Water scarcity challenges across urban regions with expanding irrigation

Irrigation expansion is often posed as a promising option to enhance food security. Here, we assess the influence of expansion of irrigation, primarily in rural areas of the contiguous United States (CONUS), on the intensification and spatial proliferation of freshwater scarcity. Results show rain-fed to irrigation-fed (RFtoIF) transition will result in an additional 169.6 million hectares or 22% of the total CONUS land area facing moderate or severe water scarcity. Analysis of just the 53 large urban clusters with 146 million residents shows that the transition will result in 97 million urban population facing water scarcity for at least one month per year on average versus 82 million before the irrigation expansion. Notably, none of the six large urban regions facing an increase in scarcity with RFtoIF transition are located in arid regions in part because the magnitude of impact is dependent on multiple factors including local water demand, abstractions in the river upstream, and the buffering capacity of ancillary water sources to cities. For these reasons, areas with higher population and industrialization also generally experience a relatively smaller change in scarcity than regions with lower water demand. While the exact magnitude of impacts are subject to simulation uncertainties despite efforts to exercise due diligence, the study unambiguously underscores the need for strategies aimed at boosting crop productivity to incorporate the effects on water availability throughout the entire extent of the flow networks, instead of solely focusing on the local level. The results further highlight that if irrigation expansion is poorly managed, it may increase urban water scarcity, thus also possibly increasing the likelihood of water conflict between urban and rural areas.


Introduction
Increasing population, dietary changes, and growing per capita income are elevating global food demand [1][2][3][4][5][6][7].Considering 2005 as base year, estimates indicate that crop production needs to be roughly doubled to satisfy the food demand by 2050 [2].Achieving this ambitious goal is further complicated by the impacts of climate change, which affect food production and pose challenges to global food security [8,9].Several strategies are being practiced or explored to increase the crop productivity and making it more resilient [10][11][12][13][14][15][16][17].Among these, a prominent option is through the expansion and intensification of irrigated agriculture [18][19][20].Irrigation can substantially increase crop yield, and reduce the risks from droughts [21,22].Given that the share of irrigated cropland in the US was only 16% in 2005, even though it accounted for 44% of the total crop production [23], there is a potential to significantly increase crop productivity through the transition of rainfed agriculture to irrigation-fed in the US.Recognizing this opportunity, several recent studies have explored its potential implications.For example, it was recently reported that transitioning 26% of the current global rainfed land to irrigation-fed can feed an extra 2.8 billion population [18].Despite its potential, rain-fed to irrigation-fed (RFtoIF) transition may not always be sustainable, especially if the transition is poorly managed.The irrigation expansion may cause river water depletion, groundwater depletion, and pose a threat to the aquatic ecosystem, thus resulting in freshwater scarcity [24][25][26][27].
In this study, we assess the potential impacts of RFtoIF transition of US croplands on blue water scarcity in the contiguous United States (CONUS).Given that RFtoIF transition is expected to increase the water demand for agriculture in the rural areas, our hypothesis is that it may have an impact on the water supply of domestic and industrial sectors in the urban areas.Here we specifically assess the proliferation of blue water scarcity, taking into account both the surface and renewable groundwater availability, in large urban clusters (LUCs) (see section 2.1) due to increased agricultural water use from RFtoIF transition in regions which are largely concentrated in rural areas.In contrast to a majority of the past studies concerned with water scarcity evaluations [28][29][30][31][32][33][34][35][36], and much like a few selected studies [37][38][39], here we explicitly consider the role of water transfer to urban areas from 316 surface water withdrawal points.While this study shares certain similarities with previous studies by Flörke et al [38] and He et al [37], which discussed future increases in water scarcity in urban areas with higher intensity of irrigation intensification in existing irrigated regions, this study, unlike the previous ones, assesses the changes in scarcity also due to spatial expansion of irrigation in previously non-irrigated areas.Furthermore, this study provides additional insights into the causes of why certain regions are more or less susceptible to the impacts of irrigation expansion on water scarcity vis-à-vis their water demand and buffering capacity of ancillary water sources.The need for this evaluation is timely especially given the latent potential for irrigation expansion in central and eastern United States, where several regions have already experienced more than 100% increase in irrigation expansion just within 20 years period [40].

