Environmental flow requirements largely reshape global surface water scarcity assessment

The inclusion of environmental flow requirements (EFRs) in global water scarcity assessments is essential to obtain a reasonable representation of the water scarcity status. However, at a global scale, the quantification of EFRs is subject to large uncertainties resulting from various methods. So far, it is unclear to what extent the uncertainties in EFRs affect global water scarcity assessments. In this study, we examined the differences between EFR estimation methods and quantified their effects on spatially explicit water scarcity assessments, based on reconstructed global water withdrawal data and naturalized streamflow simulations. The global mean EFRs estimated by different methods ranged from 129 m3 s−1 to 572 m3 s−1. Consequently, with the fulfillment of the EFRs, the area under water scarcity ranged between 8% and 52% of the total global land area, and the affected population ranged between 28% and 60% of the total population. In India and Northern China, 44%–66% and 22%–58% of the country’s land area, respectively, is affected by water scarcity; this percentage is higher than that found in other countries. The effects of different EFRs on water scarcity assessment are large in many regions, but relatively small in regions that experience intensive water use due to anthropological activities (such as Northern China and India). Through this study, we have put forth the need for the reconciliation of the estimates of EFRs to produce more reasonable and consistent water scarcity assessments.


Introduction
Water scarcity often indicates a shortage of regional water resources and may pose significant threats to food security and the human living environment. Water scarcity has increased over the past decades (Wada et al 2011 and is projected to increase further in the future due to climate change and increasing water demands (Hanasaki et al 2013, Schewe et al 2014. In addition to human water demands, environmental flow requirements (EFRs) maintaining the health of river ecosystem is also a key component in water scarcity assessment (Smakhtin et al 2004). However, EFRs often conflict with anthropological water use (Richter 2010, Pastor et al 2019, and the fulfillment of EFRs may impose large restrictions on the human appropriation of water resources from rivers , Hanasaki et al 2013, Wada et al 2016. EFRs are necessary for the calculation of the Sustainable Development Goal 6.4.2 indicator of water stress (FAO 2019). Notably, EFRs have been considered in recent water scarcity assessments (Liu et al 2016a(Liu et al , 2019; however, varying river regime conditions make the estimation of EFRs both complex and difficult and, therefore, there is a need for further studies, particularly for large-scale assessments (Poff et al 2010, Pastor et al 2014.
In previous studies, ecologically and hydrologically based approaches have been applied to estimate the EFRs on different scales (Richter et al 1997, Arthington et al 2006, Acuña et al 2020. In large-scale water scarcity assessments, the methods used to estimate EFRs are mostly hydrologically based (Pastor et al 2014), mostly relying on the streamflow/runoff (owing to the lack of information on ecological habitat conditions). Some hydrologically based approaches (Tennant 1976, Tessmann 1980 consider the ecological health status of rivers to some extent. However, for simplicity, the global EFRs estimated using these methods often assumed a 'fair to good' ecological status (Jägermeyr et al 2017, Pastor et al 2019.
In recent literature, EFRs are considered to be an explicit component in water scarcity assessment (Smakhtin et al 2004, Wada et al 2011, Liu et al 2019. This is because the appropriation of streamflow to meet EFRs is necessary to maintain the health of the river ecosystem, which is important for sustainable economic and social development. Therefore, in water scarcity assessment, EFRs should be explicitly represented as an indispensable water user with a high priority, instead of assigning them leftover water resources (after human water use). The EFRs estimated by various methods have been used in water scarcity assessments at river basin and global scales (Pastor et al 2014, Theodoropoulos et al 2018. The wide range of EFRs used in previous studies has led to large uncertainties and inconsistencies in global water scarcity assessments with respect to their magnitude and spatial distributions. The extent to which different EFR estimates can affect water scarcity assessment is unclear to date. Consequently, it is difficult to determine the affected areas and the number of people suffering from water scarcity and develop appropriate policies to address water challenges.
