Estimating the
 Value of Land
for Use In Drinking Water Cost Models








Contract # EP-B16C-00027

Draft 




October 2020






                                                                               
Prepared for:
Rajiv Khera
U.S. Environmental Protection Agency
1200 Pennsylvania Ave., NW
Washington, D.C. 20460





Submitted by:
Abt Associates 
6130 Executive Blvd.
Rockville, MD 20852 



Table of Contents
1.	Introduction	1
2.	Background	2
2.1	The Regulated Industry	2
2.2	Economic Analysis of Regulatory Actions	3
2.2.1	Social and Private Costs	3
2.2.2	Incorporating Land Costs	4
3.	Data	7
3.1	Rural Land Value	7
3.2	Urban Land Value	8
3.3	Number of Rural and Urban Systems	12
4.	Method	17
4.1	Overview	17
4.2	Estimating the Probability of a Rural or Urban System	17
4.3	Estimating the Expected Rural Land Cost	18
4.4	Estimating the Expected Urban Land Cost	19
4.5	Estimating Overall Expected Land Cost	20
5.	Comparisons to 2013 Analysis	22
5.1	Urban and Rural Probabilities	22
5.2	Expected Rural Land Value	23
5.3	Expected Urban Land Value	23
5.4	Changes in Expected Value	25
6.	Limitations and Uncertainties	26
7.	References	27
8.	Appendix A: Urban/Rural Classification Procedures	29
8.1	Data Differences	29
8.2	Method Difference	30

Table of Exhibits

Exhibit 1. Community Water Systems by Population Served and Primary Water Source	2
Exhibit 2. Estimated Ratio of Land Costs to Treatment Process Costs by Technology and System Size (service population range)	6
Exhibit 3. Pasture Land Values (2020 dollars)	7
Exhibit 4. Average Industrial Land Values by State and Place Population Size Category (2020 dollars)	9
Exhibit 5. Estimated Number of Rural Community Water Systems, by State and Population Served Size Category	13
Exhibit 6. Estimated Number of Urban Community Water Systems, by State and Population Served Size Category	14
Exhibit 7. Estimated Probability of Rural and Urban Location for Community Water Systems, by Population Served Size Category	18
Exhibit 8. Expected (Weighted Average) Value per Acre of Rural Land (2020 dollars)	19
Exhibit 9. Expected (Weighted Average) Value per Acre of Urban Land (2020 dollars)	20
Exhibit 10. Expected Value per Acre of Land for Community Water Systems (2020 dollars), by Population Served Size Category	20
Exhibit 11. Urban and Rural Probabilities, 2013 and 2020, by Population Served Size Category	23
Exhibit 12. Expected Rural Land Values, 2013 (2019 dollars) and 2020 (2020 dollars), by Population Served Size Category	23
Exhibit 13. Expected Urban Land Values, 2013 (2019 dollars) and 2020 (2020 dollars), by Population Served Size Category	25
Exhibit 14. Expected Land Values, 2013 and 2020 (2020 dollars), by Population Served Size Category	25
Exhibit 15. Sources of Uncertainty in the Expected Land Value Estimates	26
Exhibit A - 1. Schematic of GIS Analysis for Classifying Facilities as Urban or Rural	29
Exhibit A - 2. Illustration of Method for Calculating Potential Positional Errors (in meters)	31

Abbreviations and Acronyms

CWS
community water system
EPA
U.S. Environmental Protection Agency
GIS
geographic information system
NDWAC
National Drinking Water Advisory Committee
NTNCWS
non-transient non-community water system
O&M
operating and maintenance
PWS
public water system
SCADA
supervisory control and data acquisition
SDWA
Safe Drinking Water Act
SDWIS/FED
Federal Safe Drinking Water Information System
TNCWS
transient non-community water system
UA
urbanized area
USDA
U.S. Department of Agriculture
WBS
work breakdown structure

Introduction
The Safe Drinking Water Act Amendments of 1996 and numerous other statutes and executive orders require the U.S. Environmental Protection Agency (EPA, or the Agency) to estimate regulatory compliance costs as part of its rulemaking process. EPA uses these estimates in social benefit-cost analyses, as well as various distributional analyses (such as those required by the Regulatory Flexibility Act and the Small Business Regulatory Enforcement Fairness Act), to inform its decision making. Thus, improvements in the accuracy of regulatory compliance cost estimates enhance the Agency's policymaking capabilities.
EPA has several models to estimate the cost of treatment technologies to comply with drinking water rules. These models use a work breakdown structure (WBS) to generate system-level treatment cost estimates based on engineering analysis at the component level and corresponding unit costs. EPA pursued the WBS approach to derive system-level costs to address recommendations regarding its methods for estimating drinking water compliance costs (USEPA, 1996). One of the recommendations was to incorporate the cost of land associated with incremental treatment requirements.
This report describes EPA's research and resulting method for estimating the cost of land in economic analyses of drinking water regulations, and is organized as follows. Section 2 provides background information on the regulatory context for estimating drinking water treatment costs and the recommendations EPA is addressing with this research. Section 3 presents available data on the cost of land, and information on where drinking water systems are located. Section 4 describes the method for calculating expected land values from the data, and the resulting per acre estimates. Section 5 provides a comparison of the land cost estimates to a similar analysis conducted in 2013 (Leidos, 2013). Section 6 describes the limitations and uncertainties in the land cost estimates.

