US cities can manage national hydrology and biodiversity using local infrastructure policy

Significance

We introduce a unique and detailed data-driven approach that links cities’ hard infrastructures to their distal ecological impacts on streams. Although US cities concentrate most of the nation’s population, wealth, and consumption in roughly 5% of the land area, we find that city infrastructures influence habitats for over 60% of North America’s fish, mussel, and crayfish species and have contributed to local and complete extinctions in 260 species. We also demonstrate that city impacts are not proportionate to city size but reflect infrastructure decisions; thus, as US urbanization trends continue, local government and utility companies have opportunities to improve regional aquatic ecosystem conditions outside city boundaries through their hard infrastructure policies.

Abstract

Cities are concentrations of sociopolitical power and prime architects of land transformation, while also serving as consumption hubs of “hard” water and energy infrastructures. These infrastructures extend well outside metropolitan boundaries and impact distal river ecosystems. We used a comprehensive model to quantify the roles of anthropogenic stressors on hydrologic alteration and biodiversity in US streams and isolate the impacts stemming from hard infrastructure developments in cities. Across the contiguous United States, cities’ hard infrastructures have significantly altered at least 7% of streams, which influence habitats for over 60% of North America’s fish, mussel, and crayfish species. Additionally, city infrastructures have contributed to local extinctions in 260 species and currently influence 970 indigenous species, 27% of which are in jeopardy. We find that ecosystem impacts do not scale with city size but are instead proportionate to infrastructure decisions. For example, Atlanta’s impacts by hard infrastructures extend across four major river basins, 12,500 stream km, and contribute to 100 local extinctions of aquatic species. In contrast, Las Vegas, a similar size city, impacts <1,000 stream km, leading to only seven local extinctions. So, cities have local policy choices that can reduce future impacts to regional aquatic ecosystems as they grow. By coordinating policy and communication between hard infrastructure sectors, local city governments and utilities can directly improve environmental quality in a significant fraction of the nation’s streams reaching far beyond their city boundaries.

Cities are the modern world’s epicenters of sociopolitical power and economic production, but also among the primary drivers of land transformation and resource consumption across the globe. Within the United States, almost 95% of the population and household income occurs within metropolitan statistical areas (SI Methods). The world’s growing urban populations will continue to extend commodity supply chains well beyond municipal boundaries, inducing environmental stress in distal geographies (1). As they grow, global cities are shifting toward reliance on expansive infrastructure and supply chain networks (2), which are controlled through a multitude of social institutions and disparate policies (3). Historically, local government policy was typically shaped by the immediate socioeconomic context within municipal boundaries, and externalities beyond that boundary were generally ignored (4, 5). However, city leaders are increasingly taking the initiative to transform regional social and environmental issues, reflecting a desire to leverage a city’s power to improve sustainability and welfare in the city’s area of influence.

Cities’ demand for goods and services are met through consumer supply chains (soft networks) or land, energy, water infrastructures (hard networks). A city’s external soft infrastructure and supply chain (1, 6) involves shipping of commodities, and is controlled by the diffuse individual purchasing decisions of private citizens and businesses; these soft networks are naturally resistant to government policy and control. By contrast, some of a city’s hard infrastructure systems (6, 7), such as land use practices within the municipal boundary (8), water and wastewater systems, and “EnergySheds” (i.e., a region of transmission structures balancing electricity production at power plants with intense consumption in cities) collectively comprise a city’s land/energy/water (LEW) network and tend to be directly controlled by local city governments and utilities (Fig. 1). These infrastructures have wide-ranging direct and indirect impacts on natural resources, particularly aquatic ecosystems. The urban transformation of land to impervious surfaces induces dramatic storm flows (8), displacing water from natural infiltration to downstream communities (9). EnergySheds can be extensive, overlap with other cities’ EnergySheds, and be composed of many different energy production technologies with varying water use (10). Finally, public drinking water supplies can be highly extractive and require infrastructure that transports water beyond natural watershed boundaries. Thus, these hard infrastructures can in principle create pathways by which local governments and utilities can manage ecosystem integrity beyond the municipal boundary.

Fig. 1.

Mapping a city’s LEW network as impacts to hydrologic and biodiversity impacts in river networks enables communication among disparate policy sectors.

The health of aquatic ecosystems is of general interest to the public at large, and of special interest to cities that are located along streams. Understanding the major contributors of hydrologic alteration (9) and biodiversity loss (5) reveals the predominant pathways in which city planners can minimize future impacts to aquatic ecosystems (2). Furthermore, clean and hydrologically intact streams provide water supply, stormwater management, and recreational services to cities. At the same time, cities incur large costs to meet federally regulated goals for stormwater quality and wastewater quality management costs that can be mitigated or exacerbated depending on the ecological health of the stream. Moreover, although municipal boundaries are mutually exclusive, the impacts of cities’ external supply infrastructures overlap with other cities (1), so the hundreds of cities in the United States should be concerned about cooperation and competition on ecosystem and water supply concerns (2).

This study is the first application of a data-driven model to map hydrologic flow alteration and biodiversity impacts on all US streams and attribute these impacts to their anthropogenic causes, specifically those relevant to city infrastructures. Once predominant anthropogenic stressors of hydrology and biodiversity are identified, the study then employs a detailed analysis of five cities varying in geography, population size, and LEW infrastructure to quantify the impact of their hard infrastructures and visualize the pathways by which these cities can directly manage regional aquatic ecosystems using local policy. Herein, we answer the question, What is the extent of a city’s impact on hydrology and biodiversity in rivers when evaluated through its hard infrastructure network (Fig. 1)?

SI Methods

Hydrologic Alteration Indices.

Hydrologic alteration was measured using 12 hydrologic indices, 10 of which were calculated as proportional changes based on (O − E)/E, where O and E are the observed and expected (i.e., natural) values for metrics, respectively. Two of the metrics, seasonality alteration index and cumulative hydrologic alteration, were based on multidimensional measurements. Indices described different aspects of the hydrograph, including the magnitude, timing, frequency, duration, and rate of change in flow (25).