This study explicitly investigates the uncertainties in surface water scarcity assessments stemming from EFR methods, on a global, regional, and basin scale. The spatial characteristics of the EFRs related to hydroclimatological features were elaborated, and the area and population affected by surface water scarcity were quantified to demonstrate the effects of different estimates of EFRs on the assessments. This study also contributes to improving the understanding of the heterogeneity of the estimates of EFRs on a global scale and the underlying hydrological mechanisms linked to hydroclimate conditions.

Environmental flow requirements (EFRs) estimation
Seven hydrologically based methods (table S1 (available online at stacks.iop.org/ERL/16/104029/ mmedia)), namely, the flow duration curve (FDC) method (Q90 and Q50), Tennant method (1976), Tessmann method (1980, variable monthly flow (VMF) method (Pastor et al 2014), Q90/Q50 method (Pastor et al 2014), and Smakhtin method (2004) were used to estimate EFRs (Liu 2021). Most of these methods are commonly used to determine global EFRs (Jägermeyr et al 2017, Pastor et al 2019. Two FDC percentages, Q90 (Gopal 2013) and Q50 (median) (USFWS 1981) which are key elements of other methods (e.g. Q90/Q50, Smakhtin) and straightforwardly comparable to others, were considered in this study. Q50 was not used as commonly as Q90 but has been used in recent studies (Belmar et al 2011, El-Jabi andCaissie 2019). The Tennant method determines EFRs based on the mean annual flow (MAF). The Tessmann method is a modification of the Tennant method using different proportions of the MAF and mean monthly flow (MMF), considering an intermediate season (Gerten et al 2013, Pastor et al 2014, Jägermeyr et al 2017, Grantham et al 2019, Mekonnen and Hoekstra 2020. The VMF method follows the natural variability of river flow and estimates the EFRs monthly in global assessments. The Smakhtin method considers biophysical and environmental flow conditions, defines low flow requirements as Q90, and estimates high-flow requirements based on MAF and Q90. The Q90/Q50 method estimates EFRs as Q90 streamflow for lowflow seasons and Q50 for high-flow seasons (Pastor et al 2014). Notably, in the initial application of the Smakhtin method, regulated streamflow is used to calculate the EFRs (Smakhtin et al 2004), which is different from other methods. To be consistent in this study, we used the naturalized streamflow for the EFR estimation carried out using all the above methods. EFRs of 10%, 20%, …, 90% of MAF and Q80, Q70, and Q60 are also applied uniformly worldwide in the water scarcity assessment for sensitivity tests.
Inadequate ecological and management data have made it infeasible to consider the ecological status of individual rivers on a global scale. To ensure the comparability of these methods, a fair ecological condition was assumed for the Tennant, Tessmann, VMF, and Smakhtin methods. The low-flow and high-flow seasons were defined according to the MMF and MAF as per Pastor et al (2014); then, all EFR methods were applied on a monthly basis. By definition, the EFRs determined from the FDC based on all the records over the duration of 120 months were found to be constant.

Water scarcity index (WSI)
The water scarcity index (WSI), defined as the ratio of water withdrawal to water availability, is the most widely used index for global water scarcity assessment (Wada et al 2011, Hanasaki et al 2013, Liu et al 2019. Water withdrawal is the sum of withdrawals of all sectors (including agricultural, industrial, domestic, and livestock sectors). However, human beings cannot use all the available water resources because EFRs must be met to maintain river ecosystem health and for sustainable social and economic development. For this reason, EFRs have been included as a key component in the indicator for assessing the sustainable development goal (SDG) 6.4.2 on water scarcity (FAO 2019). According to the Food and Agriculture Organization (FAO) guidelines, only water in excess of EFRs can be potentially utilized for or allocated to human activities. Hence, water scarcity is defined as the water withdrawal of all sectors divided by the difference between the total renewable freshwater resources and EFRs; water availability is determined as the simulated natural streamflow with the EFRs deducted (Liu et al 2019). For comparison, we also calculated the traditional WSI without deducting the EFRs from the available water.