Background
This section addresses three topics. First, it provides background information on the types of drinking water systems potentially affected by new regulations, and EPA's approach for evaluating cost impacts to these systems. Following this is a discussion of relevant economic analysis concepts. Finally, this section addresses specific Blue Ribbon Panel comments regarding land acquisition costs.
The Regulated Industry
There are more than 148,000 public drinking water systems (PWS) in the United States (USEPA, 2020a). A PWS is a publicly or privately owned entity that regularly supplies drinking water to 25 or more people or 15 or more service connections. More than 50,000 systems are community water systems (CWS), which serve the same customers year-round. The service populations range from 25 people to several million people. Exhibit 1 shows the number of CWS by population size category and primary water source. 
Exhibit 1. Community Water Systems by Population Served and Primary Water Source
Population Served
Primarily Ground Water
Primarily Surface Water
Water Source Unknown
Total
<=100
11,179
1,011
13
12,203
101-500
13,276
2,160
11
15,447
501-1,000
4,211
1,183
3
5,397
1,001-3,300
5,555
2,481
0
8,036
3,301-10,000
2,794
2,243
1
5,038
10,001-50,000
1,345
2,004
0
3,349
50,000-100,000
161
407
0
568
>100,000
74
367
0
441
Total
38,595
11,856
28
50,479
<=100
11,179
1,011
13
12,203
Source: USEPA (2020b).
Alternatively, non-community systems provide drinking water in nonresidential settings. The majority of this type of system serve transient populations like campgrounds and gas stations. These transient non-community water systems (TNCWS) do not serving 25 or more of the same people for at least six months per year. Some, called non-transient non-community water systems (NTNCWS), treat the same population for more than six months per year (e.g., schools). 
EPA drinking water standards vary across types of systems because of variations in exposure. For example, surface water treatment rules that regulate exposure to biological contaminants apply to surface water systems of all types because there is higher exposure of water to possible biological contamination compared to systems with ground water sources. On the other hand, rules regulating carcinogens such as arsenic apply to CWS and NTNCWS, but not to TNCWS. Since CWS and NTNCWS serve the same population more frequently, the service population is a higher risk because of higher exposure to water contaminants consumed daily.
EPA's national cost analysis approach accommodates this variability in regulatory impacts because it develops system-level cost estimates for those systems that require treatment (or non-treatment) to comply with a new regulation. The system-level costs for treatment vary by technology and incorporate capital costs for equipment, construction, and engineering, as well as operating and maintenance (O&M) costs. System size, or population served, is an important determinant of costs because systems that treat large flows will incur higher capital and operating costs than systems treating smaller flows. The WBS models provide system-level costs using a flexible framework in which costs vary by system size. The WBS models include a land cost among the up-front, one-time costs based on estimated land requirements for capital items and buffer area and unit cost estimates for land (i.e., $/acre). The land requirements vary by system size and treatment technology.
--------------------------------------------------------------------------------
Economic Analysis of Regulatory Actions
When EPA promulgates a regulation affecting drinking water systems, such as a rule establishing the maximum level of a contaminant, it evaluates the rule's incremental social costs and benefits, along with its potential impacts on household water expenditures. These two types of analysis answer different policy questions. The social benefit-cost analysis addresses the question of whether a regulation improves overall well-being (i.e., even though there are distributional effects, theoretically the gainers could compensate the losers such that everyone is at least as well off as before the policy). In contrast, the economic impact analysis evaluates the distributional effects on various subpopulations affected by the policy  -  who gains and who loses, and by how much.
The types of costs EPA includes in these two analyses differ. For the social benefit-cost analysis, EPA estimates social cost, which it compares with social benefit to evaluate the net change in social welfare. For its economic impact or distributional analyses, EPA calculates private costs. This section addresses the distinction between these two types of cost estimates and its implications for including land costs in the WBS models.
Social and Private Costs
As described in EPA's (2010) Guidelines for Preparing Economic Analyses, total social cost is the sum of the opportunity costs incurred by society as a result of a new regulation. Opportunity costs are the value of the goods and services foregone by society when resources are used to comply with and implement a regulation. For example, if a regulation requires a system to install new filter basins, the social costs include the opportunity costs of the labor and materials needed to design, build, and operate the basins because these resources are not available for other uses.
These "real-resource" compliance costs are the principal component of total social costs. They come from purchasing, installing, and operating new pollution control equipment, or otherwise altering production processes to comply with regulatory requirements. Other types of social cost include costs incurred by regulatory agencies in their monitoring, administrative, and enforcement tasks (EPA, 2010).
EPA also estimates the private cost of compliance for drinking water suppliers as an input into an economic impact analysis of a policy. The impacts can be for an industry as a whole, or for individual households (EPA, 2010). Private costs are the out-of-pocket cost of compliance incurred by drinking water utilities. Because utilities pass these costs on to households as drinking water fees, private costs are used for evaluating household impacts of new rules.
As noted above, social and private costs may or may not be equal. The private or out-of-pocket costs for new treatment equipment or additional operator labor will most likely equal the social or opportunity costs. In other instances, however, an out-of-pocket expense may not be a social cost because the expense does not represent a real resource allocation that has an opportunity cost. An example is a transfer payment such as a tax. A tax transfers money from one entity to another, but it does not represent an opportunity cost because there is no accompanying use of real resources such as labor. Although transfer payments are an out-of-pocket expense and, therefore, are included in private costs, they are excluded from social cost analysis. Conversely, there are social costs that do not have corresponding out-of-pocket expenses and, therefore, are not included in private cost estimates. In particular, costs associated with the environmental impacts caused by new waste streams will not be captured in an estimate of incremental out-of-pocket expenses unless there are waste disposal fees.
The private and social costs for land may differ. For a new treatment process, the social cost of land is the opportunity cost of using space for the process equipment, housing structures, and surrounding access space. Thus, there will always be social costs associated with a new treatment process. This is true regardless of whether a system incurs the private cost of purchasing additional land for a new process. Including land costs in the WBS models will provide estimates of social cost, but may overstate private costs.
Incorporating Land Costs
EPA's Blue Ribbon Panel on Safe Drinking Water Act Costing (the "Panel") identified site-specific or local conditions as the most significant factor affecting the compliance cost for individual water systems (EPA, 1996). Land acquisition is among the more variable site-specific costs that the Panel recommended EPA address in its regulatory analyses. As an example, the Panel mentioned that many ground water systems have wells located in residential areas that provide little or no room for a treatment plant. In addition, local ordinances and restrictions may prohibit siting near existing facilities, resulting in remote operations that have to be integrated into the overall treatment process. Supervisory control and data acquisition (SCADA) and other telemetry processes reduce the cost of integrating these processes, but not to the level of those for systems with readily available low cost land.
Due to the uncertainty in the magnitude of cost for such site-specific factors, the Panel recommended that EPA develop a lower bound estimate that does not include site-specific costs, and an upper bound estimate that does include site-specific costs. Then, sensitivity analyses could be used to evaluate large differences in factors and cost scenarios. The National Drinking Water Advisory Council (NDWAC) recommended that EPA include land costs for all technologies even though land may not be a major cost driver and poses certain difficulties of estimation (NDWAC, 2001). NDWAC suggested estimating land costs as 2% to 5% of the total unit capital cost; however, it did not provide a basis for this range.
To address these comments, EPA is including land cost estimates in the WBS cost models. These models provide a detailed breakdown of costs based on specifications such as flow rate, raw water quality, treated water quality, and engineering design assumptions; installed capital cost estimates at the component level are then aggregated to obtain a total process cost estimate. The WBS models also provide cost estimates for indirect capital costs, add-on costs, and annual operating and maintenance costs. The cost of land is a line item among the add-on costs, and the estimate is the product of a unit cost per acre of land and a land acreage requirement that is calculated in the WBS engineering analysis. For social cost analyses, the market value of land or out-of-pocket costs are a reasonable proxy for relevant opportunity costs.
EPA's prior analysis of land costs (Leidos, 2013) showed that those costs were generally well below the 2% to 5% range recommended. That analysis applied the unit land cost estimates to acreage values generated by the WBS models and then divided total land cost by total process costs. Exhibit 2 shows the resulting percentages for several treatment technologies and eight system size categories (i.e., population served). For most technologies and system size categories, the land costs equaled less than 2% of process costs. The percentages tend to be higher for the smallest system size (25 to 100 people), then decline through the fourth size category (1,001 to 3,000 people). The percentages increase for the fifth size category because the WBS models include a design breakpoint of 1 million gallons per day, which leads to larger process footprints and subsequent land requirements. 
