Estimating watershed degradation over the last century and its impact on water-treatment costs for the world’s large cities

Significance

Urban water-treatment costs depend on the water quality at the city’s source, which in turn depends on the land use in the source watersheds. Here, we show that globally urban source watershed degradation is widespread, with 9 in 10 cities losing significant amounts of natural land cover in their source watersheds to agriculture and development. This watershed degradation has impacted the cost of water treatment for about one in three large cities globally, increasing those costs by about half. This increase in cost matters because increases in water-treatment costs are paid for by those living in cities, so watershed degradation has had a real quantitative cost to hundreds of millions of urbanites.

Abstract

Urban water systems are impacted by land use within their source watersheds, as it affects raw water quality and thus the costs of water treatment. However, global estimates of the effect of land cover change on urban water-treatment costs have been hampered by a lack of global information on urban source watersheds. Here, we use a unique map of the urban source watersheds for 309 large cities (population > 750,000), combined with long-term data on anthropogenic land-use change in their source watersheds and data on water-treatment costs. We show that anthropogenic activity is highly correlated with sediment and nutrient pollution levels, which is in turn highly correlated with treatment costs. Over our study period (1900–2005), median population density has increased by a factor of 5.4 in urban source watersheds, whereas ranching and cropland use have increased by a factor of 3.4 and 2.0, respectively. Nearly all (90%) of urban source watersheds have had some level of watershed degradation, with the average pollutant yield of urban source watersheds increasing by 40% for sediment, 47% for phosphorus, and 119% for nitrogen. We estimate the degradation of watersheds over our study period has impacted treatment costs for 29% of cities globally, with operation and maintenance costs for impacted cities increasing on average by 53 ± 5% and replacement capital costs increasing by 44 ± 14%. We discuss why this widespread degradation might be occurring, and strategies cities have used to slow natural land cover loss.

Humanity is experiencing the fastest rate of urbanization in history. Over the 20th century, the urban population increased from 220 million to 2.9 billion, and by 2050, another 3.4-billion increase is expected (1, 2). One of the most fundamental requirements of urban existence is a source of clean, sufficient water (3, 4). Urban water supply systems are often complex, drawing water from multiple locations, some surface and some groundwater, some close and some far from the city center (5, 6). Seventy-eight percent of large cities rely on surface water sources (6), and their urban supply systems create teleconnections (7, 8) between source watersheds and the urban users who depend on them. The world’s largest cities (>750,000 population), the focus of this paper, occupy less than 1% of the Earth’s land surface (9) but their source watersheds occupy 41% of its surface (6).

Natural land cover in urban source watersheds provides important ecosystem services that help maintain the water quality at the city’s water source, so-called raw water that will then be treated and distributed to urban residents (10, 11). Natural land cover stabilizes soil, minimizing erosion and sediment loading (12, 13). Natural land cover also has lower loading than most human land-uses of excess nutrients, such as nitrogen (N) and phosphorus (P), and other pollutants (14, 15). When humans convert natural land cover to other uses such as agriculture or housing, the loss of natural land cover decreases ecosystem service provision and the anthropogenic land uses increases pollution, which leads to a decline in water quality (16). This loss of natural land cover over time, and the resulting impacts on hydrology, is often called watershed degradation (17).

This paper focuses on the water quality at the intakes of large cities globally. This raw water quality matters because it determines the type and intensity of water treatment needed to reach drinking water standards (18–20). For instance, Alcott et al. (10) contrast the relatively minimal treatment of Boston’s water supply from the largely forested Quabbin reservoir with the more extensive treatment required from Worcester’s reservoir, which has had significant land development and water quality degradation in its source watershed. Better raw water quality reduces the need for sediment removal (e.g., addition of coagulants such as alum), makes filtration easier (or in cases of exceptional water quality, removes the need for water filtration), and reduces the need for additional processes (e.g., the need to remove disinfection byproducts). In addition, lower concentration of phosphorus and nitrogen reduce algal growth and the amount of organic matter in the water, simplifying filtration and reducing the prevalence of disinfection byproducts. Thus, water-treatment plants (WTPs) using raw water of high quality can be designed using simpler treatment technologies, which lead to a lower capital cost during construction and lower operations and maintenance (O&M) costs. Avoiding watershed degradation helps maintain raw water quality which reduces treatment costs, as the “natural capital” of natural land cover functions as an alternative to human capital invested in a WTP (13).

Although the relationship between land cover and water quality has been measured or modeled in numerous watersheds (21, 22), the scientific understanding of the global importance of watershed degradation to urban water-treatment costs has been limited to date. Until recently (6), there has been no global dataset of where cities get their water from or of the complex teleconnections between source watersheds and cities. Similarly, although there are numerous anecdotal instances of clean water leading to reduced treatment costs, there has been relatively little global study of how financially important this proves to be for the world’s water utilities. For instance, one notable study looked at 27 US water utilities and found a relationship between forest cover and O&M costs, but did not consider capital costs or changes over time in forest cover and was limited to US utilities (18).

Here, we combine our recently created global dataset of where cities get their water from (20) with datasets on the growth of human population and land use over time (23, 24), allowing us to reconstruct watershed degradation for the period 1900–2005 for the world’s source watersheds. Using empirically based data for a training dataset on sediment, N, and P loading for the United States (25), we relate water quality to land use and population in the source watershed, allowing us to estimate how sediment and nutrient pollution have changed. Note that, although our statistical approach for the estimation of pollutant loading is relatively simple, it is necessitated by the long-term period of our analysis. More complex, spatially explicit models of sediment or nutrient loading and in-stream dynamics would require global maps of prior land cover and land cover transitions over the last century at high spatial resolution (<100s meters), which is not available for most watersheds globally. Finally, we assembled a unique dataset of the water-treatment technologies used in 264 WTPs and use it to estimate the statistical relationship between water quality degradation and increased treatment costs. Our two main research questions are as follows: (i) how has the degradation of source watersheds affected water quality for the world’s largest cities; and (ii) how much has the decline in water quality affected water-treatment costs for the world’s urban water utilities?

Results

Globally, most source watersheds have increased in population density over the period 1900–2005, but there is substantial spatial variation (Fig. 1A). Some places like the Ganges/Brahmaputra basin (home to cities like Dhaka) have increased dramatically in population density, whereas others have increased only slightly in population density, such as the source watersheds of Paris. Population density in urban source watersheds in 2005 varies by four orders of magnitude. Some of the highest values are in Asia. For example, the Dong watershed that supplies Shenzhen and Hong Kong, via an interbasin transfer, had a population density of 261 people/km² in 2005 (Fig. 1B). Readers who wish to see a higher-resolution image of Fig. 1 A and B should consult Figs. S1 and S2, respectively.