Definition of urban and rural areas
The US Census Bureau delineates geographic areas identifying them as urban or rural.Urban areas represent densely developed aggregations of census blocks, and usually encompass residential, commercial, and other non-residential land uses.Areas not qualifying as urban are coined as rural.Here, the urban-rural area information is obtained from US Census Bureau as TIGER/line Shapefile [41].
In this study, we assess the impact of RFtoIF in 53 large urban clusters (also referred as LUCs henceforth) which are spread over around 11.9 million hectares and populate around 146 million people.The choice of these LUCs is partly motivated by their significant populace, exceeding 750 000, and also due to the availability of comprehensive surface water withdrawal points data for them [39].Population information for urban regions is obtained from Gridded Population of the World (GPW), SEDAC [42].

Assessment of blue and green water scarcity
Green water scarcity is assessed using the GWS index which captures the fraction of crop water requirement that is not met by green water, and is obtained as the ratio of monthly irrigation water demand (=crop water requirement-green water use) and crop water requirement [43].Green water refers to the rainwater and soil moisture consumed by crops.GWS is calculated at monthly resolution using where, CWR is crop water requirement or the amount of water required by a crop to grow optimally, and CWSG is the crop water supply from green water.CWR and CWSG for a given month are calculated by summing daily PET and AET for the month, respectively.GWS is calculated for rainfed crops, therefore, water-limited AET that is solely due to precipitation is used here (see supplementary information for more detail).A region is considered green water scarce if CWSG < 0.9 CWR or in other words, GWS > 0.1 based on Rosa et al [43].
Blue water scarcity in this study is quantified using the cumulative abstraction to demand (CAD) metric, which is the ratio of water abstraction to water demand [24].CAD is calculated at a monthly time step as the ratio of monthly water abstraction to the demand of all the sectors in the grid cell.Here water abstraction corresponds to abstracted water from both surface and subsurface sources, while the water demand quantifies the total water needed to satisfy the demands of agricultural, domestic, and industrial sectors [44].When water abstraction in a region is less than the water demand, CAD falls below unity.Generally, CAD < 1 indicates a water shortage, and an alternative source of water is needed to alleviate water scarcity.Smaller is the CAD value, more severe is the scarcity.A low, moderate, high, and severe blue water scarcity corresponds to 0.8 < CAD 0.99, 0.5 < CAD 0.8, 0.3 < CAD 0.5, and CAD 0.3, respectively.The water scarcity classification thresholds using CAD are consistent with the other widely used water scarcity indexes-water withdrawal to availability and water availability per capita [45].Herein, all reported results regarding the regions that face scarcity correspond to CAD 0.8, which indicates a moderate to high blue water scarcity, unless explicitly stated otherwise.