Exhibit 2. Estimated Ratio of Land Costs to Treatment Process Costs by Technology and System Size (service population range)
Technology
25-100
101-500
501-1,000
1,001-3,300
3,301-10,000
10,001-50,000
50,001-100,000
100,001-1 million
Activated Alumina[1]
2.2%
1.8%
1.0%
0.8%
1.1%
0.9%
0.7%
1.2%
Anion Exchange
2.0%
2.0%
1.5%
1.4%
1.6%
1.3%
0.9%
1.2%
Cation Exchange
1.9%
1.8%
1.3%
1.1%
1.4%
1.4%
1.0%
1.1%
Chlorine - Disinfection
5.1%
4.7%
3.3%
3.6%
5.1%
3.7%
2.6%
3.2%
Conventional Filtration
0.7%
0.9%
0.9%
1.7%
1.9%
0.4%
0.6%
1.5%
Direct Filtration
0.7%
0.7%
0.6%
0.8%
2.3%
0.2%
0.4%
0.9%
Granular Activated Carbon [2]
1.8%
1.2%
0.5%
0.4%
0.4%
0.3%
0.2%
0.3%
Greensand Filtration
1.6%
1.7%
1.0%
0.9%
1.2%
1.1%
0.8%
1.5%
Hypochlorite - Disinfection
1.6%
1.4%
0.9%
0.8%
1.2%
1.2%
0.9%
1.4%
Lime Softening
-
-
0.4%
0.7%
0.7%
1.0%
1.1%
2.6%
Multistage Bubbling Aeration
1.1%
1.1%
0.8%
0.8%
1.4%
1.2%
1.0%
1.7%
Packed Tower Aeration
0.9%
0.8%
0.6%
0.6%
1.0%
1.1%
0.6%
0.8%
Low Pressure Membrane Filtration (Pressurized)
0.4%
0.3%
0.2%
0.1%
0.2%
0.2%
0.2%
0.3%
Reverse Osmosis
0.6%
0.6%
0.4%
0.4%
0.4%
0.3%
0.2%
0.4%
Biological treatment (Fixed Bed Pressurized)
0.6%
0.6%
0.5%
0.5%
0.8%
0.8%
0.6%
0.9%
Source: EPA's WBS cost models (Leidos, 2013).
A dash (-) indicates that the technology would not be used for the design flow.
1. Assumes pH adjustment and disposal (rather than regeneration) of spent media.
2. Assumes disposal (rather than on-site or off-site regeneration) of spent media. Pressurized contactors for the four smallest size categories and gravity-fed basins for the four largest size categories.
Data
Abt assumed that requirements for new treatment processes would lead to construction on undeveloped land; i.e., existing industrial structures are not repurposed for a treatment plant. We reviewed several sources of land price data to find values for undeveloped land; developed land prices would double count site improvement costs that are explicit line items in the WBS models. We determined that there are relatively few sources that provide values nationwide, and selected the sources that provided the best information for rural and urban land values throughout the United States. The following discussion pertains to these sources and the land values that we obtained.
Rural Land Value
For rural areas, the U.S. Department of Agriculture's (USDA) Census of Agriculture reports the value of land with buildings, but does not provide undeveloped land value. However, in another survey, USDA (2019b) reports average farm real estate (land and buildings), cropland, and pasture land values by State and region. The USDA derives these values from a survey conducted during the first two weeks of June of a random probability sample of approximately 10,500 segments of land averaging one square mile. Enumerators contact all farmers with operations within the sampled segments to obtain current land values. State statistical offices and USDA's Agricultural Statistics Board analyze the average values for consistency and reasonability. 
We selected the pasture land values as the best proxy for the value of undeveloped rural land available for drinking water treatment. Given the relatively small land acreage requirement, we assumed it is unlikely that productive cropland will be displaced or farm structures demolished. We escalated the 2019 values to 2020 values based on the average annual increase of values from 2015 to 2019 in each state. Exhibit 3 shows the value of pasture land by State. The report does not include land values in U.S. Territories: American Samoa, Guam, the Northern Mariana Islands, Puerto Rico, and the Virgin Islands. Therefore, the rural land cost estimates do not include values for the territories.
Exhibit 3. Pasture Land Values (2020 dollars)
State
Value ($/acre)
State
Value ($/acre)
Alabama (AL)
2,695
Montana (MT)
688
Alaska[1] (AK)
766
Nebraska (NE)
1,104
Arizona[2,4] (AZ)
1,240
Nevada[2,4] (NV)
833
Arkansas (AR)
2,703
New Hampshire[3] (NH)
6,667
California (CA)
3,082
New Jersey (NJ)
13,302
Colorado (CO)
859
New Mexico[2] (NM)
436
Connecticut[3] (CT)
6,667
New York (NY)
1,545
Delaware[3] (DE)
6,667
N. Carolina (NC)
4,823
Florida (FL)
5,462
N. Dakota (ND)
818
Georgia (GA)
3,768
Ohio (OH)
3,378
Hawaii[1] (HI)
9,927
Oklahoma (OK)
1,483
Idaho (ID)
1,639
Oregon (OR)
773
Illinois (IL)
3,088
Pennsylvania (PA)
3,378
Indiana (IA)
2,433
Rhode Island[3] (RI)
6,667
Iowa (IA)
2,600
S. Carolina (SC)
3,287
Kansas (KS)
1,388
S. Dakota (SD)
1,073
Kentucky (KY)
3,077
Tennessee (TN)
3,944
Louisiana (LA)
2,899
Texas (TX)
1,694
Maine[3] (ME)
6,667
Utah[2] (UT)
1,281
Maryland[3] (MD)
6,667
Vermont[3] (VT)
6,667
Massachusetts[3] (MA)
6,667
Virginia (VA)
3,998
Michigan (MI)
2,558
Washington (WA)
761
Minnesota (MN)
1,695
West Virginia (WV)
2,140
Mississippi (MS)
2,488
Wisconsin (WI)
2,344
Missouri (MO)
2,006
Wyoming (WY)
587
Source: USDA (2019b), and USDA (2019a) as noted below.
1. Land costs for Alaska and Hawaii are farm real estate values from the 2017 Census of Agriculture adjusted to 2020 dollars using land price trends in States with comparable values for pasture land. We adjusted the value for Alaska from $718/ac to $766 based on the average annual increase of 2.2% in Mountain states (MT, ND, SD, WY, CO, and NM). We adjusted the Hawaii value from $9,328/ac to $9,927 based on the average annual increase of 2.1% from 2015 to 2019 for New England states. Data for these two States represents the market value of land and buildings because the Census of Agriculture does not contain pasture land value. Therefore, the values likely overstate the value of pasture land in Alaska and Hawaii.
2. Land costs for AZ, NV, NM, and UT exclude Native American reservation land.
3. USDA reports land costs for CT, DE, ME, MD, MA, NH, RI, and VT as a sub-regional average.
4. Pasture land value for 2019 was not published due to insufficient reports. A pasture land value was calculated using the latest published value (2010 for Arizona and 2009 for Nevada) escalated with a regional pasture inflation factor to calculate a 2020 value.
Urban Land Value
In urban areas, larger water treatment facilities requiring additional land are likely to be located in areas zoned for industrial land use. We assumed that undeveloped industrial land would be preferred as a more cost-effective option over developed industrial land, because developed land may have existing improvements such as structures that might need to be demolished or renovated. Furthermore, the WBS models include cost line items usually associated with development activities such as geotechnical and electricity costs. Therefore, developed land values would double-count other WBS costs.
Data sources for undeveloped industrial land values are limited. Data sources such as the U.S. Department of Housing and Urban Development's American Housing Survey provide the value of developed land, but do not report data for undeveloped parcels. For the 2013 update, unimproved industrial land sales transaction data for many urban areas were available on the LoopNet website (LoopNet, 2013). 
LoopNet no longer provides sales values for historical transaction data. It does, however, report prices and acres for industrial land lots for sale (LoopNet, 2020). The listed sale prices may not reflect the final transaction value, which could be higher or lower. In the absence of data comparing industrial land sale and contract values, we assumed that  -  on average  -  the for-sale value is a reasonable estimate of the final sale value. 
We extracted and processed the available industrial data. We downloaded the location, price, and acreage data for industrial lots posted for sale from April 3[rd] to 17[th] for 3,796 industrial land lots. We used the location information to merge Census Place (Census Bureau, 2020) population data with each lot record for 1,694 places. To match the population lower bound for urban clusters, we then excluded places having populations of fewer than 2,500 people. We also excluded a few outlier records that, upon review, did not meet the undeveloped industrial lot criterion. The end result was price information for 1,499 places in 50 states.
The 2020 update extends the urban land area values to places with fewer than 50,000 people. We assumed that system service populations correspond with place populations. Thus, we estimated average prices for five place population size categories that correspond with typical system size categories: 2,500 to 3,300; 3,301 to 10,000; 10,001 to 50,000; 50,001 to 100,000; and more than 100,000. In contrast, the 2013 update included the two largest size categories. We also estimated average prices by state. The result is average prices for a total of 250 strata (i.e., 50 states x 5 size categories). For each stratum, we calculated acre-weighted averages across all of the records within the specified state and population size category, so records including more acres receive more weight. Exhibit 4 shows weighted average prices by state and population size category. For some strata, there were no industrial land lots listed for sale.
Exhibit 4. Average Industrial Land Values by State and Place Population Size Category (2020 dollars per acre)
State
2,500 to 3,300
3,301 to 10,000
10,001 to 50,000
50,001 to 100,000
>100,000
AL
No data  
 $13,224 
 $34,410 
No data  
 $30,617 
AK
No data  
 $197,935 
 $103,759 
No data
No data
AZ
 $14,706 
 $20,695 
 $52,234 
 $49,475 
 $159,028 
AR
No data  
 $62,000 
 $46,074 
 $36,976 
 $14,286 
CA
 $15,347 
 $30,497 
 $107,704 
 $282,974 
 $262,360 
CO
 $122,564 
 $154,116 
 $142,645 
 $136,255 
 $357,184 
CT
 $75,829 
 $78,011 
 $78,816 
 $206,581 
 $598,592 
DE
 $250,000 
 $29,256 
 $103,390 
 $250,044 
No data  
FL
 $40,867 
 $35,339 
 $110,321 
 $163,464 
 $164,024 
GA
 $11,980 
 $53,680 
 $36,588 
 $34,750 
 $15,869 
HI
No data
No data  
 $2,021,891 
No data  
$2,441,176 
ID
No data
 $55,074 
 $19,294 
 $180,730 
 $155,163 
IL
 $54,450 
 $24,181 
 $59,886 
 $133,535 
 $462,621 
IN
 $169,955 
 $53,404 
 $48,782 
 $63,726 
 $76,356 
IA
 $25,000 
 $37,671 
 $72,594 
 $72,738 
 $134,615 
KS
No data
 $108,900 
 $56,887 
 $125,409 
 $104,166 
KY
No data
 $18,425 
 $35,965 
 $95,179 
 $97,729 
LA
 $101,713 
 $70,982 
 $101,631 
No data  
 $103,291 
ME
No data  
 $72,500 
 $14,852 
No data
No data
MD
 $166,937 
 $117,548 
 $66,977 
 $98,591 
 $113,563 
MA
 $421,622 
 $101,671 
 $187,550 
 $108,711 
 $221,799 
MI
 $29,795 
 $29,754 
 $61,613 
 $286,347 
 $138,770 
MN
 $43,560 
 $31,742 
 $44,254 
 $97,171 
 $220,069 
MS
No data  
 $40,728 
 $40,818 
 $84,187 