Water scarcity is defined as WSI > 1 for the assessments that explicitly consider EFRs, indicating that the available water is insufficient to meet water demands. In previous studies, the threshold for water scarcity was conventionally set as 0.4% or 40% (Alcamo et al 2003, Oki 2006, Wada et al 2011, Hanasaki et al 2018, irrespective of whether EFRs are included. However, there is confusion regarding the threshold because the 0.4 threshold without considering EFRs indicates that 40% of the water resources/streamflow can be appropriated for human water use. This implies that 60% of the total water resources is allocated for environmental flow and is not available for any other application. So far, there has been no elaboration for only allowing 40% of water resources for human users and maintaining 60% in the rivers (potentially for environmental use). In our study, EFRs were estimated explicitly, and the quantity was deducted from the water availability. Thus, a WSI greater than 1 indicates that water resources cannot meet the water demands of humans and the natural environment simultaneously. In this case, the conceptual meaning of the threshold of 1 is clear, although the desired threshold can be closely related to societal choices (FAO 2019). However, a threshold of 0.4 is still used for the WSI calculated without the explicit consideration of EFRs. Thus, the water scarcity assessment using threshold 1 can be compared with the traditional water scarcity assessment applied without the explicit consideration of EFRs.
In this study, the water scarcity assessment is presented as multi-year means, based on the monthly WSI at grid cells over the 2001-2010 period. The WSI values were averaged over 10 years for each month. The area under water scarcity (in percentage; A ws ) is the sum of areas of the grid cells with WSI > 1 (WSI > 0.4 for those without EFR) in specific regions/basins, portrayed as the percentage of total land areas for each month, and is then, averaged over the study period. Like A ws , the population affected by water scarcity (P ws ) is the sum of the population within the grid cells with WSI > 1 (WSI > 0.4 for those without EFR) in specific regions/basins, portrayed as the percentage of the total regional population. Greenland and the grid cells having an annual mean discharge of <5 m 3 s −1 were excluded from the calculation of A ws, P ws, and the following analysis.

Streamflow and water withdrawal data
Water availability was determined from the multimodel simulations of the monthly natural streamflow of the NOSOC experiment, that is, without the consideration of human impact (in pristine condition), at a spatial resolution of half-degree.  (Döll and Lehner 2002) was applied to the six GHMs to obtain the river flow. The uncertainty in the streamflow simulations has been documented in previous literature (Liu et al 2017, Zhao et al 2017 and is not the scope of this study. Thus, monthly streamflow medians were computed from the 18 combinations of individual GHMs and individual forcings at each grid cell to determine the water availability. For water withdrawal, we used the global water withdrawal data with a spatial resolution of halfdegree which was reconstructed based on the data reported by the FAO and the GHM simulations during 1971-2010 (Huang et al 2018). In this dataset, the water withdrawal from the FAO AQUASTAT database was used to adjust the global gridded of water withdrawals simulated by four global hydrological models (H08, LPJmL, PCR-GLOBWB, and WaterGAP2) from the ISIMIP2a dataset.

Regional analysis
Analysis was also performed based on river basins and hot spot regions to examine the impacts of EFRs on water scarcity over regions having different hydroclimatic regimes. Hot spot regions were identified according to water scarcity conditions and the intensity of water withdrawal (figure S1). A ws and P ws were calculated for each hot spot region and basin and averaged over the period. The mean aridity index (AI, the ratio of total precipitation to potential evapotranspiration) (Trabucco and Zomer 2019) was calculated to represent the climatic conditions in individual river basins. The map of global major basins (figure 1(A)) was obtained from the FAO GeoNetwork (https://data.apps.fao.org/map/catalog/srv/ eng/catalog.search?id=30914%26;currTab=simple#/ metadata/7707086d-af3c-41cc-8aa5-323d8609b2d1, accessed on 2021-7-10).

