Aqueduct 40 Technical Note

TECHNICAL NOTE
Aqueduct 4.0: Updated decision-relevant

global water risk indicators
Samantha Kuzma, Marc F.P. Bierkens, Shivani Lakshman, Tianyi Luo, Liz Saccoccia, Edwin H. Sutanudjaja, and Rens Van Beek
ABSTRACT
CONTENTS Water is essential to the progress of human societies. Food production, elec-
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .2 tricity generation, and manufacturing, among other things, all depend on it.
Water risk framework. . . . . . . . . . . . . . . . . . . . . 2 However, many decision-makers lack the technical expertise to fully under-
Hydrological model. . . . . . . . . . . . . . . . . . . . . . . 4 stand hydrological information.
Future projections. . . . . . . . . . . . . . . . . . . . . . . . .9 We present Aqueduct 4.0, the latest iteration of WRI’s water risk framework
Indicators. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10 designed to translate complex hydrological data into intuitive indicators of
Country and state aggregation. . . . . . . . . . . 33 water-related risk. We curated 13 water risk indicators—spanning quantity,
Grouped and overall water risk. . . . . . . . . . . 35 quality, and reputational concerns—into a comprehensive framework. Each
Limitations. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36 indicator is sourced from an open-source, peer-reviewed data provider
and then transformed to normalized risk score based on the severity of the
Appendix A. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
water challenge. For 5 of the 13 indicators, we used a global hydrological
Appendix B. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
model called PCR-GLOBWB 2 to generate novel datasets on sub-basin water
Appendix C . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41 supply and use.
Appendix D. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
Endnotes. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45 We also used the PCR-GLOBWB 2 model to project future sub-basin water
supply, demand, stress, depletion, and variability using CMIP6 climate forcings.
References. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
The projections centered around three periods (2030, 2050, and 2080) under
Acknowledgements. . . . . . . . . . . . . . . . . . . . . .51
three future scenarios (business-as-usual SSP 3 RCP 7.0, optimistic SSP 1 RCP
About the authors. . . . . . . . . . . . . . . . . . . . . . . .51 2.6, and pessimistic SSP 5 RCP 8.5).
The normalized indicator scores have been aggregated by category (quantity,

Technical notes document the research or analytical
methodology underpinning a publication, interactive
quality, reputational, and overall) into composite risk scores using sector-
application, or tool. specific weighting schemes. In addition, select sub-basin scores have been
aggregated into country and provincial administrative boundaries using a
weighted average approach, where sub-basins with more demand have a
Suggested Citation: Kuzma, S., M.F.P. Bierkens, S.
Lakshman, T. Luo, L. Saccoccia, E. H. Sutanudjaja, higher influence over the final administrative score.
and R. Van Beek. 2023. “Aqueduct 4.0: Updated
decision-relevant global water risk indicators.” The main audience for this technical note includes users of the Aqueduct
Technical Note. Washington, DC: World Resources tool, for whom the short descriptions on the tool and in the metadata docu-
Institute. Available online at: doi.org/10.46830/ ment are insufficient. Key elements of Aqueduct, such as overall water risk,
writn.23.00061. cannot be directly measured and therefore are not validated. Aqueduct
remains primarily a prioritization tool and should be augmented by local and
regional deep dives.
TECHNICAL NOTE | Version 1.0 | August 2023 | 1

INTRODUCTION WATER RISK FRAMEWORK
Background Overview
WRI’s Aqueduct™ information platform compiles advances in The water risk framework follows a composite index approach
hydrological modeling, remotely sensed data, and published data and allows multiple water-related risks to be combined.
sets into a freely accessible online platform.
There are three hierarchical levels, as can be seen in Figure 1.
Since its inception in 2011, the Aqueduct information platform We start with 13 indicators covering various types of water risk.
has informed companies, governments, and nongovernmental We then group the indicators and calculate the grouped water
organizations (NGOs) about water-related risks. Since then, risk scores (composite score) using default, industry-defined, or
the data have been updated regularly, making them comparable user-defined weighting schemes. Finally, the three groups are
on a global scale and accessible to decision-makers worldwide. combined into a single overall water risk score.
The Aqueduct information platform contains the follow-
ing online tools: The rationale for creating a water risk framework is described
in WRI’s earlier publication: “Aqueduct Water Risk Atlas”
▪ Aqueduct Water Risk Atlas (Reig et al. 2013):
▪ Aqueduct Food
▪ Aqueduct Floods This [water risk] framework organizes indicators into categories
of risk that allow the creation of a composite index that brings
▪ Aqueduct Country Rankings together multiple dimensions of water-related risk into compre-
hensive aggregated scores. By providing consistent scores across
This technical note covers the development of the Aqueduct the globe, the Aqueduct Water Risk Atlas enables rapid compari-
4.0 framework and serves as the basis of the updated Aqueduct son across diverse aspects of water risk. . . .
Water Risk Atlas and Country Rankings online tools.
The Aqueduct Water Risk Framework enables users to study
The updated framework, database, and associated tools improve indicators individually or collectively, as well as to quantify and
one of the most widely used and respected water risk frame- compare a variety of multidimensional water-related measures.
works. We continue to build on years of experience applying and
standardizing these concepts within the water community, while
We selected the 13 indicators in Aqueduct 4.0 in three steps:
presenting the latest advancements in hydrological data and
climate science.
▪ We reviewed literature on relevant water issues, existing
water indicators, and data sources.
Structure and scope of this
technical note
▪ We evaluated potential data sources through a com
parative analysis of their spatial and temporal coverage,
granularity, relevance to water users, consistency, and
This technical note will first introduce the updated water risk
credibility of sources.
framework (“Water risk framework”). Many indicators in the
framework are based on updated inputs to a global hydrological
model, which is covered in “Hydrological model” and “Future
▪ We consulted with industry, public sector, and academic
water experts.
projections” (baseline and future, respectively). In “Indica-
tors” we discuss how the hydrological data is transformed into We applied the following three principal criteria in select-
Aqueduct water risk indicators. “Country and state aggrega- ing indicators:
▪ They
tion” and “Grouped and overall water risk “cover how the
individual indicators are aggregated into administrative-level should cover the full breadth of water-related risks,
scores and grouped (categorical risk) scores. “Limitations” lists while minimizing overlap and potential confusion resulting
key limitations. from an overabundance of indicators.
▪ They should be actionable in the context of private and

public sector decision-making.
2 |
Aqueduct 4.0: Updated decision-relevant global water risk indicators
Figure 1 | Overview of Aqueduct framework
Updated in Aqueduct 4.0
Baseline water stress
Baseline water depletion
Interannual variability
Seasonal variability
PHYSICAL RISK QUANTITY
Groundwater table decline
Riverine flood risk
Coastal flood
OVERALL WATER RISK

Drought risk
Untreated connected wastewater
PHYSICAL RISK QUALITY

Coastal eutrophication potential
Unimproved/no drinking water
REGULATORY AND
Unimproved/no sanitation
REPUTAIONAL RISK
Peak RepRisk country

ESG risk index
Note: Baseline water stress, baseline water depletion, interannual variability, and seasonal variability use the new PCR-GLOBWB 2 data, described in “Hydrological model” and Appendix C.
Source: WRI.
TECHNICAL NOTE | August 2023 | 3

▪ They should comply with WRI’s commitment to open
data and transparency—allowing input data, code, and
At the time the indicators were developed for Aqueduct 3.0
(2016–2019), GLDAS provided information until the year
results to be publicly available (“WRI’s Open Data 2012, making it less relevant than PCR-GLOBWB 2 and
Commitment” n.d.). WaterGAP, both of which could be run for more recent years.
There are many similarities between PCR-GLOBWB 2 and
WaterGAP. For example, both models run global hydrology and
HYDROLOGICAL MODEL water resources on a global scale at a daily time step; integrate
demand, withdrawal, and return flows2 per time step; include
Five of the 13 indicators in our framework are based on the data
reservoirs; and use kinematic wave routing of river water.
from a global hydrological model called PCR-GLOBWB 2.
However, PCR-GLOBWB 2 can couple to a global two-layer
Readers interested only in the indicator definitions can proceed
groundwater model (based on MODFLOW) to better represent
directly to “Indicators.” In the present section, we describe
groundwater flow (de Graaf et al. 2017). The code for PCR-
how we have selected the hydrological model and processed
GLOBWB 2 is open source and therefore aligned with WRI’s
its data for the baseline indicators of Aqueduct. For data
Open Data Commitment (“WRI’s Open Data Commitment”
sources used in Aqueduct’s future projections, please refer to
n.d.). For these reasons, WRI chose to continue to work with
“Future projections.”
PCR-GLOBWB 2 and use it as the updated global hydrological
From the model, we make use of water use, water supply, and model underpinning Aqueduct.
groundwater data1 to calculate water stress, water depletion,
A description of the model itself and the settings used for Aque-
seasonal variability, interannual variability, and groundwater
duct 4.0 can be found in Appendix C. This includes references
table decline (see Table 1).
to the new input datasets used to run PCR-GLOBWB 2 for
Aqueduct 4.0 compared to the run in 2019 for Aqueduct 3.0.
Model selection
We considered several global hydrological models and selected Model data used by Aqueduct
the PCRaster Global Water Balance (PCR-GLOBWB 2)
PCR-GLOBWB 2 is a global, gridded hydrological model.
model (Wada et al. 2014; Sutanudjaja et al. 2018) over others,
Each grid cell has a size of 5 × 5 arc minutes. This equates
most notably Water Global Assessment and Prognosis (Water-
roughly to 10 kilometer (km) × 10 km pixels, with any variation
GAP) (Müller Schmied et al. 2014; Eisner 2016) and Global
depending on the latitude. Aqueduct indicators are calculated
Land Data Assimilation System (GLDAS) Phase 2 (Rodell et
using three PCR-GLOBWB 2 datasets: water use, water supply,
al. 2004). While these three are not the only global hydrological
and groundwater data (see Table 1).
models available (Paul et al. 2021; Wang et al. 2021), they were
short-listed because of their dynamic human-freshwater system
modeling and their potential to be used for water target setting.
Table 1 | Aqueduct indicators based on hydrological model output
AQUEDUCT INDICATOR MODEL DATA USED
Water Use Water Supply Groundwater Heads
Water stress ✓ ✓
Water depletion ✓ ✓
Interannual variability ✓
Seasonal variability ✓
Groundwater table decline ✓
Note: Aqueduct indicators are calculated using the respective outputs of a hydrological model. For example, water depletion is calculated using water consumption and available blue
water from the hydrological model.
Source: WRI.
4 |
WATER USE: Spatial aggregation

Aqueduct considers two metrics of water use: gross demand Grid cells are not an appropriate spatial unit to use as input
and net consumption. Gross demand3 is the maximum potential for the Aqueduct indicators. For water stress, water deple-
water required to meet sectoral demands, and net consumption, seasonal variability, and interannual variability, the
tion is the portion of demand that is lost in use—evaporated or preferred spatial units are hydrological sub-basins (Gassert et
incorporated into a product—and not returned to the system al. 2014). For groundwater table decline, the preferred spatial
(Gassert et al. 2014). units are aquifers.
Gross demand and net consumption4 for four sectors: domestic,
HYDROLOGICAL SUB-BASINS:
industrial, irrigation5, and livestock6. The (2 x 4=) 8 gridded data
sets are available for each month between January 1960 and A hydrological basin is an area that drains at a single point to
December 2019. an outlet such as a river, ocean or inland lake. Each basin can
be divided into smaller sub-basins at the confluence of two
WATER SUPPLY: streams (Lehner and Grill 2013). The assumption is that within
each hydrological sub-basin, water resources are pooled. Water
Aqueduct defines supply as available blue water—the total
withdrawal is satisfied using the water resources available to
amount of renewable freshwater available to a sub-basin with
the sub-basin.
upstream consumption removed (Gassert et al. 2014). We com-
pute available blue water as internal sub-basin runoff plus the Aqueduct 4.0 uses the HydroBASINS level 6 hydrological sub-
accumulated water flowing into the sub-basin from upstream, basins for a few reasons, including the following:
where upstream consumption is already removed (i.e., discharge)
(Gassert et al. 2014). This includes freshwater from the follow-
ing sources: surface flow, interflow7, and groundwater recharge.
▪ HydroBASINS are used in other tools and databases, so
comparing and collating data is easier.
Internal runoff8 monthly at each grid cell between January 1960 ▪ The HydroBASINS sub-basin data set contains 12 levels,
ranging from large basins to small sub-basins. In the future,
and December 2019.
this hierarchical model will allow flexibility when combining
Discharge9 monthly at each grid cell between January 1960 and additional data sets (Lehner and Grill 2013).
December 2019.
Of the 12 levels, we chose level 6 as the appropriate size of the
GROUNDWATER HEADS: sub-basins. Water demand is often satisfied with water from
Groundwater heads for each month and each grid cell between a nearby or slightly more distant source. The average distance
January 1990 and December 2014. These have been obtained from source to destination of water supply is the main selection
from the 5 arc minute two-layer global groundwater model of criterion of the appropriate HydroBASINS level. The goal is to
de Graaf et al (2017) coupled to PCR-GLOBWB 2. Note: this select a level large enough to minimize the nonnatural effect of
dataset has not been rerun since its original release in 2019. transfers of water (“inter-basin transfer”)10 and small enough to
capture meaningful local variations.
Processing model data Based on limitations, primarily the lack of comprehensive local
level inter-basin transfer data in PCR-GLOBWB 2, HydroBA-
To make the model data suitable for the Aqueduct indicator
SINS level 6 is deemed the most appropriate sub-basin level for
calculation, we further processed the data by spatial and tempo-
Aqueduct 4.0 analysis. For perspective, HydroBASINS level 6
ral aggregation.
has a median area per sub-basin11 of 5,318 km2 (roughly the size
▪ Spatial aggregation. Water use and water supply are

aggregated to hydrological sub-basins. Groundwater heads
of the U.S. state of Delaware or twice the size of Luxembourg).
The distribution of sub-basin areas is depicted in Figure 2.
are aggregated to aquifers. PCR-GLOBWB 2 and HydroBASINS level 6 both assume
▪ Temporal aggregation. We apply statistical methods to the
output time series to get a representative value for the recent
a strictly convergent flow. This means that it cannot model
bifurcations. This is an issue in delta regions, where rivers
situation, while correcting for annual anomalies. tend to split. To address this issue, we have identified delta
sub-basins and merged them. The methodology is explained
in Appendix D.