Fig. 1.

Spatial variation in human activity in drinking water source watersheds of large cities. (A) The increase in a city’s source watershed population density from 1900 to 2005. (B) Shenzen and its supply watersheds, including the large Dong River watershed. Note that the Dong River also supplies water to Hong Kong. (C) The increase in a city’s source watershed agricultural use, both cropland and ranchland, from 1900 to 2005. (D) New York City and its supply watersheds.

Fig. S1.

A high-resolution version of Fig. 1A.

Fig. S2.

A high-resolution version of Fig. 1B.

Agricultural expansion (cropland and ranchland) shows similarly large spatial variability (Fig. 1C). Areas like the pampas of Argentina had large increases in agricultural utilization, whereas New England had a decrease as farms were abandoned. These spatial patterns matter because they translate to spatial patterns in water quality, as we show below.

An examination of intake locations shows that cities have placed their water intakes to, in part, avoid severe water quality problems. The majority of intakes are upstream from the cities they serve. This hydrologic head of course facilitates water movement by gravity, but also potentially allows cities to source from less crowded watersheds. For instance, New York City is located along the Hudson River (Fig. 1D), a very urbanized river in its lower reaches, with population density exceeding 10,000 people/km². If New York City drew from the Hudson River, the average population density of the entire source watershed, including more rural parts of the watershed farther north, would be 123.7 people/km² and 14% agricultural utilization. By drawing water from two major reservoirs located an average of 151 km away, New York City can source from watersheds with an average population density of 65.3 people/km² and 16% agricultural utilization.

Globally, urban source watershed population density has increased significantly from a median of 22.3 to 124 people/km² over the study period (1900–2005), increasing by a factor of 5.4, with the fastest growth occurring in the last few decades of the 20th century (Fig. 2A). Population growth in urban source watersheds follows the exponential growth pattern commonly seen for human population globally during the 20th century, although the rate of increase in urban source watersheds is actually faster than for overall global population, which increased by a factor of 3.8 over the 20th century (1, 26).

Fig. 2.

Trends over time in watershed degradation and water quality in urban source watersheds. (A) Time series of median cropland and rangeland coverage, as well as population density. (B) Cropland and rangeland coverage, plus population density, predict sediment loading. Red line is the 1:1 line. Trends for N and P (not shown) are similar. (C) Estimated pollutant yield over time, relative to the average pollutant yield in 1900. (D) Proportion of WTPs by level of watershed degradation.

Trends for agricultural utilization in source watersheds are more complex temporally (Fig. 2A). Cropland area increased dramatically until 1960 then plateaued and modestly declined from 1990 on. The slight decline in global average cropland utilization in source watershed is due primarily to declines in some source watersheds in parts of the United States and Europe (Fig. 1C). Use as rangeland, in contrast, continuously expanded between 1900 and 2005. Trends over time in different regions are shown in Table S1.

Table S1.

Trends in median agricultural utilization and population density over time in region

Region	Cropland (%)		Grazing (%)		Population density (people/km2)
	1900	2005	1900	2005	1900	2005
Africa	5.3	17.4	10.9	33.1	16.2	102.8
Asia	13.0	26.5	1.7	5.1	65.0	299.1
Europe	18.8	29.9	14.5	18.7	50.6	108.2
Latin America and the Caribbean	2.2	7.8	4.8	28.3	14.8	142.5
Northern America	9.6	7.7	7.5	12.8	6.5	30.2
Australia/New Zealand	2.0	3.3	2.4	8.3	2.4	12.6

Region definitions follow those used by the UNPD.

Population density and agricultural utilization in a source watershed can be used to predict sediment (R² = 0.50, P < 0.001), N (R² = 0.58, P < 0.001), and P (R² = 0.63, P < 0.001) loading. A 10% increase in population density leads on average (holding all other variables constant) to an 0.8% increase in sediment loading, a 2.6% increase in N loading, and a 1.6% increase in P loading (Table S2). Similarly, a 10% increase in source watershed utilization for cropland leads to a 1.6% increase in sediment loading, a 1.3% increase in N loading, and a 0.1% increase in P loading. Utilization for rangeland has no statistically significant relationship to sediment, N, or P loading. Note that this is likely due to the relatively coarse global data used in our study, as there are many studies that have shown important impacts of rangeland on water quality in particular watersheds or contexts (see literature review in ref. 27).

Table S2.

Regression coefficients describing the relationship between predictor variables and sediment, N, or P loads in the SPARROW dataset for 2005

Variable	Estimate	SE	T value	P
Sediment load
Crop area	0.17	0.003	52.3	<0.001
Grazing area	0.076	0.003	21.8	<0.001
Population	0.087	0.002	47.5	<0.001
Watershed area	0.45	0.011	39.5	<0.001
RKLS factor	0.144	0.010	14.5	<0.001
RKLS factor × watershed area	0.014	0.022	6.3	<0.001
Intercept	11.64	0.049	239.7	<0.001
N load
Population	0.27	0.002	160.5	<0.001
Watershed area	0.35	0.005	76.8	<0.001
Crop area × N application rate	0.014	0.0003	53.4	<0.001
Intercept	6.76	0.02	367.6	<0.001
P load
Population	0.17	0.001	116.8	<0.001
Watershed area	0.62	0.004	157.7	<0.001
Crop area × P application rate	0.014	0.0002	53.2	<0.001
Intercept	4.29	0.015	272.6	<0.001

All variables were log-transformed to improve normality.

Based on the fitted regressions with empirical data in our training dataset (Fig. 2B), we can estimate the decline in water quality for source watersheds between 1900 and 2005. Sediment yields increased by 40% between 1900 and 2005, with a roughly linear pattern of increase (Fig. 2C). Similarly, P yields have increased by 47% over the study period. The biggest increase is for N yields, which increased 119% over the study period, with a more exponential pattern of increase.

Global average figures mask substantial variation among source watersheds. Because sediment, N, and P yields were correlated among one another, we used a principal components analysis (PCA) to describe the main axis of variability in watershed degradation (Table S3). Ninety percent of urban source watersheds had some degree of watershed degradation (Fig. 2D). Around 44% of cities had a moderate or severe decline in their source watershed. A small percentage of watersheds (10%) had an improvement in water quality over the 20th century.

Table S3.

