A Spatially Detailed Blue Water Footprint of the United States Economy

1 This paper quantifies and maps a spatially detailed and economically complete blue water 2 footprint for the United States, utilizing the National Water Economy Database version 1.1 3 (NWED). NWED utilizes multiple mesoscale (county-level) federal data resources from the 4 United States Geological Survey (USGS), the United States Department of Agriculture (USDA), 5 the U.S. Energy Information Administration (EIA), the U.S. Department of Transportation 6 (USDOT), the U.S. Department of Energy (USDOE), and the U.S. Bureau of Labor Statistics 7 (BLS) to quantify water use, economic trade, and commodity flows to construct this water 8 footprint. Results corroborate previous studies in both the magnitude of the U.S. water footprint 9 (F) and in the observed pattern of virtual water flows. Four virtual water accounting scenarios 10 scenarios were developed with minimum (Min), median (Med), and maximum (Max) 11 consumptive use scenarios and a withdrawal-based scenario. The median water footprint 12 (FCUMed) of the U.S. is 181,966 Mm3 (FWithdrawal: 400,844 Mm3; FCUMax: 222,144 Mm3; FCUMin: 13 61,117 Mm3) and the median per capita water footprint (F'CUMed) of the U.S. is 589 m3 capita-1 14 (F'Withdrawal: 1298 m3 capita-1; F'CUMax: 720 m3 capita-1; F'CUMin: 198 m3 capita-1). The U.S. hydro15 economic network is centered on cities and is dominated by use at local and regional scales. 16 Approximately (58 %) of U.S. water consumption is for the direct and indirect use by cities. 17 Further, the water footprint of agriculture and livestock is 93 % of the total U.S. water footprint, 18 and is dominated by irrigated agriculture in the Western U.S. The water footprint of the 19 industrial, domestic, and power economic sectors is centered on population centers, while the 20 water footprint of the mining sector is highly dependent on the location of mineral resources. 21 Owing to uncertainty in consumptive use coefficients alone, the mesoscale blue water footprint 22 uncertainty ranges from 63 % to over 99 % depending on location. Harmonized region-specific, 23

2008; Seager et al., 2007). Without adequate substitutes for water as an input to production, the 36 economic impact of a drought will propagate beyond local hydrological systems, and dependent 37 water-intensive industries, into the global economy. Disruptions to the production and 38 distribution of water-intensive goods, including electricity and other energy sources, have the 39 potential to spread across seemingly disparate localities over short time periods and are 40 inherently a coupled natural-human (CNH) system phenomenon (Liu et al., 2007). 41 Understanding our vulnerability to these types of events requires a synthesis of network theory, 42 hydrology, geoscience, and economic theory into a unified food-energy-water (FEW) system 43 science that is only possible through the novel fusion of comprehensive economic, commodity 44 flow, hydrologic and geospatial datasets. 45 Due to global economic connectivity, a drought that diminishes the production and trade 46 in water-intensive goods has consequences for water resources management worldwide.
Substitutes for drought-affected agricultural products will have to be cultivated elsewhere by 48 bringing new land under cultivation, intensifying production, or replacing existing crops with 49 crops no longer viable in the Western U.S. (Mann and Gleick, 2015;Castle et al., 2014;McNutt, 50 2014). Given the climatic, political, legal, geographical, and infrastructural constraints to 51 developing new water supplies, which exist to varying extents worldwide, the potential solutions 52 to systemic global water resources problems now lie in managing the scarcity, equity, and 53 distribution of existing water resources through the global hydro-economic network rather than 54 the large-scale development of new, physical sources of water (Gleick, 2003). Further, the 55 importance of managing the scarcity, equity, and distribution of blue water resources only 56 increases as rainwater becomes more variable because the majority of water used for food 57 production in the U.S. is green water (rainwater) (Marston et al., 2018). Physical hydrology and 58 water supply are mostly localized issues of "blue" physical water stocks and flows of both 59 human and natural origin. But the global emerges from the local, and actionable information 60 regarding the scarcity, equity, and distribution of global water resources is attainable only by 61 mapping the network of hydro-economic connections at a local level, associated with specific 62 cities, irrigation districts, rivers, and industries. Hydro-economic connections are created through 63 the trade of water-intensive products and can be measured through virtual water accounting and 64 water footprinting. 