H08 model simulations
To assess the impacts of RFtoIF transition on blue water scarcity, a global hydrological model, H08 [44], is used to simulate monthly water availability over the CONUS at a spatial resolution of 5 × 5 arcmin.Two scenario simulations are performed.Scenario S1 represents the status quo during 1996-2005, a period around which most of the input data for H08 are available at continental scale (e.g.crop area fraction for 19 crops [46], irrigated area fraction [47], etc).Scenario S2 simulates the transition of all rain-fed croplands that experience green water scarcity, to irrigation-fed.
The H08 consists of six submodels named land surface, river routing, crop growth, water abstraction, environmental flow, and reservoir operations.H08 was run at daily time intervals and a spatial resolution of 5-arcmin over the period 1996-2005 for the CONUS.All submodels of H08 are coupled to obtain monthly blue water demand for agricultural, industrial, and domestic sectors, and blue water availability in each cell.Blue water demand is satisfied by varied surface and groundwater sources.Surface water is supplied by rivers, canals, reservoirs, and desalination plants while groundwater is supplied from renewable and nonrenewable groundwater resources.The municipal sector is given priority in water supply, followed by the industrial and agricultural sectors, respectively.Daily meteorological forcing data of precipitation, wind speed, air temperature, air pressure, specific humidity, and longwave and shortwave radiation were obtained from NLDAS [48] at 0.125 • , hourly, and downscaled at 5 arc min, daily.Additional non-meteorological input data including irrigated area-area equipped for irrigation and area actually irrigated [47], cropland area [49], crop area fraction and spatial distribution of 18 selected crops [46], and water withdrawal for domestic and industrial sectors (FAO [50]) were obtained for the year circa 2000.Other relevant data for H08, including parameterizations, were directly obtained based on Hanasaki et al [51] The environmental flow requirements (EFRs) are determined using Shirakawa's algorithm in which all grids are classified (dry, wet, and stable) based on the monthly minimum and maximum streamflow [52].
The model divides a grid cell into four subcells for the irrigated first-crop area, irrigated secondcrop area, rainfed area, and no crop area.Irrigated areas are assumed to support a maximum of two crops, a major crop or the first-crop and a secondary crop as the second-crop.The model estimates daily irrigation water requirements using meteorological forcing, crop and agricultural information (crop intensity, crop type, irrigation efficiency, etc).Irrigation is applied to the crops to maintain 75% soil saturation.Annual national industrial and municipal water requirements are obtained from the AQUASTAT database [50] and spatially interpolated at 5 arc min according to the population density [42].
H08 incorporates two types of reservoirs, large and medium-sized.Large reservoirs have a catchment area of more than 5000 km 2 and are located on the main river streams and can control the flow.The medium size reservoirs are generally located in the tributaries and act as tank storage, it stores the water until the storage capacity is reached.Any additional water than storage capacity is released to downstream.
The canal water supply system in the H08 enables the grids to transfer water to large distances.H08 considers two types of aqueducts characterized as explicit and implicit.Explicit canals are those that are physically constructed and can be validated by literature, while implicit canals are based on the assumption that the river water is shared with the first neighboring cell.Implicit canals help prevent the artificial gap in water availability for the cells nearby rivers.Due to the unavailability of the continental scale data of explicit canals, the model may underestimate the water abstraction, especially in urban areas.This is alleviated to some extent by the use of city water map data that provides information on the water sources for 53 cities in the US [39].Large cities abstract water from urban withdrawal points (groundwater, surface water, and desalination plants), some of them are located around a few hundred kilometers away from the cities.Urban water withdrawal point information was implemented in H08 as canal origins.
For both scenarios, S1 and S2, the model [44] allocates water to a grid according to the water demand and availability at the source of water.The available water in any grid is the sum of runoff generated in the grid, renewable groundwater reserve, canal water abstraction, water abstraction from reservoirs, and water released from upstream grids after fulfilling their all-sectoral demands to the grid under consideration.The model also accounts for EFRs [53] as an additional demand, while estimating the blue water scarcity.

Assessment of intensification and proliferation of blue water scarcity
The total water demand and abstraction in a LUC is calculated by summing the demand of all LUC grids.The ratio between total monthly water abstraction, both from surface and groundwater, to demand summed over all LUC grids represents CAD for LUCs where TA LUC,m and TD LUC,m are total monthly water abstraction and demand in LUC grids, respectively, calculated by summing daily industrial and domestic water abstraction and demand.We did not consider the agricultural water demand in LUC grids due to the presence of small fraction of irrigated croplands in suburban areas.Urban water withdrawal points serve additional source of water abstraction for LUCs.It is assumed that if an urban water withdrawal point is designated to supply water to a city, all the city's grids can abstract water from it based on their demand.In this study, the intensification of water scarcity is defined as the increase in intensity of blue water scarcity following RFtoIF transition, i.e. areas facing CAD 0.99 being lower CAD in S2 than in S1.Spatial proliferation of blue water scarcity indicates expansion of areas (or model cells, used interchangeably henceforth) that do not face water scarcity to begin with, i.e.CAD > 0.99 in S1, but do so following RFtoIF transition, i.e.CAD 0.99 in S2.