Figure 2 | Area distribution of HydroBASIN Level 6
7,000
6,000
5,000
Frequency
4,000
3,000
2,000
1,000
0
100 101 102 103 104 105
Area km2
Source: WRI.
AQUIFERS
Groundwater head data are aggregated to groundwater aquifers ing through a river channel, not a flux, and therefore requires a
(BGR and UNESCO 2008). This data set of global aquifers is different spatial aggregation technique.
selected because it has global coverage and is used in the previ-
To turn discharge into inflow, we first must identify inflow and
ous versions of Aqueduct.
outflow points13 for each sub-basin using the PCR-GLOBWB
STEP 1: SPATIAL AGGREGATION OF WATER USE 2 local drainage direction (LDD14) network (see Appendix C).
Normally, a hydrologically sound sub-basin would have just one
Sectoral gross demand and net consumption are aggregated to inflow and one outflow point; however, due to the rasteriza-
HydroBASINS level 6. First, the data are resampled from 5 × tion15 of the sub-basins and the mismatched resolutions between
5 arc minute to 1 × 1 arc minute12. Then, the data, which are in PCR-GLOBWB 2’s digital elevation model and the HydroBA-
flux (m/month), are converted to volume (million m3/month) SIN sub-basins16, we have numerous inflow and outflow points
by multiplying each grid by its cell area (m2) and dividing by a per sub-basin (see Figure 3). Some of these discharge flow
million. Finally, we sum the gridded volume per sub-basin. points are considered “false”17, meaning the stream temporarily
leaves or enters the sub-basin. In each sub-basin, all flow points
STEP 2: SPATIAL AGGREGATION OF WATER SUPPLY satisfy this condition:
Supply—also known as available blue water—is also aggregated
to HydroBASIN level 6. Available blue water equals the internal true outflow = all discharge inflow–false discharge outflow +
sub-basin runoff plus the accumulated water flowing into the internal runoff–internal consumption
sub-basin from upstream, where upstream consumption is
already removed (i.e., discharge) (Gassert et al. 2014). This translates to the definition of available blue water by
removing the internal consumption term:
First, we use the internal runoff to find the renewable water sup-
ply within each sub-basin—that is, internal catchment supply available blue water = all discharge inflow–false discharge
before consumption is removed. Like the demand data, runoff is outflow + internal runoff
available as a flux, and so we follow the same spatial aggregation
process as demand. We sum the discharge values at all (true and false) inflow points
and subtract the discharge values from the false outflow points
Next, we use discharge to account for the accumulated water
(white and purple arrows in Figure 3, respectively) to calculate
flowing into each sub-basin, with upstream consumption
the accumulated water flowing into the sub-basin. We then add
already removed. Discharge is the rate (m3/sec) of water flow-
internal runoff to estimate available blue water.
6 |
Temporal aggregation
One of the advantages of the Aqueduct framework is its ease of STEP 1: TOTAL WATER USE
use. Although time series provide detailed insights, for a priori- We calculate the total gross demand and total net consumption
tization method and combined framework, summary indicators by summing up the four sectors (domestic, industrial, irrigation,
are preferred. Aqueduct provides chronic water risk information. and livestock) for each sub-basin and month ( January 1979–
This is very different from near-real-time water risk information December 2019). The results are two time series: Gross total
or a historical assessment. Ideally, each indicator is representative demand and net total consumption for January 1979–December
of the relative time period—be that baseline or future—without 2019 for each sub-basin.
anomalies (the exception being the variability indicators, in
which anomalies are intentionally captured). STEP 2: SPLIT MONTHS
We apply temporal aggregation steps to convert historical time We then break up the time series into one series for each month.
series into useful input for the baseline indicator calculations. This yields time series of all Januarys between 1979 and 2019, all
Aqueduct 4.0’s baseline represents a 40-year period (1979– Februarys between 1979 and 2019, and so on to all Decembers
2019)18. For water stress and depletion, the long-term trends between 1979 and 2019. We do this for total gross demand, total
are extracted from the noise using the methodology below. For net consumption, and available blue water.
seasonal and interannual variability, the raw time series are used.
STEP 3: REGRESSION
Temporal aggregation for future projections can be found
In most sub-basins, the water use data follow a clearly increas-
“Processing model data” in the next section. Groundwater head
ing trend. This is caused by increases in underlying drivers such
data are processed separately; see “Groundwater table decline” in
as growth in population and gross domestic product (GDP). In
the “Indicators” section for more information.
Figure 3 | Example of discharge inflow and outflow points
Source: Authors.

addition, both the demand data19 and supply data can be erratic. therefore be handled separately. We identified those sub-basins
We try to reduce noise while keeping an accurate representa using two criteria with thresholds taken from Aqueduct 2.1
tion of the present value based on the long-term trend. We first (Gassert et al. 2014):
smooth the data by applying a trailing moving average with a
window size of 10 years. We then run a Theil-Sen regression20 A sub-basin is “arid” if annual baseline available water <
on the moving average to capture the trend. We select the point 0.03 meters per year (m/yr)
intersecting with our target year—2019—to represent the base-
A sub-basin is “low water use” if annual baseline gross total
line. The independent variable is time (year), and the dependent
withdrawal < 0.012 m/yr
variable is either gross total withdrawal, net total withdrawal, or
available blue water.
Data must be below both the arid and low water use thresholds
Additionally, we restrict the predicted value to the minimum to be masked. Monthly data is masked using the annual arid and
and maximum range of the 10-year moving average window low water use categories.
values. The predicted value can never exceed the maximum of
the 10-year window values or be lower than the minimum of the Processed water use and water supply
10-year window functions. After applying the spatial and temporal aggregation steps, we
have estimates of gross demand (ww), net consumption (wn),
We opted for a window size of 10 years to capture longer
and available blue water (ba) for each sub-basin.
climatic and socioeconomic trends while filtering annual
anomalies. The temporal aggregation steps 1 through 3 for an Delta regions and arid and low water use sub-basins will be
example sub-basin are shown in Figure 4. treated accordingly in the indicator calculation. We use the
aggregated time series of gross total withdrawal, net total
STEP 4: MASK ARID AND LOW WATER USE SUB-BASINS withdrawal, and available water to calculate water stress, water
Aqueduct indicators require robust data as inputs.21 Sub- depletion, seasonal variability, and interannual variability.
basins where data are sparse or very close to zero should
Figure 4 | Theil-Sen regression for gross demand on a 10-year moving average window for July in an example basin
(Erbo Sub-basin [216041])
0.02
0.018
0.016
gross demand (m/month)
0.014
0.012
0.01
0.008
0.006
0.004
0.002
0
1979 1985 1990 1995 2000 2005 2010 2015 2019
Year
Gross demand (m/month) 10-year moving average Theil Sen Regression
Source: Authors.
8 |
FUTURE PROJECTIONS
The availability of and competition for water resources around compared to preindustrial levels (1850–1900) (Arias et al. 2021).
the world will be affected by climate change, population growth, SSP1 is characterized by sustainable socioeconomic growth:
and economic development. This analysis complements our stringent environmental regulations and effective institutions,
baseline water risk data by providing information on future rapid technological change and improved resource efficiency,
water availability that is relevant for long-term planning, adapta- and low population growth (Wada et al. 2016). SSP3-7.0 repre-
tion, and investment by both the private and public sectors. sents a “business as usual” scenario with temperatures increasing
by 2.8°C to 4.6°C by 2100. SSP3 is a socioeconomic scenario
This section details the methodology used to develop global characterized by regional competition and inequality, includ-
estimates of water supply (available blue water), water use ing slow economic growth, weak governance and institutions,
(gross demand and net consumption), water stress (the ratio of low investment in the environment and technology, and high
demand to supply), water depletion (the ratio of consumption population growth, especially in developing countries. SSP5-8.5
to supply), interannual variability, and seasonal (intra-annual) represents a “pessimistic” scenario with temperature increases
variability for three 30-year periods centered on milestone up to 3.3°C to 5.7°C. SSP5 describes fossil-fueled development:
years 2030, 2050, and 2080. Projections of climate variables are rapid economic growth and globalization powered by carbon-
driven primarily by general circulation models from the Coupled intensive energy, strong institutions with high investment in
Model Intercomparison Project phase 6 (CMIP6) project, and education and technology but a lack of global environmental
socioeconomic variables are based on the Shared Socioeconomic concern, and the population peaking and declining in the 21st
Pathways database from the International Institute for Applied century. Each scenario has varying effects on water availability in
Systems Analysis. different parts of the world.
Model data used by Aqueduct GENERAL CIRCULATION MODELS (GCMS):
The Aqueduct Future Projections are based on a new dataset For each scenario, we ran five GCMs to account for the uncer-
called PCR-GLOWB-based hydrological projection of future tainty in climate models: GFDL-ESM4, IPSL-CM6A-LR,
global water states with CMIP 6 (HYPFLOWSCI6) (Sutan- MPI-ESM1–2-HR, MRI-ESM2-0, and UKESM1-0-LL. The
udjaja et al. 2023). They use the same model structure and five GCMs were bias-corrected22 toward the observed climate
classes of data to define water use and supply as the baseline, forcing data used for the baseline (Lange 2021, Hempel et
except they are created using different climate forcing data al. 2013). They were chosen because they represent a span of
and cover greater time periods. HYPFLOWSCI6 uses climate temperature-precipitation variations (e.g., cold-wet). These five
forcing data from multiple future scenarios of socioeconomic variations provided a good sample size for model uncertainty
and climate conditions, which are each run through five separate given our computational limitations23. GCM data are converted
climate models. into PCR-GLOBWB inputs using the methodology described
in Sutanudjaja et al. (2023). GCM data are prepared for both
SOCIOECONOMIC AND CLIMATE SCENARIOS: the historic time period (1960–2014) and the three future
scenarios (2015–2100, each). In all, there are 5 historical runs
Estimates of each indicator are developed for three socio-
and (5 x 3=) 15 future runs.
economic and climate scenarios used in CMIP6 (SSP1–2.6,
SSP3-7.0, and SSP5-8.5). Shared socioeconomic pathways
(SSPs), indicated by the first number in each scenario (1, 3, and Processing model data
5), describe alternative futures of societal development and water To make the model data suitable as input for the Aqueduct indi-
use. The second number in each scenario (2.6, 7.0, and 8.5) cator calculation for future projections, we again process the data
indicates the level of radiative forcing (W m−2) through 2100. by spatial and temporal aggregations.
These drive the climate factors in general circulation models
(GCMs). The SSP pathways were used to project future water
Spatial aggregation
use, while the SSP/RCP combined pathways were used to
project future water supply. For each future indicator, we aggregate water use and water sup-
ply to HydroBASINS level 6 hydrological sub-basins using the
SSP1–2.6 represents an “optimistic” scenario limiting the rise in same methodology applied to the baseline (see under “Hydro-
average global surface temperatures by 2100 to 1.3°C to 2.4°C logical model”).

Temporal aggregation Future projections limitations
We also follow the baseline methodology (see under “Hydro- Irrigation data projects crop extents using the IMAGE dataset,
logical model”) to temporally aggregate the future projections which ends in 2050 (Doelman et al. 2018). Therefore, crop
time series of water supply, demand, and consumption to our extents beyond 2050 are assumed to remain static to the 2050
milestone years: we smooth the data using a 10-year trailing extent (though total irrigation demand may still fluctuate
moving average and isolate the trend using a Theil-Sen regres- beyond 2050 due to varying climatic conditions). This could
sion. The only difference is the time periods used: underestimate future (2051–2100) irrigation demand in loca-
▪ 2014: 1960–2014 (historic GCM data)

tions with historically low levels of irrigation that are likely to
expand over the next century (like many countries in Africa).
▪ 2030: 2015–2045 (future GCM data) Likewise, livestock water demand data ends in 2014, and is
▪ 2050: 2035–2065 (future GCM data) assumed to remain constant through 2100. Livestock water
▪ 2080: 2065–2095 (future GCM data)

demand is the daily drinking water required per animal as a
function of temperature (Wada et al. 2016) and amounts to less
than 1 percent of global demand in 2019. Ninety-eight percent
We select the regressed value for each milestone year. We
of an animal’s water footprint comes from the feed they con-
perform this step for each month in order to generate our
sume (Mekonnen and Hoekstra 2012), which Aqueduct covers
annual results.
under irrigation demand. Still, we are underestimating future
(2020–2100) livestock demand.
BIAS CORRECTION:
Each GCM uses unique estimates of climate under a given
scenario spanning from the pre-industrial age to the year 2100, INDICATORS
meaning that historical GCM climate forcing data may deviate
from what actually happened. To correct this, we must adjust the For each of the 13 indicators in our framework, this section
future monthly GCM data to our monthly baseline data—spe- offers a description, a calculation of raw values, and a conversion
cifically, total gross demand, total net consumption, and available to 0–5 scores. This enables us to aggregate the indicators into
blue water. We perform this bias correction in two steps. First, groups, as well as to provide an overall water risk score. For each
we calculate the change between the future GCM output and indicator, we also include the key limitations.
the historical GCM output:
Aqueduct 4.0 uses the United Nations Office for Disaster Risk
Reduction (UNDRR) risk element terminology of hazard, expo-
Adjusted Future Data = sure, and vulnerability. Each indicator is assigned a risk element
Baseline + (Future GCM - Historic GCM) (see Figure 5):
Processed water use and water supply ▪ HAZARD: Threatening event or condition (e.g., flood event,
water stress condition).
After applying the spatial and temporal aggregation steps and
the bias correction, we have 2030, 2050, and 2080 estimates of ▪ EXPOSURE: Elements present in the area affected by the
hazard (e.g., population, asset, economic value).
total gross demand, total net consumption, and available blue
water for each GCM for each scenario. We use these data to
calculate future water stress, water depletion, seasonal variability,
▪ VULNERABILITY: The resilience or lack of resilience of
the elements exposed to the hazard.
and interannual variability using the indicator methodology
outlined in “Indicators.” Then, we find the median, minimum,
Figure 5 | Elements of risk
maximum, and standard deviation across the GCMs per
scenario for each indicator. The median is used as the default
representation for each scenario. We estimate the uncertainty RISK HAZARD EXPOSURE VULNERABILITY
of the GCM data by calculating the coefficient of variation
of the five GCMs scores per indicator within each catchment = X X
(Luck et al. 2015).
Source: Raw data from UNDRR, modified/aggregated by WRI.
10 |
Baseline water stress

wwm,b = Total gross demand per month, per sub-basin in
GENERAL:
[million m3/month]
Name Baseline Water Stress
bam,b = Available blue water per month, per sub-basin in
Subgroup Physical risk quantity [million m3/month]
Risk Element R=HxExV
This results in 12 time series of water stress (one for each
RESULTS: month) per sub-basin. Additionally, we limit the raw values to a
Spatial resolution Hydrological sub-basin (HydroBASINS 6) maximum of 1 and a minimum of 0. Note that water resources
in delta sub-basins are pooled.
Temporal resolution Monthly and annual
SOURCE: STEP 2: CALCULATE ANNUAL WATER STRESS
Spatial resolution 5 x 5 arc minute grid cells We calculate the annual water stress by applying a weighted
average of monthly values, with total demand as the weight;
Temporal resolution Monthly
months with higher demand will have more influence on the
Temporal range 1979–2019 annual stress value. Months with more demand reflect when the
EXTRA: human need for water is greatest—it is also when socioeconomic
dependency for water is most critical (Gassert et al. 2013). The
Partner organization(s): Utrecht University annual raw value (wsy,b) is found by multiplying the monthly raw
Model PCR-GLOBWB 2 value (wsm,b) by the monthly weight (wwm,b), summing, and
dividing by the sum of the weights across the year (y).
Date of publication 2023
Description
Baseline water stress measures the ratio of total water demand
to available renewable surface and groundwater supplies. Water
demand includes domestic, industrial, irrigation, and livestock Sub-basins classified as “arid and low water use” are han-
consumptive and nonconsumptive uses. Available renewable dled separately.
water supplies include the impact of upstream consumptive
water users and large dams on downstream water availability. Conversion to risk categories
Higher values indicate more competition among users. The risk thresholds are based on Aqueduct 2.1 (Gassert
et al. 2014).
Calculation
RAW VALUE RISK CATEGORY SCORE
Baseline water stress is calculated using the postprocessed gross
demand and available blue water per sub-basin time series <10% Low 0–1
from the default PCR-GLOBWB 2 run (covered in “Hydro- 10–20% Low-medium 1–2
logical model”).
20–40% Medium-high 2–3
STEP 1: CALCULATE MONTHLY WATER STRESS 40–80% High 3–4
>80% Extremely high 4–5
Arid and low water use 5
In which,
wsm,b = Water stress per month, per sub-basin in [-]

The raw values are remapped to a 0–5 scale using the fol- Although the underlying models have been validated, the results
lowing equation: are not. Water stress remains subjective and cannot be measured
directly. The lack of direct validation makes it impossible to
assess some of the parameters in our calculation, such as the
length of the input time series, regression method, and optimal
moving window size.
Where r is the raw indicator value and score is the indica-
tor score [0–5]. The water stress indicator presented here does not explicitly take
into account environmental flow requirements,24 water quality, or
Limitations access to water. Views differ regarding what to include in a water
stress indicator (Vanham et al. 2018).
All limitations of the underlying data, including those pro
duced by the PCR-GLOBWB 2 global hydrological model and Finally, we should stress that Aqueduct is tailored to large-scale
HydroBASINS 6 hydrological sub-basin delineation, apply to comparison of water-related risks. The indicators have limited
this indicator. Please see the original publications of these data added value on a local scale.
sets for a full list of limitations.
One of the biggest assumptions is that water resources are

pooled within each sub-basin. However, in HydroBASINS 6,
coastal and island sub-basins are often grouped to make the area
of the sub-basins more homogeneous. The assumption of shared
water resources might not hold in aggregated coastal sub-basins.
12 |
Baseline water depletion

STEP 1: CALCULATE MONTHLY WATER DEPLETION
GENERAL:
Name Baseline Water Depletion

Subgroup Physical risk quantity
RESULTS: In which,
Spatial resolution Hydrological sub-basin (HydroBASINS 6)
wdm,b = Water depletion per month, per sub-basin in [-]
wnm,b = Total net consumption per month, per sub-basin in
SOURCE:
[million m3/month]
Spatial resolution 5 x 5 arc minute grid cells bam,b = Available blue water per month, per sub-basin in
Temporal resolution Monthly [million m3/month]
Temporal range 1979–2019

This results in 12 time series of water depletion (one for each
EXTRA: month) per sub-basin. Additionally, we limit the raw values to a
Partner organization(s): Utrecht University maximum of 1 and a minimum of 0. Note that water resources
in delta sub-basins are pooled.
Model PCR-GLOBWB 2
Date of publication 2023 STEP 2: CALCULATE ANNUAL WATER DEPLETION
We calculate the annual water depletion by applying a weighted
average of monthly values, with total net consumption as the
Description weight; months with higher consumption will have more
Baseline water depletion measures the ratio of total water influence on the annual depletion value. Months with more
consumption to available blue water. Total water consumption consumption reflect when the human need for water is great-
includes domestic, industrial, irrigation, and livestock consump- est—it is also when socioeconomic dependency for water is
tive uses. Available renewable water supplies include the impact most critical (Gassert et al. 2013). The annual raw value (wda,b) is
of upstream consumptive water users and large dams on down- found by multiplying the monthly raw value (wdm,b) by the
stream water availability. Higher values indicate larger impact monthly weight (wnm,b), summing, and dividing by the sum of
on the local water supply and decreased water availability for the weights across the year (y).
downstream users.
Baseline water depletion is similar to baseline water stress;

however, instead of looking at gross water demand (consumptive
plus nonconsumptive), baseline water depletion is calculated
using consumptive withdrawal only. Sub-basins classified as “arid and low water use” are han-
dled separately.
Calculation
Baseline water depletion is calculated using the processed total
net consumption and available blue water per sub-basin time
series from the default PCR-GLOBWB 2 run (covered in
“Hydrological model”).