PCA loadings for the three component variables

Axis	Variance explained (%)	Loadings
		Sediment yield	Nitrogen yield	Phosphorus yield
1	85.1	0.528	0.608	0.592
2	13.5	−0.841	0.279	0.464
3	1.4	−0.117	0.742	−0.659

Only PCA axis 1 was used in further analysis. Component variables were log-transformed to improve normality. The PCA was conducted on scaled and centered component variables.

Source watersheds with higher N yields are associated with more complex water-treatment technologies (Fig. 3A). For instance, for source watersheds in the cleanest third of watersheds globally, 42% of WTPs use two-stage filtration, direct filtration, or no filtration, technology categories that require relatively clean raw water. In contrast, source watersheds in the dirtiest third of watersheds globally have only 22% of their WTPs using these same three technologies. A similar association between water quality and technology level exists for P yield and sediment yield, and we modeled technology level as a function of the first axis of our PCA using an ordinal logistic regression (Fig. S3).

Fig. 3.

Effect of water quality on treatment costs. (A) Treatment technology as a function of population density, empirical trends. Trends for cropland look similar. (B) Average operations and maintenance (O&M) and replacement cost as a function of treatment technology, for a 250 MLD plant. Replacement cost is expressed as the annual cost, assuming a 30-y bond at 5% interest rate.

Fig. S3.

Probability of occurrence of each WTP technology, as a function of the first score of the PCA. Greater values on PCA axis 1 correspond with higher values of sediment, N, and P yields (i.e., greater watershed degradation). See Table S3 for PCA details. Probability of occurrence was modeled with an ordinal regression, with the coefficient for PCA axis 1 having an odds ratio of 1.29 (95% CI, 1.12–1.50; P < 0.001).

The association of lower water quality with more complex water-treatment technology classes matters because more complex water-treatment technologies cost significantly more (Fig. 3B). A typical 250 million liter per day (MLD) no-filtration WTP might cost $104 million in capital costs to build, plus $1.7 million per year in O&M costs, for a total annualized cost of $8.5 million. This cost is a 20% lower annualized cost than a so-called conventional filtration plant, which uses sand or gravel filtration. At the other end of the spectrum, an advanced filtration plant, such as one using membrane filtration, would have 2.1-fold greater annualized costs than a conventional filtration plant.

We estimate that 29% of cities globally have had a significant increase in water-treatment costs due to watershed degradation between 1900 and 2005. That is, these cities would likely be using a lower technology level today if the watershed could be restored to the land use patterns of 1900. The WTPs impacted fit a specific profile. First, their source watershed has had rapid degradation, often significantly exceeding the global average rate of source watershed degradation. Second, in 1900 the watershed was relatively pristine, so a lower technology level would have been hypothetically likely.

Although only approximately one in four cities have been impacted to date, for those that are impacted, there has been a significant increase in water-treatment costs (Fig. 4). Impacted cities had an estimate 53 ± 5% increase in O&M costs and a 44 ± 14% increase in capital costs. The distribution of impacts has a long right tail, with some cities having a doubling or more of treatment costs due to watershed degradation in the study period.

Fig. 4.

Estimated percent increase in costs due to degradation for impacted WTPs. Around one in four (28%) WTPs draw water from source watersheds that have been sufficiently degraded from 1900 to 2005 to have likely required more complex water-treatment technologies. In these impacted WTPs, the average WTP is 42% more expensive in capital costs and 52% more expensive in operations and maintenance (OM) costs. Error bars are bootstrapped CIs.

Discussion

This paper provides a global estimate of how much natural land cover degradation has increased treatment costs for a sample of the world’s large cities. If the results from our studied cities applied across all global cities, which house 3.6 billion people (2), we would expect that 1.0 billion people are in cities whose treatment costs have been significantly impacted by watershed degradation. One study (28) estimates US $17B annually in capital expenditures for drinking WTPs, so if 29% of all cities had their capital costs impacted by watershed degradation raised an average of 44%, this implies a US $2.2B annual increase in capital expenditures due to watershed degradation. Another study estimated global WTP O&M as US $21B per year (20), so a similar calculation implies that watershed degradation over our study period (1900–2005) has increased WTP O&M by $3.2B per year. The total cost of watershed degradation to water utilities is therefore about US $5.4B annually, which represents a net present cost to urban water utilities of roughly US $108B, assuming a 5% annual discount rate. This financial impact is important to study because it is a cost imposed on urban water-treatment utilities.

By and large, land owners in source watersheds do not consider the effects of their land use on downstream water users. The ecosystem services provided by natural land cover are nonmarket goods, so landowners receive no direct benefit for allowing natural land cover to remain. Conversely, the decision to convert natural land yields benefits to landowners but imposes a large externality on urban water utilities, because of the increase in pollution that often results and the decrease in ecosystem service provision. The conversion of natural land cover to other land uses, such as housing or agriculture, has significant economic value to not just landowners but society at large. Our analysis cannot say whether the degradation of natural land cover is a net good or bad thing, because it only quantified the costs to water utilities of watershed degradation.

Our analysis focused on large cities, but it is worth noting that 1.9 billion people globally live in small cities (<750,000 people) (2). There is reason to think water supply systems are systematically different for small cities than for the large cities we studied. Small cities are much more likely to use groundwater, at least in the United States where comprehensive data are available (29). Small cities also tend to have water intakes for smaller, more local source watersheds. More study is needed to see if the trend we show for large cities holds for small cities.

Most cities, large and small, will be expanding in the 21st century. This urban growth will lead to increased demand for urban water withdrawals, a trend that is happening as surface appears likely to continue being degraded. Cities will also increasingly have to plan for the impact of climate change, which will change water supply and timing in many watersheds globally. Cities may respond to the confluence of factors by developing new water sources, whether from surface waters, groundwater, or desalination. Cities may also try to make better use of their current supply, by limiting leakage from pipes, decreasing domestic consumption, or reusing wastewater (30).

An alternative strategy for cities is source watershed planning and conservation to limit further watershed degradation (30). The goal here is to limit water pollution, often by giving economic value to the ecosystem services that natural land cover provides. This reduction in water pollution can occur through government policy, such as zoning or other land use regulations, or through a payment for watershed services scheme. Regardless the mechanism, the goal is to give value to nature to correct the market failure. Source watershed conservation may be an important strategy to safeguard urban water supplies in the next few decades.

Materials and Methods

Mapping Urban Water Sources.