RFtoIF transition's impact on sectoral water use and blue water scarcity
In scenario S1, more than 72.8% of the total cropland area or 82.5% of the total rainfed cropland faces green water scarcity for at least one month in a year (figure S1).This is consistent with previous studies where 70% of the cropland area was reported to be facing green water scarcity in the CONUS [43] during the same period [54].To assess the reliability of our model outputs, we further evaluate the model's performance by comparing simulated runoff with the composite runoff generated by Fekete et al., 2002 [54].The Nash-Sutcliffe Efficiency (NSE) for this comparison yielded a value of 0.69, indicating a robust association between the model results and composite runoff data over the CONUS.Spatially, the GWS magnitude for any given month generally increases with the monthly aridity index (PET/P) (figure S2).Areas facing green water scarcity for at least one month a year on average in S1, are considered for RFtoIF transition in S2 (figure S3).
Given that freshwater is predominantly shared among agricultural, domestic, and industrial sectors, RFtoIF transition alters water availability, and consequently, water withdrawal by all three sectors.Specifically, irrigation expansion causes an increase in annual average agricultural water demand over the simulation period, with total water withdrawal increasing from 318 million m 3 per day to 1119 million m 3 per day after the RFtoIF transition (figure S4).The largest increase takes place in the summer (table S2).Notably, the increase in agricultural water use results in less water available for industrial and domestic water use, resulting in a reduction from 600 million m 3 per day to 587 million m 3 per day.
Next, we assess the average monthly blue water scarcity for both scenarios.The difference in CAD, after and before the RFtoIF transition shows the impact of transition on blue water scarcity (figure 1).The land area facing at least a moderate annual average blue water scarcity (CAD 0.8) increases from 71.5 million ha (∼9.33% of the total land area in CONUS) to 241.08 million ha (∼31.45% of the total land area), i.e. an increase of 169.6 million ha, after RFtoIF transition (see definitions of blue water scarcity severities in section 2.1).The spatial distribution of blue water scarcity varies monthly, and peaks in spring and summer largely because of the increased water demand during this period.The impact is maximum during the month of August, when the land area facing moderate blue water scarcity increases from 68.6 million ha (∼9% of the total land area in CONUS) to 228.7 million ha (∼30% of the total land area) after RFtoIF transition.In S1, around 27% and 66% of the CONUS face blue water scarcity that is at least moderate (CAD 0.8) and low (CAD 0.99) in intensity for at least one month, respectively.The corresponding values increase to 49% and 76% after the RFtoIF transition.The scarcity intensification is largest in High Plains, with Texas, Kansas, and Nebraska experiencing intensification in the majority of months.Significant expansion is also experienced in the eastern US, which has low or no water scarcity in scenario S1.California, Oklahoma, Iowa, Indiana, South Dakota, North Dakota, Minnesota, Illinois, and Missouri observe the spatial proliferation of blue water scarcity, mainly in the summer (figure 1).The somewhat conspicuous reduction of CAD in North Dakota in winter is due to a reduction in water availability for industrial and domestic sectors, which in turn is a result of upstream usage of water for irrigation of winter crops in Montana following RFtoIF transition.Notably, the water reduction in North Dakota is small but the change in CAD is high due to the small water demand.Some areas of Mississippi and Arkansas that contribute to the lower Mississippi river basin also show an increase in blue water scarcity in the summer after the RFtoIF transition.
RFtoIF transition is expected to generally increase blue water scarcity in areas undergoing transition because of the extra water usage in irrigation.The aggravated blue water scarcity in the transitioned area indicates that existing renewable water resources (i.e. the river discharge or reservoirs) and water transportation infrastructure (i.e.implicit and explicit aqueducts, and urban water withdrawal points, (see section 2.3)) are inadequate for fulfilling the increased water demand due to RFtoIF transition.Notably, the RFtoIF transition also causes a rise in monthly blue water scarcity in the areas untouched by the transition.This is because an increase in agricultural water withdrawal from surface water sources due to irrigation expansion in upstream areas leads to a reduction in water flow in river channels, and hence less water availability in receiving lakes and reservoirs.Notably, the average annual surface and groundwater use for irrigation increases from 162 and 156 million m 3 per day to 517.8 and 601 million m 3 per day after RFtoIF transition, respectively.

RFtoIF transition and the urban water security
RFtoIF transition, which is primarily concentrated in rural areas (see definition in rural areas in 2.1) with 97% of the RFtoIF transitioned land lying within it, may have significant impacts on the urban water security.Analyses of blue water scarcity over LUCs (see Methods: Water Supply Data of LUCs) for which detailed data of water supply infrastructure is publicly available, show evidence of both spatial proliferation and intensification of blue water scarcity in them.CAD estimates over LUCs are evaluated to assess the differential impacts of RFtoIF transition on them.The impact of RFtoIF transition is significant in LUCs, with spatial proliferation (intensification) of blue water scarcity increasing by 0.97 million hectares (8.2 million hectares), i.e. around 4.4% (37.5%) of the total area of LUCs considered in this study.Before RFtoIF transition, i.e. in scenario S1, 86.2% of LUC and 90.2% of rest of the area (henceforth referred to as ROA) model cells have CAD values greater than 0.8, which belongs to a low or no water scarcity category (figure 2).After the RFtoIF transition, the percent of cells that face no or low water scarcity reduces to 84.3% and 67.8% for LUCs and ROA, respectively.In contrast, 13% of the total ROA cells are estimated to face a moderate blue water scarcity after the transition, while it was 5.3% in S1.The LUCs also see a hike in the number of cells facing moderate blue water scarcity after transition with fractional area rising to 9.2% from 7.8%.