Conversion to risk categories Limitations
The thresholds are based on Brauman et al. (2016). See Baseline Water Stress, Limitations (above in this section).
RAW VALUE RISK CATEGORY SCORE In addition, we had to omit the categories “dry year” and “sea-
sonal” from Brauman et al. (2016) to make the indicator suitable
<5% Low 0–1
for the Aqueduct framework.
5–25% Low-medium 1–2
25–50% Medium-high 2–3
50–75% High 3–4
>75% Extremely high 4–5
Arid and low water use 5
We use linear interpolation within each category to remap the

raw values to a 0–5 scale using the following equation:

tor score [0–5].
14 |
Interannual variability
In which,
GENERAL:
Name Interannual Variability iavm,b = Interannual variability per month, per

Subgroup Physical risk quantity sub-basin in [-]
cvm,b = Coefficient of variation per month, per
sub-basin in [-]
RESULTS:
bam,b = Available blue water per month, per sub-basin in
Spatial resolution Hydrological sub-basin (HydroBASINS 6) [million m3/year]
SOURCE:
Conversion to risk categories
The risk thresholds are based on Aqueduct 2.1 (Gas-
Spatial resolution 5 x 5 arc minute grid cells
sert et al. 2014).
Temporal resolution Monthly
<0.25% Low 0–1
EXTRA:
0.25–0.50% Low-medium 1–2
Partner organization(s): Utrecht University
0.50–0.75% Medium-high 2–3
Model PCR-GLOBWB 2
0.75–1.00% High 3–4
>1.00% Extremely high 4–5
Description
The raw values are remapped to a 0–5 scale using the fol-
Interannual variability measures the average between-year lowing equation:
variability of available water supply, including both renewable
surface and groundwater supplies. Higher values indicate wider score = max(0, min(5,4r))
variations in available supply from year to year.
Calculation tor score [0–5].
Interannual variability is calculated using the available water

time series from the default PCR-GLOBWB 2 aggregated in
Limitations
space but not in time. See “Hydrological model.” See Baseline Water Stress, Limitations (in this section).
Interannual, or between year, variability is defined as the In addition, we have analyzed the full time series of PCR-
coefficient of variation (CV) of available water for each sub- GLOBWB 2; that is, 1979 to 2019. We have not analyzed the
basin. The CV is the standard deviation (SD) of the available effect of using a different range.
water, divided by the mean. The CV per sub-basin is determined
for each individual month, as well as annually.

Seasonal variability
The coefficient of variation is calculated using these 12 averages.
GENERAL:
Name Seasonal Variability

In which,
RESULTS:
Spatial resolution Hydrological sub-basin (HydroBASINS 6) sevb = Seasonal variability per sub-basin in [-]
Temporal resolution Annual baseline bam,b = Available water per month, per sub-
SOURCE:
basin in [m/month]

Temporal resolution Monthly The risk thresholds are based on Aqueduct 2.1 (Gas-
Temporal range 1979–2019 sert et al. 2014).
EXTRA: RAW VALUE RISK CATEGORY SCORE
Partner organization(s): Utrecht University <0.33 Low 0–1
Model PCR-GLOBWB 2 0.33–0.66 Low-medium 1–2
Date of publication 2023 0.66–1.00 Medium-high 2–3
1.00–1.33 High 3–4
Description >1.33 Extremely high 4–5
Seasonal variability measures the average within-year variability
of available water supply, including both renewable surface and
groundwater supplies. Higher values indicate wider variations
lowing equation:
of available supply within a year.
score = max(0, min(5,3r))
Calculation Where r is the raw indicator value and score is the indica-
Seasonal variability is calculated using the available water time tor score [0–5].
series from the default PCR-GLOBWB aggregated in space but
not in time. See “Hydrological model.” Limitations
First, the available water per month, per sub-basin, is calculated See Baseline Water Stress, Limitations (in this section).
over the entire time series 1979–2019 (40 years).
Additionally, the effect of using different lengths of the input
time series is not examined. The human and climatic influence
on available water is likely to be more profound in recent years.
In which,
bam,b = Average available blue water per month, per sub-

basin in [million m3/year]
bay,m,b = Available blue water per year month, per sub-basin
in [million m3/year]
16 |
Groundwater table decline

Mountainous areas are determined by comparing the height of
GENERAL:
the floodplain within a cell with the average elevation of that
Name Groundwater Table Decline same cell. The elevation of the floodplain is derived from the 30
Subgroup Physical risk quantity × 30 arc second digital elevation data from HydroSheds (Lehner
et al. 2008). The flood plain elevation is simply the mini-
Risk Element R=HxExV mum of the input.
RESULTS:
Spatial resolution Groundwater aquifer (WHYMAP)

Temporal resolution Annual baseline In which,
SOURCE: hfloodplain,5’ = Elevation of floodplain in meters for each 5 x 5

Spatial resolution 5 x 5 arc minute grid cells arc minute cell
Temporal resolution Monthly hDEM 30” = Elevation derived from 30 x 30 arc second digital
elevation model (DEM) in meters
EXTRA: The average elevation for each 5 arc minute cell is taken directly
from the HydroSheds data. If the difference between the flood-
Partner organization(s): Deltares, Utrecht University
plain elevation and the average elevation is greater than 50m,
Model PCR-GLOBWB 2 + MODLFLOW the cell is classified as mountainous.
Description
Groundwater table decline measures the average decline of the In which,
groundwater table as the average change for the period of study
(1990–2014). The result is expressed in centimeters per year hfloodplain,5’ = Elevation of floodplain in meters for each 5 x 5
(cm/yr). Higher values indicate higher levels of unsustainable arc minute cell
groundwater withdrawals. hDEM 5’ = Elevation derived from 5 x 5 arc minute DEM in
meters (approximately 11km at the equator)
Calculation
Groundwater table decline is calculated using the groundwater The threshold of 50m was chosen as it proved to include 70
heads time series from the PCR-GLOBWB 2 run coupled with percent of the unconsolidated sediments mapped in the Global
MODFLOW to account for lateral groundwater flow processes. Lithological Map (Hartmann and Moosdorf 2012).
This indicator is based on the gridded25 monthly groundwater After masking the mountainous areas, results are aggregated to
heads between January 1990 and December 2014.26 groundwater aquifers derived from the Worldwide Hydrogeo-
The groundwater aquifers contain several geomorphological logical Mapping and Assessment Programme (WHYMAP)
features, which for practical reasons can be divided into sedi- data set (BGR and UNESCO 2018).
mentary basins and mountain ranges. In mountainous areas, The monthly results at the aquifer scale are fitted with a first-
most materials are hard rock and eventually weathered. In the order regression. The slope of this regression line (cm/yr)
PCR-GLOBWB 2 model coupled with MODFLOW, very indicates the existence of a downward (or upward) trend. The
deep groundwater influences the averages in mountainous following estimators are used to further assess the trend: (1)
cells and is not representative. These cells are therefore dis- coefficient of determination and (2) the p value.
carded from the calculations following the method in de Graaf
et al. (2015).

The coefficient of determination is used to determine whether Limitations
the trend is linear or erratic. A minimum threshold of 0.9 is
The limitations of PCR-GLOBWB 2, MODFLOW,
applied to mask out erratic and error-prone trends.
WHYMAP, climate forcing, and other input data sets are
For the p value, a maximum threshold of 0.05 is used. propagated to these results. The results are only validated using
a literature review of selected aquifers and by comparing the
results to coarse remote-sensing data.
The threshold for masking out mountainous areas was set once
Conversion to risk categories without a sensitivity analysis. The temporal range [1990–2014]
The risk category thresholds are based on a combination of was selected on the basis of expert judgment and differs from
expert judgment and a literature review (Galvis Rodrí some of the other water quantity indicators that use [1960–
guez et al. 2017). 2014] as the input time series.

See Galvis Rodríguez et al. (2017) for additional limitations.
<0 cm/y Low 0–1

0–2 cm/y Low-medium 1–2
2–4 cm/y Medium-high 2–3
4–8 cm/y High 3–4
>8 cm/y Extremely high 4–5
Within each category, we use linear interpolation to convert the

raw values to a 0–5 using the following equation:

tor score [0–5].
18 |
Riverine flood risk

GENERAL: Calculation
Name Riverine Flood Risk Data on the population impacted by riverine floods are provided
by Aqueduct Floods at the state/HydroBASIN 6 intersect
Subgroup Physical risk quantity scale (Ward et al. 2020). The data set estimates the average
Risk Element R=HxExV number of people to be impacted annually for several flood
event magnitudes (2, 5, 10, 25, 50, 100, 250, 500, and 1,000 in
RESULTS:
return periods).
The expected annual affected population is calculated using a risk
Temporal resolution Annual baseline curve (Meyer et al. 2009). To create the curve, the return periods
SOURCE: are first converted into probabilities (i.e., 1/return period) and
then plotted on the x axis against the impacted population (Fig-
ure 6). Next, flood protection is added to the graph. The current
Temporal resolution Annual level of flood protection, given in return years, comes from the
Flood Protection Standards (FLOPROS) model (Scussolini et
Temporal range 2010
al. 2016). All impacts that fall to the right of the flood protec
EXTRA: tion line (i.e., impacted by smaller floods) are assumed to be
Partner organization(s): Deltares, IVM, PBL, Utrecht University protected against floods and are removed from the calculation.
The expected annual affected population is calculated by integrating
Model GLOFRIS (Ward et al. 2020) the area under the curve to the left of the flood protection line.
The expected annual affected population is calculated for each
state/HydroBASIN 6 intersect, then aggregated up to the
HydroBASIN 6 scale. The total population in each state/Hydro-
Additional data source Existing flood protection levels
BASIN 6 intersect is also summed to the HydroBASIN 6 scale
Input spatial resolution State (Ward et al. 2020). Finally, the raw riverine flood risk score—the
percentage of population expected to be affected annually by riverine
Input temporal resolution Annual
floods per HydroBASIN 6—is calculated:
Input temporal rance 2016
Source FLOPROS (Scussolini et al. 2016)
Description In which,
Riverine flood risk measures the percentage of population
expected to be affected by riverine flooding in an average year, rfr = Riverine flood risk raw values in [-]
accounting for existing flood-protection standards. Flood risk popexp,r,y = Expected annual affected population by riverine
is assessed using hazard (inundation caused by river overflow), flooding in [number of people]
exposure (population in flood zone), and vulnerability.27 The
existing level of flood protection is also incorporated into the
risk calculation. It is important to note that this indicator
represents flood risk not in terms of maximum possible impact
but rather as average annual impact. The impacts from infre-
quent, extreme flood years are averaged with more common, less
newsworthy flood years to produce the “expected annual affected
population.” Higher values indicate that a greater propor
tion of the population is expected to be impacted by riverine
floods on average.

Figure 6 | Risk Curve Used to Calculate Expected Annual Affected Population from Floods
300
Probability-impact curve
1/10 year flood
protection Flood Protection
Affected population (millions)
200 Non-protected/
expected damage
100
Protected/damage avoided
0
0 10% 20% 30% 40% 50% 60%
Probability of event occurring
Source: WRI.

The thresholds are based on quantiles, with the exception of the The raw values are remapped to a 0–5 scale using the fol-
basins with no riverine hazard. Basins without a flood hazard are lowing equation:
given the lowest risk score, 0, and are removed from the rest of
the data set before the quantiles are calculated.
0 to 1 in 1,000 Low 0–1

1 in 1,000 to 2 in 1,000 Low-medium 1–2
2 in 1,000 to 6 in 1,000 Medium-high 2–3
6 in 1,000 to 1 in 100 High 3–4
More than 1 in 100 Extremely high 4–5
tor score [0–5].
Limitations
Riverine and coastal flood risks must be evaluated and used
separately, as the compound risks between river and storm
surges are not modeled. The data also assume that flood events
are entirely independent of each other, so the impact from
overlapping flood events is not considered. Finally, the data do
not include any indirect impacts from flooding (e.g., disrupted
transportation, loss of work, etc.).
20 |
Coastal flood risk

Name Coastal Flood Risk Data on the population impacted by coastal floods are provided
by Aqueduct Floods at the state/HydroBASIN 6 intersect
scale (Ward et al. 2020). The data set estimates the average
Risk Element R=HxExV number of people to be impacted annually for several flood
RESULTS: event magnitudes (2, 5, 10, 25, 50, 100, 250, 500, and 1,000 in
return periods).
Temporal resolution Annual baseline The expected annual affected population is calculated using a risk
curve (Meyer et al. 2009). To create the curve, the return periods
SOURCE: are first converted into probabilities (i.e., 1/return period) and
Spatial resolution 30 x 30 arc minute grid cells then plotted on the x axis against the impacted population
(Figure 6). Next, vulnerability—or flood protection—is added
Temporal resolution Annual
to the graph as a vertical line. The current level of flood protec-
Temporal range 2010 tion, given in return years, comes from the FLOPROS model
EXTRA: (Scussolini et al. 2016). All impacts that fall to the right of
the flood protection line (i.e., impacted by smaller floods) are
Partner organization(s): Deltares, IVM, PBL, Utrecht University
assumed to be protected against floods and are removed from
Model GLOFRIS (Ward et al. 2020) the calculation. The expected annual affected population is calcu-
Date of publication 2019 lated by integrating the area under the curve to the left of the
flood protection line.
The expected annual affected population is calculated for each

Additional data source Existing flood protection levels state/HydroBASIN 6 intersect and then aggregated up to the
Input spatial resolution State HydroBASIN 6 scale. The total population in each state/Hydro-
BASIN 6 intersect is also summed to the HydroBASIN 6 scale
(Ward et al. forthcoming). Finally, the raw coastal flood risk
Input temporal rance 2016 score—the percentage of population expected to be affected annually
Source FLOPROS (Scussolini et al. 2016)
by coastal floods per HydroBASIN 6—is calculated:
Description
Coastal flood risk measures the percentage of the population
expected to be affected by coastal flooding in an average year, In which,
accounting for existing flood protection standards. Flood risk is
assessed using hazard (inundation caused by storm surge), expo- cfr = Coastal flood risk raw values in [-]
sure (population in flood zone), and vulnerability.28 The existing popexp,r,y = Expected annual affected population by coastal
level of flood protection is also incorporated into the risk flooding in [number of people]
calculation. It is important to note that this indicator represents
flood risk not in terms of maximum possible impact but rather
as average annual impact. The impacts from infrequent, extreme
flood years are averaged with more common, less newsworthy
flood years to produce the “expected annual affected population.”
Higher values indicate that a greater proportion of the popu
lation is expected to be impacted by coastal floods on average.