Differences in the estimated environmental flow requirements (EFRs)
The estimated mean monthly EFRs (figure 1(A)) generally show patterns similar to those of streamflow (figure S1), that is, EFRs are high in wet regions and low in arid regions. Specifically, EFRs are generally large in the Amazon River, Congo River, Yangtze River, and several rivers in tropical regions and high latitudes. In our study, the minimum (figure 1(B)) and maximum (figure 1(C)) EFRs showed similar spatial patterns to the mean EFRs. The coefficient of variation (CV) indicates the relative differences between these EFRs across the world, which is larger than 0.5 in more than 70% of the global land areas and ∼1 in most arid regions ( figure 1(D)). The mean monthly EFRs showed significant differences among the EFR methods in many regions (see figures S2 and S3 in the supplementary information (SI) for the EFRs for high, low, and intermediate flow).
The estimated global mean EFRs by different methods were quite distinct from each other in terms of their mean values and seasonal cycles (figure 2(A)). The Q50 and Q90 EFRs were constant in all months because they were determined from the FDC estimated for the entire period. The EFRs estimated using the Q90/Q50, Tessmann, VMF, and Tennant methods showed similar seasonal patterns, with larger values from March to July; the global mean EFRs estimated using the Smakhtin method were larger from May to November. Global mean EFRs calculated using Q50 were the largest (428 m 3 s −1 ), followed by those calculated using Q90/Q50 (308 m 3 s −1 ). The Tennant method provided the smallest global mean EFRs (96 m 3 s −1 ), followed by the VMF EFR (174 m 3 s −1 ). The Tessmann, Smakhtin, and Q90 methods had relatively similar global mean EFRs of 243 m 3 s −1 , 221 m 3 s −1 , and 213 m 3 s −1 , respectively.
The basin EFRs estimated using Q50 were always the largest, while the smallest basin EFRs are often obtained using the Smakhtin, Q90, and Tennant methods. The EFRs estimated using the VMF method often lay between those estimated using all the methods. The basin EFRs generally increased with the AI (figure 2(B)), that is, the EFRs of wetter basins were larger than those of the drier ones. This trend, estimated from the EFRs obtained using all the methods, is more significant for dry basins (AI < 0.65) than wet basins (AI ⩾ 0.65).

Water scarcity assessments with and without the consideration of environmental flow requirements (EFRs)
The mean monthly WSIs with and without the consideration of EFRs over the 2001-2010 period are shown in figure 3. Water scarcity indicated by the WSI without EFRs was mostly found in Central Asia, South Asia, East Asia, Western North America, and North Africa ( figure 3(A)). In our study, the extent of water scarcity estimated was generally consistent with previous studies (Oki 2006, Wada et al 2011, except for larger areas found in North Africa and Central Asia. Compared to the traditional WSI, a larger extent of water scarcity was indicated by the Q50 and Tessmann WSIs, while less water scarcity was indicated by other WSIs with EFRs. Notably, WSIs with EFRs showed distinct spatial patterns across different EFR methods. The WSI with Q50 EFRs appears to be the largest, showing WSI values >1 in more than half of global land areas covering both humid and arid regions (figure 3(C)), and most WSI values range between 0.8 and 1.2 (see the inner plot). Although the EFRs estimated by the Tessmann method showed spatial patterns similar to those of Q50, the WSI with Tessmann EFRs shows a smaller extent of water scarcity compared to the WSI with Q50 EFRs. The WSI values with Tessmann EFRs showed a wider histogram than Q50, and high values were largely found in arid regions (figure 3(F)). WSIs calculated using Q90 ( figure 3(B)), Smakhtin (figure 3(F)), and Q90/Q50 (figure 3(I)) EFRs were similar in terms of their spatial patterns and relative frequencies. The relative frequency of the WSI using Tennant EFRs was different from the WSI obtained using the EFRs by Q90, Smakhtin, and Q90/Q50; specifically, the Tennant WSI showed more values less than 0.2 and more values greater than 1 ( figure 3(D)). The WSIs estimated using VMF EFRs (figure 3(G)) showed similar spatial patterns and relative frequency distributions compared to the WSIs estimated without EFRs ( figure 3(A)).