This study focused on a stratified sample (6) of urban agglomerations greater than 750,000 people, which were surveyed by the World Urbanization Prospects (WUP 2011) report conducted by the United Nations Population Division (2). For each target city, we geolocated freshwater withdrawal points and aligned them with the HydroSHEDS (31) digital elevation model (SI Materials and Methods).

The full geodatabase of urban water sources, called the City Water Map (v2.2), is publically available online at the KNB Data Repository (32).

Changes in Anthropogenic Activities over Time.

Our information on population density, as well as human land use, was the History Database of the Global Environment (HYDE) version 3.1 (33). Details on the HYDE methodology are available online (34) and in SI Materials and Methods. We extract for each urban source watershed the HYDE predictions of population density and agricultural land use for 1900, 1910, 1920, 1930, 1940, 1950, 1960, 1970, 1980, 1990, 2000, and 2005 (Fig. S4).

Fig. S4.

Comparison of cropland area between the HYDE dataset used in this study (A) and the SAGE dataset (B) for the year 1900. See text for details.

Relating Anthropogenic Activities to Water Quality.

There is a large literature on how anthropogenic activities affect water quality, with agricultural land uses (35) and population density (36) affecting sediment and nutrient loading. Many papers have constructed spatially explicit models of sediment (37, 38) or nutrient loading (39, 40) for the contemporary time period, based on the hydrology and mechanistic processes that lead to water pollution. However, the goal of this paper was to construct estimates of water quality over more than a century (1900–2005). More complex, spatially explicit would require detailed global maps of prior land cover and land cover transitions over this 105-y period, which is not available for most watersheds globally. Accordingly, we built a statistical model to predict sediment and nutrient loading based on agricultural land uses and population density, for which we do have estimates over the past century.

To quantify the relationship between the HYDE measures of anthropogenic activity and water quality, we related the HYDE measures to data for the United States from the SPARROW (SPAtially Referenced Regressions on Watershed attributes) database. The SPARROW models structure is described by Schwarz et al. (41). For all urban source watersheds in our sample of US cities, we extracted SPARROW estimates of sediment, nitrogen, and phosphorus loading, using the SPARROW national interpolated grids (42).

These empirically based water quality estimates for the United States were statistically compared with the HYDE estimates of population density, crop land use, and ranchland land use in the source watersheds, using linear regression. The total pollutant loading of sediment, N, and P was modeled as a function of the total crop area in the upstream contributing watershed, the total ranchland area in the watershed, the total human population in the watershed, and the watershed area.

In addition, for sediment we included the watershed average of the RKLS component of the universal soil loss equation (43), which represents the total erosion (not accounting for land use practices) as a product of rainfall erosivity (R), soil erodibility (K), and topography (LS). Our values for RKLS were taken from McDonald et al. (20).

Similarly, for N and P, we included information on the contemporary application rates of these nutrients on cropland and grazing. This information was taken from the global grids of the Global Fertilizer and Manure (GFD), version 1, dataset. Agricultural land was assumed to have both manure and fertilizer applied at the rates specified by the GFD, whereas grassland/pasture was assumed to have only manure applied at the rates specified by the GFD.

To validate the predictions of our statistical model, we compared for circa 2005 our estimate of N loading with that predicted by the Water Balance Model (39, 40), as downloaded from the World Water Development Report II website. The correlation between our statistical model predictions and the Water Balance Model was high (R = 0.71), and generally follows the 1:1 line (Fig. S5).

Fig. S5.

For large source watersheds, the correlation between the estimates of N loading in this study with that from another published global model, the Water Balance Model. The dotted line is the 1:1 line. See text for details.

Please see SI Materials and Methods for more detail on our statistical analysis.

WTP Technologies.

We collected information on treatment technologies used by 264 WTPs for around 100 cities in the United States and around 30 international cities. For more detail on the collection of this dataset, please see SI Materials and Methods and McDonald and Shemie (20). Table S4 lists the WTPs.

Table S4.