The thresholds are based on quantiles, with the exception of the Riverine and coastal flood risks must be evaluated and used
basins with no coastal hazard. Basins without a flood hazard are separately, as the compound risks between river and storm
given the lowest risk score, –1, and removed from the rest of the surges are not modeled. The data also assume that flood events
data set before the quantiles are calculated. are entirely independent of each other, so the impact from
overlapping flood events is not considered. Finally, the data do
RAW VALUE RISK CATEGORY SCORE not include any indirect impacts from flooding (e.g., disrupted
0 to 9 in 1,000,000 Low 0 transportation, loss of work, etc.).
9 in 1,000,000 to 7 Low-medium 1–2

100,000
7 in 100,000 to 3 in Medium-high 2–3
10,000
3 in 10,000 to 2 in 1,000 High 3–4
More than 2 in 1,000 Extremely high 4–5
lowing equation:

tor score [0–5].
22 |
Drought risk
GENERAL:
The hazard, exposure, vulnerability, risk, and no-data mask data
available at 5 × 5 arc minute resolution are averaged for each
Name Drought Risk hydrological sub-basin.
RESULTS:
In which,
Temporal resolution Annual baseline drsub-basin = Drought risk per sub-basin
SOURCE: npix = Number of pixels per sub-basin
Spatial resolution 5 x 5 arc minute grid cells drpix = Drought risk per pixel
Temporal resolution Annual
Temporal range 2000–2014 Conversion to risk categories
EXTRA: The risk categories are derived from Carrão et al. (2016):
Partner organization(s): IRC RAW VALUE RISK CATEGORY SCORE
Model Various 0.0–0.2 Low 0
Date of publication 2016 0.2–0.4 Low-medium 1–2
0.4–0.6 Medium 2–3
0.6–0.8 Medium-high 3–4
0.8–1.0 High 4–5
Description
Drought risk measures where droughts are likely to occur, the
population and assets exposed, and the vulnerability of the
lowing equation:
population and assets to adverse effects. Higher values indicate
higher risk of drought. score = 5r
The drought risk indicator is based on Carrão et al. (2016) and
is used with minimal alterations. Drought risk is assessed for Limitations
the period 2000–2014 and is a combination of drought hazard, Many of the indicators in the Aqueduct water risk framework
drought exposure, and drought vulnerability. represent a hazard. Some indicators, including drought risk,
add exposure and vulnerability. Aqueduct combines these risk
Risk = hazard × exposure × vulnerability elements into a single framework.
The methodology is explained in Carrão et al. (2016): The drought risk indicator does not consider hydrological
drought and excludes associated risks such as unnavigable rivers.
Drought hazard is derived from a non-parametric analysis of
historical precipitation deficits at the 0.5 [degree resolution]; Other Aqueduct risk categories are typically skewed toward
drought exposure is based on a non-parametric aggregation of the higher side, with the category “extremely high” as the top
gridded indicators of population and livestock densities, crop category. The drought risk indicator has not been interpreted yet
cover and water stress; and drought vulnerability is computed as and is therefore presented at a low–high scale instead of low–
the arithmetic composite of high level factors of social, economic extremely high.
and infrastructural indicators, collected at both the national and
See Carrão et al. (2016) for limitations of the different risk ele-
sub-national levels.
ments (hazard, exposure, vulnerability) and the input data sets.

Untreated connected wastewater
GENERAL:
on country-level connection and treatment rates into one data
set through a hierarchical methodology. The data include the
Name Untreated Connected Wastewater percentage of households connected to sewerage systems (percent
Subgroup Physical risk quantity connected), and the percentage of wastewater connected left
untreated (i.e., not treated using primary, secondary, or tertiary
Risk Element R=HxExV treatments) (percent untreated).
RESULTS:
The calculation is based on the Environmental Performance
Spatial resolution Country Index’s Wastewater Treatment (WWT) indicator
Temporal resolution Annual baseline (Wendling et al. 2018):
SOURCE: WWT = percent treated to at least primary * percent connected

Spatial resolution Country
WWT examines the performance of wastewater treatment
Temporal resolution Annual (Wendling et al. 2018). The untreated, connected wastewater
Temporal range 2000–2010 indicator reverses the WWT to instead examine the hazard:
EXTRA:
Partner organization(s): IFPRI, Veolia

Model Various
Date of publication 2016 In which,
UCW = Unimproved/connected wasterwater

Description raw value in [%]
Untreated connected wastewater measures the percentage of c = Percent connected wastewater in [%]
domestic wastewater that is connected through a sewerage u = Percent untreated wastewater in [%]
system and not treated to at least a primary treatment level.
Wastewater discharge without adequate treatment could expose Conversion to risk categories
water bodies, the general public, and ecosystems to pollutants
such as pathogens and nutrients. The indicator compounds The risk thresholds are based on quantiles, with the exception of
two crucial elements of wastewater management: connection the “low to no wastewater connected” threshold. All data
and treatment. Low connection rates reflect households’ lack marked in this category are given the highest risk score and
of access to public sewerage systems; the absence of at least removed from the rest of the data set before the quantiles
primary treatment reflects a country’s lack of capacity (infra are calculated.
structure, institutional knowledge) to treat wastewater. Together
these factors can indicate the level of a country’s current capacity
to manage its domestic wastewater through two main pathways: <30% Low 0
extremely low connection rates (below 1 percent), and high 30–60% Low-medium 1–2
connection rates with little treatment. Higher values indicate
higher percentages of point source wastewater discharged 60–90% Medium-high 2–3
without treatment. 90–100% High 3–4
100% Extremely high 4–5
Calculation
Low to no wastewater 5
Sewerage connection and wastewater treatment data come connected
from a white paper published by the International Food Policy
Research Institute (IFPRI) and Veolia (Xie et al. 2016). In
brief, Xie et al. aggregate three of the leading research papers
24 |
The raw values are remapped to a 0–5 scale using the fol- Limitations
lowing equation:
Important sources of water pollution, such as industrial waste
and agricultural runoff, are not included. Wastewater that may
be treated on-site, such as with private septic tanks, is also not
captured due to a lack of available data. In addition, the severity
of water pollution, which depends on the magnitude of loadings
of pollutants and dilution capacity of receiving water bodies,
is not represented (from a 2017 personal communication with
Xie). This indicator also does not account for all water pollution
sources, as it is focused primarily on household connection rates.

tor score [0–5].

Coastal eutrophication potential
indicate higher levels of excess nutrients with respect to silica,
GENERAL:
creating more favorable conditions for harmful algal growth
Name Coastal Eutrophication Potential and eutrophication in coastal waters downstream.
Subgroup Physical risk quality
Calculation
The calculation described below is based on Billen and Garnier’s
RESULTS:
(2007) Indicator of Coastal Eutrophication Potential (ICEP)
Spatial resolution Hydrological sub-basin (HydroBASINS 6) methodology. The nutrient data come from Bouwman et al.
(2015). In short, the data are based on the Global NEWS 2
Temporal resolution Annual baseline
model (Mayorga et al. 2010) and aligned to Simulated Topo-
SOURCE: logical Network basins (Vörösmarty et al. 2000). The NEWS
Spatial resolution Simulated Topological Network (STN) 2 model uses biophysical, natural, and anthropogenic (both
point and nonpoint) nutrient sources, along with in-watershed
Temporal resolution Annual and in-river removal processes, to derive global nutrient yields
Temporal range 2000 (Mayorga et al. 2010). Total N and P fluxes are calculated by
summing NEWS 2 nutrient yield data for dissolved organic,
EXTRA:
dissolved inorganic, and particulate nutrients. Si fluxes are
Partner organization(s): Utrecht University, Washington State simply the dissolved inorganic Si yields in the basin.
University
The calculation is based on the Redfield molar ratio (C:N:P:Si =
Model NA
106:16:1:20), which is a representation of the approximate
Date of publication 2016 nutrient requirement of marine diatoms (Billen and
Garnier 2007):
Additional data source STN Basins

Input spatial resolution 30 x 30 arc seconds
Input temporal rance 2000 In which,
Source Vörösmarty et al. (2000)
CEP = Coastal eutrophication potential [kg
C-equivalent km2/day]
Description
jn = Mean flux of total nitrogen delivered at the outlet of the
Coastal eutrophication potential (CEP) measures the potential river basin [kg N/km2/yr]
for riverine loadings of nitrogen (N), phosphorus (P), and silica
jp = Mean flux of total phosphorus delivered at the outlet of
(Si) to stimulate harmful algal blooms in coastal waters. The
the river basin [kg P/km2/yr]
CEP indicator is a useful metric to map where anthropogenic
activities produce enough point-source and nonpoint-source jSi = Mean flux of dissolved silica delivered at the outlet of
pollution to potentially degrade the environment. When N and the river basin [kg Si/km2/yr]
P are discharged in excess over Si with respect to diatoms, a N = Molar mass of nitrogen [14g/mol]
major type of algae, undesirable algal species often develop. The
stimulation of algae leading to large blooms may in turn result Si = Molar mass of silica [28g/mol]
in eutrophication and hypoxia (excessive biological growth and P = Molar mass of phosphorus [14g/mol]
decomposition that reduces oxygen available to other organ-
isms). It is therefore possible to assess the potential for coastal A negative value indicates that silica is present in excess over the
eutrophication from a river’s N, P, and Si loading. Higher values limiting nutrient and thus suggests the absence of eutrophica-
tion. A positive value indicates an excess of nutrients over the
26 |
potential for diatom growth, suggesting suitable conditions for The raw values are remapped to a 0–5 scale using the fol-
the growth of harmful algae (Garnier et al. 2010). lowing equation:
As a final step, the results are aggregated to HydroBASIN level

6 to align the indicator with the remainder of the framework.

The thresholds used to convert raw values into risk scores are
based on the suggested risk categories of the Transboundary
Water Assessment Programme (TWAP) for ICEP (IOC-
UNESCO and UNEP 2016), with one adjustment: the
boundary between TWAP’s low and medium categories was
increased from –1 to 0 to better reflect the elevated risk warn-
ing in Aqueduct.
RAW VALUE RISK CATEGORY SCORE tor score [0–5].
<-5 Low 0–1

Limitations
-5–0 Low-medium 1–2
Eutrophication can also impact freshwater, but a global data
0–1 Medium-high 2–3 set for freshwater eutrophication potential is not currently
available. Therefore, the indicator does not reflect the risk of
1–5 High 3–4
eutrophication upstream of the coastal zone. In addition, the
>5 Extremely high 4–5 index calculation does not account for shifts in seasonality or the
characteristics of the receiving water body.

Unimproved/No drinking water
are summed to represent the percentage of the population
GENERAL:
with access to improved drinking water. The improved rate is
Name Unimproved/No Drinking Water then inverted into the unimproved/no access rate by subtracting
Subgroup Regulatory and reputational risk improved from 100 percent. This is done for the national, rural,
and urban averages in each country. The national average is used
Risk Element R=HxExV to fill in any missing rural or urban averages.
RESULTS:
The unimproved/no access rate is matched to each Aqueduct
Spatial resolution Hydrological sub-basin (HydroBASINS 6) geometry (intersect of states, HydroBASIN 6, and aquifers; see
Temporal resolution Annual baseline 6.1) using the International Organization for Standardization
(ISO) codes provided by the Database of Global Administrative
SOURCE:
Areas (GADM) (“GADM Metadata” n.d.).
Spatial resolution Country (rural/urban)
Rural and urban populations are calculated for each Aqueduct
Temporal resolution Annual geometry. Rural and urban populations come from a gridded
Temporal range 2015 2010 population data set produced by the Netherlands Envi-
ronmental Assessment Agency (PBL) (van Vuuren et al. 2007).
EXTRA:
The gridded population data set is parsed into rural and urban
Partner organization(s): JMP populations using a 2010 urban extent data layer (van Huijstee
et al. 2018) and then summed by Aqueduct geometry.
Model NA
Date of publication 2017 The rural and urban unimproved/no access rate is multiplied by
the rural and urban populations, respectively, to find the number of
people with unimproved/no access to drinking water in each
Additional data source Urban extents Gridded population
Aqueduct geometry. The rural and urban totals are then summed
and aggregated to the HydroBASIN 6 scale, along with total
Input spatial resolution 30 arc seconds 30 arc seconds population. Finally, the raw score—the weighted percentage of
Input temporal resolution Annual Annual population with unimproved/no access per HydroBASIN 6—
is calculated:
Input temporal rance 2010 2010
Source van Huijstee et al. (2018) van Vuuren et al. (2007)
Description
Unimproved/no drinking water reflects the percentage of the In which,
population collecting drinking water from an unprotected dug
well or spring, or directly from a river, dam, lake, pond, stream, UDW = Unimproved/no drinking water raw value in [-]
canal, or irrigation canal (WHO and UNICEF 2017). Specifi- rrural = Rural unimproved/no access to drinking
cally, the indicator aligns with the unimproved and surface water water rate in [-]
categories of the Joint Monitoring Programme ( JMP)—the
lowest tiers of drinking water services. Higher values indi rurban = Urban unimproved/no access to drinking
cate areas where people have less access to safe drinking water rate in [-]
water supplies. poprural = Rural population in [number of people]
popurban = Urban population in [number of people]
Calculation
poptot = Total population in [number of people]
Data for this indicator come from the 2015 drinking water
access rates published by JMP (WHO and UNICEF 2017).
The statistics from JMP’s “at least basic” and “limited” fields
28 |

The risk thresholds are based on Aqueduct 2.1 (Gas- The unimproved/no drinking water indicator is presented at
sert et al. 2014). a finer resolution than originally published by JMP under the
assumption that access rates among rural and urban communi-
RAW VALUE RISK CATEGORY SCORE ties are consistent throughout a country. The methodology
<2.5% Low 0–1 fails to account for regional and local differences in access
within countries.
2.5–5.0% Low-medium 1–2
5.0–10.0% Medium-high 2–3
10.0–20.0% High 3–4
lowing equation:

tor score [0–5].

Unimproved/No sanitation
Name Unimproved/No Drinking Sanitation Data for this indicator come from the 2015 sanitation access
rates published by JMP (WHO and UNICEF 2017). Statistics
Subgroup Regulatory and reputational risk from JMP’s “at least basic” and “limited” fields are summed
Risk Element R=HxExV to represent the percentage of the population with access to
improved sanitation. The improved rate is then inverted into the
RESULTS:
unimproved/no access rate by subtracting improved from 100
Spatial resolution Hydrological sub-basin (HydroBASINS 6) percent. This is done for the national, rural, and urban averages
in each country. The national average is used to fill in any miss-
Temporal resolution Annual baseline
ing rural or urban averages.
SOURCE:
The unimproved/no access rate is matched to each Aqueduct
Spatial resolution Country (rural/urban)
geometry (intersect of states, HydroBASINS 6, and aquifers; see
Temporal resolution Annual 6.1) using the International Organization for Standardization
(ISO) codes provided by GADM (“GADM Metadata” n.d.).
Temporal range 2015
EXTRA: Rural and urban populations are calculated for each Aqueduct
geometry. Rural and urban populations come from a gridded
Partner organization(s): JMP
2010 population data set produced by PBL (van Vuuren et al.
Model NA 2007). The gridded population data set is parsed into rural and
urban populations using a 2010 urban extent data layer (van
Huijstee et al. 2018), and then summed by Aqueduct geometry.
The rural and urban unimproved/no access rate is multiplied by

Additional data source Urban extents Gridded population the rural and urban populations, respectively, to find the number
Input spatial resolution 30 arc seconds 30 arc seconds of people with unimproved/no access to sanitation in each
Aqueduct geometry. The rural and urban totals are then summed
Input temporal resolution Annual Annual and aggregated to the HydroBASINS 6 scale, along with total
Input temporal rance 2010 2010 population. Finally, the raw score—the weighted percentage of
population with unimproved/no access per HydroBASINS 6—
Source van Huijstee et al. (2018) van Vuuren et al. (2007)
is calculated:
Description
Unimproved/no sanitation reflects the percentage of the popula-
tion using pit latrines without a slab or platform, hanging/
bucket latrines, or directly disposing human waste in fields,
In which,
forests, bushes, open bodies of water, beaches, other open spaces,
or with solid waste (WHO and UNICEF 2017). Specifically,
USA = Unimproved/no sanitation raw value in [-]
the indicator aligns with JMP’s unimproved and open defecation
categories— the lowest tier of sanitation services. Higher values rrural = Rural unimproved/no access to sanitation rate in [-]
indicate areas where people have less access to improved rurban = Urban unimproved/no access to drinking
sanitation services. water rate in [-]
poprural = Rural population in [number of people]
popurban = Urban population in [number of people]
poptot = Total population in [number of people]
30 |

The risk thresholds are based on Aqueduct 2.1 (Gas- Unimproved/no sanitation is presented at a finer resolution
sert et al. 2014). than is originally published by JMP under the assumption that
access rates among rural and urban communities are consistent
RAW VALUE RISK CATEGORY SCORE throughout a country. The methodology fails to account for
<2.5% Low 0–1 regional and local differences in access within countries.
2.5–5.0% Low-medium 1–2

5.0–10.0% Medium-high 2–3
10.0–20.0% High 3–4
lowing equation:

tor score [0–5].