Variables that influence the spatial distribution and intensity of blue water scarcity, and changes in it due to RFtoIF transition
The spatial distribution of CAD is found to be largely controlled by the relative availability of water from upstream.Locations (or model cells) receiving high incoming lateral flow, which refers to locally generated runoff and water transferred through canals from adjacent cells, or runoff generally have higher CAD values or low blue water scarcity (figure 4(a)).For example, among cells with CAD 0.5, around 55% of them have a lateral flow of less than 0.0001 MCM per day.For non-urban cells, CAD distribution is affected by crop area as well, as water demand increases with higher crop area.Grids in Midwestern US and Central Valley region have higher crop area and hence often have low CAD values.Despite having higher lateral flow, many grids experience lower CAD values due to high crop fraction (figure 4(a)).Further analysis shows a large number of such grids are situated in the dense cropped regions of Midwestern US and lower Missouri river basin.In contrast, for LUC cells that usually do not have any significant fraction of croplands, the water demand and consequently the CAD is influenced by the human population.Cells with less population and large lateral flow tend to show higher CAD values or less blue water scarcity (figure 4(b)).Cells with high population and low lateral flow but still higher CAD are the regions that receive water from distant withdrawal points.
The change in blue water scarcity is quantified by ∆CAD, which is either zero or negative.Of the LUC cells that experience a change in CAD, most observe ∆CAD between 0 to −0.2 (table S3 and figure S6).The same is true for non-transitioned ROA cells.Among the ROA cells that undergo transition, a large fraction of them (>70%) have ∆CAD <−0.2.Around 5.7% of the transitioned grids have ∆CAD ranging from −1 to −0.8.Overall, when all cells are considered, more than 24% of the cells experience ∆CAD <−0.2.
To understand the influences on the spatial distribution of ∆CAD, ∆CAD for LUC and ROA cells are expressed as: (3) where, A1 and D1 (A2 and D2) are the water abstraction and demand in scenario S1 (S2), respectively.∆abs (=A2-A1) and ∆dem (=D2-D1) represent the change in water abstraction and demand due to transition, respectively.∆dem is either zero or positive while ∆abs is either negative or positive depending on the availability of excess water available for abstraction following RFtoIF transition.For LUCs, since the water demand remains the same in both scenarios because of the absence of RFtoIF transition in them, ∆dem is zero and the equation ( 4) reduces to: As indicated in equation ( 5), ∆CAD increases as the magnitude of ∆abs increases for LUC cells (figure 5(a)).For a given ∆abs, high-demand LUCs that are primarily the areas with high population density or industrialization experience smaller change in CAD or blue water scarcity.Conversely, ∆CAD is generally higher for urban areas which experience a higher reduction in water abstraction (figure S7).LUCs with higher ∆abs experience a reduction in CAD value after the transition.For example, Houston, TX receives water from Lake Livingston on the Trinity River, and Lake Houston and Lake Conroe on the San Jacinto River, for its daily domestic and industrial needs and does not face water scarcity in S1.After irrigation expansion in scenario S2, predicted water availability in current surface water sources reduces and the existing water transport infrastructure is unable to meet the city water demands.Thus, the number of water-scarce Phoenix-Mesa and Tucson experience blue water scarcity for 6 and 4 months, respectively, in both scenarios S1 and S2, as relatively small increase in water withdrawal from RFtoIF transition can be supplemented by water supply from Central Arizona Project reservoirs (Tucson and Phoenix-Mesa) and Salt Lake Project (Phoenix-Mesa).In a few circumstances, number of months experiencing changes in scarcity may be zero because they may already be scarce in all months in scenario S1.For example, New York-Newark, Oklahoma City, and San Antonio, already experience a full year of water scarcity, indicating that no further months are added in those cities.It is to be noted that for several LUCs, such as Milwaukee, Kansas City, Chicago, and St. Louis, while RFtoIF transition in the upstream contribution area decreased river flow causing negative change in abstraction from it (figure 6), local water sources in the neighborhood that supply water through canals are able to cushion this reduction (as indicated by positive change in abstraction from canals).In contrast, LUCs that experience increase in number of months of scarcity generally experience negative change in abstraction from both rivers and the canals.This highlights that the impact of RFtoIF transition on scarcity can be mediated by ancillary water sources that are not directly or significantly impacted by RFtoIF transition.
For ROA cells, in addition to ∆abs and D1, the spatial distribution of ∆CAD is controlled by additional variables including ∆dem and A1 (see equation ( 4)).Notably, among the ROAs, most transitioned locations have positive ∆abs, while the non-transitioned cells either have zero or negative ∆abs (figure 5(b)).This suggests that transitioned cells withdraw more water to match the demand after RFtoIF transition, while cells that do not participate in transition withdraw less or the same amount of water depending on the extent of reduction in water availability at the location.∆CAD for nontransitioned ROA cells behaves like that of LUC cells, with its absolute value increasing with an increase in ∆abs.In contrast, at the transitioned ROA locations, the absolute value of ∆CAD decreases with an increase in ∆abs magnitude for a given ∆dem.In other words, if the increase in abstraction does not match the increase in water demand, transitioned locations experience higher ∆CAD.To conclude, figure 5 shows the characteristics of the urban and rural cells based on their change in demand, abstraction and CAD after RFtoIF transition.It captures the intricate interplay of water demand and availability factors that shape water scarcity under land use change.The analysis identifies cells prone to heightened scarcity from irrigation expansion, aiding targeted sustainable management.