Estimates of natural flow were generated for nonreference gages using models (from reference gages), predam hydrologic records, or by assigning gages to natural hydrologic classes (25). Using 2,249 reference USGS stream gages, random forest models were developed to predict natural flow values at 4,839 nonreference gages for 10 hydrologic metrics based on climate and physiographic information (25). Predisturbance hydrologic information was available for 250 of the nonreference gages. Reference condition values for the 10 metrics were used to calculate hydrologic alteration in nonreference gages using the (O − E)/E formula as described in Methods: mean daily flow, runoff per catchment area, daily Cv flow, 1-d low-flow, 1-d high-flow, low-flow frequency (occurrences < 25th percentile), high-flow frequency 1 (occurrences > 75th percentile), high-flow frequency 6 (occurrences > 3 × median flow), fall rate of flow events, and reversals in flow (25). A seasonality alteration index was calculated using differences in observed (O) and expected (E) values for all mean monthly flows using the formula:

$${\displaystyle \sum _{m=i}^{12}{\left(\frac{(O_i-E_i)}{E_i}\right)}^2},$$

where differences are measured for the i^th month (m). To calculate the cumulative hydrologic alteration index, principal components analysis (PCA) was conducted for reference and nonreference gages combined, and significant components were identified using the broken-stick method. PC scores (S) were partitioned by hydrologic classes (25), and 90th confidence intervals of scores for reference gages were calculated for each significant component representing the interval (a…b), where a and b are lower and upper bounds, respectively. Cases of nonreference gages where $S_i\ne a_i\le S_i\le b_i$ were assigned the eigenvalue (V_i) for the i^th significant PC. The cumulative hydrologic alteration statistic was calculated for each nonreference gauge using

$${\displaystyle \sum _{i=1}^n\left|S_i-a_i\right|*V_i}\hspace{0.17em}\text{for}\hspace{0.17em}{S}_i\le a_i,\text{and}\hspace{0.17em}{\displaystyle \sum _{i=1}^n\left|S_i-b_i\right|*V_i}\hspace{0.17em}\text{for}\hspace{0.17em}{S}_i\ge a_i,$$

where n is the number of significant PCs.

Hydrologic Alteration Models.

Cumulative hydrologic alteration was the only metric used for examining hydrologic alteration for the entire United States, whereas all 12 metrics were used for city case studies. For the entire United States, random forest explained 46.2% of the out-of-bag variance (i.e., cross-validation) in cumulative hydrologic alteration and had low mean square error = 0.068. Random forest model performance is provided in Table S1. Variables used in models predicting hydrologic alteration, their relative importance, and their relevance to different sectors are provided in Fig. S1 and described in Table S2. Eight variables were used in estimating stream hydrologic impacts of ULT, eight variables for EP, and three variables for WS (Fig. S1).

Table S1.

Performance of random forest models predicting hydrologic alteration in 12 metrics

Metric	Tennessee and Atlantic Gulf		Lower Colorado
	% Var	MSE	% Var	MSE
Cum hydrologic alteration	52.7	0.056	32.9	0.092
Mean daily flow	57.1	0.041	24.7	0.073
Daily Cv	43.3	0.019	35.0	0.032
Runoff per km2	50.6	0.077	37.3	0.113
Low-flow frequency	59.2	0.054	38.2	0.077
High-flow frequency 1	49.0	0.029	35.8	0.054
High-flow frequency 6	47.4	0.036	42.1	0.051
1-d low flow	42.4	0.093	32.5	0.107
1-d high flow	54.7	0.042	24.3	0.075
Fall rate	61.7	0.047	26.5	0.067
Reversals	68.0	0.024	45.6	0.041
Seasonality alteration	58.5	0.003	22.3	0.014

The abbreviations % Var and MSE refer to percent variation and mean squared error, respectively, in out-of-bag sample (cross-validation sample).

Fig. S1.

Relative importance of variables used in random forests for the United States, the Tennessee River Basin (TRB) and Atlantic Gulf Basins (ATL), and the Lower Colorado Basin (Lower CO). Error bars represent SD in relative importance values among models for 12 hydrologic alteration metrics.

Table S2.

Descriptions of variables used in random forest models

Variable	Description
ECOHYDR	Ecohydrologic regions (combination of freshwater ecoregions and HUC2 regions)
DRAIN_SQKM*	Upstream drainage area of stream reach in km2
MAFLOWU*	Mean annual flow (m3⋅s−1)
PPT30MEAN*	30-y average precipitation averaged for entire upstream network
nidStorSQK	Cumulative sum of US Army Corps of Engineers National Inventory of Dams storage (megaliters) divided by upstream drainage area (km2)
DOR	Degree of regulation (% of annual runoff stored by dams; see nidStorSQK)
N_prcnt_watr†	Percentage of water land cover
N_prcnt_brrn†	Percentage of barren land cover
N_prcnt_frst†	Percentage of forest land cover
N_prcnt_wetl†	Percentage of wetland land cover
L_prcnt_11†	Percentage of open water land cover
L_URBAN†	Percentage of low, medium, and high development intensity land cover
L_URBANL‡	Percentage of low development intensity land cover
L_URBANM‡	Percentage of medium development intensity land cover
L_URBANH‡	Percentage of high development intensity land cover
L_AGR†	Percentage of all agricultural land cover types
L_PASTURE‡	Percentage of pasture/hay land cover
L_CROPS‡	Percentage of crop land cover
L_POPDENS‡	Population density
L_ROADCR‡	Road crossings per km2
L_ROADLEN	Road length (km) per km2
L_DAMS‡	Number of dams per km2
L_NPDES‡	Number of National Pollutant Discharge Elimination Systems (NPDESs) per km2
N_URBAN	Percentage of low, medium, and high development intensity land cover
N_URBANLC‡	Percentage of low development intensity land cover
N_URBANMC‡	Percentage of medium development intensity land cover
N_URBANHC‡	Percentage of high development intensity land cover
N_AGR	Percentage of all agricultural land cover types
N_PASTUREC‡	Percentage of pasture/hay land cover
N_CROPSC‡	Percentage of crop land cover
N_POPDENSC‡	Population density
N_ROADCRC‡	Road crossings per km2
N_ROADLENC‡	Road length (km) per km2
N_DAMSC‡	Number of dams per km2
N_MINESC‡	Number of mines per km2
N_TRIC‡	Number of Toxic Release Inventory Sites per km2
N_NPDESC2‡	Number of National Pollutant Discharge Elimination Systems (NPDESs) per km2
N_CERCC‡	Number of Superfund National Priorities List sites from the Compensation and Liability Information System
L_DistIndx‡	Local Disturbance Index
NDistIndx‡	Network Disturbance Index
CumDistInd‡	Cumulative disturbance index, calculated from LDistIndx and NDistIndx
N_adrain_sum§	Cumulative area (square meters) subject to artificial drainage
N_irrig_sum§	Cumulative area (square meters) subject to irrigation
N_tiles_sum§	Cumulative area (square meters) of tile drains
N_ditch_sum§	Cumulative area (squared meters) subject to the practice of ditches
N_mw_sum§	Sum of megawatt capacity of all power plants
N_mwh_sum¶	Sum of megawatt hour generation of all power plants
N_Fdivmgd_sum¶	Proportion of mean annual flow diverted for power plant generation
N_Fwthmgd_sum¶	Proportion of mean annual flow withdrawn for power plant generation
N_Fconmgd_sum¶	Proportion of mean annual flow consumed for power plant generation
N_Fsdivmgdsum¶	Proportion of mean summer flow diverted for power plant generation
N_Fintake_sum#	Proportion of mean annual flow withdrawn for public water supply

Variables preceded by L indicate values are summarized only for the local NHDPlus catchment; N indicates the variable is summarized for the entire upstream network.