65 A water footprint is defined as the volume of surface water and groundwater consumed 66 during the production of a good or service and is also called the virtual water content of a good 67 or service (Mekonnen and Hoekstra, 2011a). Virtual water, also known as indirect water or 68 embodied water, has been studied as a strategic resource for two decades as it allows geographic 69 areas (country, state, province, city) to access more water than is physically available (Allan, 70 1998 Marston et al., 2015). Using NWED data, water footprints of production and consumption can be 72 calculated for U.S. counties, metropolitan areas, and states. A water footprint of production is the 73 total volume of water consumed with a geographic boundary, including water consumption for 74 local use less virtual water export (Mekonnen and Hoekstra, 2011b). A water footprint of 75 consumption is water consumption for local use in addition virtual water import (Mekonnen and 76 Hoekstra, 2011b). 77 This paper presents the first spatially-detailed and economically-complete blue water 78 footprint database of a major country, the U.S., using data from the National Water Economy 79 Database (NWED), version 1.1. The methodological innovations of NWED lie in trade flow 80 downscaling through the novel data fusion of multiple U.S. Federal datasets. This process yields 81 a complete, network-based water footprint database of surface water and groundwater with 82 flexible geographic aggregation from the county-level to international-level for multiple transit 83 modes and trade metrics. NWED is economically complete, to the extent possible, since it 84 utilizes input water data that covers the vast majority of U.S. water withdrawal activities 85 (Maupin et al., 2014). The service industry is included in NWED although we assume virtual 86 water flows resulting from the service industries are de minimus compared to the commodity-87 producing sectors of the economy and thus do not estimate these flows (Rushforth and Ruddell, 88 2015). NWED contains four consumptive use scenarios -a withdrawal-based scenario, in 89 addition to minimum, median, and maximum consumptive use scenarios. Currently, NWED is 90 constrained to blue virtual water flows to focus on potential human-mediated intervention points 91 quarterly, water withdrawal data are not. However, annual water withdrawal and consumption 184 data could be disaggregated to the month scale using median monthly demand curves (Archfield 185 et al., 2009;Weiskel et al., 2010). This lack of data availability does present challenges because 186 there are substantial sub-annual fluctuations in water withdrawal and consumption. Water 187 demands for agriculture and power are highly seasonal and neither the beginning nor the end of a 188 drought coincides with calendar years. For example, the Texas-North Mexico drought began in 189 the latter half of 2010 (Seager et al., 2014). As we further develop NWED, we will develop 190 methods to address this shortcoming, but for now are limited to the annual timescale. 191 192

Geography of NWED 193
The county-scale of geography and annual-scale of time are the appropriate scales of 194 aggregation for a nationally-scoped water footprint analysis in the U.S. given the available water 195 withdrawal and commodity flow data. For the purposes of planning, policy, and law, especially 196 in the absence of larger cities, counties and county equivalents are socio-political units that 197 effectively define the "local" scale of U.S. society and the economy. Additionally, most services 198 are consumed locally within the county where they are produced. In rural areas, a county is an 199 aggregation of socio-economically similar small towns and agricultural areas. In urban areas, a 200 county is more socio-economically diverse, but its statistical data are dominated by a single 201 major metropolitan area and the county is, therefore, representative of that metropolitan area. 202 While the largest metropolitan areas in the U.S. cover several counties and range from a half 203 million people to over 10 million, counties can still capture the economic diversity within the 204 metropolitan area. 205 The FAF FAZ is a group of counties that roughly comprise a metropolitan area, reflecting 206 the fact that the commodity distribution infrastructure of the United States is organized as a 207 spoke-and-hub network with major metropolitan areas and their distribution centers as hubs, thus 208 necessitating the need to develop a disaggregation method. FAZ were disaggregated to the 209 county level using best practices from the literature: population as an attraction factor on the 210 demand side and employment levels, the number of agricultural and livestock operations, and the 211 number of commodity-specific mining facilities on the production side (Viswanathan et  Standardized water use data and water stress data are available nationwide at the county-216 scale but do not typically exist at finer scales. A spatial unit coarser than the county will fail to 217 capture the dominant hydrological and socio-economic patterns in the water footprint, and a finer 218 spatial unit of analysis is not yet possible due to a fundamental lack of consistent, national data at 219 those scales. If finer scale or more up-to-date data do exist, those data may not be consistent with 220 national data, so consistency becomes a primary quality control issue (Mubako et al., 2013). 221 Nonetheless, sub-annual and sub-county scale water use, economic production, water stress, and 222 trade data are all needed to achieve a higher level of detail in the water footprint. 223 224