The A ws showed evident zonal patterns, indicated by large variations across latitudes, for all the WSI estimations ( figure 4(A)). The A ws was generally the largest between latitudes 10 • N and 45 • N, home to a large human population, followed by the zones between 40 • S and 20 • S. Relatively large percentages were also found at high latitudes, for example, between 60 • S and 40 • S and between 70 • N and 80 • N, for the WSIs estimated using Tessmann and Q50 EFRs.
The seasonality of global A ws showed distinct differences across the WSIs ( figure 4(B)). The A ws showed little variation over months for the WSIs calculated using VMF and Tennant EFRs and the WSI calculated without EFRs. The WSIs calculated using Q50 and Tessmann EFRs indicated larger A ws values from October to January (compared to other months). The WSIs calculated using Q90, Smakhtin, and Q90/Q50 EFRs showed almost the same A ws values, with relatively large A ws from January to March and August to September. The WSI calculated without EFR estimated the global A ws as 16% on average. The A ws indicated by the WSI calculated using Q50 EFRs was the largest (54%), followed that calculated using Tessmann EFRs (45%). The WSIs calculated using Q90, Q90/Q50, Tennant, and Smakhtin EFRs indicated A ws values of 21%-23%, and the WSI calculated using VMF EFRs indicated the same global A ws (16%) as that calculated without EFRs.
The seasonality of A ws showed regional differences, as well as differences between the EFR methods, as demonstrated by the six hot spot regions ( figure 4(B)). In America and Europe (in North Africa, Arabia, and Australia), the A ws values indicated by the WSIs calculated using Q50 EFRs were larger (smaller) than those calculated using Tessmann EFRs; however, the A ws values for Northern China and India were similar. The A ws calculated using Q90 EFRs was slightly lower than that calculated using Q90/Q50 EFRs, and both were very close to the A ws values calculated using Smakhtin EFRs. The A ws calculated using Tennant EFRs was small in Europe (9%), while the A ws calculated using VMF EFRs was relatively small in all regions. A w calculated without EFR was often large in arid/semiarid regions, such as North Africa, Arabia, and Northern China.
The WSIs without EFRs provided P ws estimations of 34% of the global population, while the WSIs with EFRs indicated that 30%-60% of the global population is experiencing water scarcity, on average ( figure 4(C)). The WSIs with Q50 (61%) and Tessmann (50%) EFRs indicated the largest population experiencing water scarcity, while the WSIs with Q90, Tennant, Smakhtin, VMF, and Q90/Q50 indicated 35%, 30%, 35%, 31%, and 36% of the global population experiencing water scarcity, respectively. Notably, on a global scale, less population experienced water scarcity in boreal summer (July-September) than other seasons.
The impact of the different EFRs on water scarcity assessment in terms of the affected area (A ws ) and affected population (P ws ) at the basin level (figure 5(A)) showed similar differences to those at the global and regional levels. The A ws and P ws were large (40%-60%) in most basins, indicated by the WSIs calculated using Q50 and Tessmann EFRs, and relatively small for those calculated using Q90, Smakhtin, Tennant, and VMF EFRs. The A ws and P ws were relatively larger in arid basins (AI < 0.65) and wet basins (AI ⩾ 0.65). The relationships between the ∆A ws (changes in A ws compared to those without EFRs) and AI are distinct for the FDC-based methods (e.g. Q90) and MAF/MMF-based methods (e.g. Tennant and Tessmann) at the basin level ( figure 5(B)). The ∆A ws increased with AI when the Q90, Q50, Smakhtin, and Q90/Q50 methods were used and decreased with the AI when the Tennant and Tessmann methods were used (see also table S2). Water scarcity in terms of affected areas and population increased with EFRs, with respect to the proportions of the MAF (figure 5(C)) and quantiles ( figure 5(D)). In particular, the A ws and P ws values estimated by individual FDC-based methods showed less differences between basins than individual MAF-based methods, and they were more distinct between the quantiles than the MAF, even in wet basins.