^*NHDPlus.

^†Derived from https://www.mrlc.gov/.

^‡Esselman et al. (13).

^§Obtained from US Geological Survey (https://water.usgs.gov/GIS/metadata/usgswrd/XML/nhd_adrain.xml) and then values accumulated in stream networks.

^¶Power plant information obtained from US Energy Information Administration (https://www.eia.gov/electricity/data/eia923) and then values accumulated in stream networks.

^#Water supply intakes obtained from reports or direct requests from state agencies and then values accumulated in stream networks.

Mapping US Infrastructure Impacts.

ULT was summarized as developed land categories (low, medium, or high intensity; https://www.mrlc.gov/) within catchments around each stream reach and the entire drainage basin upstream of each reach. Besides land development, ULT serves as a proxy for unobserved stormwater management. We collected information on locations and water use for all power plants in the United States (Table S3), and then linked power plants to reservoirs on which they depend (e.g., hydropower, nuclear). We then used network accumulation to summarize the total water use, reservoir storage, MW, and MWh associated with electricity production.

Table S3.

Sources of data for mapping city infrastructures (Table 1)

Sector	Data description	Source
Urban land transformation	Urban land cover within urban area	National Land Cover Database 2011. Multiresolution Land Characteristics Consortium (www.mrlc.gov/)
Electricity production	Electricity demand in block groups	Utility network coverage and electricity consumption. S&P Global Platts
		Household electricity consumption estimates. US Census Bureau American Community Survey (https://www.census.gov/programs-surveys/acs/)
		Business locations and categories. Pitney Bowes Spatial Data
	Electrical grid infrastructure and characteristics (transmission lines and capacity, substations and capacity, power plants, utility networks)	S&P Global Platts
	Power plant location, capacity, generation, and water use	Energy Information Administration (https://eia.gov/electricity/data/eia923)
	Reservoirs associated with power plants	National Hydrography Dataset (https://nhd.usgs.gov/)
Water supply	Water demand in block groups	Household water consumption estimates. US Census Bureau American Community Survey (https://www.census.gov/programs-surveys/acs/)
		Business locations and categories. Pitney Bowes Spatial Data
		USGS Water Use in the United States (https://water.usgs.gov/watuse/)
	Knoxville public water supply intakes	Tennessee Department of Environment and Conservation (https://tennessee.gov/environment/article/tdec-dataviewers)
	Atlanta public water supply intakes	Georgia Environmental Protection Branch (https://epd.georgia.gov/watershed-protection-branch)
		Metropolitan North Georgia Water Planning District Water Metrics Report. Prepared by Atlanta Regional Commission, February 2011 (http://documents.northgeorgiawater.org/2010_Water_Metrics_Report_FINAL%281%29.pdf)
	Las Vegas public water supply intakes	20, 21, 34
	Phoenix public water supply intakes	Arizona Department of Water Resources (2017) ADWR GIS data and maps (http://www.azwater.gov/azdwr/gis/)
		Arizona Water Resources Development Commission (2010). Final Report, Vol 1. Arizona Department of Water Resources, Phoenix
		18
	Tucson public water supply intakes	See sources for Phoenix
		City of Tucson (2004) Water Plan: 2000–2050. City of Tucson Water Department. Final Report. (https://www.tucsonaz.gov/files/water/docs/waterplan.pdf)
		City of Tucson (2017) Tucson Water Plans, Reports, Organizational Documents (https://www.tucsonaz.gov/water/documents-and-links)
	Reservoirs associated with intakes	National Hydrography Dataset (https://nhd.usgs.gov/)

Wastewater impacts were included in random forest models via National Pollutant Discharge Elimination System (NPDES) permits, but had negligible statistical effects on hydrologic alteration and were excluded from further consideration. Fuels for electricity production were excluded because water use for this production is still highly uncertain, but more importantly, because local city government and utilities do not usually control fuels production for electricity. Roads and railroads are excluded because most are not under control by cities or are lumped in with urban land use at the local level. Ultimately, this suggests our estimates of city pathways to hydrologic alteration and biodiversity are minimums that could be substantially exceeded if loosely connected infrastructures were included.

Mapping City Infrastructure Impacts.

Although city-specific impacts were informed by the US analysis, mapping city infrastructures required more specific attention, because we were isolating impacts of each city from other anthropogenic sources of alteration (e.g., other cities) in the landscape. Once LEW infrastructures were identified (see paragraph below), they were geographically isolated and summarized as variables (Table S1) for dendritic stream reaches using network path analysis. Using Eq. 1 (main text), we mapped hydrologic impacts for ULT, EP, and WS.

Data sources used for mapping city infrastructures are provided in Table S3. ULT was created by summarizing 2011 national land cover classes within urban areas, as defined by the Census Bureau. EP and WS were developed using the following methods. Using penalized maximum entropy dasymetric modeling (30), downscaled estimates of annual water and energy use were generated for block groups within urban areas (31) by combining data from Public Use Microdata Areas, utility-level electricity consumption, American Community Survey electricity consumption for households, and business point data (Table S3). Water supply intakes, their withdrawal estimates, and associated reservoirs were assembled for surrounding regions (from requests to state agencies) and attributed to cities by reviewing metropolitan reports and balancing urban consumption with intake withdrawal (Table S3). EnergySheds were delineated based on the spatial extent to which urban energy consumption was balanced by energy production in the surrounding grid relative to consumption in the surrounding rural region. Electric grid networks were developed using Utility Network Analyst (ArcMap 10.2) from powerline coverages, respective voltage categories, and their linkages to power plants and substations through network amendments. Connections between census block group centroids and nearest substations were included in grid networks. Energy consumption estimates were aggregated to the nearest substation. Starting at the urban epicenter, incremental network path analysis was conducted to delineate utility networks as concentric rings, which were created through preferentially weighting power lines and power plants of higher voltage or higher energy production, respectively. The most distal utility network at which consumption balanced production was established as the outermost edge of the EnergyShed. Power plants falling within the EnergyShed, their respective water use, and associated reservoirs were assembled.

Biodiversity Impacts.

Based on almost 870,000 georeferenced observations of species, we used several online databases to determine the most up-to-date information for species names, conservation status, and native status (to the United States and basins surrounding cities) (28). After correcting for changes in taxonomic nomenclature and excluding saltwater species (n = 757), our fish list included 1,302 freshwater species, 1,204 and 98 of which were indigenous and nonindigenous to the United States, respectively. The list of crayfish and mussels included 391 and 199 indigenous species (excluding 538 saltwater species), respectively, and 9 nonindigenous clam species. Lists of total indigenous North American freshwater fauna include 1,252 fish, 391 crayfish, and 342 mussels.

US Metropolitan Statistical Area Population, Income, and Urban Electricity Estimates.