 $44,145 
MO
 $33,825 
 $25,334 
 $93,867 
 $54,915 
 $132,121 
MT
No data  
 $149,382 
No data  
 $20,981 
 $144,608 
NE
No data  
 $180,799 
 $71,696 
 $57,087 
 $47,603 
NV
 $158,472 
 $95,026 
 $37,742 
 $262,887 
 $121,961 
NH
No data  
 $213,920 
 $65,772 
No data  
No data  
NJ
 $62,987 
 $139,014 
 $169,701 
 $252,864 
$2,217,195 
NM
No data  
 $31,169 
 $167,578 
No data  
 $138,096 
NY
 $35,145 
 $94,076 
 $318,115 
 $10,653 
$1,540,820 
NC
 $90,000 
 $35,171 
 $38,778 
 $34,441 
 $54,096 
ND
 $23,032 
 $129,597 
 $96,356 
 $72,796 
No data  
OH
 $51,163 
 $39,175 
 $43,129 
 $36,180 
 $76,624 
OK
No data  
 $17,449 
 $26,154 
 $139,569 
 $93,438 
OR
 $209,934 
 $23,363 
 $71,416 
 $201,531 
 $195,404 
PA
 $64,555 
 $74,874 
 $52,737 
 $39,104 
 $542,056 
RI
No data  
No data  
 $94,502 
No data
No data
SC
 $11,097 
 $29,394 
 $46,680 
 $31,976 
 $117,466 
SD
No data
No data
 $111,236 
 $101,751 
No data  
TN
No data  
 $14,615 
 $61,631 
 $42,761 
 $55,460 
TX
 $15,000 
 $27,565 
 $20,234 
 $122,125 
 $56,245 
UT
No data
 $156,824 
 $68,547 
 $349,738 
 $352,104 
VT
No data
No data  
 $85,000 
No data  
No data  
VA
 $24,055 
 $42,094 
 $51,755 
 $108,834 
 $136,537 
WA
 $207,961 
 $209,434 
 $216,328 
 $190,414 
 $182,681 
WV
No data  
 $35,000 
No data
No data
No data
WI
 $58,032 
 $57,865 
 $82,663 
 $252,671 
 $110,424 
WY
 $112,809 
 $142,715 
 $187,673 
 $72,727 
No data  
Source: Land values are from LoopNet (2020); Census place population estimates are from the 2018 1-year American Community Survey (Census Bureau, 2020).
No data = no industrial lots in LoopNet for this stratum
Averages shown are acre-weighted averages of all industrial land sales in each stratum.
Applying industrial land values to all small urban systems is a source of uncertainty. There are some small CWS that are not municipal systems serving Census-designated places. Instead, they serve a discrete subset of residential customers (e.g., serving a mobile home park and a single residential subdivision). To minimize distribution costs, these systems would install new treatment processes near existing treatment facilities or wells, which may be located in areas zoned for residential use, not industrial use. Based on survey data collected in October of 2019, the REALTORS(R) Land Institute and the National Association of REALTORS(R) (2019) reported a national median price of $25,000 per acre for residential land.  
Nevertheless, even if water treatment processes are added to land zoned for residential use, residential land prices may not represent the opportunity cost of doing so. For example, in a subdivision served by a ground water well, the well may be surrounded by open space not owned by any a single homeowner. Even if the land is zoned residential, it could not be sold as a residential lot if it is commonly owned by all subdivision residents served by the well. Therefore, we assumed that industrial land values better approximate the opportunity costs of space used to treat water at these types of systems than residential lot values. The industrial land values may over- or under-state the opportunity cost of land.
Number of Rural and Urban Systems
As described in section 4, we combine the state-level values with system counts to estimate weighted rural and urban land values. To do this, we need estimates of the number of rural and urban systems. To identify whether systems are in rural or urban locations, we conducted an analysis of system location using geographic information system (GIS) data. 
We used two sources of system spatial data for the GIS coordinates needed to categorize systems as either rural or urban. The first source was a dataset of facility latitude and longitude data that EPA provided in a 2009 extract from the Federal Safe Drinking Water Information System (SDWIS/FED), which Leidos used for the 2013 land value update (Leidos, 2013). The latitude and longitude data  -  one pair per system  -  are coordinates for treatment plants or other water system facilities such as intakes or wells for systems reporting coordinates in SDWIS/FED. Facility coordinates are available for approximately 84% of systems. The second source was the current inventory of CWS from EPA's SDWIS/FED (USEPA, 2020b), summarized in Exhibit 1. The 2020 version of downloaded SDWIS/FED data does not contain latitude or longitude fields. It does, however contain a ZIP code field. We mapped system ZIP codes to ZIP area coordinates published by the Census Bureau (2019b) for systems in the current inventory (USEPA, 2020b) that did not report GIS coordinates in the 2009 SDWIS/FED database.
Given the system latitude and longitude coordinates, we used a point-to-polygon intersection analysis to identify which coordinates are in urban areas. To define urban areas, we used two types recognized by the Census Bureau: urbanized areas and urban clusters. The Census Bureau classifies an urban area as a densely settled core of census tracts and/or census blocks that meet a minimum population density requirement (1,000 people per square mile), along with adjacent territory containing non-residential urban land uses as well as territory with low population density included to link outlying densely settled territory with the densely settled core (Census Bureau, 2019a). An urbanized area contains 50,000 or more people; an urban cluster contains at least 2,500 and less than 50,000 people (Census Bureau, 2019a). To account for reported uncertainty in polygon boundaries in Census GIS files, we included a 51-meter buffer around each system coordinate pair (see Appendix A). When a system location, including the buffer, intersected either an urban cluster or urbanized area polygon, we categorized the system as "urban." Otherwise, we classified the system as "rural." 
Exhibit 5 shows the estimate of the number of rural systems by state and system size category. The results indicate that some systems that serve large populations nevertheless have facilities located in areas considered rural for the purpose of this analysis.
Exhibit 5. Estimated Number of Rural Community Water Systems, by State and Population Served Size Category
State
25-100
101-500
501-1,000
1,001-3,301
3,301-10,000
10,001-50,000
50.001-100,000
>100K
Total
AK
87
140
36
18
10
4
0
0
 295 
AL
6
21
38
144
149
63
3
2
 426 
AR
41
139
132
183
69
21
3
1
 589 
AZ
163
303
100
137
70
32
6
3
 814 
CA
673
619
151
182
81
58
10
6
 1,780 
CO
192
250
78
83
53
27
7
4
 694 
CT
94
77
13
8
4
8
0
3
 207 
DC
1
0
0
0
0
0
0
0
 1 
DE
30
42
7
7
3
1
0
0
 90 
FL
116
196
65
102
47
21
7
9
 563 
GA
409
420
98
121
60
27
6
5
 1,146 
HI
12
25
9
16
22
8
2
2
 96 
IA
156
348
165
179
44
12
3
0
 907 
ID
283
210
55
45
21
9
0
0
 623 
IL
98
324
217
236
64
21
1
1
 962 
IN
98
84
98
150
74
31
4
1
 540 
KS
110
310
123
168
22
6
0
1
 740 
KY
13
15
21
91
104
55
3
1
 303 
LA
57
146
110
218
94
13
5
0
 643 
MA
64
43
17
27
36
28
4
3
 222 
MD
81
104
37
26
6
6
1
1
 262 
ME
169
76
32
47
18
10
0
1
 353 
MI
263
251
96
157
49
28
6
2
 852 
MN
169
262
126
126
26
7
1
1
 718 
MO
292
399
155
204
66
19
2
2
 1,139 
MS
37
204
163
375
98
6
0
0
 883 
MT
279
229
48
53
10
4
0
0
 623 
NC
373
360
99
114
81
72
7
10
 1,116 
ND
66
115
51
55
23
4
0
0
 314 
NE
96
253
91
80
12
2
0
1
 535 
NH
204
168
24
45
12
5
0
1
 459 
NJ
35
46
21
15
11
7
1
1
 137 
NM
110
215
56
63
25
12
1
0
 482 
NV
34
54
18
23
11
3
1
0
 144 
NY
531
479
166
215
66
23
0
5
 1,485 
OH
117
182
121
145
50
36
0
2
 653 
OK
115
303
147
226
78
26
3
0
 898 
OR
235
225
53
75
33
19
8
2
 650 
PA
417
364
106
179
76
38
5
7
 1,192 
RI
13
19
1
6
6
3
0
1
 49 
SC
78
62
47
67
48
34
1
4
 341 
SD
131
184
58
53
22
8
0
0
 456 
TN
16
35
30
90
101
62
6
3
 343 
TX
452
993
410
615
243
93
11
9
 2,826 
UT
92
137
67
60
37
20
3
1
 417 
VA
316
309
60
123
45
18
3
3
 877 
VT
128
152
36
44
22
5
0
0
 387 
WA
612
506
91
138
38
20
3
3
 1,411 
WI
192
192
118
137
17
14
3
2
 675 
WV
40
81
54
112
39
7
0
0
 333 
WY
73
114
20
31
11
7
1
0
 257 
Total
8,469
10,785
4,135
5,814
2,407
1,063
131
104
32,908 
Source: GIS analysis of USEPA (2009 and 2020b) system coordinates and Census Bureau (2020) urban areas
Exhibit 6 shows the resulting number of community water systems located in urban areas based on the Census definitions. The results show that there are many systems serving small populations located in urban areas.
Exhibit 6. Estimated Number of Urban Community Water Systems, by State and Population Served Size Category
                                     State
                                    25-100
                                    101-500
                                   501-1,000
                                  1,001-3,301
                                 3,301-10,000
                                 10,001-50,000
                                50.001-100,000
                                   >100K
                                     Total
AK
50
45
14
7
4
2
1
1
 124 
AL
3
2
3
7
30
35
6
3
 89 
AR
1
9
4
25
35
23
3
1
 101 
AZ
43
53
12
25
19
20
4
7
 183 
CA
261
320
65
101
156
196
85
82
 1,266 
CO
39
61
19
28
33
35
4
6
 225 
CT
94
117
18
27
16
19
8
1
 300 
DC
0
0
0
0
0
2
0
1
 3 
DE
19
43
14
17
17
13
0
3
 126 
FL
236
335
92
130
102
124
45
32
 1,096 
GA
177
193
58
36
67
60
6
13
 610 
HI
6
3
2
3
4
3
1
0
 22 
IA
35
41
14
23
41
25
5
3
 187 
ID
57
38
5
8
10
8
4
2
 132 
IL
61
154
84
146
160
173
21
9
 808 
IN
46
48
12
41
50
39
7
4
 247 
KS
18
18
9
31
35
19
2
4
 136 
KY
2
2
0
14
28
33
3
2
 84 
LA
43
110
28
50
75
46
6
6
 364 
MA
52
51
3
19
56
107
21
3
 312 
MD
49
60
18
28
33
14
5
4
 211 
ME
7
9
1
6
4
2
0
0
 29 
MI
94
127
45
83
110
79
16
5
 559 
MN
33
39
24
39
66
66
14
2
 283 
MO
58
64
22
53
65
43
4
5
 314 
MS
10
21
13
34
51
44
3
1
 177 
MT
91
62
7
8
15
1
2
1
 187 
NC
274
401
75
54
60
39
5
7
 915 
ND
5
5
0
2
5
3
2
1
 23 
NE
16
22
7
10
16
9
1
1
 82 
NH
124
90
10
9
10
11
1
0
 255 
NJ
54
82
28
61
70
125
11
15
 446 
NM
46
46
14
18
15
16
2
2
 159 
NV
30
23
5
4
11
4
0
4
 81 
NY
178
267
72
112
98
97
13
12
 849 
OH
72
123
38
82
111
97
14
10
 547 
OK
40
52
12
24
17
19
4
2
 170 
OR
110
81
4
22
24
24
2
2
 269 
PA
161
203
63
139
110
74
13
9
 772 
RI
9
13
2
3
5
7
4
0
 43 
SC
72
65
14
20
33
28
8
2
 242 
SD
13
17
3
6
11
5
1
1
 57 
TN
1
5
2
17
28
50
11
7
 121 
TX
262
423
168
352
449
189
22
29
 1,894 
UT
12
13
3
6
17
29
7
3
 90 
VA
37
66
17
39
33
30
5
13
 240 
VT
7
16
1
2
4
1
0
0
 31 
WA
369
304
44
81
59
73
21
7
 958 
WI
109
89
10
55
85
56
5
2
 411 
WV
5
12
14
48
28
14
2
1
 124 
WY
20
24
8
7
8
0
1
0
 68 
Total
3,611
4,467
1,200
2,162
2,589
2,231
431
331
17,022 
Source: GIS analysis of USEPA (2009 and 2020b) system coordinates and Census Bureau (2020) urban areas
Because the rural and urban values in preceding sections did not include data for U.S. Territories, the system distribution data in Exhibit 5 and Exhibit 6 also exclude systems in U.S Territories. As of 2020, there are 235 rural systems and 314 urban systems in the territories based on the USEPA (2020b) inventory data and the GIS analysis described above.