WTPs used in this study

WTP_ID	First_city	WTP_name	Technology category
1	Kansas City	Kansas City WTP	Conventional filtration
2	Dublin	Vartry Works Roundwood	Conventional filtration
3	Dublin	Dodder Works Ballyboden	Conventional filtration
4	Dublin	Liffey Works Ballymore Eustace	Conventional filtration
5	Dublin	Lexilip TP	Conventional filtration
11	Bridgeport	Aquarion WTP 1	Conventional filtration
12	Bridgeport	Aquarion WTP 2	Conventional filtration
13	Bridgeport	Aquarion WTP 3	Conventional filtration
14	Bridgeport	Aquarion WTP 4	Conventional filtration
15	Bridgeport	Aquarion WTP 5	Conventional filtration
16	Bridgeport	Aquarion WTP 6	Conventional filtration
17	Bridgeport	Aquarion WTP 7	Conventional filtration
18	Bridgeport	Aquarion WTP 8	Conventional filtration
19	Bridgeport	Aquarion WTP 9	Conventional filtration
20	Multiple MA cities	John J. Carroll WTP	No filtration
23	New York	Croton Water Filtration Plant	Filtration plus
24	New York	Catskill and Delaware UV Disinfection Facility	No filtration
26	Nairobi	Kabeteá WTP	Conventional filtration
27	Nairobi	Ngethu WTP	Conventional filtration
40	Atlanta	Chattahoochee Water Treatment Plant	Conventional filtration
41	Atlanta	Hemphill Water Treatment Plant	Conventional filtration
44	Baltimore	Montebello Filtration Plant No. 1	Conventional filtration
45	Baltimore	Montebello Filtration Plant No. 2	Conventional filtration
46	Baltimore	Ashburton Filtration Plant	Conventional filtration
47	Cape Town	Steenbras Water Treatment Plant	Conventional filtration
48	Cape Town	Blackheath WTP	Conventional filtration
49	Cape Town	Faure WTP	Conventional filtration
50	Cape Town	Voelvlei WTP	Conventional filtration
51	Cape Town	Wemmershoek WTP	Direct filtration
54	Cape Town	Kloof Nek WTP	Conventional filtration
55	Cape Town	Constantia Nek WTP	Conventional filtration
56	Cape Town	Brooklands WTP	Conventional filtration
57	Cape Town	Helderberg WTP	Conventional filtration
59	Rio de Janeiro	ETA Guandu	Conventional filtration
60	Rio de Janeiro	ETA Laranjal	Conventional filtration
62	Tokyo	Ozaku Purification Plant	Conventional filtration
63	Tokyo	Asaka Purification Plant	Filtration plus
65	Tokyo	Misono Purification Plant	Filtration plus
66	Tokyo	Misatno Purification Plant	Filtration plus
67	Tokyo	Kanamachi Purification Plant	Filtration plus
68	Tokyo	Nagasawa Purification Plant	Conventional filtration
69	Tokyo	Kinutashimo Purification Plant	Advanced filtration
70	Tokyo	Kinuta Purification Plant	Advanced filtration
122	Seattle	Cedar Water Treatment Facility	No filtration
123	Seattle	Tolt Water Treatment Facility	Filtration plus
127	Kolkata	Palta WTP	Conventional filtration
136	Chennai	Kilpauk WTP	Conventional filtration
138	Bengaluru	TG Halli WTP	Conventional filtration
139	Bengaluru	TK Halli WTP	Conventional filtration
140	Multiple CA cities	F. E. Weymouth Treatment Plant	Conventional filtration
141	Multiple CA cities	Robert B. Diemer Treatment Plant	Conventional filtration
142	Los Angeles	Los Angeles Aqueduct Filtration Plant	Filtration plus
143	Sao Paulo	ETA TAIA	Conventional filtration
144	Sao Paulo	ETA Guará	Conventional filtration
145	Sao Paulo	ETA Alto da Boa Vista	Conventional filtration
146	Sao Paulo	ETA Casa Grande	Conventional filtration
147	Shanghai	Changqiao WTP	Advanced filtration
149	Shanghai	Pudong WTP	Filtration plus
150	Shanghai	Yangshupu WTP	Conventional filtration
151	Shanghai	Zhabei WTP	Advanced filtration
152	Beijing	Tiancunshan Water Treatment Plant	Filtration plus
153	Beijing	Ninth WTP Surface	No filtration
165	Karachi	COD Filter Plants	Conventional filtration
166	Karachi	Pipri (new) Filter Plant	Conventional filtration
167	Karachi	Pipri (old) Filter Plant	Conventional filtration
168	Karachi	NEK (old) Filter Plant	Conventional filtration
169	Karachi	NEK (new) Filter Plant	Conventional filtration
170	Karachi	Hub Filter Plant	Conventional filtration
171	Karachi	Gharo Filter Plants	Conventional filtration
172	Buenos Aires	Planta Potabilizadora Gral. San Martín	Conventional filtration
173	Buenos Aires	Planta Potabilizadora Gral. Belgrano	Conventional filtration
174	Buenos Aires	Planta Dique Luján	Conventional filtration
175	Manila	La Mesa Treatment Plant 1	Conventional filtration
177	Manila	Putatan Treatment Plant	Advanced filtration
178	Birmingham	Birmingham Water Board	Conventional filtration
179	Birmingham	Birmingham Water Board WTP 2	Conventional filtration
180	Birmingham	Birmingham Water Board WTP 3	Conventional filtration
181	Birmingham	Birmingham Water Board WTP 4	Conventional filtration
185	Multiple US cities	Phoenix Munic Water Sys WTP 1	Two-stage filtration
186	Multiple US cities	Phoenix Munic Water Sys WTP 2	Two-stage filtration
188	Multiple US cities	Phoenix Munic Water Sys WTP 4	Two-stage filtration
189	Multiple US cities	Phoenix Munic Water Sys WTP 5	Two-stage filtration
190	Multiple US cities	Mesa, Munic Water Dept. WTP 1	Two-stage filtration
385	Multiple US cities	Mesa, Munic Water Dept. WTP 4	Filtration plus
386	Multiple US cities	Mesa, Munic Water Dept. WTP 5	Filtration plus
389	Los Angeles	Los Angeles-City, Dept. of Water & Power WTP 1	Filtration plus
391	Richmond	Richmond, City of WTP 3	No filtration
427	Sacramento	Sacramento, City of WTP 1	Two-stage filtration
428	Sacramento	Sacramento, City of WTP 2	Two-stage filtration
429	San Diego	San Diego - City of WTP 1	Two-stage filtration
430	San Diego	San Diego - City of WTP 2	Two-stage filtration
431	San Diego	San Diego - City of WTP 3	Two-stage filtration
433	San Diego	San Diego - City of WTP 2	Two-stage filtration
434	Los Angeles	Los Angeles-City, Dept. of Water & Power WTP 3	Filtration plus
469	San Jose	Santa Clara Valley Water District WTP 1	Two-stage filtration
470	San Jose	Santa Clara Valley Water District WTP 2	Two-stage filtration
471	San Jose	Santa Clara Valley Water District WTP 3	Two-stage filtration
479	Multiple CO cities	Aurora, City of WTP 1	Direct filtration
480	Multiple CO cities	Aurora, City of WTP 2	Direct filtration
483	Multiple CO cities	Denver Water Board . WTP 1	Two-stage filtration
484	Multiple CO cities	Denver Water Board . WTP 2	Two-stage filtration
485	Multiple CO cities	Denver Water Board . WTP 3	Conventional filtration
489	Multiple CO cities	Aurora, City of . WTP 4	Direct filtration
492	Multiple CO cities	Pueblo, Board of Water Works WTP 1	Two-stage filtration
533	Multiple FL cities	City of Tampa-Water Department WTP 1	Two-stage filtration
554	Louisville	Louisville Water Company WTP 1	Conventional filtration
555	Louisville	Louisville Water Company WTP 2	Conventional filtration
560	Multiple US cities	Detroit WTP 1	Conventional filtration
561	Multiple US cities	Detroit WTP 2	Conventional filtration
562	Multiple US cities	Detroit WTP 3	Conventional filtration
563	Multiple US cities	Detroit WTP 4	Conventional filtration
564	Multiple US cities	Detroit WTP 5	Conventional filtration
586	Minneapolis	Minneapolis WTP 1	Two-stage filtration
587	Saint Paul	Saint Paul WTP 1	Two-stage filtration
599	Raleigh	Raleigh, City of WTP 1	Filtration plus
601	Omaha	Metropolitan Utilities District WTP 1	Two-stage filtration
602	Omaha	Metropolitan Utilities District WTP 2	Two-stage filtration
672	Rochester	Rochester City WTP 1	Direct filtration
675	Multiple US cities	Cleveland, City of-Baldwin Plt. WTP 1	Conventional filtration
676	Multiple US cities	Cleveland, City of-Baldwin Plt. WTP 2	Two-stage filtration
677	Multiple US cities	Cleveland, City of-Baldwin Plt. WTP 3	Two-stage filtration
678	Multiple US cities	Cleveland, City of-Baldwin Plt. WTP 4	Two-stage filtration
679	Cincinnati	Cincinnati, City of-Miller WTP 1	Filtration plus
686	Oklahoma City	Okc Draper WTP 1	Two-stage filtration
693	Philadelphia	Philadelphia Water Department WTP 1	Two-stage filtration
694	Philadelphia	Philadelphia Water Department WTP 2	Two-stage filtration