Mean total population per metropolitan statistical area (MSA) and mean aggregate household income for past 12 mo (in 2012 inflation-adjusted dollars) of all counties falling inside and outside of MSAs were obtained from the 2008–2012 American Community Survey (ACS) summary tables (Table S3).

Total residential and commercial/industrial electricity consumption per census block group was estimated by multiplying the number of businesses and occupied houses in each block group by the average kilowatt hour per household or business. Averages were calculated from the utility service areas (Table S3) overlapping each household or business, which were then summed to derive the total kilowatt hour demand in each block group. Total numbers of occupied houses and businesses were estimated using the 2008–2012 ACS summaries and the business points provided by the Pitney Bowes Business Points Dataset (Table S3). Electricity consumption of all block groups falling inside and outside US Census-defined urban areas were summed.

Results and Discussion

US Urban Land Transformation and Electricity Production Impacts.

Streams with hydrology departing from natural or reference conditions are termed hydrologically altered, which we represent as changes in any one of 12 different hydrologic indices (Methods and SI Methods). Using a presumptive threshold of 20% hydrologic alteration (11), we estimate that almost 31% of streams (1.56 × 10⁶ km) in the contiguous United States are hydrologically altered based on our cumulative hydrologic alteration index (Fig. 2A and SI Methods). These estimates are congruent with other national assessments depicting hydrologic alteration in stream gages (25%) (12) or characterizing streams habitats using surrogates of hydrologic alteration (39%) (13). However, a more conservative threshold of 10% suggests that almost 80% of streams show some sign of hydrologic alteration. Our results suggested that the most influential anthropogenic drivers of hydrologic alteration in the United States were urban land cover and reservoir storage, whereas other variables related to city infrastructure, such as waste water discharges, were not as significant (Fig. S1). Thus, for the entire United States, we subsequently focused on impacts of urban land transformation (ULT) and electricity production (EP), i.e., indices representing the combined effects of multiple variables related to those sectors (SI Methods).

Fig. 2.

Hydrologic and biodiversity impacts of ULT and EP in the contiguous United States. (A) Cumulative hydrologic alteration mapped to stream reaches and distribution of stream length by degree of alteration. (B) ULT and (C) EP impacts on hydrologic alteration in the nation’s streams. (D) Stream distance and size characteristics impacted by ULT and EP sectors. (E) Biodiversity impacts (fish, crayfish, and bivalves) of each sector consider current (C), historically present (H) but locally extinct, and nonindigenous (NI) species and global conservation ranking (SI Methods). Low (blue bars) and high (red bars) estimates generated by accounting for detection probability.

Impacts from ULT include ∼6.2% of streams (3.14 × 10⁵ km), whereas 1.3% of streams (6.58 × 10⁴ km) are impacted from EP (Fig. 2 B–D). When considered jointly, ULT and EP impact 7% of US streams. Although these estimates may not seem extensive, they result in very large biodiversity impacts. In total, ULT and EP have impacted 1,223 fish, mussel, or crayfish species, 260 of which are locally extinct and 970 of which are currently extant. Of the extant species, 27% are imperiled or vulnerable to extinction (Fig. 2E). On an individual basis, ULT impacts 1,118 fish, mussel, or crayfish species (current or locally extinct), whereas EP impacts 938 species (Fig. 2E). This suggests ULT and EP impact 59% and 50% of all freshwater species found in North America, respectively (SI Methods). Additionally, as much as 192 (20%) species and 216 (19%) species are estimated to be locally extinct due to EP and ULT impacts, respectively. Although ULT impacts occupy far more of the nation’s stream mileage, EP tends to impact far larger systems, with average upstream drainage areas and mean annual flows, 5.6 and 6.7 times greater, respectively, than ULT-impacted streams (Fig. 2D). Likewise, EP impacts 14.2 species per 1,000 km of stream compared with 3.56 species per 1,000 km impacted by ULT (Fig. 2E). As a result, cumulative biodiversity impacts of EP in the United States approximate that of ULT.

Our results clearly display that EP propagates hydrologic impacts within most large river systems in the United States. Electricity production, especially related to reservoir operation, can alter hydrology for extensive river distances (e.g., >10² km) (14). In contrast, ULT is typically compact, intensive, and inherently tied with population density, which suggests urban impacts are localized and transformative of river environments proximate to impervious surfaces (9). Although our results suggest this is true to an extent, the map of ULT hydrologic impacts extend well beyond urban boundaries in many cases (Fig. 2B) and is likely dependent upon the nature and extent of impervious surfaces and exceedance of hydraulic thresholds (9).

We estimate that 92% of US residential and commercial electricity consumption occurs in urban areas (SI Methods). Additionally, more than one third of the streams regulated by power plants (1.9 × 10⁴ km) in the contiguous United States are also recipients of hydrologically modified stream flows from upstream urbanization. This suggests cities not only offset their resource burdens on distal ecosystems (1), but they also compound stress on external regulations. For example, US power plant operations must be responsive to power load demands while minimizing environmental impacts and serving other purposes (e.g., flood control). Hence, irregular flows from urbanization are likely to place additional stress on energy operations, yet there is no federal regulation of storm flows beyond pollution control (15).

Quantifying City Infrastructure Impacts.

The national-scale analysis yielded important insights into the primary drivers of hydrologic alteration relevant to city infrastructures. Here, we transition to assessing the individual impacts of cities on regional hydrology and biodiversity by linking cities, their utilities, and surrounding resources via hard infrastructure mapping. We selected five rapidly growing cities in two groups representing the water-stressed southeastern and southwestern United States, eastern and western power grid interconnections, and “old” (eastern United States) and “new” (western United States) ages and styles of infrastructure and institutions. Due to rapid population growth combined with water stress, cities in these regions have strong potential to cease increased ecosystem impacts and to create cross-competition between cities’ hinterlands via the water and power infrastructure. Cities were similar in that large federal water managers were present in all regions. Capturing city LEW infrastructures requires establishing the city as the hub of networks linking energy demand, water demand, and associated resources in the surrounding landscape. From these interdependent relationships, we derived geospatial data relevant to capturing hydrologic alteration among the ULT, EP, and water supply (WS) infrastructures (Fig. 3). For instance, we identified power plants and water intakes (and associated water use and reservoirs) contributing to each city’s EnergyShed and water supply network, respectively (Table 1). Collectively, we term a city’s ULT, EP, and WS infrastructure the LEW network.

Fig. 3.

Examples of geographic data used to isolate the relative roles of different city infrastructure sectors in altering hydrology in stream networks for Atlanta (Upper) and Las Vegas (Lower). (Left) Developed land cover is summarized within urban areas to represent urban land transformation. (Center) EnergySheds are developed as utility network regions along the electric grid and balance energy demand with production from power plants. (Right) Water supply intakes and power plants supporting city demands are associated with reservoirs and summarized for cities. Sources of data are provided in Table S3.