Method
This section provides a description of the method used to develop the national average land values for use in EPA's WBS cost models using the data reported in section 3.
Overview
As noted above, the WBS models include national average land values differentiated by system size (population served) categories. Therefore, we used the following formula to estimate the national average value for each size category:
Prural,jxExpValueAcrerural,j+ Purban,jxExpValueAcreurban,j
where: 	
      Prural,j = probability a system is rural for size category j
      ExpValueAcrerural,j = expected rural land cost per acre for size category j
      Purban,j = probability a system is urban for size category j
      ExpValueAcreurban,j = expected urban land cost per acre for size category j.
This approach is consistent with the national cost modeling approach. The national cost model randomly selects systems from within a size category and then randomly applies a contaminant concentration and, if necessary, a treatment technology. The contaminant concentration is based on a national occurrence distribution and the treatment technology is based on a probability distribution across applicable technology options in the compliance forecast. None of these compliance forecast probabilities depends on a rural or urban system classification and because of this independence, the probability weighted average land value is appropriate. The following sections describe how we derived each of the terms in the equation.  
Estimating the Probability of a Rural or Urban System
Exhibit 5 and Exhibit 6 in Section 3.3 provide the estimated number of rural and urban systems by size category, respectively. We used the total rows from these exhibits to estimate the probability that a system is either rural or urban, by size category, as shown in Exhibit 7. 
As expected, the probability that a system is urban is greatest for the largest systems, but is also somewhat higher for the smallest size categories than for systems serving 501 to 3,300 people. This reflects the presence of small systems that serve subpopulations in urban areas such as mobile home parks and subdivisions. Exhibit 7 also shows that there are rural systems serving populations greater than 2,500 people. This result indicates that there are systems serving urban populations that have their facilities located in rural areas. For example, a large surface water system may use water from a protected watershed located far from the urban service area. 
The probabilities for urban systems are slightly higher than those estimated for the Leidos (2013) analysis. This is expected because the 2013 analysis did not include systems in urban clusters in the estimate of urban systems.
Exhibit 7. Estimated Probability of Rural and Urban Location for Community Water Systems, by Population Served Size Category
Item
25-100
101-500
501-1,000
1,001-3,300
3,301-10,000
10,001-50,000
50.001-100,000
>100,000
Total
Number of Rural Systems
                                     8,469
                                    10,785
                                     4,135
                                     5,814
                                     2,407
                                     1,063
                                      131
                                      104
                                    32,908
Number of Urban Systems
                                     3,611
                                     4,467
                                     1,200
                                     2,162
                                     2,589
                                     2,231
                                      431
                                      331
                                    17,022
Probability of Rural System
                                      70%
                                      71%
                                      78%
                                      73%
                                      48%
                                      32%
                                      23%
                                      24%
                                      66%
Probability of Urban System
                                      30%
                                      29%
                                      22%
                                      27%
                                      52%
                                      68%
                                      77%
                                      76%
                                      34%
Source: Number of rural and urban systems shown in Exhibit 5 and Exhibit 6. Probability of being a rural (urban) system calculated by dividing the number of rural (urban) systems by the total number of systems in each size category.
Estimating the Expected Rural Land Cost
The expected rural land cost is a system-weighted average of the rural land values across states. Exhibit 3 provides rural land costs, as approximated by the cost of pasture land, by State, and Exhibit 5 provides the estimated number of rural systems in each State by size category. Thus, these exhibits provide the data needed to calculate the following for each size category:

ExpValueAcrerural,j=i,iRSi,jxRLVii,iRSi,j 
where:
      ExpValueAcrerural,j = the expected per acre value of rural land for size category j
      RSi,j = the estimated number of rural systems in State i for size category j
      RLVi = the estimated average value per acre of pasture land in State i.
Exhibit 8 shows the results by size category. We estimated the expected value of rural land by size category  -  instead of estimating a single national value  -  to incorporate additional information on how the distribution of systems across states differs across size categories. Systems serving fewer than 1,000 may tend to be concentrated in some states while larger systems are concentrated in others. 

The expected value calculation is able to give higher weight to the states with the most systems in each size category. For example, in the size category serving greater than 100,000 people, the calculation for expected rural land value weighted North Carolina, Florida, and Texas the highest. These states had the most rural systems serving over 100,000 people, with 10, 9 and 9, respectively.

Exhibit 8. Expected (Weighted Average) Value per Acre of Rural Land (2020 dollars)
Population Served
Value[1]
25-100
$2,701
101-500
$2,518
501-1,000
$2,493
1,001-3,300
$2,531
3,301-10,000
 $2,796 
10,001-50,000
 $3,036 
50,001-100,000
 $2,955 
>100,000
 $3,576 
1. Weighted by location of rural facilities (excluding U.S. Territory systems).
 Estimating the Expected Urban Land Cost
Using an approach similar to the one described in section 4.3, we applied the urban system estimates in Exhibit 6 as weights to the industrial land values in Exhibit 4. Because the acreage-weighted land values in Exhibit 4 vary by place size category, we calculated weighted-average values for five size categories:

ExpValueAcreurban,j=i,iUSi,jxULVi,ji,iUSi,j 
where: 
      ExpValueAcreurban, j = the expected value of undeveloped urban land in size category j
      USi,j = the estimated number of urban systems in State i in size category j
      ULVij = the estimated average value per acre of undeveloped urban land in State i and size category j.
Some states did not have any price data for a size category. The systems in that category were removed from the summation. In total, this excluded 358 systems from the analysis.
Exhibit 9 shows the results by size category. We estimated the expected value of urban land by size category  -  instead of estimating a single national value  -  to incorporate additional information on how the distribution of systems across states differs across size categories. The expected value calculation is able to give higher weight to the states with more systems in each size category. For example, in the size category serving 1,000 to 3,300 people, the expected urban land value calculation weighted Texas, Pennsylvania, and Illinois the highest. These states had the most urban systems, with 352, 146 and 139, respectively. 
For system size categories below 1,000, we adopted the same approach as Leidos (2013) and used the value for the 1,001 to 3,300 size category as a proxy for urban land values for smaller system sizes. As noted above, some of small systems may be located on land zoned for residential use located with the residential service area such as mobile home parks or subdivisions. The industrial land value of $61,296 may overstate land value in these instances. 
Exhibit 9. Expected (Weighted Average) Value per Acre of Urban Land (2020 dollars)
Population Served
Value[1]
25-100
$61,296 
101-500
$61,296
501-1,000
$61,296
1,001-3,300
$61,296
3,301-10,000
$54,942
10,001-50,000
$95,789
50,001-100,000
 $160,151 
>100,000
 $328,772
1. Weighted by location of urban facilities (excluding U.S. Territory systems).
 Estimating Overall Expected Land Cost
Combining the probabilities of a system being urban or rural (Exhibit 7) and the expected rural and urban land values produces the expected value per acre of land by system size category. As Exhibit 10, shows, the expected value is higher for larger systems, and is also higher for the smallest systems when compared to systems serving 501 to 3,300 people. These higher values likely reflect that many of the smallest systems serve subpopulations in urban areas such as mobile home parks and subdivisions.
Exhibit 10. Expected Value per Acre of Land for Community Water Systems (2020 dollars), by Population Served Size Category
                                     Item
25-100
101-500
501-1,000
1,001-3,300
3,301-10,000
10,001-50,000
50.001-100,000
>100,000
Probability of Rural System
                                      70%
                                      71%
                                      78%
                                      73%
                                      48%
                                      32%
                                      23%
                                      24%
Probability of Urban System
                                      30%
                                      29%
                                      22%
                                      27%
                                      52%
                                      68%
                                      77%
                                      76%
Expected Rural Land Value
                                    $2,701
                                    $2,518
                                    $2,493
                                    $2,531
                                    $2,796
                                    $3,036
                                    $2,955
                                    $3,576
Expected Urban Land Value
                                    $61,296
                                    $61,296
                                    $61,296
                                    $61,296
                                    $54,942
                                    $95,789
                                   $160,151
                                   $328,772
Expected Land Value
                                    $20,216
                                    $19,733
                                    $15,720
                                    $18,460
                                    $29,819
                                    $65,857
                                   $123,509
                                   $251,024
Source: Rural and urban weights are from Exhibit 7. Weighted average land values are from Exhibit 8 and Exhibit 9. 
Comparisons to 2013 Analysis
Leidos (2013) described the method used to estimate the value of land for inclusion in drinking water cost modeling. Abt's updated analysis, described in this document, used many of the same analysis methods and calculations. Before comparing the results of these analyses, this section will first detail the similarities and differences between them. These analyses utilize the same methodology of creating an expected value for an acre of land by first estimating value per acre of both rural and urban land and the probabilities each will occur for different system size categories. Both also used the same land value sources  -  USDA and LoopNet. Despite their many similarities, there are key differences that will drive the changes we see between the results. The 2020 analysis has expanded upon its 2013 counterpart to include more and newer data. This has allowed a more detailed look at systems serving between 2,500 and 50,000 people. 
Urban and Rural Probabilities
The probabilities for each size category that land is either urban or rural for both the 2013 and 2020 analyses are created by looking up the coordinates of water system facilities (e.g., water treatment facility, well-head, or intake). This approach allowed a GIS analysis to determine whether those coordinates fell within Census Bureau Urbanized Area polygons. Both analyses accounted for error in the coordinate information in different manners, which is outlined in Appendix A: Urban/Rural Classification Procedures. They both use latitude and longitude data from a 2009 SDWIS/FED extract as a source. Additionally, the 2020 analysis included an additional spatial data source from the current inventory of CWS EPA's SDWIS/FED (USEPA, 2020b). The inventory does not have latitude and longitude information, but does include a ZIP code, which Abt used to look up a pair of GIS coordinates for every system that did not have coordinate data in the 2009 SDWIS/FED data. 
The probabilities are based on an expanded definition of urban land in 2020. Instead of just looking for intersections with urbanized area polygons, it additionally looks for intersections with urban cluster polygons. Urbanized areas cover places with populations over 50,000 people, while urban clusters cover places with populations over 2,500 people in proximity to Urbanized Areas. These clusters connect densely populated suburban areas to densely populated city centers. 
Exhibit 11 shows the probabilities for both rural and urban land in 2013 and in 2020. The probability of being an urban system increased from 2013 across all categories. Because urban land values tend to be higher than rural land values, this change will tend to increase overall land values in 2020 compared to 2013.
Exhibit 11. Urban and Rural Probabilities, 2013 and 2020, by Population Served Size Category
Year
System Location
25-100
101-500
501-1,000
1,001-3,300
3,301-10,000
10,001-50,000
50.001-100,000
>100,000
2013
                                     Rural
74%
74%
81%
77%
55%
41%
34%
31%

                                     Urban
26%
26%
19%
23%
45%
59%
66%
69%
2020
                                     Rural
70%
71%
78%
73%
48%
32%
23%
24%