Peak RepRisk country ESG risk index
GENERAL:
index (RepRisk n.d.). The results of the screening process are
delivered to the RepRisk team of analysts, who are responsible
Name Peak RepRisk Country EDG Risk Index for curating and analyzing the information. They hand select the
Subgroup Regulatory and reputational risk items, give each risk incident a score (based on severity, source,
and novelty), and write a risk summary. Before the risk incident
Risk Element R=HxExV is published, a senior analyst runs a quality check to ensure that
RESULTS: the process has been completed in line with RepRisk’s strict,
rules-based methodology. After the senior analyst has given her
Spatial resolution Country
or his approval, the final step in the process, the quantification
Temporal resolution Annual baseline of the risk, is performed through data science. The Peak RepRisk
SOURCE: country ESG risk index takes into consideration the impact of a
country’s risk incidents within the last two years and the average
Spatial resolution Country of a country’s Worldwide Governance Indicators. The data used
Temporal resolution Annual in Aqueduct 3.0 cover October 2016 through October 2018. To
learn more about RepRisk, please visit https://www.reprisk.com/
our-approach or contact RepRisk.
EXTRA:
Partner organization(s): RepRisk Conversion to risk categories

Model NA The risk thresholds are based on guidance from RepRisk
(RepRisk n.d.).
<25% Low 0–1

Description
25–50% Low-medium 1–2
The Peak RepRisk country ESG risk index quantifies business
conduct risk exposure related to environmental, social, and gov- 50–60% Medium-high 2–3
ernance (ESG) issues in the corresponding country. The index 60–75% High 3–4
provides insights into potential financial, reputational, and com-
pliance risks, such as human rights violations and environmental >75% Extremely high 4–5
destruction. RepRisk is a leading business intelligence provider
that specializes in ESG and business conduct risk research for
companies, projects, sectors, countries, ESG issues, NGOs, and
lowing equation:
more, by leveraging artificial intelligence and human analysis
in 20 languages. WRI has elected to include the Peak RepRisk
country ESG risk index in Aqueduct to reflect the broader
regulatory and reputational risks that may threaten water
quantity, quality, and access. While the underlying algorithm is
proprietary, we believe that our inclusion of the Peak RepRisk
country ESG risk index, normally unavailable to the public, is
a value-add to the Aqueduct community. The peak value equals
the highest level of the index in a given country over the last two
years. The higher the value, the higher the risk exposure.
RepRisk screens over 80,000 media, stakeholder, and third-party
sources daily to identify and analyze ESG-related risk incidents
and quantify them into the Peak RepRisk country ESG risk
32 |
COUNTRY AND STATE

AGGREGATION
The updated country and state aggregations of the baseline and Inputs
future projection data follow the weighted aggregation method-
ology published in Gassert et al. 2013. We provide an excerpt of This section details the methodology to aggregate Aqueduct
their rationale and methods here for your convenience: indicators from the sub-basin level to state and country borders.
The methodology requires three inputs: gridded weights to
define where water is being used, an Aqueduct indicator score to
Most water-related decisions are made across political or rescale, and target regions (state and country borders).
administrative boundaries, creating a demand for simple and
robust water information to support decision making at the
Gridded weights
administrative level. Governments devise policies to manage
water resources within their borders and can use country indica- The gross demand data is used to indicate where human need
tors as a statistic against which to benchmark themselves. Many for water is greatest—it is also where socioeconomic dependency
financial institutions divide their portfolios by country, and thus for water is most critical (Gassert et al. 2013). Areas with higher
require national-level water data to evaluate portfolio exposure water demand will have more influence over the final aggre-
to water-related risks. gated score. We use the following gridded demand data sets
from PCR-GLOBWB 2:
However, accurately assessing the state of water resources across
administrative boundaries is a significant challenge; and simple, Baseline gross demand for four sectors: domestic, industrial,
comparable, and robust water information to support decision irrigation, and livestock. The 4 gridded data sets are available for
making at that level remains sparse. The spatial variation of each month between January 1960 and December 2019.
water resources complicates the development of meaningful
country and basin-level indicators. As opposed to other resources Projected gross demand for four sectors—domestic, industrial,
such as forests and agricultural lands, whose stationarity simpli- irrigation, and livestock—available for each month between
fies measurement and management, water cannot be accounted January 1960 and December 2100. Domestic, industrial, and
for by using only administrative boundaries. Even within small livestock demand vary among the three future scenarios, mean-
administrative regions, hydrological conditions may vary from ing there are 3 data sets per demand type for every time step.
lush rainforest to dry prairie. Transboundary lakes and rivers Irrigation29 also varies by GCM (5 total) within each scenario,
further complicate water accounting, as special efforts must be meaning there are (5 x 3=) 15 irrigation data sets available for
made to avoid double counting the water supply, they provide every time step. See “Future projections” for more details on the
across regions. future projections data.
The Aqueduct Water Risk Atlas (Aqueduct) first models global

water-risk indicators at a relatively granular hydrological
Aqueduct indicators
catchment scale. In this analysis WRI then employs a weighted Water stress measures the competition over water resources and
aggregation methodology that brings Aqueduct’s catchment-level is an effective metric to evaluate water-related risks at the state
information up to the country and state levels. This methodology and country levels. We take the water stress score—the normal-
addresses each of the challenges described above by starting with ized version of the raw data—for the baseline, 2030, 2050, and
indicators that were computed within basic hydrological units, 2080. See the first section of “Indicators” for more details on
and assigning spatially explicit weights to reflect the importance how the water stress score is calculated.
of the specific areas based on where water is being used. From
these calculations, WRI generated estimates of the average Target regions
level of exposure to Aqueduct’s baseline water stress indicator Our target geometries are state and country borders. We use
for all countries. GADM Level 1 data (GADM Metadata n.d.) to define the
borders. These are state borders, which inherently also contain
country borders. We intersect the GADM Level 1 data with
the HydroBASIN level 6 data (sub-basins used by Aqueduct) to
traverse between the two systems.

Figure 7 | Examples of the Three Spatially Explicit Inputs
A B C
Notes: a. Source indicators: baseline water stress; b. Gridded weights: total water withdrawal; c. Target regions: country boundaries.
Source: WRI Aqueduct.
Weighted aggregation
Processing gridded weights
First, we sum the gridded gross demand monthly layers into In the end, we have five measures of gross demand (total,
annual data. Then, we follow the same methodology described domestic, industrial, irrigation, livestock) for five time periods
in the Spatial aggregation of water use section (in 2.3.1)—we (baseline, 2014, 2030, 2050, and 2080) for the three SSP sce-
resample, convert flux to volume, and sum per target geometry. narios (for the future periods).
The only difference is we use the state/sub-basin intersect as the
target geometry rather than the sub-basins. Next, we then sum- Applying the weighted average
marize the data by the following milestone years:
We compute the average water stress score for every target
▪ Baseline: 1979–2019 (historic observed climate forcing) region (i.e., states and countries) using a weighted average
approach, where the weight is water demand. Following Gassert
▪ 2014: 1960–2014 (historic GCM climate forcing) et al. 2013, the weighted mean indicator value (sr) is found by
▪ 2030: 2015–2045 (future GCM climate forcing) multiplying the sub-basin indicator (sp) by the state-sub-basin
▪ 2050: 2035–2065 (future GCM climate forcing) weight (wp), summing, and dividing by the sum of the weights
across the entire administrative region (r).
▪ 2080: 2065–2095 (future GCM climate forcing)
To do this, we follow the Temporal Aggregation steps outlined in
“Hydrological model”—we calculate total demand by sum-
ming sectoral demand, smooth the data using a 10-year moving
average, and apply a Theil Sen regression—with two exceptions:
(1) we do not split the data by months because we are only
using annual data, and (2) we perform the temporal aggregation
of the individual sectors as well as total demand. Finally, we
bias-correct the projected demand data to the baseline using the
methodology outlined in “Future projections.” For the projected
demand, we use the median value of the five GCMs to represent
each scenario.
34 |
GROUPED AND OVERALL Table 2 | Industry or user relevance weights and

their descriptions
WATER RISK
After calculating the 13 indicators and converting them to a LEGEND WEIGHT INTERPRETATION
uniform 0–5 scale, we can calculate the grouped and overall
water risks (composite indices). See Figure 1 for an over- No weight 0 Not relevant
view of the groups. Represents very low relevance to
Very low 0.25
the industry or user
Geometries Low 0.5
Represents low relevance to the
industry or user
Each of the 13 indicators is calculated at one of three different
spatial scales: hydrological sub-basin, country, or groundwater Represents medium relevance to
Medium 1
aquifer. See “results: spatial resolution” in the summary table the industry or user
of each indicator. To combine the indicators into one frame- Represents high relevance to the
High 2
work, we take the union of the three geometries. The resulting industry or user
geometries are a unique combination of a hydrological basin,
Represents very high relevance
groundwater aquifers, and an administrative boundary. Very high 4
to the industry or user
Weighted aggregation Source: WRI.
The subgroups (physical risk quantity, physical risk quality, and

regulatory and reputational risk) and overall risks are calculated to water risks and the subjective nature of indicator weights
by taking a weighted average of the indicators that belong to made this technique an ideal fit. The results of the default
each subgroup. weighting scheme can be found in the first column of Table 3.
Exposure to water-related risks varies with the characteristics Industry-specific weighting scheme
of water users. To obtain aggregated water risk scores, users can
modify the weight of each indicator to match their exposure Additionally, we developed preset weighting schemes for nine
to the different aspects of water risk. There are five weights, or industry sectors on the basis of information provided in corpo-
descriptors of relevance, on a base 2 exponential scale. This is rate water disclosure reports and input from industry experts
preferred over a linear scale because of the human tendency to to reflect the risks and challenges faced by each water-intensive
categorize intensity by orders of magnitude of difference (Tri- sector. For each industry, we modified the default indicator
antaphyllou 2010). Users can also exclude indicators completely weights on the basis of the relative importance of each indicator
from aggregation. See Table 2 for an overview of the weights. to the industry using information disclosed by companies on
their exposure to, and losses from, water-related risks. To vali-
Users have three options for the weighting scheme: default, date the industry-sector preset weighting schemes, we presented
industry-specific, or custom. preliminary weighting schemes to industry representatives
from the nine sectors and solicited feedback on the relative
Default weighting scheme importance of each indicator for their sector. The results can be
found in Table 3.
To determine a default set of indicator weights, we used input
from six staff water experts following the principles of the
Delphi technique. This technique uses a series of intensive
Custom weighting scheme
questionnaires interspersed with controlled opinion feedback In the online tool, users can specify their own custom
to obtain the most reliable consensus of opinion from a group weighting scheme.
of experts (Rowe and Wright 1999). The Delphi technique is
intended for use in judgment situations; that is, ones in which Using the weighting schemes, grouped water risk scores can be
pure model-based statistical methods are not practical or pos- calculated. The relative weight of each indicator is illustrated in
sible because of the lack of appropriate historical data, and thus Figure 7. The definition for each subgroup is listed below:
some form of human judgment input is necessary (Dalkey and
Helmer 1963). The lack of consistent information on exposure

Physical Risk Quantity and three indicators in the quality and regulatory and reputa-
tional groups, respectively. This makes these groups sensitive to
Physical Risk Quantity measures risk related to too little or
errors in the underlying data. We provide industry and custom
too much water by aggregating all selected indicators from the
weighting to mitigate this limitation, but this requires the user
physical risk quantity category. Higher values indicate higher
to understand the data. The framework’s water quality indicators
water quantity risks.
do not reflect the full range of water quality threats but focus
Physical Risk Quality on nutrient pollution. The framework does not endorse framing
water-quality solutions solely around coastal eutrophication or
Physical Risk Quality measures risk related to water that is unfit municipal wastewater. A fourth limitation of the framework
for use by aggregating all selected indicators from the Physi- approach is the mixing of risk types. The framework is incon-
cal Risk Quality category. Higher values indicate higher water sistent in including the exposure and vulnerability layers for
quality risks. all indicators.
Regulatory and Reputational Risk In addition to the limitations of the framework approach,
each indicator’s baseline and future projections come with its
Regulatory and Reputational Risk measures risk related to
own limitations. For the indicator-and-projection-specific
uncertainty in regulatory change, as well as conflicts with the
limitations, please see the relevant sections above and the
public regarding water issues. Higher values indicate higher
associated literature. Since many of the indicators rely on the
regulatory and reputational water risks.
PCR-GLOBWB 2 hydrological model and HydroBASINS
Finally, the three grouped water risk scores can be used to 6 (hydrological sub-basins), some of these specific limitations
determine the overall water risk score. The sums of the weights are copied below.
are used to calculate the relative contribution of each group.
Coastal sub-basins and islands in HydroBASINS 6 are often
Overall Water Risk grouped for various reasons explained in Lehner et al. (2008).
This grouping is coarse and results in inaccuracies, primarily
Overall Water Risk measures all water-related risks, by aggre- when water demand can be satisfied using remote water supply.
gating all selected indicators from the Physical Risk Quantity,
Physical Risk Quality, and Regulatory and Reputational Risk PCR-GLOBWB 2 has no means to model interbasin transfer.
categories. Higher values indicate higher water risk. Interbasin transfer happens when demand in one sub-basin
is satisfied with supply from another sub-basin that is not
upstream. Many major metropolitan areas source their water
LIMITATIONS from adjacent sub-basins. Thus, baseline water stress in a given
sub-basin may at times appear worse than it is where interbasin
Not every aspect of water risk has usable global data sets transfers are available to meet demand in that sub-basin. Alter-
enabling it to be incorporated into our framework. Certain natives to the moving window size and regression method used
important elements are partially missing from the framework, to process the PCR-GLOBWB 2 results could not be assessed
such as water management and governance.30 due to the lack of validation data.
The local social dimensions of water risks are not incorpo Direct validation of the aggregated grouped water risks and
rated into this framework and database. Policy, regulation, and overall water risk is not possible. The perception of water risk
response to water crises are paramount in estimating water risks is subjective, and robust validation methods for multi-indicator
and fully understanding their impacts. In the end, each region frameworks are unavailable.
or location’s ability to cope with water-related issues will affect
its water risk. It is crucial to understand what the Aqueduct 4.0 framework
and database can and cannot do. Aqueduct 4.0 is tailored to
Several limitations are associated with the framework (com- comparing regions on a larger scale. It has limited application at
posite index) approach. First, it requires mapping the indicators a local level. The presented results should therefore be used as a
to comparable (0–5) scale, thereby losing information such as prioritization tool, after which deeper dive assessments should
absolute values. The second limitation, linked to the first, is that be used to understand local conditions with greater accuracy.
we combined data with various spatial and temporal resolutions
and ranges into a single framework. Third, there are only two
36 |
Table 3 | Industry or user relevance weights and their descriptions
DEFAULT AGRICULTURE CHEMICALS CONSTRUCTION ELECTRIC FOOD AND MINING OIL AND SEMICONDUCTOR TEXTILE
MATERIALS POWER BEVERAGE GAS
Baseline water
1 4 4 2 2 4 4 2 1 2 2
stress
Baseline water
2 4 4 2 2 4 4 2 1 2 2
depletion
Interannual
3 0.5 2 1 2 1 1 2 1 2 2
PHYSICAL RISK QUANTITY
variability
Seasonal
4 0.5 0.5 1 0.5 2 0.5 1 0.5 1 0.5
variability
Groundwater
5 4 4 2 2 0.5 4 2 1 2 1
table decline
Riverine flood
6 1 1 4 1 2 0.5 4 1 1 1
risk
Coastal flood
7 1 1 4 1 4 0.5 4 4 1 2
risk
8 Drought risk 2 4 2 1 4 2 4 0.5 1 1
Untreated
9 connected 2 1 2 1 0.25 2 0.5 0.25 4 4
PHYSICAL RISK QUALITY
wastewater
Coastal
10 eutrophication 1 4 0.25 0.5 1 2 0.25 0 2 1
potential
Unimproved/
11 no drinking 2 2 2 1 0.25 1 4 4 1 2
REGULATORY AND REPUTATIONAL RISK
water
Unimproved/
12 2 2 2 1 0.25 1 4 4 1 2
no sanitation
Peak RepRisk
13 country ESG 0.5 0.25 2 0.5 0.25 2 4 4 2 4
risk index
Source: WRI.