Table 1.

Characteristics of urban, energy, and water supply sectors for each city used to isolate sector-specific roles in hydrologic alteration models

Characteristic	Knoxville	Atlanta	Las Vegas	Phoenix	Tucson
Population (103)	559	4,515	1,886	3,629	843
Developed land (km2)	704	3,979	827	2,286	584
Public water demand (106 L⋅d−1)	201	1,548	1,416	2,025	19
Per capita water demand (L⋅d−1⋅ind−1)	360	344	750	556	23
Number of intakes	22	87	3	43	16
Reservoir storage public water supply (103 megaliters)	3,424	4,055	37,297	322	0
Energy demand (GWh⋅y−1)	11,717	69,792	17,435	35,633	8,098
EnergyShed area (km2)	18,354	67,922	61,704	23,766	47,391
Per capita energy demand (MWh y−1⋅ind−1)	21	15	9	10	10
Number of power plants	25	142	44	43	30
Reservoir storage power plants (103 megaliters)	9,609	21,443	37,297	4,886	0
Energy Efficiency Score (city rank)*	48.5 (–)	51.5 (18)	33.5 (32)	57 (14)	–(–)

Data used to map city infrastructures are provided in Table S3. Ind, individual.

^*2017 Energy Efficiency Scorecard, American Council for an Energy-Efficient Economy (aceee.org/local-policy/city-scorecard). Scores out of 100. Higher scores and lower ranking indicate superior energy efficiency.

Stream mileage and associated biodiversity impacted from altered hydrology was not strongly related to population size (Fig. S2), per-capita energy demand, or energy efficiency (Table 1), but generally reflected an east-to-west pattern, primarily driven by regional differences in water availability and faunal richness. After accounting for stream network density, we found that relationships between impacts and city population size remained weak (Fig. S2). LEW impacts ranged from 867 km for Tucson to almost 12,500 km for Atlanta (Figs. 4 and 5), and biodiversity impacts included 523 indigenous species for Atlanta but only 2 for Tucson (Fig. 5 and Fig. S3). Streams impacted by western cities had biological communities dominated by nonindigenous species relative to eastern cities (Fig. 5 and Fig. S3) (16). Hydrologic impacts for individual infrastructures also ranged dramatically. For all cities, ULT consistently impacted more stream length than EP and WS sectors; however, EP impacted the most species in Knoxville, Atlanta, and Phoenix (Fig. 5). In comparison with eastern cities, WS impacts approximated those of EP in Phoenix and Las Vegas, a likely result of energy production and water supply infrastructure using the same reservoirs (Figs. 3 and 5).

Fig. S2.

Relationships among population size, city demands, and the degree of hydrologic impacts from ULT, EP, public WS, and the entire LEW network. Pearson’s correlation coefficient (r) and associated significance values are provided.

Fig. 4.

Hydrologic impacts of each city based solely on the ULT sector (in panels) vs. the entire LEW (not in panels).

Fig. 5.

Length of stream and number of fish, crayfish, and mussels species impacted by individual sectors and cumulative urban energy–water nexus footprints for each city. Dots above the bar plot represent relative stream mileage impacted depending upon which hydrologic metric was considered. Percentiles were calculated based on length of stream impacted across 12 different hydrologic metrics. Length of stream and number of species impacted is based on maximum values for the 12 metrics.

Fig. S3.

Biodiversity impacts from individual sectors and cumulative urban energy–water nexus footprints for each city. Species include fish, mussels, clams, and crayfish, and relative proportions of conservation ranking is provided (global ranking from NatureServe). Low and high estimates of species represent species counts corrected for detection probabilities.

Competing Cities and Sectors.

Mapping systemic impacts on river environments reveals competition among cities and the potential to develop cooperative transbasin agreements between local city governments. Undoubtedly, urban geography has considerable relevance to aquatic ecosystem impacts (9, 17) and subsequent city competition. For instance, Atlanta’s ULT extends across the headwaters of three major basins and propagates hydrologic impacts for almost 9,600 river km, which intersect 21 other cities (Fig. 4). In other cases, human–environmental infrastructure results in complex and unexpected water competition without respect to geography. For example, Phoenix and Tucson are geographically proximate to one another, yet share no ecologically relevant overlap in each other’s impacts (Fig. 4). However, Phoenix and Tucson coordinate management of water supplies through the Central Arizona Project (CAP) (18). Las Vegas, however, occurs over 480 km from Phoenix, but exerts hydrologic impacts on 474 km of the lower Colorado River, which directly competes with public water supplies of the CAP. Natural hydrography also plays a large role in urban-generated hydrologic alteration (17). In comparison with water-rich eastern US cities, sparse dendritic stream networks in the western United States promote competition via more intensive water abstraction at fewer locations (Table 1).

Irrespective of geography, cities can impact far reaching areas due to the sheer intensity of resource demands. Atlanta’s EnergyShed impacts 569 km of the Savannah River Basin and 982 km of the Tennessee River Basin (Fig. 4). These impacts only compound the hydrologic alteration resulting from cities more proximate to those watersheds. Additionally, Georgia legislature is renegotiating their state boundary with Tennessee to claim part of the Tennessee River to support Atlanta’s water demand (19). Our framework challenges the prevailing viewpoints of city-to-city water competition and policy governance in two main ways. First, we suggest that city competition does not necessarily follow the traditional upstream-to-downstream model. Indeed, cities occurring downstream or in adjacent basins can inflict just as much, if not more, water competition on other cities than if they had occurred upstream. Second, the only monetary compensation for water use relates to the physical movement of water through interbasin transfers and not virtual water movement, i.e., electricity production. For example, in 2009, Georgia proposed purchasing 379 million L⋅d⁻¹ from South Carolina in the upper Savannah River to support Atlanta (19). Our analysis suggests, however, that Atlanta is already impacting the Savanah River and its tributaries, because the basin provides over 20% of Atlanta’s electricity demand.

Translating LEW networks into metrics of hydrologic alteration offers a template to examine sector-to-sector competition and provide clarity to complex disagreements over water. The 30-y water conflict between Florida and Georgia over flows in the Apalachicola Chattahoochee Flint (ACF) River reached a climax in 2013 with Florida requesting the US Supreme Court create an equitable apportionment of water between the two states (19). Florida’s suit claims that Georgia overuses water for Atlanta’s public water supply and Georgia’s agriculture industry (19). Although withdrawals undoubtedly impact flows in the ACF, our analysis suggests that, by far, the largest hydrologic and biodiversity impacts of Atlanta stem from ULT and EP, not WS (Figs. 4 and 5). Unless the sectors exerting the largest influence on hydrology are abated, we suggest there is little hope to expect drastic improvements in water sustainability in the ACF. To our knowledge, the water conflict has remained tangential to EP impacts.