                                     Urban
30%
29%
22%
27%
52%
68%
77%
76%

Expected Rural Land Value
The expected rural land values for both 2013 and 2020 analyses use the same methodology and utilize data from the USDA Census of Agriculture. Average pasture land values for each state are used as a proxy for undeveloped industrial land values. The GIS analysis used to create the urban and rural probabilities also provided the number of rural systems in each state and size category. Using this information, an expected value is created for each size category.   
The only difference between the 2013 and 2020 analyses was the underlying data. The original analysis used the 2013 USDA Census of Agriculture while the 2020 analysis used the 2019 USDA Census of Agriculture. Abt extrapolated the 2019 data to 2020 using the trend from 2015 to 2019 for each state. After converting the 2013 values from 2012 dollars to 2020 dollars, we see in Exhibit 12 that the rural land values have stayed relatively constant. 
Exhibit 12. Comparison of Expected Rural Land Values, by Population Served Size Category (2020 dollars)
                                     Year
25-100
101-500
501-1,000
1,001-3,300
3,301-10,000
10,001-50,000
50.001-100,000
>100,000
2013[1,2]
$2,818 
$2,636 
$2,573 
$2,607 
$2,844 
$3,338 
$3,079 
$3,643 
2020[2]
$2,701 
$2,518 
$2,493 
$2,531 
$2,796 
$3,036 
$2,955 
$3,576 
Percent Change
-4.2%
-4.5%
-3.1%
-2.9%
-1.7%
-9.0%
-4.0%
-1.8%
1. Escalated from 2012 dollars to 2020 dollars using a GDP deflator (Federal Reserve, 2020); the April 2012 value is 99.713 and the April 2020 value is 112.817.
2. Weighted by location of systems in the respective analysis years.
Expected Urban Land Value
Similar to expected rural land value, the expected urban land value reflects the same methodology in both 2013 and 2020. The GIS analysis used to create the urban and rural probabilities also provided the number of urban systems in each state and size category. Abt gathered data from LoopNet for industrial land values to create the average urban land value in each state and size category. Together, the number of systems and average urban land value in each state and size category is used to create an expected urban land value for each size category. 
The differences are in the LoopNet data. In 2013, LoopNet provided average industrial lot values for cities with populations over 50,000 people. These values represented final sale prices. In 2020, these data is no longer available from LoopNet. Instead, data is collected from LoopNet's raw listings for industrial lots in each state. These values represent the listing sale price, which may be higher or lower than the final sale price. We matched the listing addresses to 2018 population information from the Census Bureau. We classified places with population over 2,500, consistent with the minimum size of urban clusters, as urban and included industrial land values in the expected urban value calculation. Due to this difference, the 2020 analysis is able to better analyze land values for systems serving between 2,500 and 50,000 people. 
In Exhibit 13, the results show a lower value of urban land for systems serving fewer than 10,000 people; while, land values are higher for systems serving over 10,000 people. For smaller systems, we expect the values to decrease in 2020 because the 2020 data collection allowed for analysis of places with fewer than 50,000 people. Following the same methodology as in 2013, we use the urban land value for the smallest population category we are able to calculate for all smaller size categories  -  2,500 to 3,300 people. In 2013, however, the small system value was based on places with population between 50,000 and 100,000 people.
The increase seen in urban land values for systems over 10,000 people, and particularly the doubling of urban land values for systems over 50,000 people can potentially be explained by the underlying data. Since the 2020 data are prices of current listings rather than final sale prices, there is the potential for both positive (i.e., properties eventually sell for less than the listing price) and negative bias (i.e., the properties sell for more than the listing price). 
Exhibit 13. Comparison of Expected Urban Land Values, by Population Served Size Category (2020 dollars)
Year
25-100
101-500
501-1,000
1,001-3,300
3,301-10,000
10,001-50,000
50.001-100,000
>100,000
2013[1][,2]
$81,691 
$81,691 
$81,691 
$81,691 
$81,691 
$81,691 
$81,691 
$168,339 
2020[2]
$61,296 
$61,296 
$61,296 
$61,296 
$54,942 
$95,789 
$160,151 
$328,772 
Percent change
-25%
-25%
-25%
-25%
-33%
17%
96%
95%
1. Escalated from 2012 dollars to 2020 dollars using a GDP deflator (Federal Reserve, 2020); the April 2012 value is 99.713 and the April 2020 value is 112.817.
2. Weighted by location of systems in respective study years.

--------------------------------------------------------------------------------
Changes in Expected Value
Lastly, the increasing urban percentages work to amplify the differences we see in urban land values between the two studies (Exhibit 14). Expected land values for systems serving more than 10,000 people are higher, while values for systems serving over 50,000 people more than double. Systems under 10,000 see decreases in expected land value, which is consistent with both rural and urban land values decreasing. 
Exhibit 14. Comparison of Expected Land Values, by Population Served Size Category (2020 dollars)
Year
25-100
101-500
501-1,000
1,001-3,300
3,301-10,000
10,001-50,000
50.001-100,000
>100,000
2013[1]
$23,390 
$23,365 
$17,512 
$20,896 
$38,352 
$49,312 
$55,131 
$118,091 
2020[1]
$20,216 
$19,733 
$15,720 
$18,460 
$29,819 
$65,857 
$123,509 
$251,024 
Percent Change
-14%
-16%
-10%
-12%
-22%
34%
124%
113%
1. Weighted by location of systems in respective study years.
Limitations and Uncertainties
Exhibit 15 describes the sources of uncertainty in the expected land cost estimates. As indicated, the overall direction of bias is unknown (i.e., neither clearly over or under estimating potential land costs). 
Exhibit 15. Sources of Uncertainty in the Expected Land Value Estimates
Source
Potential Impact on Values
Comments
Latitude and longitude information in SDWIS may not accurately identify location
+/-
Estimated average urban and rural land costs are weighted by the number of urban and rural systems in each State. Systems are classified as urban or rural based on latitude and longitude of treatment facilities and other facilities or on ZIP code coordinates when coordinates are missing from SDWIS. Both types of coordinates may contain error, resulting in an over-count or under-count of urban systems.
LoopNet data is nonrandom sample
+/-
A random sample is needed to generate unbiased estimates. In particular, the LoopNet values may be higher than national averages because the observations may be from urban regions with more profitable or more developed land markets. Additionally, the data are for-sale undeveloped industrial lots and the actual sale price may be higher or lower than the listing price.
LoopNet data are lots on the market instead of final sales values
+/-
LoopNet no longer makes national sales data available. Lots on the market may sell for more or less than the list values. 
Urban land cost data in LoopNet are limited
+/-
For some States, LoopNet data were not available for one or more size categories. Systems in these states were excluded from the weighted average urban value across states. It is unknown whether this approach over- or under-estimates weighted average costs.
Rural and urban land values are system-weighted averages across states
+/-
There may be geographic or other site-specific variations in rural and urban land values that are not captured by the use of system weighted averages. There is no clear direction of bias. 
+/- = impact of assumption on cost estimates is unknown.

References
Bigelow, Daniel P., and Allison Borchers. 2017. Major Uses of Land in the United States, 2012. EIB-178, U.S. Department of Agriculture, Economic Research Service. Retrieved from https://www.ers.usda.gov/webdocs/publications/84880/eib-178.pdf?v=0.draft. 
Census Bureau. 2003. TIGER Frequently Asked Questions. Washington, D.C.: Economics and Statistics Administration, Bureau of the Census. 
Census Bureau. 2019a. 2010 Census Urban and Rural Classification and Urban Area Criteria. Retrieved from https://www.census.gov/programs-surveys/geography/guidance/geo-areas/urban-rural/2010-urban-rural.html.
Census Bureau. 2019b. Gazatteer Files: ZIP Code Tabulation Areas. Retrieved from https://www.census.gov/geographies/reference-files/time-series/geo/gazetteer-files.html.
Census Bureau. 2020. 2018 American Community Survey Table B01003: Total Population for All Places. Retrieved from https://data.census.gov/cedsci/table?q=all%20places%20population&g=0100000US.160000&tid=ACSDT1Y2018.B01003&vintage=2018&hidePreview=true&moe=false&tp=true.
Dana, P. H. 1999. "Geodetic Datum Overview." Boulder, CO: The Geographer's Craft Project, Department of Geography, University of Colorado at Boulder. Online at http://www.colorado.edu/geography/gcraft/notes/datum/datum.html.
Environmental Systems Research Institute (ESRI). 2000. Understanding Map Projections. Redlands, CA: Environmental Systems Research Institute.
Federal Reserve. 2020. Gross Domestic Product: Implicit Price Deflator, Seasonally Adjusted. Retrieved from https://fred.stlouisfed.org.
Leidos. 2013. Estimating the Value of Land for Use in Drinking Water Cost Models. Reston, VA: Leidos.
LoopNet. 2013. LoopNet - #1 in Commercial Real Estate Online. Online at http://www.loopnet.com/.
LoopNet. 2020. LoopNet Industrial Land for Sale. Retrieved from https://www.loopnet.com/search/industrial-land/for-sale/?sk=ca68b38bd4d71ef17362b43bac2b64dd.
National Drinking Water Advisory Council (NDWAC). 2001. Report of the Arsenic Cost Working Group to the National Drinking Water Advisory Council. Final.
REALTORS(R) Land Institute and the National Association of REALTORS(R). 2020. The 2019 Land Market Survey. Retrieved from https://www.rliland.com/about-realtors-land-institute/land-markets-survey/. 
U.S. Department of Agriculture (USDA). 2019a. Census of Agriculture 2017. Washington, D.C.: National Agricultural Statistics Service. Retrieved from https://www.nass.usda.gov/Publications/AgCensus/2017/Full_Report/Volume_1,_Chapter_1_US/usv1.pdf.
U.S. Department of Agriculture (USDA). 2019b. Land Values  -  2019 Summary. Washington, D.C.: National Agricultural Statistics Service. Retrieved from https://www.nass.usda.gov/Publications/Todays_Reports/reports/land0819.pdf. 
U.S. Environmental Protection Agency (USEPA). 1996. Proceedings of the Blue Ribbon Panel on Safe Drinking Water Act Costing. Panel held July, 1996 in Cincinnati, Ohio.
USEPA. 2009. SDWIS/FED Database Extract, 8/3/2009. Provided by USEPA, Office of Ground Water and Drinking Water. 
USEPA. 2010. Guidelines for Preparing Economic Analyses. Updated in 2014 and 2016. Washington, D.C.: U.S. EPA, Office of the Administrator. EPA 240-R-00-003. 
USEPA. 2020a. Information about Public Water Systems. Retrieved from https://www.epa.gov/dwreginfo/information-about-public-water-systems.
USEPA. 2020b. SDWIS Federal Reports Advanced Search. Retrieved from https://ofmpub.epa.gov/apex/sfdw/f?p=108:1:::NO:1::.