Discussion and synthesis
It is well known that transitioning from rainfed to irrigation-fed agriculture boosts crop yields and improves food security.Our continental hydrologic simulation, however, shows that RFtoIF transition over croplands that experience green water scarcity for an average of at least one month a year intensifies freshwater scarcity in both transitioned and nontransitioned areas.Notably, increase in scarcity in urban areas is not dominated by those situated in arid settings.Instead, it depends on a multitude of factors ranging from increase in water abstraction in the upstream, magnitude of water demand in the urban region, and buffer amount available in nearby water withdrawal points.Our simulation results show that among just the 53 considered LUCs, around 16 million additional urban residents will get affected by such a transition.This may increase the risk of water conflict between urban and the surrounding upstream rural water users, as it is being realized in many water stressed situations throughout the world [55][56][57][58][59]. Notably, large urban areas with improved water infrastructure are better equipped to meet the increased water demand resulting from additional water usage for irrigation, mitigating the impact of water scarcity in these regions.The analysis was conducted assuming all rainfed areas facing green water scarcity for at least one month are converted to irrigation fed, which while being an unlikely scenario in terms of its implementation, helps highlight the degree of impact that may be incurred.Additional sensitivity analysis to assess the impact of GWS threshold and number of months on the total green water scarce area shows that the total area facing green water scarcity does not change much for low GWS threshold values (GWS <= 0.3) or the number of minimum number of months (<= 3) considered (figure S8).Ideally, identification of GWS threshold or related drivers that trigger RFtoIF transition or determine the extent of RFtoIF intensification would allow for a more realistic RFtoIF transition scenario.
However, such a determination remains challenging due to the influence of a range of factors including economic conditions of farmers, incentives provided by local/federal government, penetration of agricultural extension education among the farming community, farmers' social networks, water availability, political considerations, water use by other sectors, water infrastructure, and more [43,[60][61][62][63][64].The study does not account for water consumption by poultry and livestock in agricultural sectors.The impact of RFtoIF transition on the increase in the number of livestock and consequently water use by them [65] is also not considered here.A model facilitating livestock and other farm water consumption may be used to assess the overall impact.However, the water use by livestock is minimal as compared to other sectors (less than 1% of total freshwater withdrawals in 2000) [66].
The water scarcity evaluations performed here are based on the historical datasets, i.e. crop area distribution and irrigated area maps circa 2000 and 2005, respectively.Given that new and better data is continuously being generated, the reported population facing blue water scarcity in the status quo and transition scenario are expected to change with their usage.It is also to be noted that by the year the full RFtoIF transition (as simulated in S2) may get realized, if it ever does, the climate is likely to be different.However, given the uncertainty in timeline of this transition, the current study does not consider the concomitant impacts of changes in climate on evapotranspiration, precipitation, water availability, and water demand [67,68].During the transition, other socioeconomic changes such as urban and rural demographics, water infrastructure technology, economic changes, changes in water withdrawal efficiency for all three sectors, cropping patterns, agricultural management practices, land cover change, etc. are subject to change and may affect the water scarcity in an area as well.These factors are not explicitly considered in the scenario simulations performed here.We further acknowledge that the RFtoIF transition may disturb local hydrological cycle and affect the precipitation, evapotranspiration, surface temperature, and other land atmospheric interactions [69][70][71].These factors can affect the water scarcity estimates as well.
It is to be noted that just as is the case with most model implementations, the H08 model results, which have been used to obtain the scenario simulations in this study, suffer from model structure and data uncertainty.For example, as the model does not consider lateral groundwater flow between cells, it may have impacted the estimates of the spatial distribution of CAD and ∆CAD as groundwater withdrawals may directly impact the surface water resources [72,73].Uncertainties also exist in terms of accounting for all the possible surface water supplies.An effort has been made in this study to reduce this uncertainty by including urban water withdrawal points [39].A more accurate dataset on distant water supply may improve the results.Notably, only surface urban withdrawal points were considered, due to their ability to be incorporated as canal origin points in the current version of H08.This may result in an overestimation of blue water scarcity in cities that rely primarily on groundwater or desalinization for their municipal and industrial water demand.Another source of uncertainty may arise due to the current representation of urban water withdrawal points in H08.Specifically, while all LUC grids are able to abstract water from urban withdrawal points, the grids that come first in the pre-defined sequence are prioritized for withdrawal.This could potentially impact water scarcity at the grid scale, although the overall effect on the LUC level is likely to be minimal as the abstraction from all grids is aggregated to calculate the total water abstracted by the LUCs.Furthermore, the study assumes that the domestic sector is the first to extract water, followed by the industrial and then the agricultural sectors.Assuming agriculture has the lowest priority in water abstraction is a pragmatic choice given the lack of information on which regions prioritize which sectors.However, the results of water scarcity can be sensitive to this assumption.According to Flörke et al [38], climate change affects the surface water deficit in urban areas in significantly different ways depending on the water extraction priority assigned to the urban population.Here, we performed an additional analysis by assigning agriculture as the first priority and the domestic sector as the last for water abstraction.The altered hierarchy in water use resulted in increase in water scarcity in 7 LUCs in S2 (previously, the increase was noted in 6 LUCs).Notably, the average change in CAD in urban areas after RFtoIF transition for this new water abstraction priority configuration rose by around 30%, indicating more urban grids with water scarcity.These results highlight that establishing appropriate water withdrawal priorities in river basins can help mitigate RFtoIF transition's impact.