Figure 8 | Indicator weights per industry
PHYSICAL RISK QUANTITY PHYSICAL RISK QUALITY REGULATORY & REPUTATIONAL RISK
0 0.2 0.4 0.6 0.8 1.0
Default
Agriculture
Chemicals
Construction
materials
Electric Power
Food & Beverage
Mining
Oil & Gas
Semiconductor
Textile
Baseline water stress Seasonal variability Coastal flood risk Untreated connected wastewater Unimproved/no drinking water
Baseline water depletion Groundwater table decline Drought risk Coastal eutrophication potential Unimproved/no sanitation
Interannual variability Riverine flood risk Peak RepRisk country ESG risk index
Notes: Weights are based on data availability. Masked or NoData values are excluded from the aggregated weighting. Please see the online tool for the results. The data are also
available for download.
Source: WRI.
38 |
APPENDIX A: DEMAND, Figure A1 | Schematic of demand, gross and net

withdrawal, and return flow
WITHDRAWAL, AND RETURN FLOW
This appendix describes the hydrological terminology used in
PCR-GLOBWB 2. An overview is shown in Figure A1. PCR-GLOBWB
2 determines water demand. Withdrawal is demand limited by
available water.
Withdrawal consists of two components: Consumptive withdrawal WATER CONSUMPTION

and nonconsumptive withdrawal. Gross withdrawal refers to
consumptive plus nonconsumptive withdrawal. Net withdrawal refers
to only the consumptive withdrawal.
The nonconsumptive withdrawal will return to the water body, usually

downstream, and is also referred to as return flow.
WATER WITHDRAWALS RETURN FLOWS
WATER RESOURCES
(e.g., river, aquifer, lake, ocean)
Source: WRI.

APPENDIX B: GEOGRAPHIC
CONVERSION TABLE
Table B1 is intended to provide a quick and
approximate sense of scale
Table B1 | Common arc lengths
ARC LENGTH DECIMAL DEGREES DISTANCE AT EQUATOR (KM) APPROXIMATE DISTANCE AT EQUATOR (KM)
360 arc degrees 360 40,030.17 40,000

1 arc degree 1 111.19 110
30 arc minutes 0.5 55.60 55
5 arc minutes 0.08333 9.27 10
1 arc minute 0.016667 1.85 2
30 arc seconds 0.008333 0.93 1
15 arc seconds 0.004167 0.46 .5
Source: WRI
40 |
APPENDIX C: PCR-GLOBWB 2
Water stress, water depletion, interannual variability, seasonal Model structure
variability, groundwater table decline, and elements of the flood
risk indicators are all based on the PCRaster Global Water PCR-GLOBWB 2 is a grid-based, modular global hydrological model.
Balance 2 model (PCR-GLOBWB 2) (Sutanudjaja et al. 2018; The world is represented by a 4,320 × 2,610 grid with a resolution
Sutanudjaja et al. 2023). of 5 × 5 arc minutes. For each of these cells, the model contains the
following modules:
This appendix covers the basic model structure of PCR-GLOBWB 2
and the settings used for the Aqueduct run. ▪ Meteorological forcing
For baseline water stress, baseline water depletion, interannual ▪ Land surface
variability, and seasonal variability we used a setup with default
▪ Groundwater
groundwater configuration. We will refer to this run as the default
PCR-GLOBWB 2 run. ▪ Surface water routing
For the groundwater table decline indicator, we used a setup of PCR-

▪ Irrigation and water use
GLOBWB 2 with an advanced representation of groundwater based See Figure C1 for a schematic representation of the model.
on MODFLOW (de Graaf et al., 2017). We will refer to this setup as the
PCR-GLOBWB 2 + MODFLOW run. Meteorological forcing module
To model key weather elements that affect hydrology, the
Digital elevation model meteorological forcing of PCR-GLOBWB 2 uses daily time series of
The starting point of almost any hydrological model and analysis spatial fields of precipitation, temperature, and reference evaporation.
is a digital elevation model (DEM). The DEM will determine the
The default run is forced using data from two data sources: Global
runoff direction; that is, the way the water flows. Aqueduct uses the
Soil Wetness Project Phase 3 (GSWP3) v1.09 (Dirmeyer et al.
same DEM as PCR-GLOBWB 2 and is a combination of the 30 ×
2006) for the period 1960–78 and W5E5 (which merges WATCH
30 arc second HydroSheds data (Lehner et al. 2008) with the 3 ×3
Forcing Data with ERA531 (WFDE5)) to extend the analysis to 2019
arc second Multi-Error-Removed Improved-Terrain Hydro Digital
(Lange et al. 2021).
Elevation Model (MERIT Hydro DEM) (Yamazaki et al. 2019). Lakes
and wetlands from the Global Lakes and Wetlands Database (GLWD) Reference evapotranspiration is calculated using Penman-Monteith,
(Lehner and Döll 2004a) are extracted. Finally, reservoirs and dams according to the FAO guidelines (Allen et al. 1998).
from the Global Reservoir and Dam (GRanD) database have been
For groundwater, the PCR-GLOBWB 2 + MODFLOW run is forced
used (Lehner et al. 2011). In short, lakes and reservoirs are part of
using combined Climatic Research Unit (CRU) and Era-Interim (Harris
PCR GLOBWB 2’s drainage network, meaning their storage is actively
et al. 2014; Dee et al. 2011). Although the model ran for 1959–2015,
updated through the routing network. Lake outflow uses a standard
only the results for 1990–2014 have been used to calculate the
storage-outflow relationship (Bos, 1989); reservoir flow follows a
groundwater table decline indicator (Verdin and Greenlee 1996).
release strategy based on the average passing discharge (limited by
the minimum and maximum storage per each reservoir’s construction
year). Lake and reservoir storage is subject to abstraction from both
Land surface model
evaporation and human withdrawals. The result is a hydrologically This is the central module of PCR-GLOBWB 2 and connects directly
corrected data set of elevation, resampled to the PCRGLOBWB to all other modules. It consists of multiple vertically stacked layers:
resolution of 5 × 5 arc minutes (approximately 10 km at the equator). canopy, snow, soil layer 1 (S1), and soil layer 2 (S2). See Figure C1.
There are vertical fluxes between the stacked layers (e.g., S1 to S2
Local drainage direction and vice versa), as well as with the climate forcing module (e.g.,
precipitation and evaporation) and the groundwater module (e.g., S2
The local drainage direction, or the way water flows from one grid
to groundwater). Furthermore, there are horizontal fluxes to the runoff
cell to the next, is derived from the DEM and assumes a strictly
convergent flow. This means that in PCR-GLOBWB 2 and Aqueduct,
bifurcations and river deltas are modeled as one stream instead of
splitting rivers.

Figure C1 | PCR-GLOBWB 2 schematic overview
IRRIGATION AND
WATER USE MODULE
Irrigation Livestock Industrial Domestic
METEOROLOGICAL
FORCING MODULE
Precipitation Evaporation Qchannel
Desalination
Canopy
Snow
Qdr
LAND SURFACE
MODULE
Soil 1
Soil 2
Qsf Reservoirs
Qbf
GROUNDWATER SURFACE WATER
Groundwater
MODULE ROUTING MODULE
lnf
Surface water withdrawal Return flows from surface water use Groundwater abstraction Return flows from groundwater use surface
Note: “Schematic overview of a PCR-GLOBWB 2 cell and its modeled states and fluxes. S1, S2 (soil moisture storage), S3 (groundwater storage), Qdr (surface runoff—from rainfall
and snowmelt), Qsf (interflow or stormflow), Qbf (baseflow or groundwater discharge), and Inf (riverbed infiltration from to groundwater). The thin red lines indicate surface water
withdrawal, the thin blue lines groundwater abstraction, the thin red dashed lines return flows from surface water use, and the thin dashed blue lines return flows from groundwater
use surface. For each sector, withdrawal − return flow = consumption. Water consumption adds to total evaporation. In the figure, the five modules that make up PCR-GLOBWB 2 are
portrayed on the model components” (Sutanudjaja et al. 2018).
Source: Based on raw data from Sutanudjaja et al. (2018), modified/aggregated by WRI.
42 |
module. Within each grid cell, subgrid variability is modeled using a Finally, additional monthly vegetation properties, including leaf area
land-use class approach. This means that each grid cell is assigned a index (LAI) and crop factors, are derived from the MIRCA 2000 data
fraction of four land-use classes: set (Portmann et al. 2010) and the Global Crop Water Model (Siebert
and Döll 2010). For each of the four land-use classes, the following
▪ Tall natural vegetation evaporative fluxes are defined:
▪ Short natural and nonnatural (rainfed crops) vegetation

▪
▪
Interception evaporation
▪
Nonpaddy-irrigated crops
▪
Bare soil evaporation
▪
Paddy-irrigated crops (e.g., wet rice)
Snow sublimation
For instance, a grid cell might consist of 20 percent tall natural
vegetation, 25 percent short vegetation, 40 percent nonpaddy- ▪ Vegetation-specific transpiration
irrigated crops, and 15 percent paddy-irrigated crops (total 100 Another main building block in the land surface model is runoff
percent). Soil and vegetation parameters are obtained for each class and infiltration modeling. There are two runoff components in the
and for each grid cell. Hence the soil and vegetation conditions are land surface module: (1) direct runoff from soil layer 1 combined
spatially distributed. with snowmelt from the snow layer and (2) stormflow19 runoff
from soil layer 2.
The Global Land Cover Characteristics Data Base, version 2.0 (“GLCC
2.0” 2010) and land surface parameter data set (Hagemann 2002) are Direct and stormflow runoff are determined by excess infiltration
used to assign the four land-use classes to each 5 arc minute grid according to the advanced ARNO scheme approach (Todini 1996;
cell as well as obtain a few soil and vegetation parameters. Hagemann and Gates 2003). This scheme determines which fraction
will transfer vertically (infiltration) or horizontally (runoff).
For each of the four land-use classes and for each soil layer (S1 and
S2), the remaining soil parameters are defined using SoilGrids250
(Hengl et al., 2017) and the WISE data set on global soil properties
(Batjes 2012). SoilGrids250 was used to replace the Digital Soil Map
of the World (Nachtergaele et al. 2009) because of its finer resolution.

APPENDIX D: DELTA SUB-BASINS
The underlying digital elevation models of PCR-GLOBWB 2 and To classify HydroBASINS into delta regions, a spatial join was
HydroBASINS assume a strictly convergent flow, which in some performed between HydroBASIN level 6 and GDBD with delta
cases leads to erroneous results. Rivers will sometimes bifurcate, classifier information. The HydroBASINS that intersect the GDBD
especially in flat delta areas. Available water resources and water delta basins are put on a shortlist for further inspection.
withdrawal are pooled within each sub-basin. Therefore, sub-basins
The second step is to count the number of separate GDBD
that are part of the same delta need to be grouped and assumed to
streams in each sub-basin. Multiple streams are an indication for
belong to a common hydrological unit.
delta sub-basins.
The previous version of Aqueduct uses sub-basins derived from the
As a third step, each shortlisted delta sub-basin is manually
Global Drainage Basin Database (GDBD) (Masutomi et al. 2009).
checked by comparing the shortlisted sub-basins with all water
By default, GDBD does not contain information about which basins
bodies extracted from OpenStreetMaps and the flow direction
were grouped. According to the author of GDBD, it is not possible
and flow accumulation of both HydroBASINS and PCR-GLOBWB
to replicate the delta grouping using the HydroBASINS data set.
2 (OpenStreetMap contributors 2018; Lehner and Grill 2013;
Therefore, additional information regarding the 67 delta basins in
Sutanudjaja et al. 2018).
GDBD was obtained directly from the authors and joined to the
original database. In total, 196 HydroBASINS are grouped into 63 delta basins. Eighty-
nine delta regions have been examined. A column containing delta
The process of finding and grouping delta HydroBASINS includes
information is added to the final Aqueduct database.
a semiautomated way to create a shortlist for potential delta sub-
basins, and a manual step to ensure the correct classification.
44 |
ENDNOTES
1. We used time series of groundwater heads. Groundwater head is 12. Aqueduct 3.0 rescaled data to a 30 arc second resolution. How-
a measure of pressure and can be linked to groundwater tables. ever, with the inclusion of the 15 future projections runs, Aqueduct
See the groundwater table decline indicator for more information. 4.0 had about 43x more data to process; therefore, we experi-
mented with different resampling sizes, assessing both the run
2. See Appendix A for the terminology. time and change in results. We found that by using a 1 arc minute
resolution, we could run the spatial reduction twice as fast while
3. Gross demand is the maximum potential water requirement,
introducing minimal changes in the results.
compared to withdrawals, which represents the actual amount of
water used. For example, a given year may use fewer withdrawals 13. An inflow point is where the stream enters the catchment for the
if heavy precipitation satisfied the water requirements. See Ap- first time and is completely upstream of the catchment (there is
pendix A for more details. no presence of the catchment in its upstream path); an output
point is where the stream leaves the catchment, and there is no
4. Water use estimates for domestic, industrial, and livestock sectors
presence of the catchment in its downstream path.
are inputs to PCR-GLOBWB 2. Irrigation is the only sectoral de-
mand that is an output of PCR-GLOBWB 2. Irrigation demand is a 14. LDD is at the same 5 x 5 arc minute resolution as discharge data.
function of crop extents and climatic conditions such as tempera-
ture and humidity, and is therefore calculated on-the-fly during 15. Rasterization is the process of turning a polygon into a grid (i.e., a
the modeling process (to adjust to the climate forcing data). For series of squares). Polygons may have rounded edges, which can
example, a crop field will require more water during a hotter sum- get lost in the rasterization process. Location boundaries, such as
mer than a cooler summer. sub-basins and state borders, are polygons, whereas hydrological
data, such as discharge, exist in a gridded format.
5. Consumption from irrigation is calculated by taking the portion of
precipitation lost to evaporation. 16. PCR-GLOBWB 2 uses a 5 x 5 arc minute spatial resolution,
whereas HydroBASINS sub-basins are derived from a much
6. We assume that all water withdrawn for livestock is consumed. finer digital elevation model (3 ×3 arc seconds) resampled to 15
Therefore, livestock net consumption equals livestock gross de- x 15 arc second resolution. The result is that the larger 5 × 5 arc
mand. minute grid cells might (partially) overlap adjacent sub-basins,
thereby erroneously making water available to that sub-basin.
7. Interflow is the flow of water in the unsaturated ground below the
surface but above the groundwater level. It discharges to above- 17. Discharge represents cumulative flow. Although a false output
ground streams rather than infiltrates into groundwater. point may temporarily show flow into an adjacent sub-basin, it is
not contributing to that basin’s accumulated flow because they
8. The PCR-GLOBWB2 output is called “runoff”. It is the sum of direct
are not hydrologically connected.
runoff, interflow, and base flow within the catchment. Consump-
tion is not removed. For more detail, see Appendix C. 18. Years prior to 1979 (1960–1978) are no longer included in the
baseline because they use a different reference climate forcing
9. The PCR-GLOBWB2 output is called “river discharge”. It is the
dataset than the more recent years. Data from 1979 to 2019 use
accumulative flow of direct runoff, interflow and base flow, minus
W5E5 (Lange et al. 2021) for its observed atmospheric climate
water consumption (i.e., surface water abstraction—return flow).
forcing data. Data from 1960 to 1978 use Global Soil Wetness
Evaporation from river and infiltration from river to groundwater
Project Phase 3 (GSWP3) v1.09 (Dirmeyer et al. 2006).
are also removed from the accumulated flow. For more detail, see
Appendix C. 19. The irrigation is especially sensitive to climate forcing and evapo-
transpiration algorithm limitations.
10. Water that is being transported from one basin to another other
than natural flow. 20. The Theil-Sen regressor is a non-parametric method for linear
regression that uses the median slope between points to estimate
11. Using the World Eckert IV projection.
trend (Sen 1968). It is particularly powerful at handling outliers,
which is why we use it over the Ordinary Least Sum regressor.