695	Philadelphia	Philadelphia Water Department WTP 3	Two-stage filtration
698	Pittsburgh	Pittsburgh Water & Sewer Auth WTP 1	Two-stage filtration
702	Providence	Providence-City of WTP 1	Conventional filtration
715	Dallas	Dallas Water Utility WTP 1	Two-stage filtration
716	Dallas	Dallas Water Utility WTP 2	Filtration plus
717	Dallas	Dallas Water Utility WTP 3	Two-stage filtration
719	El Paso	El Paso Water Utilities-Pub Serv B WTP 1	Filtration plus
720	El Paso	El Paso Water Utilities-Pub Serv B WTP 2	Filtration plus
721	Houston	Houston City of - Public Works Dep WTP 1	Two-stage filtration
722	Houston	Houston City of - Public Works Dep WTP 2	Two-stage filtration
723	Houston	Houston City of - Public Works Dep WTP 3	Two-stage filtration
724	Dallas	Dallas Water Utility WTP 4	Two-stage filtration
725	McAllen	Mcallen City of WTP 1	Two-stage filtration
726	McAllen	Mcallen City of WTP 2	Two-stage filtration
744	Austin	Austin City of -Water & Wastewater WTP 1	Two-stage filtration
745	Austin	Austin City of -Water & Wastewater WTP 2	Two-stage filtration
746	Austin	Austin City of -Water & Wastewater WTP 3	Two-stage filtration
747	Salt Lake City	Salt Lake City Water Sys WTP 1	Two-stage filtration
748	Salt Lake City	Salt Lake City Water Sys WTP 2	Conventional filtration
749	Salt Lake City	Salt Lake City Water Sys WTP 3	Two-stage filtration
750	Richmond	Richmond, City of WTP 1	Two-stage filtration
761	Multiple US cities	Milwaukee Waterworks WTP 1	Filtration plus
762	Multiple US cities	Milwaukee Waterworks WTP 2	Filtration plus
764	Moscow	Rublevskaya WTP	Filtration plus
765	Moscow	West WTP	Advanced filtration
766	Moscow	North WTP	Conventional filtration
767	Moscow	East WTP	Filtration plus
768	Osaka	Kunijima Purification Plant	Filtration plus
769	Osaka	Niwakubo Purification Plant	Filtration plus
770	Osaka	Toyono Purification Plant	Filtration plus
774	Paris	Orly WTP	Filtration plus
775	Paris	Joinville WTP	Filtration plus
777	Guangzhou	Nanzhou Water Supply Plant (NWSP)	Filtration plus
780	Guangzhou	Shimen Water Plant	Conventional filtration
786	Shenzhen	Meilin Water Works	Filtration plus
792	Seoul	Guui Water Purification Plant	Filtration plus
793	Seoul	Amsa WTP	Filtration plus
794	Seoul	Gangbuk WTP	Filtration plus
795	Seoul	Gwangam WTP	Filtration plus
797	Seoul	Yeongdeungpo WTP	Advanced filtration
798	Seoul	Ttukdo WTP	Filtration plus
799	Jakarta	Pejompongan WTP 1	Conventional filtration
800	Jakarta	Pejompongan WTP 2	Conventional filtration
801	Jakarta	Cilandak WTP	Conventional filtration
804	Jakarta	Buaran I WTP	Conventional filtration
805	Jakarta	Buaran II WTP	Conventional filtration
806	Jakarta	Pulo Gadung WTP	Conventional filtration
807	Multiple Asian cities	Hong Kong WTP 1	Conventional filtration
808	Multiple Asian cities	Hong Kong WTP 2	Conventional filtration
809	Multiple Asian cities	Hong Kong WTP 3	Conventional filtration
810	Multiple Asian cities	Hong Kong WTP 4	Conventional filtration
811	Multiple Asian cities	Hong Kong WTP 5	Conventional filtration
812	Multiple Asian cities	Hong Kong WTP 6	Conventional filtration
813	Multiple Asian cities	Hong Kong WTP 7	Conventional filtration
814	Multiple Asian cities	Hong Kong WTP 8	Conventional filtration
815	Multiple Asian cities	Hong Kong WTP 9	Conventional filtration
816	Multiple Asian cities	Hong Kong WTP 10	Conventional filtration
817	Multiple Asian cities	Hong Kong WTP 11	Conventional filtration
818	Multiple Asian cities	Hong Kong WTP 12	Conventional filtration
819	Multiple Asian cities	Hong Kong WTP 13	Conventional filtration
820	Multiple Asian cities	Hong Kong WTP 14	Conventional filtration
821	Multiple Asian cities	Hong Kong WTP 15	Conventional filtration
822	Multiple Asian cities	Hong Kong WTP 16	Conventional filtration
823	Multiple Asian cities	Hong Kong WTP 17	Conventional filtration
824	Multiple Asian cities	Hong Kong WTP 18	Conventional filtration
825	Multiple Asian cities	Hong Kong WTP 19	Conventional filtration
826	Multiple Asian cities	Hong Kong WTP 20	Conventional filtration
827	Multiple Asian cities	Hong Kong WTP 21	Conventional filtration
828	Bogotá	Planta Wiesner	Direct filtration
829	Bogotá	Yomasa WTP	Conventional filtration
830	Bogotá	El Dorado WTP	Conventional filtration
831	Bogotá	Vitelma WTP	Conventional filtration
832	Bogotá	Tibitoc WTP	Conventional filtration
833	Bogotá	Planta la Laguna	Conventional filtration
834	Santiago	El Complejo Las Vizcachas	Conventional filtration
835	Santiago	Vizcachitas WTP	Conventional filtration
836	Santiago	Ingeniero Antonio Tagle WTP	Conventional filtration
837	Santiago	Planta de Producción La Florida	Conventional filtration
838	Brasília	ETA Rio Descoberto	Conventional filtration
839	Brasília	ETA Lago Sul	Conventional filtration
840	Brasília	ETA Brasilia	Conventional filtration
841	Brasília	ETA Paranoa	Conventional filtration
842	Medellín	Planta de Potabilización Manantiales	Conventional filtration
844	Medellín	Planta de Potabilización Villa Hermosa	Conventional filtration
854	Monterrey	La Boca	Conventional filtration
855	Cali	Cali River Treatment Plant	Conventional filtration
857	Cali	Puerto Mallarino Treatment plant	Conventional filtration
858	Cali	Reform Treatment Plant	Direct filtration
859	Guayaquil	Conventional WTP	Conventional filtration
860	Guayaquil	Lurgi WTP	Conventional filtration
861	Guayaquil	Planta Nueva WTP	Conventional filtration
862	Barranquilla	AAA WTP 1	Conventional filtration
863	Barranquilla	AAA WTP 2	Conventional filtration
864	Barranquilla	AAA WTP 3	Conventional filtration
865	Barranquilla	AAA WTP 4	Conventional filtration
866	Barranquilla	AAA WTP 5	Conventional filtration
868	Quito	Bellavista	Conventional filtration
872	Quito	Puengasí	Conventional filtration
874	Quito	El Placer	Conventional filtration
880	Quito	El Troje	Filtration plus
902	Tegucigalpa	Planta Potabilizadora Los Laureles	Conventional filtration
903	Tegucigalpa	Planta Potabilizadora La Concepción	Conventional filtration
904	Multiple US cities	James W. Jardine Plant	Filtration plus
905	Multiple US cities	South Water Purification Plant	Filtration plus
906	London	Ashford Common Water Treatment Works	Filtration plus
907	London	Kempton Park Water Treatment Works	Filtration plus
908	London	Hampton WTW	Two-stage filtration
909	London	Walton WTW	Two-stage filtration
911	Kinshasa	Ngaliema WTP	Conventional filtration
912	Kinshasa	N’djili WTP	Conventional filtration
913	Kinshasa	Lukunga WTP	Conventional filtration
914	Kinshasa	Lukaya WTP	Conventional filtration
915	Bangkok	Bangkhen WTP	Conventional filtration
916	Bangkok	Samsen WTP	Conventional filtration
917	Bangkok	Thonburi WTP	Conventional filtration
918	Bangkok	Mahasawat WTP	Conventional filtration
919	Hyderabad	Kodandapur WTP	Conventional filtration
920	Hyderabad	Peddapuram Water Works	Conventional filtration
921	Hyderabad	Rajampet Treatment Works	Conventional filtration
922	Tehran	Jalalieh WTP	Conventional filtration
923	Tehran	Kan WTP	Conventional filtration
924	Tehran	Tehran Pars WTP	Conventional filtration
927	Manaus	ETA Mauazinho	Conventional filtration
928	Manaus	ETA 1	Conventional filtration
929	Manaus	ETA 2	Direct filtration
930	Madrid	Torrelaguna	Filtration plus
933	Madrid	Navacerrada	Filtration plus
934	Madrid	La Jarosa	Filtration plus
935	Madrid	Santillana	Filtration plus
936	Madrid	Colmenar	Filtration plus
937	Madrid	Valmayor	Filtration plus
938	Madrid	Rozas de Puerto Real	Conventional filtration
939	Madrid	Pinilla	Filtration plus
940	Madrid	La Aceña	Filtration plus
946	Multiple US cities	R. C. Harris Water Treatment Plant	Conventional filtration
947	Multiple US cities	F. J. Horgan Water Treatment Plant	Conventional filtration
948	Multiple US cities	R. L. Clark Water Treatment Plant	Conventional filtration
949	Multiple US cities	Island Water Treatment Plant	Conventional filtration