Uncertainties in the estimates of environmental flow requirements (EFRs)
The EFRs are subject to considerable uncertainty, as indicated by the large differences in the results obtained using the different methods. The differences between the EFRs in terms of CV showed distinct spatial and seasonal patterns; in particular, they were larger in dry regions (compared to wet regions) (figure 1). This pattern resembled the temporal variation in the streamflow (figure S1(B)), that is, a high CV of EFRs was found in regions having highly variable monthly streamflow. The CV of the EFRs was most likely associated with streamflow variation, which is a key factor in some EFR methods.
The single ecological status ('fair' condition) across the world may also bring uncertainty into the EFR estimates in regions that expect other ecological conservation goals (Poff et al 2010). However, given the difficulties in determining the ecological conditions of rivers worldwide, using a uniform ecological condition across the world would be a practical choice for global assessment. The ecological management classes may be a promising framework for estimating the EFRs that relate to the current or desired condition of a river and are perceived as scenarios of the environmental state of rivers (Sood et al 2017, FAO 2019. It considers the ecological condition and management perspectives for both natural (unmodified) and human-disturbed river ecosystems. The EFR estimates could be more reasonable by setting different classes for regulated and unregulated rivers. Meanwhile, increasing global ecological data (Sayre et al 2014, Ghiggi et al 2019, Jeliazkov et al 2020 will favor global water scarcity assessments with consideration of ecological conditions. According to the differences in EFRs (figure 2) and their definitions, the EFR methods can be roughly grouped into two types. One type is MAF/MMFbased, for example, Tennant, Tessmann, and VMF. These methods show different levels of EFRs in terms of the absolute values and the proportions of MAF (figures 1 and S5), and the proportions of MAF show little spatial heterogeneity ( figure S4). The other group is FDC-based, that is, the Q50, Q90, and Q90/Q50 methods. The EFRs estimated by these methods often largely differ from each other because of the different percentiles used and show large spatial variability in terms of the proportion of MAF ( figure  S5). The Smakhtin method uses the information of both Q90 and MAF, but has similar patterns as the Q90 method. We noticed that, for the FDC-based methods (i.e. Q90/Q50 and Smakhtin methods), the ratios of EFRs to MAF calculated in this study were different from the results of Pastor et al (2014). This might be because we used the medians of multimodel simulations of natural streamflow, while Pastor et al (2014) used the LPJmL simulation only.

Effects of uncertainties in environmental flow requirements (EFRs) on water scarcity assessment
The large discrepancy between the EFR methods is the major source of uncertainty in the WSI estimates. The uncertainty in terms of the standard deviation (STD) across the WSI estimates with different EFRs ( figure S6(A)) is larger than the STD across the WSIs estimated from the streamflow simulation of individual combinations of GHMs and global meteorological forcing ( figure S6(B)). The uncertainty arising from EFRs dominates the uncertainty in the WSI estimates in most areas of the world ( figure S6(C)).
The area and population affected by water scarcity showed small differences between the WSIs that used the FDC-based EFR methods, such as Q90, Q90/Q50, and Smakhtin. These methods produced different EFRs (figures 2 and S5), but the area and population experiencing water scarcity showed similar spatial patterns (figure 4). In contrast, the differences between the WSIs with MAF (or MMF)-based EFRs were more significant. The ∆A ws estimated using FDC-based and MAF-based methods showed distinct responses to climate conditions at the basin level ( figure 5). This may be because the EFRs estimated by the MAF-based methods accounted for spatially uniform proportions of MAF across regions, irrespective of their climate conditions, while the FDCbased EFRs accounted for different proportions of MAF worldwide (figures S4 and S5). The sensitivity test (figures 5(C) and (D)) showed that water scarcity conditions were more different between FDC-based percentiles than between the proportions of MAF. Compared to each other, the FDC-based methods estimated larger EFRs in wet basins and smaller EFRs in dry basins, whereas the MAF/MMF-based methods estimated smaller EFRs in wet basins and larger EFRs in dry basins.