21. Baseline water stress and baseline water depletion use a fraction
approach. When the denominator is very close to zero, the value
REFERENCES
will be extremely high and is often based on only a few data Allen, R.G., L.S. Pereira, D. Raes, M. Smith, and others. 1998. “Crop
points. Please see the “Indicator” section and the respective equa- Evapotranspiration: Guidelines for Computing Crop Water Require-
tions. ments.” FAO Irrigation and Drainage Paper 56.
22. Bias-corrected date for these GCMs are available under the ISI- “AQUASTAT.” n.d. http://www.fao.org/nr/water/aquastat/main/index.
MIP 3b protocol (ISIMIP 3 Protocol 2021). stm. Accessed February 13, 2019.
23. The computation resources for running the model and storing its Arias, P., N. Bellouin, E. Coppola, R. Jones, G. Krinner, J. Marotzke,
outputs limited us to 5 GCM runs per scenario. V. Naik, M. Palmer, G.-K. Plattner, J. Rogelj, M. Rojas, J. Sillmann,
T. Storelvmo, P. Thorne, B. Trewin, K. Rao, B. Adhikary, R. Allan, K.
24. Environmental flow requirements are implicitly considered in the Armour, K. Zickfeld, 2021. IPCC AR6 WGI Technical Summary. pp.
thresholds. 33–144. https://doi.org/10.1017/9781009157896.002.
25. 5 x 5 arc minute. Batjes, N.H. 2012. “ISRIC-WISE Derived Soil Properties on a 5 by 5
Arc-Minutes Global Grid (Ver. 1.2).” ISRIC-World Soil Information.
26. Our research partners at Deltares and Utrecht University ran the
model for 1960–2014, but only 1990–2014 has been used to cal- BGR and UNESCO (Bundesanstalt für Geowissenschaften und
culate the indicator. The reason is that groundwater development Rohstoffe and United Nations Educational, Scientific, and Cul-
began to increase in the late 1980s and 1990s in some countries tural Organization). 2018. “Groundwater Resources of the World.”
that use groundwater intensively. We assume this period to be https://www.whymap.org/whymap/EN/ Maps_Data/Gwrw/
representative of the current trend; however, further optimization gwrw_node_en.html.
might provide better insights as to the best range. Although this is
different from some other indicators, we are consistently calcu- Billen, G., and J. Garnier. 2007. “River Basin Nutrient Delivery to
lating baseline scores, thereby making it possible to aggregate the Coastal Sea: Assessing Its Potential to Sustain New Produc-
various temporal ranges. tion of Non-siliceous Algae.” Marine Chemistry 106 (1–2): 148–60.
doi:10.1016/j.marchem.2006.12.017.
27. The vulnerability of people to floods is assessed as a binary con-
dition: they are either flooded or they are not. Bouwman, A., T. Kram, and K. Klein Goldewijk. 2006. “Integrated Mod-
elling of Global Environmental Change: An overview of IMAGE.4.” PBL
28. The vulnerability of people to floods is assessed as a binary con- Netherlands Environmental Assessment Agency, September 11.
dition: they are either flooded or they are not.
Bouwman, A.F., A.H.W. Beusen, J.A. Harrison, and D.C. Reed. 2015.
29. Water use estimates for domestic, industrial, and livestock sectors “Nutrient Release in Global Coastal Marine Ecosystems, and Model-
are inputs to PCR-GLOBWB 2. Irrigation is the only sectoral de- ing of Impacts (Hypoxia, Harmful Algal Blooms and Fisheries) in Rela-
mand that is an output of PCR-GLOBWB 2. Irrigation demand is a tion to Coastal Conditions.” Global Foundations for Reducing Nutrient
function of crop extents and climatic conditions such as tempera- Enrichment and Oxygen Depletion from Land Based Pollution, in
ture and humidity and is therefore calculated on-the-fly during Support of Global Nutrient Cycle. Global Environment Facility, United
the modeling process (to adjust to the climate forcing data). For Nations Development Programme, and Intergovernmental Oceano-
example, a crop field will require more water during a hotter sum- graphic Commission. https://www.thegef.org/project/global-founda-
mer than a cooler summer. tions-reducingnutrient-enrichment-and-odflb-pollution-support-gnc.
30. These elements are indirectly covered in the Regulatory and Brauman, K.A., B.D. Richter, S. Postel, M. Malsy, and M. Flörke. 2016.
Reputational Risk group. “Water Depletion: An Improved Metric for Incorporating Seasonal and
Dry-Year Water Scarcity into Water Risk Assessments.” Elementa: Sci-
31. WATCH (WATer and global CHange); ERA5 (ECMWF Re-Analysis
ence of the Anthropocene 4 (January): 000083. doi:10.12952/journal.
fifth generation); ECMWF (European Centre for Medium-Range
elementa.000083.
Weather Forecasts).
46 |
Carrão, H., G. Naumann, and P. Barbosa. 2016. “Mapping Global “FAOSTAT.” 2012. FAO (Food and Agriculture Organization of the
Patterns of Drought Risk: An Empirical Framework Based on United Nations). http://faostat.fao.org/. Accessed February 12, 2019.
Sub-national Estimates of Hazard, Exposure and Vulnerability.” “GADM Metadata.” n.d. https://gadm.org/metadata.html. Accessed
Global Environmental Change 39 (July): 108–24. doi: 10.1016/j.gloenv- February 11, 2019.
cha.2016.04.012.
Galvis Rodríguez, S., E.H. Sutanudjaja, and M. Faneca Sànchez. 2017.
Dalkey, N., and O. Helmer. 1963. “An Experimental Application of the “Memo: Update on the Groundwater Risk Indicators.” Memo. Delft, the
DELPHI Method to the Use of Experts.” Management Science 9 (3): Netherlands: Deltares.
458–67. doi:10.1287/ mnsc.9.3.458.
Garnier, J., A. Beusen, V. Thieu, G. Billen, and L. Bouwman. 2010.
Dirmeyer, P., X. Gao, M. Zhao, Z. Guo, T. Oki, N. Hanasaki, 2006. “N:P:Si Nutrient Export Ratios and Ecological Consequences in
“GSWP–2: Multimodel Analysis and Implications for Our Perception of Coastal Seas Evaluated by the ICEP Approach.” Global Biogeochemi-
the Land Surface.” Bulletin of The American Meteorological Society 87 cal Cycles 24 (4). doi:10.1029/2009GB003583.
(10): 1381--98. https://doi.org/10.1175/BAMS-87–10–1381.
Gassert, F., M. Luck, M. Landis, P. Reig, and T. Shiao. 2014. “Aqueduct
Dee, D.P., S.M. Uppala, A.J. Simmons, P. Berrisford, P. Poli, S. Ko- Global Maps 2.1: Constructing Decision-Relevant Global Water Risk
bayashi, U. Andrae, et al. 2011. “The ERA-Interim Reanalysis: Configu- Indicators.” World Resources Institute. https://wriorg.s3.amazonaws.
ration and Performance of the Data Assimilation System.” Quar- com/s3fs-public/Aqueduct_ Global_Maps_2.1-Constructing_De-
terly Journal of the Royal Meteorological Society 137 (656): 553–97. cicion-Relevant_Global_Water_Risk_ Indicators_final_0.pdf.
doi:10.1002/qj.828.
Hagemann, S. 2002. “An Improved Land Surface Parameter Dataset
de Graaf, I., E. Sutanudjaja, L. Van Beek, and M. Bierkens. 2015. “A for Global and Regional Climate Models.” Max-Planck-Institut für
High-Resolution Global-Scale Groundwater Model.” Hydrology and Meteorologie, January. doi:10.17617/2.2344576.
Earth System Sciences 19 (2): 823–37.
Hagemann, S., and L.D. Gates. 2003. “Improving a Subgrid Runoff
de Graaf, I.E.M., R.L.P.H. van Beek, T. Gleeson, N. Moosdorf, O. Parameterization Scheme for Climate Models by the Use of High
Schmitz, E.H. Sutanudjaja, and M.F.P. Bierkens. 2017. “A Global-Scale Resolution Data Derived from Satellite Observations.” Climate Dynam-
Two-Layer Transient Groundwater Model: Development and Applica- ics 21 (3–4): 349–59. doi:10.1007/s00382-003-0349-x.
tion to Groundwater Depletion.” Advances in Water Resources 102
(April): 53–67. doi:10.1016/j. advwatres.2017.01.011. Harris, I., P.D. Jones, T.J. Osborn, and D.H. Lister. 2014. “Updated
High-Resolution Grids of Monthly Climatic Observations: The CRU
Doelman, J.C., E. Stehfest, A. Tabeau, H. van Meijl, L. Lassaletta, TS3.10 Dataset.” International Journal of Climatology 34 (3): 623–42.
D.E.H.J. Gernaat, K. Hermans, M. Harmsen, V. Daioglou, H. Biemans, doi:10.1002/joc.3711.
S. van der Sluis, D.P. van Vuuren, 2018. “Exploring SSP land-use
dynamics using the IMAGE model: Regional and gridded scenarios of Hartmann, J., and N. Moosdorf. 2012. “The New Global Lithological
land-use change and land-based climate change mitigation.” Global Map Database GLiM: A Representation of Rock Properties at the
Environmental Change 48: 119–35. https://doi.org/10.1016/j.gloen- Earth Surface.” Geochemistry, Geophysics, Geosystems 13 (12). https://
vcha.2017.11.014 agupubs.onlinelibrary.wiley.com/ doi/10.1029/2012GC004370.
Doorenbos, J., and W. Pruitt. 1977. “Crop Water Requirements.” Irriga- Hempel, S., K. Frieler, L. Warszawski, J. Schewe, F. Piontek, “A trend-
tion and Drainage Paper No. 24. Rome: FAO (Food and Agriculture preserving bias correction—the ISI-MIP approach.” Earth System
Organization of the United Nations). Dynamics 4: 219–36 (2013).
EOG and NOAA (Earth Observation Group and National Oceanic and Hengl, T., J.M. de Jesus, G.B.M.Heuvelink, M.R. Gonzalez, M. Kili-
Atmospheric Administration). n.d. “Nighttime Lights.” NOAA. https:// barda, A. Blagotić, W. Shangguan, M.N. Wright, X. Geng, B. Bauer-
ngdc.noaa.gov/eog/viirs/ download_dnb_composites.html. Accessed Marschallinger, M.A. Guevara, R. Vargas, R.A. MacMillan, N.H. Batjes,
February 13, 2019. J.G.B. Leenaars, E. Ribeiro, I. Wheeler, S. Mantel, B. Kempen, 2017.
“SoilGrids250m: Global gridded soil information based on machine
Eisner, S. 2016. “Comprehensive Evaluation of the WaterGAP3 Model learning.” PLOS ONE 12, e0169748. https://doi.org/10.1371/jour-
across Climatic, Physiographic, and Anthropogenic Gradients.” PhD nal.pone.0169748
diss., University of Kassel.