WTP_ID is a unique identifier for the WTP that joins with the City Water Map database, available online (40). First_city is the main city served by the WTP, whereas WTP_name is its name. Technology category is our assigned category of treatment technology for the plant. See SI Materials and Methods for details. ETA, Estação de Tratamento de Água.

For the purpose of this project, WTPs were classified into seven categories, based on the categories in McGiveney and Kawamura (44): no filtration; no filtration with additional processing; direct filtration; two-stage filtration; conventional filtration; filtration with additional processing; and advanced filtration (e.g., membrane filtration). O&M and capital costs were estimated following McGiveney and Kawamura for all 264 WTPs in our sample, based on the size of the plant, the treatment category, and the presence of any additional processing steps. We adjusted all costs to US$2015, using the Engineering News-Record Construction Cost Index (ENR-CCI). The methodology of McGiveney and Kawamura produces preliminary design estimates, which they report vary from actual O&M and capital costs by +50% to –30% (44). Note that Table S4 contains information on the treatment technology class of each WTP.

Water-Treatment Costs and Water Quality.

Treatment technology categories were compared with our estimated sediment, N, and P loads using ordinal regression, specifically a proportional odds logistic regression. See SI Materials and Methods for more details on the ordinal regression analysis.

SI Materials and Methods

Mapping Urban Water Sources.

This study focused on urban agglomerations greater than 750,000 people, which were surveyed by the World Urbanization Prospects (WUP 2011) report conducted by the United Nations Population Division (2). The WUP lists the past and current population of each urban agglomeration greater than 750,000 people, which contained 1.8 billion people in 2015 (26). In the first phase, we targeted for data collection the 50 cities with largest population in the WUP 2011 or primary cities if they were larger than 750,000 people. In the second phase, because it was not feasible to collect information on all large cities, we targeted a stratified sample of cities (6).

For each target city, we used web searches in the primary language used in the city to find the names of the water utilities or agencies that supply water, as well as the name of water sources and the amount of water withdrawn. Once the place names of water sources were identified, we geolocated the sources using Google Maps or other geographical atlases.

After geolocation, surface freshwater withdrawal points had their location adjusted (“snapped”) to match the underlying hydrographic river system, as represented by the global high-resolution hydrographic dataset HydroSHEDS (31). The HydroSHEDS digital elevation model was created from NASA’s Shuttle Radar Topographic Mission (SRTM) and further processed to ensure correct hydrographic flow paths. If the snapping adjustment step is not performed, small spatial errors in the location of a point could lead to large errors in the delineation watershed. More detail on the snapping algorithm can be found in McDonald et al. (6).

The full geodatabase of urban water sources, called the City Water Map (v2.2), is publically available online at the KNB Data Repository (32).

Changes in Anthropogenic Activities over Time.