In this study, areas facing water scarcity are derived using the CAD index.The index is similar to other commonly employed water scarcity indices such as the criticality ratio [74], Falkenmark index [29], and water footprint based index [75][76][77].Previous studies [24,45] have shown that estimates of population experiencing scarcity are only mildly sensitive to the choice of water scarcity metrics (table S1(a)), and there is a close correspondence between these indices.This is unsurprising as majority of these metrics use two similar primary variables, namely the abstracted water amount or the water available for abstraction, and demand or water footprint (table S1(b)).Differences between metrics may arise from the inclusion of additional variables, such as accounting of water-use losses in water abstraction term or the numerator of CAD.The magnitude of these additional variables can lead to variations in the estimates of water scarcity across metrics.It is worth noting that each metric often applies subjective thresholds to classify the severity of scarcity, which can also contribute to disparities in results between metrics.
Apart from these, uncertainties may also occur due to the use of spatially uniform irrigation and, domestic and industrial water use efficiencies, which are set to 0.6, 0.15, and 0.1, respectively, over all the cells.Furthermore, the results are based on a temporally static distribution of irrigation area, croplandpasture fraction, and areas of different crops over the simulation period.The industrial and domestic water withdrawal used in this study is obtained from AQUASTAT and downscaled to the modeling scale of 5 × 5 arcmin based on the population distribution.Notably, the temporal variation of domestic and industrial water withdrawal is also not taken into consideration.This issue can be attenuated if the model is supplied with more accurate data on water withdrawal by domestic and industrial sectors.CAD and ∆CAD estimates are likely to be also affected by uncertainties in EFR, which here is defined based on the monthly average river discharge [52].Studies [78] have previously reported that the EFR estimation method may determine water scarcity assessment, although Mekonnen and Hoekstra [28] also noted that the population living under moderate blue water scarcity does not change significantly for the uncertainty range of EFR.Another source of uncertainty is from the current AET and PET parameterizations, which do not account for crop-specific stomatal conductances.Given that these conductances may vary with crops and cultivars [79], uncertainties in PET and ET estimates can be reduced by performing calibration and validation against remotelysensed evapotranspiration estimates [80].
Despite the aforementioned methodological limitations, the analysis clearly shows that the controls on changes in blue water scarcity (or ∆CAD) are different between non-transitioned and transitioned locations.The trend of changes in ∆CAD vis-à-vis changes in abstraction is also contrasting between non-transitioned and transitioned locations.In addition, the effects of RFtoIF transition on urban areas, especially in regards to additional months being affected by scarcity is dependent both on antecedent scarcity state before transition and presence of ancillary water supply sources to cities either from reservoirs or locations that are not directly impacted by RFtoIF transition.Overall, the results indicate that upstream of urban areas with low irrigation expansion (e.g.LUCs like Los Angeles-Long Beach-Anaheim, CA; Las Vegas-Henderson, NV; Birmingham, AL; Atlanta, GA; Salt Lake City-West Valley City, UT; Hartford, CT; and Raleigh, NC) or those equipped with more robust water infrastructure with ample buffer supply sources (e.g., cities like St. Louis, MO-IL; Cincinnati, OH-KY-IN; Phoenix-Mesa, AZ; etc.), may be more suitable for undergoing a transition to irrigated agriculture (figure S9).It is to be noted that just because some LUCs experience a large change in ∆CAD doesn't necessarily imply that transitions in the neighboring ROA are unlikely.Factors such as political support, economic soundness of LUCs, the significance of water for both LUC and ROA users, impacts on water quality, and other considerations may ultimately determine the feasibility and support for RFtoIF transition.Future studies may perform an integrated wholistic assessment of all the drivers, stresses, impacts and responses at each LUCs to help identify cities that can best adapt to unsustainable irrigation expansion.
Overall, the study indicates that the irrigation expansion, if not properly managed, is unsustainable.Furthermore, irrigation expansion can enhance water scarcity in large urban areas and could be a conflict agent between urban and rural water users.Given the existing significant divide in the urban-rural electorate in US [81], these conflicts are likely to get aggravated and spur social and administrative challenges regarding water allocation and access.Alterations in water resources due to rapid urbanization, and socioeconomic and climate change, are likely to further pose challenges for water managers [82][83][84][85][86][87].Since the impacts of RFtoIF transition propagate downstream, constraining urban-rural conflicts [55][56][57][58][59] may require update and/or formulation of innovative basin-scale water apportioning doctrines and compacts.Alternative solutions to mitigate these negative impacts include changing cropping pattern and practices [17,88,89], and enhancing the sustainability of virtual water trade [90][91][92].