Mapping competition among sectors also reveals vulnerabilities in a city’s LEW network. Las Vegas’s public water supply impacts are spatially synonymous with its energy impacts because the primary source of hydrologic alteration is withdrawals and operations within Lake Mead, located on the Colorado River (Fig. 3). Las Vegas relies heavily on Lake Mead for both public water supply and hydropower generation. With persistent drought conditions (16), water levels in Lake Mead have remained >70% below full pool (20), and Las Vegas recently completed the construction of a third intake extending deeper into the reservoir (21). Increases in water abstractions from increasing demands come at the expense of losses in hydroelectric generation at Hoover Dam (20). Additionally, limited storage in Lake Mead reduces the flexibility to support environmental flows for protection of endangered species and preventing native species replacement by nonindigenous species (16). Assuming no changes in water allocation strategies, Lake Mead has a 50% probability of losing all usable storage in the next 4 y, which would lead to complete collapse of the agricultural industry and public water supply for the entire region (20).

Conclusions

Where state, federal, or global regulations have failed to ensure future water sustainability, cities provide alternative platforms to make the necessary changes, including implementing local regulations and energy taxes, incentivizing renewable investments, and coordinating policies among cities and utilities (22) (SI Discussion). Our analysis shows that holistic impacts of cities on the water cycle are also not implicitly tied to population size, as others have found for land expansion (23). This suggests that growing cities have a choice in attaining water sustainability by adopting strategies to minimize reliance on infrastructures imposing significant hydrologic alterations to rivers, such as reducing thermoelectric power, or remediation alterations, such as properly managing storm flows (SI Discussion). Attaining future water sustainability for cities will require large-scale, transformative, and expensive solutions (24). This includes novel policy considerations, such as creating new basin treaties merging city governance of hard infrastructures with external institutions managing water infrastructure (SI Discussion).

Holistic and integrated approaches to understand and manage urban systems as complex human–environmental systems are desperately needed (2). The fundamental challenges of translating energy–water nexus science into practice include identifying practical solutions to sustainable water management from the minutia of complex interactions and enabling communication across disparate policy sectors (SI Discussion). Because water is the media by which we measure impacts of the LEW network, city and utility communication should be centralized around the scale at which the water policy operates. Apportioning city- and infrastructure-level environmental impacts, such as biodiversity loss, provides a platform to quantify relative responsibility of different entities in managing shared, but limited, resources.

Methods

Mapping hydrologic alteration across the United States required developing hydrologic alteration models (25) and extrapolating estimates of hydrologic alteration to stream reaches (26). Hydrologic alteration models were constructed using streamflow information from 7,088 US Geological Survey (USGS) stream gages partitioned into reference and nonreference condition (25). Estimates of natural hydrologic conditions were generated for nonreference gages (25), and measures of hydrologic alteration were quantified for 12 hydrologic metrics (SI Methods). Negative and positive changes for all metrics were scaled from 0 to 1 to represent probability of alteration (26). Fifty-two variables that influence the water cycle were assembled for basins contributing to USGS stream gages and for entire networks draining all US NHDPlusV1 stream reaches (SI Methods). Random forests (27) were used to predict measures of hydrologic alteration (Table S1) and extrapolated estimates to 2.6 million stream reaches within the United States.

Random forest models considered a comprehensive and diverse set of variables (Table S2) allowing us to isolate specific anthropogenic causes, such as ULT and EP (SI Methods). Isolating the relative roles of ULT and EP on hydrologic alteration required (i) identifying the individual effects of model variables on hydrologic alteration and (ii) summarizing hydrologic impacts for entire sectors (e.g., ULT and EP). For example, ULT and EP indices were comprised of eight and seven individual anthropogenic stressors, respectively. To identify individual roles of variables in hydrologic alteration (HA), partial dependency predictions (PDPs) were extracted from random forests by holding all other variables constant in the forest, and then predicting responses by varying values for only the variable under consideration. Data from PDPs were scaled from 0 to 1 and then used to develop partial dependency functions (PDFs) using locally adaptive polynomial regressions. PDFs represent the relative direction and magnitude of changes in HA-based values of a given disturbance variable, but this does not yield a measure of relative importance (RI) of variables on HA. RIs were derived from random forest models and scaled from 0 to 1. To calculate the relative hydrologic impacts of a given metric (M) for an entire sector, s, for the j^th stream segment, we used the following equation:

$$M_{sj}=\frac{{\displaystyle \sum _i^{n}PD{F}_{ij}*R{I}_i*\overline{H{A}_j}}}{{\displaystyle \sum _i^{n}R{I}_i}},$$

where hydrologic impacts for i to n individual variables are summed within the respective infrastructure (e.g., ULT, EP) (SI Methods). ${\overline{HA}}_j$ represents estimated hydrologic alteration metric for each stream segment based on predictions from hydrologic alteration models. We then extrapolated estimated impacts of ULT and EP on hydrologic across all streams in the United States.

Characterizing the hydrologic and biodiversity impacts of city LEW networks (ULT, EP, and WS) required that we develop spatial linkages between cities, their resource demands, and distal infrastructures, and then isolate these infrastructures from other unrelated anthropogenic impacts in the landscape (SI Methods). We first created separate hydrologic alteration models for the Tennessee River and South Atlantic Gulf Basins combined (Atlanta and Knoxville) and the Lower Colorado Basin (Las Vegas, Phoenix, and Tucson) and extrapolated $\overline{HA}$ to each stream reach. Establishing network connections between ULT, EP, and WS sectors required balancing resource demands in urban areas with surrounding electricity and water supply sources (SI Methods). Geographic features impacted only by a given city’s LEW network were isolated from the remainder of the landscape (Table S3) and network path analysis was used to summarize those variables in river networks. Using hydrologic alteration models for a respective region, hydrologic impacts for all 12 metrics were calculated for each sector individually and for the entire LEW network. Mapping hydrologic impacts for cities required establishing ecologically relevant alteration thresholds. As little as 10% hydrologic alteration can result in significant ecological degradation (11); thus, streams with ≥0.1 values for hydrologic alteration were assumed to result in biological impacts.

Biodiversity impacts included indigenous and nonindigenous fish, bivalve, and crayfish species either currently present or locally extinct (i.e., historical) within reaches exceeding the hydrologic alteration threshold. Using a database on geographical locations of species presences (28), we spatially joined species occurrence records with stream reaches and partitioned records into historical (pre-1990) and current (post-1990), as justified elsewhere (29). For the entire United States and each city, historical and current species detections falling within water footprints for individual sectors were summarized into species lists along with their conservation status (SI Methods). Comparisons of historical and current species lists yielded the total number of indigenous or nonindigenous species detected in both historical and current records (R_c and NR_c, respectively), the number of indigenous or nonindigenous species currently present but historically undetected (U_c and NU_c, respectively), and the number of indigenous species historically present but currently undetected (U_h). For each taxon, we calculated detection probabilities for indigenous species (p_i) and nonindigenous species (p_n), where p_i = U_c/R_c and p_n = NU_c/NR_c. We then corrected for false absences by inflating species richness estimates for current indigenous species ($\widehat{R_c}$) and nonindigenous species ($\widehat{N{R}_c}$), but deflating locally extinct indigenous species richness ($\widehat{U_h}$) using the following: $\widehat{R_c}$ = U_c/p_i + (R_c − U_c), $\widehat{N{R}_c}$ = NU_c/p_n + (NR_c − NU_c), and $\widehat{U_h}$ = U_h × p_i.