Appendix A: Urban/Rural Classification Procedures
This appendix describes the differences in the 2013 and 2020 GIS analysis approaches. There are two differences in source data and one method difference. The sections below describe these differences.
Data Differences
The first data difference is the system GIS coordinates. The 2013 analysis limited the coordinates to those available in the 2009 SDWIS/FED data. In the 2020 analysis, Abt supplemented these data with ZIP code coordinates for systems that could not be matched to the 2009 coordinates. Some of these systems are new systems; others did not report spatial data in the 2009 SDWIS/FED database. In 2020, 84% of the CWS coordinates are from the 2009 SDWIS/FED data; 16% are from ZIP code data. 
The second data difference is the urban area polygons. The 2013 analysis defined urban areas as the Census urbanized areas. Leidos (2013) described these as areas of 50,000 or more people located in densely settled territory, which consists of:
a cluster of one or more Census blocks or block groups, each of which has a density of at least 1,000 people per square mile, plus
surrounding Census blocks or block groups, each of which has a density of at least 500 people per square mile, plus
less densely settled blocks that form enclaves, indentations, or are used to connect discontiguous areas with qualifying densities.
As a result of the 2013 approach, LoopNet data collection was limited to urbanized areas and urban land values for systems serving fewer than 50,000 people were based on sales of industrial land in Urbanized Areas with populations of 50,000 or more. This extrapolation potentially overstated land values for smaller systems.
For the 2020 analysis, Abt used the Census' extended definition of urban area. The three bullets above define urban areas. Urbanized areas are urban areas with 50,000 or more people. Urban clusters are urban areas with 2,500 to 49,999 people, of which at least 1,500 are not in institutionalized group quarters (Census Bureau, 2019a).   
The effect of the extended urban area definition should be an increase in the number of systems categorized as urban systems. The effect of using ZIP code coordinates as proxies for system location is uncertain, although including the additional 15% of systems in the analysis should also increase the number of urban systems. 
Method Difference
The 2013 and 2020 methods differed in their approach to incorporating uncertainty in the special analysis. The spatial data for systems and the Census polygons are uncertain. Sources of uncertainty include:
Measurement error  -  imprecision in the coordinates reported in SDWIS/FED
Conversion error  -  caused by converting coordinates from one measurement system to another
Census boundary error  -  imprecision in the Census UA boundary files
ZIP code error  -  unknown error of using ZIP code coordinates in lieu of system coordinates (2020 analysis only). 
To address overall uncertainty of the three applicable errors in 2013, Leidos estimated `total potential positional error' for each system and used that estimate as the radius of a circle around the system coordinate. Then, Leidos used the circle in a circle-to-polygon intersection to categorize systems as urban. Exhibit A-1 provides an illustration of the method. The image has a square representing an urbanized area (UA) boundary. Coordinate pairs for System A and System B are the centers of respective circles. The circle radii vary to reflect the aggregate uncertainty of the coordinates. For example, System A has a total potential positional error of 500 meters. The circle around System A defines the potential locations for System A. The circle does not intersect the UA polygon, so System A is "rural." Treatment Plant B has a total potential positional error of 2,000 meters. The circle defining the potential locations for System B does intersect the UA polygon, so System B is "urban."
Exhibit A - 1. Schematic of 2013 GIS Analysis for Classifying Facilities as Urban or Rural
CENSUS UA
500 m
A
2,000 m
B






Source: Leidos (2013)
With regard to measurement error, Leidos (2013) notes that each facility with latitude and longitude information in the SDWIS/FED database could also have a value of the potential measurement error or imprecision associated with the reported coordinates. Coordinate values in the SDWIS/FED database may be inaccurate for several reasons. For instance, the person recording the coordinates may have located the facility on a map, where complete precision down to the number of minutes and seconds was impossible. In that case, the error associated with the coordinate measurement would correspond to the map scale and the resolution of the latitude and longitude markings available on the map. When available, these errors were included in the total positional error estimate for radius length.
The conversion error arises from the fact that the GIS coordinates reported in the 2009 SDWIS/FED data reflected a variety of spatial measurement systems. Leidos (2013) noted that the records of latitude and longitude coordinates for a facility in SDWIS/FED could include information on the coordinate datum. Datums define the size and shape of the earth and the origin and orientation of the coordinate systems used to map the earth (Dana, 1999). Leidos used standard routines to convert the reported coordinate datums to the same datum as is used by the Census for the urbanized area boundaries. In cases where the coordinate datum was unclear or unreported in the SDWIS/FED database, Leidos assumed the datum was the same as that used for the Census urbanized area boundaries. To address the uncertainty of this assumption, however, Leidos included either 200 meters (for facilities located in the continental U.S.) or 1,000 meters (for facilities located in Alaska and Hawaii) to the radius estimate of a system to account for potential conversion error.
Finally, Leidos (2013) included the potential positional error associated with Census UA boundaries. The map accuracy standards for the sources used to create the Census maps and boundaries allow a maximum positional error of 167 feet (51 meters) (Census Bureau, 2003). Leidos added 51 meters to the radii of all the circles around the drinking water system facilities. This approach is equivalent to enlarging the boundaries of UAs by 51 meters in all directions.
Thus, the radius of the circle around each system GIS coordinate pair is the sum of the 51 meters plus the conversion error, if any, plus the measurement error, if reported. For computational ease, Leidos then classified all facilities into three categories based on the estimated magnitude of error: 
less than or equal to 500 meters (0.3 miles)
501 to 2,000 meters (0.3 to 1.2 miles)
2,001 to 11,051 meters (1.2 to 6.9 miles).
Leidos used the maximum value in each category as the estimate of error (i.e., 500, 2,000, and 11,051 meters) for all facilities in the category. 
Leidos provided examples to illustrate the overall method. Exhibit A-2 shows those examples, which include the two systems shown in Exhibit A-1. Leidos acknowledged that the actual positional error was likely to be far less, for every facility, because the approach employs conservative measures (e.g., using the maximum potential error from each source), and because positional shifts may be not be in the same direction (and may even be in opposite directions and, thus, cancel each other out). The examples illustrate the potential for substantial over-estimation.
Exhibit A - 2. Illustration of 2013 Method for Calculating Potential Positional Errors (in meters)
System
Potential Position Error from Datum Conversion[1]
Potential Position Error from SDWIS Coordinate Measurement
Potential Position Error from Census Boundary Imprecision
Total Potential Position Error
Estimate Used for Category[2]
System A
0
30
51
81
500
System B
200
500
51
751
2,000
System C
200
200
51
451
500
System D
1,000
1,000
51
2,051
11,051
Source: Leidos (2013)
1. Potential positional error from datum conversion equals 200 meters for points in unknown datum located in continental U.S.; equals 1,000 meters for points in unknown datum in Alaska or Hawaii.
2. Potential position error equals 500 meters for points with total potential position error of 0 to 500 meters; 2,000 meters for points with total potential error of 501 to 2,000 meters; and 11,051 meters for points with total potential error of 2,001 to 11,051 meters.
Abt's analysis in 2020 did not use a circle-to-polygon intersection. Some of the data needed to estimate the circle radii was not available. The summary data from the 2009 SDWIS/FED GIS data analysis was a latitude-longitude coordinate pair associated with each system identification number. The summary data did not include estimates of the datum conversion error or coordinate measurement error. Furthermore, the error introduced by using ZIP code coordinates is also unknown in terms of magnitude and direction. 
Instead Abt used a point-to-polygon intersection with a 51-meter tolerance to account for uncertainty in the Census polygon boundaries. While the Leidos method of analysis will tend to overstate the number of urban systems, it is unclear whether the Abt method results in a clear direction of bias in using the 2009 spatial data. This is because it is unclear whether the measurement or conversion uncertainty is more likely to shift systems into or out of urban areas, on average. 