NWED Naming Convention 225
The general form of a trade linkage (T) in the FAF database is a commodity (c) that flows 226 from an origin FAZ (Oo) to a destination FAZ (Dd) over a domestic transport mode (kdom) 227 represented as tons (t), currency ($), and ton-miles (tm), where o and d are indices for the 123 228 FAZ. Additionally, each c is associated with a broader economic sector (s) that corresponds to 229 the USGS water withdrawal categories. International imports and exports originate from and 230 terminate at one of 8 international origin (OI) and destination (DE) zones via an international 231 transport mode (kint). For an import, a c is produced in an international region (OI) and flows 232 through a port of entry (Oo) and then to a Dd of final consumption. For an export, a c is produced 233 in a Oo and then exits the U.S. through a port of exit (Dd) for consumption in an international 234 region (DE). Domestic, import and export trades can be also classified by a trade type index (f) 235 Therefore, a trade linkage of a commodity in terms of t, $, and tm between an origin zone and 236 destination, which may not include a foreign region, can be represented as 237 " # ," % ,& ' ,& ( ,) *+, ,) '%-,.,/ ( , $, ). NWED builds upon FAF by further disaggregating Oo and Dd to 238 origin (In) and destination counties (Jm), respectively, and by adding virtual water, represented 239 generally as (VW). Each row in NWED is trade linkage, " # ," % ,4 + ,5 -,& ' ,& ( ,) *+, ,) '%-,.,/ , with a 240 corresponding flow of t, $, tm, and VW that can be aggregated by any combinations of index 241 4 → . However, we drop all of these subscripts for a simpler derivation of the NWED 242 disaggregation algorithm. NWED retains data for transport mode, tons, currency, and ton-miles 243 as there are NWED use cases outside of virtual water accounting that may utilize mode-specific 244 data or data on $ or tm flows. 245 246

Water Footprint of a Geographic Area 247
The water footprint of a geographic area ( :;<=> ) is the sum of the direct water use ( ), 248 virtual water inflows ( 4C ), and virtual water outflows ( "D< ) ( The disaggregation method proceeds from the origin side (O), disaggregating to origin 259 counties (I), and then to the destination side (D), disaggregating to destination counties (J). Each 260 O contains a distinct set of one or multiple origin counties (In), where C ∈ , and each D 261 contains a distinct set of multiple destination counties (Jm), where P ∈ . Further, each county 262 (n or m) within each O and D has a unique production factor (PF) and attraction factor (AF) for 263 each economic sector and, where supported by data, each commodity produced in that county. 264 Each I and J can be defined as distinct set of unitless PF or AF factors for each commodity, 265 { C : .U , .V , . . . , .XY } and { P : .U , .V , . . . , .XY }, repectively. Therefore, any Oo or Dd 266 can be represented by a column vector of . or . corresponding to the In or Jm that belong to 267 Oo or Dd. Given that the . or . define the proportion of production capacity and demand 268 attraction a county has within a Oo or Dd, the sum of the . or . for a given Oo or Dd must be 269 equal to 1 to conserve mass. Therefore, for a given commodity (c) with an associated sector (s)  . County-specific, sector-303 level water intensities ( 4 + ,n,I o%,pq ) were calculated as the quotient of county-specific, sector-304 level water withdrawals ( 4 + ,n,I o%,pq ) and county-level, sector-specific commodity production 305 ) and have the units Mm 3 t -1 . In the initial step of calculating 4 + ,n,I o%,pq , 306 groundwater and surface water withdrawals are summed to a total sector-level water withdrawal 307 figure for each county ( 4 + ,n,I o%,pq ). Virtual water flows are disaggregated back to groundwater 308 and surface water fractions in a later step. 309 (4) 4 + ,n,I o%,pq = 4 + ,n,I o%,pq ∑ The resulting 4 + ,n,I o%,pq can be interpreted as the average withdrawal-based water 311 intensity of sector-level production. 312 Next, 4 + ,n,I o%,pq were multiplied by the corresponding 4 + ,5 -,. to arrive at the virtual 313 water flows by county and commodity by transport mode. 314 (5) 4 + ,5 -,.,I o%,pq = 4 + ,n,I o%,pq × 4 + ,5 -,. 315 The 4 + ,5 -,. that results from this process assigns water withdrawals to a commodity 316 based on the tons of a c within a county according to the disaggregated FAF data. Future 317 versions of NWED will refine this process with additional commodity specific water intensities, 318 as explained further in section 2.4. 319 For notational clarity, when 4 + ,5 -,.,I o%,pq is summed for all unique origin counties (In) 320 the term is simplified to "D<,:;<=> . Conversely, when summed for all unique destination 321 counties (Jm) the term is simplified to 4C,:;<=> . Additionally, 4 + ,n,:;<=> summed over all 322 sectors for all unique counties becomes I o%,pq . This notation also holds true for 323 consumption-based virtual water flows. ⁄ 330 (7) 4 + ,5 -,.,st uv',o%,pq = 4 + ,n,st uv',o%,pq × 4 + ,5 -,. 