SI Discussion

Fundamental Challenges of Policy Convergence in the Energy–Water Nexus.

The absence of energy–water nexus dynamics in penetrating the water policy sector has likely stemmed from a scientific focus on nonmeaningful scales or soft networks that remain outside the scope of institutional operations (32). Additional complications emerge from a fundamental communication problem across many disparate US policy sectors governed by many independent regulatory agencies and utilities (33). Visualizing and conceptualizing human–environmental systems and their interdependencies is important for enabling effective communication regarding the management of those systems. Mapping city- and sector-level impacts to hydrology within river systems enables communication across disparate policy sectors operating within a city’s infrastructures (Fig. 1). Because water use is not equivalent to hydrologic alteration, gross classifications of water use or even water stress have little relevance to state water policies or biodiversity conservation. To be useful for policy or management, these metrics need to be placed within a meaningful local context. Quantifying the impacts of city infrastructures through the lens of stream networks provides a communication platform for influencing water policies by appropriating a sector’s role in changing the water cycle and freshwater ecological communities.

Transforming City Policies.

Forging city policies that manage regional ecosystems and environments using hard infrastructure requires understanding the spectra of local to regional policies, their constraints, and their potential to create positive impact. Adequately addressing issues of water sustainability will require holistic transformation of urban infrastructure designs (24). These transformations not only have implications for ecosystem impacts, but also city resilience. In an unfortunate coincidence, most global urban regions anticipated to see severe water stress or to face environmentally triggered economic decline are also the most ill-prepared to adapt to such crises (19–21). Through inter- and intracity policy coordination, cities can achieve these expensive changes in their impact trajectory, such as divestment of funds from water-intensive energy producers, water taxes (similar to carbon), and incentives to encourage water-saving technologies (22). With increasing populations, the tendency is to assume two alternative options for addressing city resource needs: reduce demand or increase supply (34). We suggest a third and superior option: achieving both reduced demand and resilient supplies by transforming city infrastructures into low-impact designs. Although increasing efficiency is a commendable practice, efficiency has little benefit to diminishing regional and global ecosystem impacts if new infrastructure designs are not adopted (24). Based on our analysis, city-level impacts on regional hydrology and biodiversity are driven by infrastructural designs within three sectors: urban land cover, electricity production, and water supply. We review city policies surrounding each of these sectors.

Urban land transformation.

Globally, US cities are among the lowest-density urban designs, typically consuming large areas of land per person (24). The model of urban growth in the southeastern United States is particularly low density and highly expansive (35). For example, Atlanta and Knoxville occupy 1.5 and 2.1 times, respectively, the amount of developed land per person than the southwestern US cities and several times more land per person than most global cities. This area is directly proportionate to the ecosystem impacts of ULT. Additionally, urbanization is far less dense in the eastern cities with developed land:urban area ratio being 0.57 and 0.61 for Atlanta and Knoxville, respectively, compared with more dense urbanization in the western United States (0.64–0.77). Our analysis suggests that at the regional scale, urban extensification—especially in the case of Atlanta—leads to greater impacts in streams than intensification. This finding corroborates the general effectiveness of modern urban stormwater management infrastructures, but also highlights the fundamental impact that land transformation creates on aquatic ecosystems regardless of management practices. Because as little as 10% urban land cover can induce significant hydrologic and ecological impacts in streams (9), less-intensive but more extensive urban development leads to more widely dispersed impacts in more streams and more species.

Electricity production.

Altering the trajectory of city impacts on ecosystems stemming from electricity production will require policies transforming hard infrastructure to minimize its impacts, more than by reducing demand through efficiency upgrades (24). For example, Atlanta has many initiatives to increase energy efficiency and ranks close to Phoenix in national energy efficiency scores (Table 1). However, Atlanta has the most extensive water-dependent hard infrastructure to support its demands, and creates more ecosystem impacts compared with other cities in our analysis. Infrastructure must be fine-tuned to mitigate impacts, or the gains of efficiency are lost.

Novel policies start with ambitious goals. Renewable energy sources like wind, solar, and photovoltaics have water use and aquatic ecosystem impact advantages over thermoelectric fossil fuel electricity sources. Only one (small) US city, Burlington, VT, has 100% of its electricity demand met by renewable sources as of 2012 (36). Multilevel governmental incentives play a large role in fostering investments in renewables. For example, cities with local finance incentives and state renewable portfolio standards (RPSs) deploy 69% and 295% more solar photovoltaic (PV) capacity, respectively, than cities without those incentives (37). Phoenix, Tucson, and Las Vegas all have local solar finance incentives and their respective states all have RPSs (37). In contrast, Atlanta and Knoxville lack both city-level financing incentives and state RPSs. The price per watt for solar PV is far cheaper in the western compared with eastern United States (37), owing to economies of scale, regulatory conveniences, and to a lesser extent diminished cloud cover in the west. Not surprisingly, western cities in our study have far more solar farms either currently operating or under development within their EnergySheds (Table S4). Although solar energy capacity is substantial across the United States, generation from solar sources only contributes to 4% and 19% of the overall electricity production in Phoenix and Tucson, but only 1.5% and 0.1% of production in Las Vegas and Atlanta. This suggests that solar and wind renewable capacities and associated infrastructures will require considerable expansion to contribute to a larger portion of electricity consumption in cities to reduce aquatic ecosystem impacts of the hard power infrastructure. Most cities pursuing high renewable penetration have the political flexibility to do so and are motivated to decrease or offset carbon emissions from fossil-based energy sources (38). Because of synergies between emissions reduction and aquatic ecosystem impact mitigation, we suggest that cities explicitly add regional aquatic ecosystem impact and water use management to carbon emissions reduction as power infrastructure policy goals, if they have not done so already.

Table S4.

Number of total and large (>10 MW) solar farms currently operating or under construction within each city’s EnergyShed and their cumulative solar capacity

City	Solar farms	Large solar farms	Solar MW	Solar	Coal	Percent generation by source
						Nuclear	Hydro	Natural gas	Wind	Other
Atlanta	9	5	290	0.1	48	17	5.3	28	0.0	0.8
Knoxville	1	0	0	0.0	35	39	26	0.0	0.5	0.0
Las Vegas	33	23	2,875	1.5	14	0.0	16	68	0.0	0.2
Phoenix	32	19	1,564	4.0	0.0	73	0.1	21	0.0	1.7
Tucson	26	10	323	19	28	0.0	0.0	46	0.0	6.2

Percent generation is based on currently operating plants within each city’s EnergyShed with information available from Energy Information Administration. Information obtained from Solar Energy Industries Associated Major Solar Projects List, www.seia.org/research-resources/major-solar-projects-list. Accessed June 8, 2017.

