditions, as well as the ability of the cost variables to represent construction costs accurately. The current recommended review period is every 5–7 years. The review period is loosely based on long-term inflation rates in the US.

Changes in BLS-Reported Areas

Every 10 years, the BLS reevaluates defined urban area centers based on census data (BLS 2020). Therefore, on a decadal basis, there is a potential change in the representative urban centers. The NSWC regionalization approach is based on data that are derived from 20 + years of information from one location. When BLS locations change, revisions are required to accommodate the new reporting areas; however, these data may be inadequate to develop reliable and representative regressions for the first few years at new locations. Tools that rely on published cost indices typically suffer similar limitations and often lag current-year calculations by at least 1 or more years.

Limited Recognition of Cost Variability in Nonmetropolitan Centers

The BLS reports CPI and PPI data for major population centers. Regionalization indices derived from BLS data are, therefore, most applicable to those urban areas. Several strategies exist for extrapolating the indices to cover more rural regions. The current approach applies the regionalization factor within a 160.9-km (100-mi) radius of the urban area’s center for which these data were collected. Regions that are more than 100 mi away are assigned the national value by default. The 160.9-km (100-mi) radius is relatively arbitrary and may be too large or too small for representing all urban areas. In some cases, there may be more than one urban center within the 160.9-km (100-mi) radius, creating overlapping coverages. The NSWC provides users the ability to select from several urban centers, to use the national value, or to override the regional cost index with a user-defined value. This provides flexibility for users that are outside of urban centers to make customized adjustments based on known relative construction costs in their region.

Appropriate Benchmarks

The regression model used to develop regional cost indices required data for benchmarking the model output. The NSWC used published historic CACI and other case studies as verification benchmarks during development. As more data become available across the US for construction and maintenance costs, testing of the NSWC with other indices and case studies are recommended at a similar reevaluation interval, as mentioned above. A database to track changes in cost across time and geographical regions is recommended to improve this aspect of cost estimating tools in the future.

Conclusions

The NSWC’s cost estimation module advances the science and capacity of the stormwater sector to estimate regional planning-level capital and maintenance costs of GI controls in the United States. While the cost estimation module is intended for estimating planning-level costs, it provides users a relative magnitude of regional stormwater project costs. It also allows for simple cost comparisons among GI control types, to select the most appropriate GI control. The module delivers estimated cost based on the region in which the GI control is implemented or national-level costs. Regression cost equations for each GI control was developed using a buildup of GI construction unit costs and other related planning and construction information in the cost estimate procedure. These regression equations were developed based on project cost data, peer-reviewed literature, and cost tools. Cost estimation and regionalization relies on BLS PPI and CPI data. The regional cost estimation methodology was validated with five regional case studies that included actual construction data. Additionally, Clary and Piza (2017) conducted a municipal survey that validated the maintenance costs in the NSWC. Using annually updated BLS cost data enables estimates to remain fairly current and supports nationwide and regionally based cost estimates for construction and operation and maintenance of GI controls.

The reliability and accuracy of the cost estimation values produced by the NSWC may be further validated with sharing or publication of GI project implementation data including costs. USEPA is focused on the development of various capabilities within life cycle costs, tools, and models for stormwater management controls (USEPA 2015). There is a need to advance the understanding of regional costs of GI controls as they become more commonly used throughout the world and as the body of knowledge grows. The cost estimation module of the NSWC has expanded the ability of stormwater and planning professionals to simply estimate regional planning-level costs of GI controls throughout the US and in major metropolitan areas.

Data Availability Statement

Some or all data, models, or code generated or used during the study are available in a repository online, found in Bernagros (2020), in accordance with funder data retention policies.