331 Owing to these consumption coefficients being developed for the Great Lakes Region, and 332 climatically similar states, the consumption-based virtual water flows in NWED are preliminary 333 and serve as placeholders until region-or county-specific and sector-level consumption 334 coefficients have been developed for the U.S. 335 Since the USGS water withdrawal data contains data on groundwater and surface water 336 withdrawals for each sector within each county, 4 + ,5 -,.,st upz,o%,pq , 4 + ,5 -,.,st uv',o%,pq , and 337 4 + ,5 -,.,st u*+,o%,pq are split into groundwater and surface water components be multiplying each 338 by the county-specific, sector-specific groundwater withdrawal percentage ( 4 + ,n,|.< ) and 339 surface water percentage ( 4 + ,n,|.< ). The process is shown below for 4 + ,5 -,.,n,<,),st upz . 340 In addition to error introduced in disaggregation, power wheeling within balancing 395 regions is a significant portion of power flow, and this is another source of error (Bialek,396 1996a;Bialek, 1996b; Bialek and Kattuman, 2004). To help compensate for the effect of wheeling 397 on the water footprint of electricity, the water intensity of a power outflows from each balancing 398 area was taken as the source-weighted average of the water intensity of power generation and 399 power inflows. Therefore, virtual water outflows from a county in NWED 1.1 is the virtual water 400 outflow associated with wheeled power through a balancing area (including power originating 401 from this area's generation) in addition to virtual water outflows associated with power 402 generation within that county. Taking into account these modifications to the standard virtual 403 water methods employed elsewhere, virtual water flows were estimated according to the methods 404 in sections 2.5 -2.6. 405 406

Urban-Rural Classification 407
Each county in the U.S. can be categorized using numerous classification schemes. For this 408 paper, and for the purpose of understanding rural-to-urban transfers of virtual water in the U.S., 409 we have classified each county in NWED by the National Center for Health Center for Health 410 Statistics (NCHS) Urban-Rural Classification Scheme for Counties (Ingram and Franco, 2012). Neglecting non-freshwater sources would underestimate the water intensity of the power grid. 432 Reclaimed water is a direct substitute for fresh water, and brackish water is a substitute in some 433 cases, so it is difficult to draw a clear line between included and excluded water withdrawals. 434 Considering the entire U.S. hydro-economy, 15 % of water withdrawals are saline. However, the 435 inclusion of non-freshwater sources does not impact the agricultural virtual water flows as no 436 saline water withdrawals are reported in this sector. For simplicity in this paper, commodity-437 based virtual water flows are reported as 'blue water' even though we incorporate additional 438 types of water beyond freshwater. Power flow-based virtual water flows are presented summed 439 over all water types -not just freshwater. The freshwater footprint of electricity is somewhat 440 smaller than the total water footprint, and this difference is larger on the coasts and in the West. 441 The current version of NWED uses national average U.S. water use efficiencies to 442 estimate international virtual water flows. The first reason for this choice is data consistency. 443 While the USGS water use data does contain some interstate variability due to data reporting 444 methods, the variability is no doubt far smaller than international variability in data reporting 445 methods among countries that mostly lack formal water census programs. Secondly, the U.S. is a 446 large, and geographically, agronomically, climatically, and economically diverse country; water 447 use efficiencies vary dramatically from region-to-region and sector-to-sector. ( 4 + ,4Cf,I o%,pq ), which is estimated by taking the sum of industrial withdrawals and the 464 difference between water withdrawal for public supplies and domestic uses by water systems; 465 mining ( 4 + ,x"C,I o%,pq ); and livestock, which includes livestock and aquaculture withdrawals 466 ( 4 + ,…" †,I o%,pq ). 4 + ,I o%,pq is also known as the Water Metabolism of a county (Kennedy et 467 al., 2015). Total, surface water, and groundwater water footprints within a county match the 468 standard Water Footprint Accounting definition of the water footprint of a geographic area 469 (Hoekstra et al., 2012). For withdrawal-based water footprints, we assume 100 % consumptive 470 use (consumption coefficient CU = 1), forcing USGS-estimated water withdrawals equal to the 471 direct use component of the Water Footprint, WU. Sector-level consumption coefficient data do 472 exist, but these data are specific to the Great Lakes region of the U.S., and climatically similar 473 states, and have large uncertainty ranges (Shaffer and Runkle, 2007). Due to the large 474 uncertainties involved with the consumption coefficients, we have attempted to estimate the 475 uncertainty associated with consumption by using three consumption coefficients for each sector 476 -a minimum (Min), median (Med), and maximum (Max) ( Table 1) With respect to (4), this specifically includes flows of services and labor across county or 490 regional lines (Rushforth and Ruddell, 2015). There is a substantial absolute error introduced by 491 zeroing virtual water flows out from counties that export services and FAF-ignored goods, and 492 this error causes urban areas' net water footprints to be overestimated (and rural areas' to be 493 underestimated by exactly the same amount). Water balances WU are unchanged. However, this 494 error is small in relative terms because these sectors are a small part of total virtual water flows 495 when compared with agriculture, power, and major industry. Labor and services are consumed 496 largely within their county of production. Important exceptions may possibly include the 497 financial services sector, which tends to be national and global in its trading patterns. 