There are three important policy and technology-related caveats to achieving a large mix of renewable energy. First, US cities that operate their own municipal utility have greater leverage in renewable energy investments as opposed to investor-owned utilities or independent system operators (39). Because municipal utilities have monopolies over transmission and power plant infrastructure and captive customer bases, they are free to rearrange their energy market (39). Cities without this market structure have far less flexibility, but can still impose taxes on utilities and provide incentives, both of which promote investments in renewables (22). A second important caveat is that the mix of available renewable resources will determine the ability of a city to attain high renewable penetration. For example, Burlington, VT, has an energy portfolio of 50% hydropower, 30% biomass, and 20% wind energy (36). Likewise, Norway has achieved near-100% renewable energy, relying primarily on hydropower energy (97%) (40). Hydropower is a flexible energy resource that can provide energy storage, while also meeting peak demand. In contrast, large integration of intermittent renewables, such as wind generation, is not viable unless paired with a mix of flexible energy resources (e.g., natural gas) or energy storage technologies (38). The final caveat is that underlying the desire of cities to promote renewable penetration lies a fundamental assumption: minimizing city impacts on the environment requires renewable energy technologies. This is certainly not always the case; in fact, carbon-neutral renewables may not translate into water-neutral technologies. For example, whereas hydropower expansion can displace fossil fuel-based technologies, hydropower has significant effects on hydrology and biodiversity. For instance, Norway’s pursuit of near zero-emission status has come at great cost to the nation’s aquatic ecosystems, such as inducing large impacts on Atlantic salmon (Salmo salar) populations (41). Solar power farms sometimes have significant water use (42), whereas natural gas combined cycle turbines use very little water (43). City governments should work with their utilities to think creatively about how to minimize total environmental impacts using the conventional power technologies in their portfolio. For example, much of Phoenix’s electricity consumption is derived from the Palo Verde nuclear plant, which is cooled with treated sewage from Phoenix, with municipal economic benefits (44).

Water supply.

Similar to electricity production, water supply impacts on surrounding ecosystems can be minimized through increasing efficiency and/or implementing new infrastructure. Water efficiency does not always yield straightforward results in reducing impacts to regional aquatic ecosystems and flows. Understanding the nuances of complex human–environmental systems may require system-level perspectives and modeling. This should lead to more informed and prioritized investments, such as whether efficiency policies need investment vs. upgraded hard infrastructures. For example, system dynamics modeling revealed that Las Vegas policies on reducing residential outdoor use would be more beneficial than reducing indoor use by the same amount because treated water (from indoors) is returned to the Colorado Rivers system and available for reuse or aquatic ecosystems (34). Water uses in the southeastern United States during the wet season may have little to no ecosystem impacts, whereas water uses during the late summer season may have large impacts. Some groundwater-fed supplies create minimal aquatic impacts. Maximizing the benefits of efficiency policies will require that cities take deep introspective evaluations of their complex infrastructure and system interdependencies.

Although cities have taken initiatives to transform water efficiency policies across sectors, cities continue to use outdated, energy-intensive, and leaky systems for water distribution. Likewise, approaches to water infrastructure investments have remained resource intensive, i.e., more intakes or more storage. For example, in response to growing water demands and lower lake levels, Las Vegas has incrementally deployed new intakes (currently three) at progressively deeper levels in Lake Mead (21). To mitigate Atlanta’s water demand, we estimate that 19 reservoirs have either been constructed or are under construction, expansion has been proposed for 16 existing reservoirs, and 8 new reservoirs are in various stages of permitting in north Georgia since the late 1990s (45). Our analysis suggests that cities should attempt to meet water demands without highly impactful hard infrastructures (e.g., reservoir construction).

The Regional Policy Context.

State and regional regulations and policies provide the context and set boundaries on the flexibility of city-level policies, yet cities’ hard infrastructures and utilities may have the freedom to cross state boundaries and operate cooperatively without involving the state government. In the case of Arizona and Nevada, prior appropriation places strict allocations of use based on purpose, which typically leads to intense water use and restricted flexibility in minimizing ecosystem impacts if appropriations remain unchanged (46). Intense use of water and strict allocations among many parties resulted in the development of basin treaties (e.g., Colorado River Compact), which can be beneficial in managing water for natural ecosystem needs. Additionally, prior appropriation provides a clear legal framework for transferring rights to environmental flows to benefit aquatic ecosystems. Some western states have coordinated policies across sectors to balance water use. For example, Arizona has several integrated energy–water policies, including appropriation of water for electricity generation, but also specific cooling tower restrictions to limit water consumption by plants.

Water rights in the eastern states, such as Georgia and Tennessee, are governed by riparian law and determined based on land ownership (including municipality governance) adjacent to water bodies (19). Water use rights are permitted typically on a case-by-case basis for a specific purpose; however, these rights cannot be transferred. State policies allow permits issued for use within acceptable thresholds that prevent ecosystem degradation; however, these thresholds, and metrics used to quantify them, vary considerably among states. This can lead to overuse and detrimental effects for downstream communities, as in the case of the ACF basin (19). Water abuse and piecemeal approaches to water management created by independent state water rights has questioned the need for “water federalism,” where water allocations are primarily controlled by congressional authorization of federal water infrastructure (47). Large-scale water projects and river basin agreements have great potential to sustainably manage water resources assuming federal water managers can serve as authoritative entities. However, basin treaties, such as the Colorado River Compact, may present obstacles, rather than assistances, to sustainable management if they are not amended with changing environmental conditions (48). Cities can work within these frameworks or cooperate with other cities to work around the state and regional frameworks.

The most effective solutions to achieving urban water sustainability would likely include a combination of city utility governance of hard infrastructures, basin treaties, cooperative agreements between regionally competitive cities, and congressionally authorized federal infrastructure allocating water for humans and the environment. In some cases, cities may have opportunities to coordinate large infrastructures with other cities (e.g., Central Arizona Project) or holistically manage their own water infrastructure for balanced human and environmental needs. For example, Phoenix almost exclusively manages the Salt River Project, which includes both the Salt and Verde Rivers, and hence, could work to restore these systems without the complexity of coordination with other jurisdictions. In the future, we may find that cities want to take direct and exclusive control over regional watersheds so as to more effectively restore and manage those aquatic systems. The hard infrastructure provides a compelling tool to exert this control, because it carries with it the authority and the locally originated financing needed to act independently of state or federal authorities.