Our information on population density, as well as human land use, was HYDE version 3.1 (33). HYDE provides spatially explicit estimates of global population over time on a 5-min grid, based on historical estimates of population, the fraction of people living in cities, and urban density. Similarly, HYDE provides spatially explicit estimates of global cropland and rangeland using an allocation model on a 5 arc-minute grid, synthesizing information from subnational agricultural production statistics with other geospatial data on factors that control where agriculture occurs (e.g., slope). Much more details on the HYDE methodology are available online (34). As a check on the accuracy of HYDE, we compared it to another commonly used dataset, the Sustainability and the Global Environment (SAGE) dataset (45), which provides estimates of crop area at 30-min resolution. The two datasets are highly correlated (R ∼ 0.6 in all time periods), with similar spatial patterns (Fig. S4).

We extract for each urban source watershed the HYDE predictions of population density and agricultural land use for 1900, 1910, 1920, 1930, 1940, 1950, 1960, 1970, 1980, 1990, 2000, and 2005 using ArcGIS 10.1.

Relating anthropogenic activities to water quality.

There is a large literature on how anthropogenic activities affect water quality, with agricultural land uses (35) and population density (36) affecting sediment and nutrient loading. Many papers have constructed spatially explicit models of sediment (37, 38) or nutrient loading (39, 40) for the contemporary time period, based on the hydrology and mechanistic processes that lead to water pollution. However, the goal of this paper was to construct estimates of water quality over more than a century (1900–2005). More complex, spatially explicit estimates would require detailed global maps of prior land cover and land cover transitions over this 105-y period, which are not available for most watersheds globally. Accordingly, we built a statistical model to predict sediment and nutrient loading based on agricultural land uses and population density, for which we do have estimates over the last century.

To quantify the relationship between the HYDE measures of anthropogenic activity and water quality, we related the HYDE measures to data for the United States from the SPARROW database. SPARROW models are calibrated off of 375 empirical measurements of water quality in the United States, and one of their primary purposes is to allow accurate interpolation of water quality at other points along the US network of streams. The SPARROW models structure is described by Schwarz et al. (41). For all urban source watersheds in our sample of US cities, we extracted SPARROW estimates of sediment, nitrogen, and phosphorus loading, using the SPARROW national interpolated grids (42).

These empirically based water quality estimates for the United States were statistically compared with the HYDE estimates of population density, crop land use, and ranchland land use in the source watersheds, using linear regression. The total pollutant loading of sediment, N, and P was modeled as a function of the total crop area in the upstream contributing watershed, the total ranchland area in the watershed, the total human population in the watershed, and the watershed area.

In addition, for sediment we included the watershed average of the RKLS component of the universal soil loss equation (43), which represents the total erosion (not accounting for land use practices) as a product of rainfall erosivity (R), soil erodibility (K), and topography (LS). Our values for RKLS were taken from McDonald et al. (20). Because we were fitting a log-log regression, the RKLS term was allowed to enter in interaction with watershed size. Given the multiplicative nature of the log-log regression, this allows for the cropland and grazing terms to interact in a way analogous to the CP (cropping factor and practices factor) terms in the standard universal soil loss equation.

Similarly, for N and P, we included information on the contemporary application rates of these nutrients on cropland and grazing. This information was taken from the global grids of the GFD, version 1, dataset. Agricultural land was assumed to have both manure and fertilizer applied at the rates specified by the GFD, whereas grassland/pasture was assumed to have only manure applied at the rates specified by the GFD.

To improve normality, all terms were log-transformed. All regressions were highly significant (P < 0.0001, R² > 0.5), and an analysis of errors using quantile–quantile (Q–Q) plots suggested that the normality assumptions of linear regression were generally met.

To validate the predictions of our statistical model, we compared for circa 2005 our estimate of N loading with that predicted by the Water Balance Model (39, 40), as downloaded from the World Water Development Report II website. Because this version of the Water Balance Model outputted predictions at the 0.5⁰ resolution, we restricted our validation analysis only to large source watersheds (>20,000 km²). The correlation between our statistical model predictions and the Water Balance Model was high (R = 0.71) and generally follows the 1:1 line (Fig. S5).

Our statistical relationship for one point in time was used to predict pollutant loading over time, a so-called space-for-time substitution. This substitution assumes that the relationship between the explanatory and predicted variables has been constant over time. This assumption is likely to be approximately true; for instance, increased land conversion over time has been historically associated with greater erosion and nutrient pollution, as would be correctly projected by our approach. If, however, the signs of the parameters in our regression equation seem unlikely to have changed over the time period of our analysis, their exact value may have changed somewhat over time. For instance, our space-for-time substitution does not account for changes in fertilizer application rates over time, which would tend to change the slope of the relationship between crop land use area and nutrient loading.

WTP Technologies.

We collected information on treatment technologies used by 264 WTPs for around 100 cities in the United States and around 30 international cities. Note that cities often have multiple sources and multiple WTPs, which may treat water from one or more sources. Information collected on each WTP follows that used by the EPA in its surveys of water utilities. Data on WTPs collected by the EPA under the requirements of the Safe Water Drinking Act was obtained through a Freedom of Information Act Request by J.P. It includes more than 30 fields documenting the presence or absence of specific treatment processes, as well as information on the quantity of water treatment. We supplemented this information by collecting information on WTPs for international cities in an equivalent format, based on publically available information in engineering reports about WTPs or financial documents for the water utility. For more detail on the collection of this dataset, please see McDonald and Shemie (20). Table S4 lists the WTPs.

For the purpose of this project, WTPs were classified into seven categories, based on the categories in McGiveney and Kawamura (44): no filtration; no filtration with additional processing; direct filtration; two-stage filtration; conventional filtration; filtration with additional processing; and advanced filtration (e.g., membrane filtration). Examples of additional processing include iron and manganese removal, lime and soda ash water softening processing, dissolved air filtration, preozonation, or granular activated carbon (GAC) filters. O&M and capital costs were estimated following McGiveney and Kawamura for all 264 WTPs in our sample, based on the size of the plant, the treatment category, and the presence of any additional processing steps. We adjusted all costs to US$2015, using the ENR-CCI. The methodology of McGiveney and Kawamura produces preliminary design estimates, which they report vary from actual O&M and capital costs by +50% to –30% (44). It is simply not feasible to collect information on actual O&M and capital costs directly from hundreds of water utilities, because many water utilities do not provide information on the value of their WTPs or list book value (the cost of previous construction of the plant, sometimes far in the past), which makes comparison among WTPs difficult. The advantage of using the McGiveney and Kawamura methodology is that we can consistently estimate O&M and capital costs for all of the WTPs in our sample. Note that Table S4 contains information on the treatment technology class of each WTP.