Figure 1 .
Figure 1.Spatial distribution of the difference in CAD (∆CAD) after and before the RFtoIF transition.Negative values indicate a decrease in CAD (or an increase in blue water scarcity) with RFtoIF transition.

Figure 2 .
Figure 2. Cumulative distribution function of CAD values for ROA (a), and LUC cells (b) for scenarios S1 (blue) and S2 (red).The numbers indicate the percent of cells belonging to different blue water scarcity classes i.e. low (0.8 < CAD 0.99), moderate (0.5 < CAD 0.8), high (0.3 < CAD 0.5), or severe (CAD 0.3).Fraction of the cells with low or no water scarcity reduces after the RFtoIF transition in both LUCs and ROA.The reduction in fraction of ROC cells is relatively larger.

Figure 3 .
Figure 3. Blue water scarcity for 53 LUCs for scenarios S1 (a) and S2 (b).The size of the circle represents the population, and the color represents the number of months a LUC faces water scarcity.

Figure 4 .
Figure 4. (a) CAD for ROA cells as a function of their lateral flow input and crop area percentage.Cells with high crop area and low lateral flow have lower CAD values.(b) CAD for LUC cells vis-à-vis lateral flow and population.(c) and (d) are same as (a) and (b), respectively, but zoomed-in for low lateral flow values.The figure illustrates the interplay of factors controlling water demand (e.g.crop area fraction and population) and supply (lateral flow input) in agriculture and urban settings.

Figure 5 .
Figure 5. (a) Scatter plot between annual average ∆CAD and ∆abs for LUC cells.Cells with higher demand (shown in million cubic meter per year (MCM/year)) have smaller ∆CAD for the same ∆abs, (b) annual average ∆abs and ∆dem scatter plot for ROA cells.Some of the non-transitioned cells show change in CAD even after no change in demand in scenario S2.For transitioned grids, change in CAD is higher for the cells with less increase in abstraction i.e. low ∆abs, while cells with high ∆abs have less change in CAD.

Figure 6 .
Figure 6.53 LUCs considered in this study and their average annual water demand and ∆abs.The ∆abs is specified separately based on the source of water abstraction i.e. river and canal.Water abstraction from water withdrawal points is included in canal water abstraction.Column on the right indicates number of months of water scarcity in scenarios S1 and S2 for each considered LUC.