Implications for the choice of environmental flow requirements (EFRs) in global water scarcity assessment
EFR is a key component of the indicator for assessing the SDGs on water scarcity (FAO 2019). The large uncertainty in WSIs arising from different EFRs suggests an urgent need to develop a more versatile method or to reconcile the estimates of EFRs to derive a consistent water scarcity assessment across regions and river basins having varied hydrological regimes worldwide. Currently, hydrological methods or desktop approaches using global datasets, such as the methods used in this study, are the most appropriate for a global assessment (FAO 2019). However, there is often no unique EFR method having wide applicability to all hydrological regimes or habitat types (Pastor et al 2014). Furthermore, the determination of EFRs could be more complex in the real world, as they are associated with desired conservation goals, water quality, water withdrawals, socioeconomic status, and hydroclimatological conditions (Döll et al 2009, Acreman andArthington 2018). In particular, human intervention, of which the most impacts should be attributed to dams and reservoirs, can alter hydrological regimes, result in significant losses of river connectivity, impair riverine ecosystems worldwide (Grill et al 2019); however, they may also regulate streamflow to maintain the designed EFRs during different seasons. Overall, the methods explicitly addressing seasonality would have a large potential for producing appropriate EFRs because the intra-annual hydrological variation is essential for maintaining river ecosystem health (Richter et al 1996), while fixed unique percentages (e.g. Q90 or 60%) seem inappropriate for large-scale assessment (Armstrong and Nislow 2012).
The results of this study can help in the selection of EFR estimation methods by applying the rule of thumb (concerning hydrological regimes) that large EFR estimates are acceptable if they do not cause aggravated water scarcity. In this study, we assumed that larger EFRs would generally favor better conditions of the river ecosystem. However, EFR estimates are often low in arid and semiarid regions, where rivers are often ephemeral and intermittent streams. In these regions, river ecosystems face more serious challenges to maintain their normal conditions and diversity with limited water (Gopal 2003). In this case, it is imperative to address the seasonal variability of the EFRs, which can be inferred from hydrological regimes. Thus, the methods that produce fixed small EFRs (e.g. Q90, see figure 2(B)) may be a practical choice but might not reveal the potential water stress suffered in these regions. Q90/Q50 would be more appropriate than Q90 because the former estimates larger EFRs than the latter and results in a slight aggravation of the water scarcity condition. The Q50 method estimates the largest EFRs, indicating widespread water scarcity, even in humid regions ( figure 5). This implies that Q50 may not be suitable for wet regions because it may overstate the water scarcity status during the dry season. In contrast, the Tessmann method could be a good choice for most wet regions because it estimates large EFRs and does not cause significant water scarcity. This may be due to the consideration of the seasonality of natural streamflow in the Tessmann method. The similar impacts on water scarcity calculated using the Q90, Smakhtin, Tennant, and VMF methods suggest that the method that estimates the largest EFRs would be a better choice in wet regions, but the one addressing seasonality would be more appropriate in dry regions. In addition, the Tennant method (MAF-based) is often more suitable for rivers with stable flow regimes, whereas the VMF method (MMF-based) is appropriate for those with larger hydrological variability. This is in line with the findings of Pastor et al (2014). These implications would favor future investigations regarding the reconciliation of EFR methods or the design of new methods by taking into account hydrological regimes, as well as ecological conditions, on a global scale.

Limitations in the estimates of water scarcity index (WSI)
This study focuses on surface water scarcity and estimates the surface water availability from streamflow. The total water availability might be underestimated in some regions where streamflow, groundwater, desalination water, and rainwater harvesting are available. As the world's largest distributed store of freshwater, groundwater plays a critical role in sustaining ecosystems (Taylor et al 2013) and is a major source of water availability in some regions. The interactions between groundwater and climate/surface water are very complex and vary over space (Cuthbert et al 2019a); thus, the contributions of groundwater to total water availability could differ significantly across regions. In the regions where the water table is topography-controlled, groundwater is often recharged by precipitation/river flow during wet seasons and, in turn, recharges the river flow during dry seasons. The long-term dynamics of groundwater can be represented by river flow to some degree, but in these regions, the seasonality would not be well captured in the water availability based on naturalized streamflow. In rechargecontrolled regions, such as the North China Plain and the southwestern United States, groundwater recharge is more episodic and often dominated by losses from ephemeral flows or intense rainfall/flooding events, with limited interactions between groundwater and streamflow/climate (Cuthbert et al 2019a(Cuthbert et al , 2019b. In these regions, water supply significantly relies on groundwater, and thus, neglecting groundwater may underestimate water availability and result in biased WSI estimates. However, it should be noted that high groundwater abstraction and lack of recharge from precipitation/river flow would result in overexploitation or persistent groundwater depletion (Wada et al 2010), which would not be sustainable and can lead to chronic severe water scarcity. Overall, the WSI estimated without the consideration of groundwater should be treated with caution in regions having limited interactions between groundwater and streamflow. Thus, incorporating groundwater in future studies related to EFRs and water scarcity assessment is a challenging but important task.