498 A limitation in the underlying FAF data is that an assumption must be made that 499 commodity production occurs at the origin and commodity consumption occurs at the 500 destination. Therefore, we must assume that there are no pass-through commodity flows. To the 501 extent possible in the underlying data, this is controlled for at international ports because pass-502 through commodity flows are identifiable from commodity flow to or from the city in which the 503 port is located. However, domestic pass-through commodity flows are not identified in the 504 current version of NWED. A method to estimate pass-through commodity flows using input-505 output methods is under development and will be included in the next version of NWED. 506 Future iterations of the NWED power flow dataset will utilize purpose-built node-507 network power flow models developed at the county-level to differentiate between power 508 outflows into generated power and wheeled power for each county. 509 Texas. In general, we find that the water supply chain, especially the step of the chain bringing 544 agricultural products from the farm to handling and processing facilities where these products 545 become 'food' is mostly local and regional with a smaller but still significant transnational and 546 international water supply chain. 547 548

Urban Dependencies on Rural Virtual Water 549
Circular virtual water flows -virtual water flows that originate and terminate within the 550 same county -are highest for urban counties (Fig. 2). Conversely, rural counties often have 551 small water footprints regardless of the presence of a large water-intensive industry, because 552 rural populations do not consume the majority of the goods produced in those regions. If such an 553 industry were present in a rural county, much of the water withdrawn flows out of the county as 554 virtual water, thus counterbalancing the large withdrawals. Counties that are in the middle of the 555 urban-rural spectrum, often a medium-to-small metropolitan area, rely heavily on agricultural 556 products as an economic input and tend to have the largest virtual water inflows of all U.S. 557 counties. Medium to small cities tend to be food processing hubs where farm goods are 558 transformed into 'food.' and NWED assigns irrigated agricultural blue water footprints to these 559 hubs. We recognize that this framing of the economy emphasizes different parts of the supply 560 chain than previous studies and are developing methods for supply chain harmonization. 561 The central counties of large metropolitan areas (Central) tend to source virtual water 562 equally across the urban-rural spectrum with a slight increase in virtual water sourcing from 563 more medium metropolitan areas and rural counties. However, there is a comparatively small 564 return flow of virtual water from large metropolitan areas back to counties with smaller 565 populations (Table 3). Instead, virtual water originating from counties associated with large 566 metropolitan areas tend to remain within that county as a circular flow or flow to other large 567 metropolitan areas, enlarging the net VW inflow of large metropolitan areas. 568 One such county is Maricopa County, the central county of the Phoenix metropolitan 569 area, which a "local water" hotspot where most of the water used in the community "stays local" 570 in the form of locally consumed virtual water flowing to other users in the same community. 571 This means the community is employing its blue water resources primarily for the hydro-572 economic benefit of its local consumers and businesses. It also means that this community's 573 dependency on its own local water resources is amplified through self-dependence, so any 574 disruption to local water supplies in Phoenix will have a positive feedback loop on that city's 575 economy (Rushforth and Ruddell, 2015). The Phoenix metropolitan area is notable as a major 576 city and population center that is simultaneously a large user of irrigation water for the 577 production of agricultural commodities, including locally consumed food products. Phoenix is 578 also relatively isolated geographically from other metropolitan areas and therefore keeps more of 579 its metropolitan area's virtual water within the local boundary, unlike east coast cities where 580 intra-metro trade and virtual water flows are more prevalent. 