IGRAC (International Groundwater Resources Assessment Centre). Mekonnen, M.M. and A.Y. Hoekstra, 2012. “A Global Assessment of the
n.d. “Global Groundwater Information System (GGIS).” https://www. Water Footprint of Farm Animal Products.” Ecosystems 15, 401–415.
un-igrac.org/globalgroundwater-information-system-ggis. Accessed https://doi.org/10.1007/s10021-011-9517-8.
February 13, 2019.
Meyer, V., D. Haase, and S. Scheuer. 2009. “Flood Risk Assessment in
IOC-UNESCO and UNEP (Intergovernmental Oceanographic Com- European River Basins—Concept, Methods, and Challenges Exempli-
mission and United Nations Environment Programme). 2016. “Large fied at the Mulde River.” Integrated Environmental Assessment and
Marine Ecosystems: Status and Trends.” Nairobi: UNEP. http://geft- Management 5 (1): 17–26. doi:10.1897/ IEAM_2008-031.1.
wap.org/water-systems/largemarine-ecosystems.
Müller Schmied, H., S. Eisner, D. Franz, M. Wattenbach, F.T. Portmann,
Kraijenhoff van de Leur, D.A. 1958. “A Study of Non-steady Ground- M. Flörke, and P. Döll. 2014. “Sensitivity of Simulated Global-Scale
water Flow with Special Reference to a Reservoir Coefficient.” De Freshwater Fluxes and Storages to Input Data, Hydrological Model
Ingenieur 19: B87–B94 Structure, Human Water Use and Calibration.” Hydrology and Earth
System Sciences 18 (9): 3511–38. doi:10.5194/ hess–18-3511–2014.
Lange, S., 2021. ISIMIP3b Bias Adjustment Fact Sheet. ISIMIP.
Nachtergaele, F., H. van Velthuizen, L. Verelst, N. Batjes, K. Dijkshoorn,
Lange, S., C. Menz, S. Gleixner, M. Cucchi, G.P. Weedon, A. Amici, N. V. van Engelen, G. Fischer, et al. 2009. “Harmonized World Soil Data-
Bellouin, H. Müller Schmied, H. Hersbach, C. Buontempo, C. Cag- base.” Wageningen, the Netherlands: ISRIC.
nazzo, 2021. WFDE5 over land merged with ERA5 over the ocean
(W5E5 v2.0). https://doi.org/10.48364/ISIMIP.342217. OpenStreetMap contributors. 2018. Planet Dump. https://
Planet.Osm.Org.
Lehner, B., and P. Döll. 2004b. “Development and Validation of a
Global Database of Lakes, Reservoirs and Wetlands.” Journal of Hy- Paul, P.K., Y. Zhang, N. Ma, A. Mishra, N. Panigrahy, R. Singh, 2021.
drology 296 (1–4): 1–22. doi:10.1016/j.jhydrol.2004.03.028. “Selecting hydrological models for developing countries: Perspective
of global, continental, and country scale models over catchment scale
Lehner, B., and G. Grill. 2013. “Global River Hydrography and Network models.” Journal of Hydrology 600, 126561. https://doi.org/10.1016/j.
Routing: Baseline Data and New Approaches to Study the World’s jhydrol.2021.126561.
Large River Systems.” Hydrological Processes 27 (15): 2171–86.
Portmann, F.T., S. Siebert, and P. Döll. 2010. “MIRCA2000-Global
Lehner, B., K. Verdin, and A. Jarvis. 2008. “New Global Hydrography Monthly Irrigated and Rainfed Crop Areas around the Year
Derived from Spaceborne Elevation Data.” Eos, Transactions, Ameri- 2000: A New High-Resolution Data Set for Agricultural and
can Geophysical Union 89 (10): 93. doi:10.1029/2008EO100001. Hydrological Modeling.” Global Biogeochemical Cycles 24 (1).
doi:10.1029/2008GB003435.
Lehner, B., C.R. Liermann, C. Revenga, C. Vörösmarty, B. Fekete, P.
Crouzet, P. Döll, et al. 2011. “High-Resolution Mapping of the World’s Reig, P., T. Shao, and F. Gassert. 2013. “Aqueduct Water Risk Frame-
Reservoirs and Dams for Sustainable River-Flow Management.” Fron- work.” Working paper. Washington, DC: World Resources Institute.
tiers in Ecology and the Environment 9 (9): 494–502. https://wriorg.s3.amazonaws. com/s3fs-public/aqueduct_water_
risk_framework.pdf.
Luck, M., M. Landis, F. Gassert. 2015. “Aqueduct Water Stress Projec-
tions: Decadal Projections of Water Supply and Demand Using RepRisk. n.d. “RepRisk ESG Data Science and Quantitative Solutions.”
CMIP5 GCMs.” Technical Note. Washington, DC: World Resources www. reprisk.com.
Institute. wri.org/publication/aqueduct-water-stress-projections.
Rodell, M., P.R. Houser, U. Jambor, J. Gottschalck, K. Mitchell, C.-J.
Masutomi, Y., Y. Inui, K. Takahashi, and Y. Matsuoka. 2009. “Develop- Meng, K. Arsenault, et al. 2004. “The Global Land Data Assimila-
ment of Highly Accurate Global Polygonal Drainage Basin Data.” tion System.” Bulletin of the American Meteorological Society 85 (3):
Hydrological Processes 23 (4): 572–84. doi:10.1002/hyp.7186. 381–94. doi:10.1175/BAMS-85-3-381.
Mayorga, E., S.P. Seitzinger, J.A. Harrison, E. Dumont, A.H.W. Beusen, Rohwer, J., D. Gerten, and W. Lucht. 2007. Development of Functional
A.F. Bouwman, B.M. Fekete, et al. 2010. “Global Nutrient Export from Irrigation Types for Improved Global Crop Modelling. PIK Report.
WaterSheds 2 (NEWS 2): Model Development and Implementation.”
Environmental Modelling & Software 25 (7): 837–53. doi:10.1016/j.
envsoft.2010.01.007.
48 |
Rowe, G., and G. Wright. 1999. “The Delphi Technique as a Forecast- Vanham, D., A.Y. Hoekstra, Y. Wada, F. Bouraoui, A. de Roo, M.M. Me-
ing Tool: Issues and Analysis.” International Journal of Forecasting 15 konnen, W.J. van de Bund, et al. 2018. “Physical Water Scarcity Metrics
(4): 353–75. doi:10.1016/S0169–2070(99)00018-7. for Monitoring Progress towards SDG Target 6.4: An Evaluation of In-
dicator 6.4.2 ‘Level of Water Stress.’” Science of the Total Environment
Sen, P.K., 1968. Estimates of the Regression Coefficient Based on Ken- 613–14 (February): 218–32. doi:10.1016/j. scitotenv.2017.09.056.
dall’s Tau. Journal of the American Statistical Association 63: 1379–89.
https://doi.org/10.2307/2285891 van Huijstee, J., B. van Bemmel, A. Bouwman, and F. van Rijn. 2018.
“Towards an Urban Preview: Modelling Future Urban Growth
Scussolini, P., J.C.J.H. Aerts, B. Jongman, L.M. Bouwer, H.C. Winse- with 2UP.” 3255. The Hague: PBL Netherlands Environmental As-
mius, H. de Moel, and P.J. Ward. 2016. “FLOPROS: An Evolving Global sessment Agency. https://www.pbl.nl/en/ publications/towards-
Database of Flood Protection Standards.” Natural Hazards and Earth an-urban-preview.
System Sciences 16 (5): 1049–61. doi:10.5194/nhess–16–1049–2016.
van Vuuren, D., P. P.L. Lucas, and H. Hilderink. 2007. “Downscaling
Shiklomanov, I. 1997. “Comprehensive Assessment of the Fresh- Drivers of Global Environmental Change: Enabling Use of Global
water Resources of the World.” Stockholm: World Meteorologi- SRES Scenarios at the National and Grid Levels.” Global Environmen-
cal Organization. tal Change: Uncertainty and Climate Change Adaptation and Mitiga-
tion 17 (1): 114–30. doi:10.1016/j.gloenvcha.2006.04.004.
Siebert, S., and P. Döll. 2010. “Quantifying Blue and Green Virtual
Water Contents in Global Crop Production as Well as Potential Verdin, K.L., and S. Greenlee. 1996. “Development of Continental Scale
Production Losses without Irrigation.” Journal of Hydrology 384 (3–4): Digital Elevation Models and Extraction of Hydrographic Features.” In
198–217. doi:10.1016/j. jhydrol.2009.07.031. Proceedings, Third International Conference/Workshop on Integrating
GIS and Environmental Modeling, Santa Fe, New Mexico, 21–26.
Steinfeld, H., P. Gerber, T.D. Wassenaar, V. Castel, M. Rosales, and
C. de Haan. 2006. Livestock’s Long Shadow: Environmental Issues Vörösmarty, C.J., B.M. Fekete, M. Meybeck, and R.B. Lammers. 2000.
and Options. Rome: FAO (Food and Agriculture Organization of the “Global System of Rivers: Its Role in Organizing Continental Land
United Nations). Mass and Defining Land-to-Ocean Linkages.” Global Biogeochemical
Cycles 14 (2): 599–621. doi:10.1029/1999GB900092.
Strahler, A.N. 1957. “Quantitative Analysis of Watershed Geomorphol-
ogy.” Eos, Transactions, American Geophysical Union 38 (6): 913–20. Vörösmarty, C.J., C. Léveque, C. Revenga, R. Bos, C. Caudill, J. Chilton,
E. Douglas, et al. 2005. Millennium Ecosystem Assessment Volume 1:
Sutanudjaja, E.H., R. van Beek, N. Wanders, Y. Wada, J.H.C. Bosmans,
Conditions and Trends, chap. 7, “Freshwater Ecosystems.” Millennium
N. Drost, R.J. van der Ent, et al. 2018. “PCR-GLOBWB 2: A 5 Arcmin
Ecosystem Assessment 1: 165–207.
Global Hydrological and Water Resources Model.” Geoscientific
Model Development 11 (6): 2429–53. doi:10.5194/gmd–11–2429–2018. Wada, Y., L.P.H. van Beek, and M.F.P. Bierkens. 2011a. “Modelling
Global Water Stress of the Recent Past: On the Relative Importance of
Sutanudjaja, E.H., R. van Beek, E. Saccoccia, S. Kuzma, T. Luo, and
Trends in Water Demand and Climate Variability.” Hydrology and Earth
M.F.P Bierkens. 2023. “Hydrological projection of future global water
System Sciences 15 (12): 3785–808. doi:10.5194/hess–15-3785–2011.
states with CMIP6.” Utrecht University. https://public.yoda.uu.nl/geo/
UU01/YM7A5H.html. Wada, Y., L.P.H. van Beek, D. Viviroli, H.H. Dürr, R. Weingartner, and
M.F.P. Bierkens. 2011b. “Global Monthly Water Stress: 2. Water Demand
Todini, E. 1996. “The ARNO Rainfall: Runoff Model.” Journal of Hydrol-
and Severity of Water Stress, 2.” Water Resources Research 47 (7).
ogy 175 (1): 339–82. doi:10.1016/S0022–1694(96)80016-3.
doi:10.1029/2010WR009792.
Triantaphyllou, E. 2010. Multi-criteria Decision Making Methods: A
Wada, Y., D. Wisser, and M.F.P. Bierkens. 2014. “Global Modeling
Comparative Study. Applied Optimization 44. Dordrecht, the Neth-
of Withdrawal, Allocation and Consumptive Use of Surface Water
erlands: Kluwer.
and Groundwater Resources.” Earth System Dynamics 5 (1): 15–40.
UNEP (United Nations Environment Programme). n.d. “UNEP.” https:// doi:10.5194/esd-5–15–2014.
www. unenvironment.org/. Accessed on February 28, 2019.

Wada, Y., M. Flörke, N. Hanasaki, S. Eisner, G. Fischer, S. Tramberend, WHO and UNICEF (World Health Organization and United Nations
Y. Satoh, M.T.H. van Vliet, P. Yillia, C. Ringler, P. Burek, D. Wiberg, Children’s Fund). 2017. Progress on Drinking Water, Sanitation and
2016. “Modeling global water use for the 21st century: the Water Hygiene: 2017 Update and SDG Baselines. https://washdata.org/.
Futures and Solutions (WFaS) initiative and its approaches.” Geo-
scientific Model Development 9: 175–222. https://doi.org/10.5194/ WRI (World Resources Institute). n.d. WRI’s Open Data Commitment:
gmd-9–175–2016. World Resources Institute. https://www.wri. org/about/open-data-
commitment. Accessed February 27, 2019.
Wang, D., K. Hubacek, Y. Shan, W. Gerbens-Leenes, J. Liu, 2021. “A
Review of Water Stress and Water Footprint Accounting.” Water 13, WRI, UNEP, UNDP, and World Bank, eds. 1998. World Resources,
201. https://doi.org/10.3390/w13020201. 1998–99: A Guide to the Global Environment Environmental Change
and Human Health. New York: Oxford University Press.
Ward, P.J., H.C. Winsemius, S. Kuzma, T. Luo, M.F.P. Bierkens, A. Bouw-
man, H. de Moel, et al. 2020. “Aqueduct Floods Methodology.” Techni- Xie, H., C. Ringler, and G. Pitois. 2016. “Assessing Global BOD, Nitro-
cal Note. Washington, DC: World Resources Institute. gen and Phosphorus Loadings under Socioeconomic and Climate
Change Scenarios.” Technical note. IFPRI (International Food Policy
Wendling, Z.A., J.W. Emerson, D.C. Esty, M.A. Levy, and A. de Research Institute).
Sherbinin. 2018. “2018 Environmental Performance Index.” New
Haven, CT: Yale Center for Environmental Law and Policy. https://epi. Yamazaki, D., D. Ikeshima, J. Sosa, P. Bates, G. Allen, T. Pavelsky, 2019.
envirocenter.yale.edu/2018-epi-report/ introduction. “Hydro: A High‐Resolution Global Hydrography Map Based on Lat-
est Topography Dataset.” Water Resources Research 55. https://doi.
Wint, W., and T.P. Robinson. 2007. Gridded Livestock of the World, org/10.1029/2019WR024873.
2007. Rome: Food and Agriculture Organization of the United Nations.
World Bank. n.d. “World Bank National Accounts Data.” World Bank.
https://data. worldbank.org. Accessed February 13, 2019.
50 |
ACKNOWLEDGEMENTS ABOUT THE AUTHORS

We are pleased to acknowledge our institutional strategic partners, Samantha Kuzma is the data lead and senior manager for Global
who provide core funding to WRI: Netherlands Ministry of Foreign Water Data Products at World Resources Institute.
Affairs, Royal Danish Ministry of Foreign Affairs, and Swedish Contact: [email protected]
International Development Cooperation Agency.
Marc F. P. Bierkens is a professor of hydrology at Utrecht University,
This publication was made possible thanks to the support of where he holds the Chair of Earth Surface Hydrology in the
Microsoft, Moody’s RMS, and Cargill Inc., and the Aqueduct Alliance Department of Physical Geography. He is also a senior scientist at
(Accenture, Ansell, Cargill Inc., Ecolab Inc., Dutch Ministry of Deltares.
Infrastructure and Water Management, Gap, International Paper,
Shivani Lakshman is a corporate water stewardship research
Mars, Meta, Microsoft, Proctor & Gamble Company, Tyson Foods Inc., analyst at the World Resources Institute.
3M). Additionally, we thank the World Bank for supporting parts of
Aqueduct Floods. Tianyi Luo is a director at Bluerisk.
The authors would like to thank the following people for providing Liz Saccoccia is a water security and geospatial associate at the
invaluable insight and assistance in their reviews of this report: World Resources Institute.
Zablon Adane (WRI), Courtnae Baily (WRI), Brittany Craig (Tyson
Edwin H. Sutanudjaja is the core developer of PCR-GLOBWB 2 in
Foods Inc.) and Homero Paltan Lopez (World Bank).
the Department of Physical Geography at Utrecht University.
The authors would also like to thank our research partners at
Rens van Beek is an associate professor of large-scale hydrology
Deltares, IFPRI, IVM, RepRisk, and Utrecht University, as well as
and Earth surface processes at Utrecht University.
our WRI colleagues Nancy Sanchez, Carlos DelReal, Renee Pineda,
Carlos Muñoz, Laura Malaguzzi Valeri, and Romain Warnault, for
their extensive guidance and feedback during the design and
development of this study.
Opinions or points of view expressed in this report are those of the

authors and do not necessarily reflect the position of the reviewers,
research partners, or the organizations they represent.

ABOUT WRI
World Resources Institute is a global research organization that Our approach
turns big ideas into action at the nexus of environment, economic
COUNT IT
opportunity, and human well-being.
We start with data. We conduct independent research and draw on
Our challenge
the latest technology to develop new insights and recommendations.
Natural resources are at the foundation of economic opportunity and Our rigorous analysis identifies risks, unveils opportunities, and
human well-being. But today, we are depleting Earth’s resources at informs smart strategies. We focus our efforts on influential
rates that are not sustainable, endangering economies and people’s and emerging economies where the future of sustainability
lives. People depend on clean water, fertile land, healthy forests, will be determined.
and a stable climate. Livable cities and clean energy are essential
CHANGE IT
for a sustainable planet. We must address these urgent, global
challenges this decade. We use our research to influence government policies, business
strategies, and civil society action. We test projects with communities,
Our vision
companies, and government agencies to build a strong evidence
We envision an equitable and prosperous planet driven by the wise base. Then, we work with partners to deliver change on the ground
management of natural resources. We aspire to create a world where that alleviates poverty and strengthens society. We hold ourselves
the actions of government, business, and communities combine to accountable to ensure our outcomes will be bold and enduring.
eliminate poverty and sustain the natural environment for all people.
SCALE IT
We don’t think small. Once tested, we work with partners to adopt

and expand our efforts regionally and globally. We engage with
decision-makers to carry out our ideas and elevate our impact. We
measure success through government and business actions that
improve people’s lives and sustain a healthy environment.
Copyright 2023 World Resources Institute. This work is licensed under the Creative Commons Attribution 4.0 International License.
To view a copy of the license, visit http://creativecommons.org/licenses/by/4.0/
10 G Street, NE | Washington, DC 20002 | WRI.ORG

Aqueduct 40 Technical Note

Uploaded by

Copyright:

Available Formats

Aqueduct 40 Technical Note

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Aqueduct 40 Technical Note

Uploaded by

Copyright:

Available Formats

TECHNICAL NOTE

Aqueduct 4.0: Updated decision-relevant

The normalized indicator scores have been aggregated by category (quantity,

TECHNICAL NOTE | Version 1.0 | August 2023 | 1

▪ Aqueduct Water Risk Atlas (Reig et al. 2013):

▪ They should be actionable in the context of private and

Figure 1 | Overview of Aqueduct framework

Updated in Aqueduct 4.0

Baseline water stress

Baseline water depletion

PHYSICAL RISK QUANTITY

Groundwater table decline

Riverine flood risk

OVERALL WATER RISK

Untreated connected wastewater

PHYSICAL RISK QUALITY

Unimproved/no drinking water

Peak RepRisk country

TECHNICAL NOTE | August 2023 | 3

Table 1 | Aqueduct indicators based on hydrological model output

AQUEDUCT INDICATOR MODEL DATA USED

Water Use Water Supply Groundwater Heads

WATER USE: Spatial aggregation

▪ Spatial aggregation. Water use and water supply are

TECHNICAL NOTE | August 2023 | 5

Figure 3 | Example of discharge inflow and outflow points

TECHNICAL NOTE | August 2023 | 7

Gross demand (m/month) 10-year moving average Theil Sen Regression

Model data used by Aqueduct GENERAL CIRCULATION MODELS (GCMS):

TECHNICAL NOTE | August 2023 | 9

▪ 2014: 1960–2014 (historic GCM data)

▪ 2080: 2065–2095 (future GCM data)

(Luck et al. 2015).

Source: Raw data from UNDRR, modified/aggregated by WRI.

Baseline water stress

wsm,b = Water stress per month, per sub-basin in [-]

TECHNICAL NOTE | August 2023 | 11

One of the biggest assumptions is that water resources are

Baseline water depletion

Name Baseline Water Depletion

Temporal range 1979–2019

Baseline water depletion is similar to baseline water stress;

TECHNICAL NOTE | August 2023 | 13

We use linear interpolation within each category to remap the

Where r is the raw indicator value and score is the indica-

Name Interannual Variability iavm,b = Interannual variability per month, per

Interannual variability is calculated using the available water

TECHNICAL NOTE | August 2023 | 15

Name Seasonal Variability

Spatial resolution 5 x 5 arc minute grid cells

bam,b = Average available blue water per month, per sub-

Groundwater table decline

Spatial resolution Groundwater aquifer (WHYMAP)

SOURCE: hfloodplain,5’ = Elevation of floodplain in meters for each 5 x 5

TECHNICAL NOTE | August 2023 | 17

RAW VALUE RISK CATEGORY SCORE

<0 cm/y Low 0–1

Within each category, we use linear interpolation to convert the

Where r is the raw indicator value and score is the indica-

The subgroups (physical risk quantity, physical risk quality, and