Desalination water, rainwater harvesting, and treated wastewater reuse could also be important sources of water availability in some regions. Desalination water is often used in coastal arid regions. The total desalination water use is estimated to be approximately 1%-4% of the global total water withdrawal (Wada et al 2011, Hanasaki et al 2016, Gude 2017. Rainwater harvesting is considered to be an interim or primary water source for domestic use in arid and semiarid areas (Lo and Gould 2015), and the implementation of rainwater harvesting systems is strongly influenced by economic constraints and local regulations (Campisano et al 2017). Thus, overlooking the contribution of desalination and rainwater to water availability may introduce uncertainty to the WSI estimates in these regions. In addition, the reuse of wastewater (∼1% of global total water withdrawal) would increase water availability, particularly in Asia .
In this study, and most water scarcity assessments, surface water availability is determined by natural streamflow. It should be noted that surface water availability can be reallocated by human activities, such as reservoir regulations. Thus, the calculation of water availability without considering reservoir regulations may affect the WSI estimates. However, at present, it is difficult to obtain accurate estimates of water availability from regulated streamflow because data on reservoir operations are generally not accessible in most countries. In the future, an online assessment (similar to that conducted by Wada et al 2011), which considers improved reservoir regulations, is necessary to reduce the uncertainty in water availability quantification.

Conclusions
In this study, we examined the differences in the EFRs estimated by several commonly used methods and quantified the effects of the inclusion of EFRs on global surface water scarcity. We used the FDC-based (Q50 and Q90), Smakhtin, Tennant, Tessmann, VMF, and Q90/Q50 methods. The results suggested large discrepancies in the EFRs estimated using the different methods. We found that the inclusion of EFRs led to significantly different spatial and temporal patterns of water scarcity. The global mean EFRs ranged from 96 m 3 s −1 to 428 m 3 s −1 , and the WSIs calculated using different EFRs indicated that the area experiencing water scarcity (WSI > 1) could vary from 8% (VMF) to more than 52% (Q50) of the global land areas, while the population experiencing water scarcity could be underestimated by 28% (Tennant) to 60% (Q50) of the global total. The areas experiencing water scarcity indicated by the WSIs calculated using the MAF/MMF-based and FDC-based EFRs increased in many basins compared to those calculated without considering EFRs. Moreover, the increments of the areas calculated using the FDCbased (MAF/MMF-based) EFRs increase (decrease) with the AI of the basins. Overall, we deduced that the representation of seasonality in hydrological regimes is essential for producing appropriate EFRs in water scarcity assessment. Thus, in this study, we underscored considerable discrepancies in global surface water scarcity assessments caused by different EFR methods.
However, a comprehensive and appropriate assessment of surface water scarcity considering EFRs remains a challenge to date. The uncertainty in EFR estimates is one of the major underlying factors that limit surface water scarcity assessment and hydrological modeling. By addressing the discrepancies in EFR estimates from different methods, this study aims to improve surface water scarcity assessment and has instrumental value for water resource management concerning competing water demands between humans and the environment. We call for more attention to the reconciliation of EFRs estimated by different methods for more reasonable global water scarcity assessments.

Data availability statement
The data that support the findings of this study are openly available at the following URL/DOI: https:// doi.org/10.5281/zenodo.4746379.