581 Counties that are associated with medium-sized metropolitan areas (Medium) break from 582 large cities' and their fringes and take on a different role in the system. While medium 583 metropolitan areas are by no means small, with a population between 250,000-999,999, they are 584 often co-located with large agricultural areas. For example, Ada County, Idaho (Boise metro 585 area), Fresno County, California (Fresno metro area), or Kern County, California (Bakersfield 586 metro) are all counties that contain medium-size metropolitan areas that are co-located with 587 intense agricultural production. In these counties, virtual water tends to be sourced from counties 588 that are as rural as the place of consumption or more rural. Medium-sized metropolitan areas, in 589 particular, are the largest destination of virtual water from rural America while also being one of 590 the largest sources of virtual water for the U.S., especially large metropolitan area -effectively 591 linking rural and urban counties. The medium-medium urban connection is the largest link in the 592 U.S. virtual water flow network, and this link is dominated by the heavy industrial and bulk 593 agricultural and processed food goods that do not tend to be produced by highly rural or densely 594 urban areas. On a per capita basis, the Medium class of city is the core of the U.S. hydro-595 economic network. County-level virtual water flow data show that there is an urban-rural divide, 596 suggesting that there is a fundamental difference in the roles of large urban areas, medium urban 597 areas, and more rural communities in the U.S. hydro-economic network. 598 In the U.S. hydro-economy, economic sectors have different structural roles as either a 599 virtual water sink or source depending on the degree to which a county is rural or urban. 600 Structurally, the agricultural sector is the bulk of the rural-to-urban transfer of virtual water 601 (59,119 Mm 3 ), but rural-to-rural and urban-to-urban virtual water flows are also significant 602 (53,731 Mm 3 and 27,743 Mm 3 , respectively). While similar, the livestock sector constitutes a 603 minority of the rural-to-urban transfer of virtual water (6,100 Mm 3 ) but has little to no impact on 604 virtual water exports. Due to the structure of the underlying commodity flow dataset, the 605 livestock sector only includes on-site water consumption at livestock operations. Inclusion of 606 water usage for livestock feed would, no doubt, increase virtual water transfers related to the 607 livestock sector and a method to do so is under development for the next NWED version. The 608 mining sector is more geographically-dependent and regional on the location of resources and 609 infrastructure. Therefore, while rural-to-urban virtual water flows are the largest within this 610 sector (337 Mm 3 ), rural-to-rural and urban-to-urban virtual water flows are also prominent (175 611 Mm 3 and 165 Mm 3 , respectively). In the power sector, the largest virtual water flow is from 612 rural-to-rural (159 Mm 3 ) followed by urban-to-urban (22 Mm 3 ) and rural-to-urban (13 Mm 3 ). 613 While there are large water withdrawals associated with the power sector, water consumption is 614 relatively low compared to other sectors. Since the results presented are for the CUMed scenario, 615 the power sector virtual water flows are small relative to the other sectors. Finally, the industrial 616 sector is primarily urban-to-urban virtual water transfers. Rural-to-urban virtual water transfers 617 would only become more pronounced if Medium metropolitan areas were considered to be rural 618 counties. While there is subjectivity to whether a county is rural or urban, especially in the 619 middle of the urban-rural spectrum, the predominant flow of virtual water is from rural counties 620 to urban counties. 621 622

U.S. International Virtual Water Imports and Exports 623
Overall, the U.S. is a net virtual water exporter, which qualitatively agrees with the 624 findings from previous international virtual water flow studies (Water Footprint Network, 2013); 625 the virtual water balance of the United States is -4,693 Mm 3 . However, while our virtual water 626 balance results agree qualitatively with previous studies, the magnitude of virtual import and 627 export in NWED is an order of magnitude lower than previously published international virtual 628 water trade data (Water Footprint Network, 2013). Potential reasons for this discrepancy are 629 discussed in Section 3.6. Of the 8 world regions in NWED, the U.S. is a net virtual water 630 exporter to each region, indicated by the negative virtual water balance (Table 4) groundwater sources, and southern Arizona (Fig. 4). Mexico, Africa, and Southwest and Central 656 Asia are the only world regions that received more virtual water in that originated as 657 groundwater (Table 5; Fig 5); suggesting that exports to these regions are potentially vulnerable 658 to unsustainable, long-term groundwater management in the U.S. than annual fluctuations in 659 surface water availability and drought (Marston et al., 2015). 660 While we do not address surface or ground water sustainability, vulnerability, or 661 overdraft specifically in this paper, it is certainly desirable to combine these results with 662 quantification of water storage and water availability, for the purpose of policy analysis. 663 Conversely, Canada, Latin America, Europe, and Asia and Oceania have more exposure to 664 surface water fluctuations and drought but are less exposed to unsustainable groundwater 665 management in the U.S. Given that the U.S. is a large hydrologically, agronomically, and 666 climatically diverse country, it is not surprising that the type of water, surface water or 667 groundwater, which an international trading partner may depend on varies based on which part 668 of the U.S. is accessed and thus potentially causing two trading partners to have vastly different 669 virtual water risk profiles. where food is grown and where irrigation is a requisite for growing crops (Fig. 6a). Where 681 irrigated agriculture is not as prevalent, urban centers are moderate water footprints as they serve 682 as regional distribution for food (Omaha, Nebraska; Wichita, Kansas; Dallas, Houston, and 683 Brownsville, Texas; New Orleans, Louisiana; Northwest Arkansas; and Central Florida). The 684 U.S. livestock footprint is more concentrated on the west coast U.S. and Snake River Valley of 685 Idaho; however, on the east coast, the Carolinas have the largest livestock water footprint (Fig.  686   6c). Outside these areas, the U.S. livestock water footprint is concentrated around cities where 687 there is a relatively large inflow of virtual water with little to no virtual water outflows. 688 Unlike the U.S. water footprint of agriculture and livestock, in which both rural and 689 urban counties play significant roles, the U.S. industrial water footprint (Fig. 6b), and to the same 690 extent the U.S. water footprint of and power production and flow and domestic water 691 consumption ( Fig. 6e and 6f The U.S. mining water footprint is highly dependent on the location of mineral resources 700 in addition to processing facilities and distribution hubs. Some geographic regions with 701 substantial mining water footprint do not have a significant water footprint in other sectors; for 702 example, northern Alaska; west Texas; the Gulf Coast; Oklahoma; North Dakota; northern 703 Michigan and Minnesota; and parts of Nevada, Montana, Utah, New Mexico, and Wyoming 704 (Fig. 6d). Southern California, and to a lesser extent Southern Arizona, is an exception to this 705 because these are regions with substantial mining activity -oil and gas in Southern California 706 and hard rock mining in Arizona -that are co-located with agricultural and industrial production 707 in addition to high domestic water consumption. 708 The net export status of a county matters because a net virtual water exporter may have a 709 very different approach to national water policy discussions than a net importer (Fig. 7). The 710 (usually medium-sized) communities that sit in between the net-importing and net-exporting 711 categories may take a distinct and more balanced position on national policy. Agricultural 712 western communities tend to be net exporters, urban communities tend to be net importers, and 713 rural eastern communities tend to be relatively neutral; midsize urban communities, such as those 714 commonly found in the Midwest and East, may be relatively neutral as well. 715 716

Uncertainty Introduced by Consumption Coefficient Estimates 717
At the county-level, blue water footprint uncertainties introduced by consumption 718 coefficients range several orders of magnitude in Mm 3 and relative percent (Fig. 8). The small 719 rural counties of Bristol Bay Borough, Alaska and Kenedy County, Texas have the smallest 720 water footprint uncertainties (<0.50 Mm 3 ). Los Angeles County, California has the largest water 721 footprint uncertainty (4,050 Mm 3 ). After Los Angeles, 3 counties have a water footprint 722 uncertainty between 3,000 -4,000 Mm 3 ; 7 counties have a water footprint uncertainty between 723 2,000 -3,000 Mm 3 ; 42 counties have a water footprint uncertainty between 1,000 -2,000 Mm 3 ; 724 and 79 counties have a water footprint uncertainty between 500 -1,000 Mm 3 . In relative terms, 725 county-level water footprint uncertainty is 58.2 % -99.9 % of a county's total water 726 withdrawals. Relative water footprint variation tends to increase in the Eastern United States. 727 However, in absolute terms, consumption coefficient variation is more important in the western 728 U.S. due to the potentially large variation in virtual water outflows from the U.S.'s largest virtual 729 water sources. 730 A community's role in the hydro-economic network, and its perspective on hydro-economic 731 policy issues, can qualitatively change depending on our uncertainty. Uncertainties introduced by 732 the consumption coefficients, which are quite large in absolute terms, roughly 17 % of U.S. 733 counties can switch between roles as a net virtual water importer and exporter (+ or -VWBalance) 734 depending on the consumptive use assumptions (Fig. 9). 735 Results using the withdrawal-based (CU = 1) scenario are located in the Supplemental 736 Information (Table SI 4 (Water Footprint Network, 2013). Secondly, despite controlling for port influences, it is likely 749 that more virtual water is attributed to ports than necessary, which would dampen international 750 virtual water flows in NWED. NWED has difficulty handling 'flow through' virtual waters flow 751 that would be otherwise assigned to a point of final consumption. In this case, a flow through 752 entity may be assigned virtual water flow at the port or another distribution hub. Lastly, previous 753 international virtual water studies included the water use of inputs in the virtual water flow of a 754 commodity, e.g., the water consumption for animal feed as part of animal products related virtual 755 water flow. A method to handle this is under development for the next version of NWED. While 756 there are disadvantages to the current method in which international trade is modeled in NWED, 757 methods to improve this aspect of the data product are ongoing and there is data structure in 758 place to merge additional international trade flow datasets with the current NWED data structure. 759 760

Temporal Uncertainty 761
As mentioned previously, the NWED data are limited in representativeness to roughly the 762 2010 -2012 post-recession timeframe but are not precisely linked to a single year. Temporal 763 uncertainty is introduced by utilizing annual timescale data. Given this, NWED data are more 764 directly relevant to surface water management than to groundwater management because surface 765 water has months to a few years of storage, and groundwater has centuries of storage, but in the 766 future we could use this data to analyze sustainability and vulnerability of water usage. 767 Depending on the assumptions about consumptive use at the economic-sector level, these two 779 datasets are in rough agreement regarding the magnitude of the U.S. water footprint. 780 The uncertainty introduced by water use data and consumption coefficients demonstrate 781 the great need for the development of region-specific, sector-level water use data and 782 consumption coefficients for the entire U.S. For example, water footprint uncertainty is roughly 783 58 % to over 99 % of a county's total water footprint, which increases in the eastern United 784 States. However, in absolute terms, consumption coefficient variation is more important in the 785 western U.S. due to the potentially large variation in virtual water outflows from the agricultural 786 sector with largest blue water withdrawals. While we have presented results for the CUMed 787 scenario in this paper, we must recognize the potentially large variation in water consumption 788 that could exist compared to what is reported. Therefore, conclusions drawn from NWED data, 789 as well as those drawn from the underlying water data, must recognize the large range of 790 uncertainty with respect to water withdrawal and consumption in the U.S. Nevertheless, there are 791 still general observable trends in U.S. virtual water flows and water footprints, which are 792 presented below. 793 The U.S. hydro-economic network is centered on cities and is dominated by the local and 794 regional scales of trade, with medium-sized cities playing a disproportionate role. The proper 795 framing of water governance and policy may be proportional to the structure of that network. 796 Large cities source from all sizes of communities, but small and rural communities mostly source 797 from other small communities, leading to a structural difference between the diversity and 798 connectivity of urban and rural water supply chains. hydrological hazards in another area in its supply chain and must be addressed through the 855 development of county-or region-specific and economic sector-specific consumption 856 coefficients. We suggest starting with cities and irrigated agriculture in the Western U.S. due to 857 the major influence that consumption coefficients have on water footprints, and because we lack 858 locally accurate consumption coefficients to distinguish between regions this prevents us from 859 accurately assessing local water balances or scarcity. 860 Despite basic limitations imposed by the primary data sources, NWED is a robustly 861 quantified blue water footprint; future refinements to NWED will seek to address these 862 limitations and add additional functionality, such increased resolution on pass-through 863 commodity flows. The empirical basis of this analysis, along with its economic completeness 864 and spatial detail, make this result a landmark resource in the scientific discussion of water 865 footprints, virtual water flow, and the sustainability and resilience of a nation's water resources 866 in the connected global economy. • If updated disaggregation and attraction factors were available, these factors were 882 updated. 883 • Specifically, agricultural disaggregation factors were updated at the crop level 884 using the latest USDA NASS. 885 • Additionally, the mining sector been updated to have commodity code specific 886 disaggregation factors using the location of mines and mineral production as 887 disaggregation factors rather than employment. 888 • The power sector and domestic sector has been added to NWED version 1.1. 889 • Export virtual water flows have been disaggregated from virtual water flows to 890 port cities. 891 • Import virtual water flows have been added to NWED version 1.