![Likelihood of the occurrence of hydro-political interactions in the main transboundary river basins (from the top-left [normalized likelihood of hydro-political issues, min = 0 and max = 1]: Ganges-Brahmaputra [1.000], Nile [0.761], Indus [0.675], Euphrates-Tigris [0.592], Danube [0.499], Mekong [0.492], Aral Sea [0.455], Niger [0.447], Congo [0.432], Zambezi [0.431], Senegal [0.372]). In the radar chart the normalized score of the main factors determining the likelihood in the specific river basins. Not all the variables explicitly used for the model are represented in the radar chart: the non-included factors, however, are derived from the climatic variables displayed.](https://www.researchgate.net/profile/Fabio-Farinosi/publication/327466480/figure/fig1/AS:667888813895689@1536248510829/Likelihood-of-the-occurrence-of-hydro-political-interactions-in-the-main-transboundary_Q320.jpg)
Content uploaded by Fabio Farinosi
Author content
All content in this area was uploaded by Fabio Farinosi on Sep 06, 2018
Content may be subject to copyright.
Contents lists available at ScienceDirect
Global Environmental Change
journal homepage: www.elsevier.com/locate/gloenvcha
An innovative approach to the assessment of hydro-political risk: A spatially
explicit, data driven indicator of hydro-political issues
F. Farinosi
a,⁎
, C. Giupponi
c
, A. Reynaud
b
, G. Ceccherini
a
, C. Carmona-Moreno
a
, A. De Roo
a
,
D. Gonzalez-Sanchez
a
, G. Bidoglio
a
a
European Commission, DG Joint Research Centre, Ispra, Italy
b
Toulouse School of Economics - National Institute for Research in Agriculture (INRA) –University of Toulouse Capitole, Toulouse, France
c
Department of Economics, Venice Centre for Climate Studies (VICCS), Ca’Foscari University of Venice, Venice, Italy
ARTICLE INFO
Keywords:
Hydro-political risk
Water cross-border issues
Transboundary water interactions
Random Forest regression
ABSTRACT
Competition over limited water resources is one of the main concerns for the coming decades. Although water
issues alone have not been the sole trigger for warfare in the past, tensions over freshwater management and use
represent one of the main concerns in political relations between riparian states and may exacerbate existing
tensions, increase regional instability and social unrest. Previous studies made great efforts to understand how
international water management problems were addressed by actors in a more cooperative or confrontational
way. In this study, we analyze what are the pre-conditions favoring the insurgence of water management issues
in shared water bodies, rather than focusing on the way water issues are then managed among actors. We do so
by proposing an innovative analysis of past episodes of conflict and cooperation over transboundary water
resources (jointly defined as “hydro-political interactions”). On the one hand, we aim at highlighting the factors
that are more relevant in determining water interactions across political boundaries. On the other hand, our
objective is to map and monitor the evolution of the likelihood of experiencing hydro-political interactions over
space and time, under changing socioeconomic and biophysical scenarios, through a spatially explicit data
driven index. Historical cross-border water interactions were used as indicators of the magnitude of corre-
sponding water joint-management issues. These were correlated with information about river basin freshwater
availability, climate stress, human pressure on water resources, socioeconomic conditions (including institu-
tional development and power imbalances), and topographic characteristics. This analysis allows for identifi-
cation of the main factors that determine water interactions, such as water availability, population density,
power imbalances, and climatic stressors. The proposed model was used to map at high spatial resolution the
probability of experiencing hydro-political interactions worldwide. This baseline outline is then compared to
four distinct climate and population density projections aimed to estimate trends for hydro-political interactions
under future conditions (2050 and 2100), while considering two greenhouse gases emission scenarios (moderate
and extreme climate change). The combination of climate and population growth dynamics is expected to impact
negatively on the overall hydro-political risk by increasing the likelihood of water interactions in the trans-
boundary river basins, with an average increase ranging between 74.9% (2050 –population and moderate
climate change) to 95% (2100 - population and extreme climate change). Future demographic and climatic
conditions are expected to exert particular pressure on already water stressed basins such as the Nile, the
Ganges/Brahmaputra, the Indus, the Tigris/Euphrates, and the Colorado. The results of this work allow us to
identify current and future areas where water issues are more likely to arise, and where cooperation over water
should be actively pursued to avoid possible tensions especially under changing environmental conditions. From
a policy perspective, the index presented in this study can be used to provide a sound quantitative basis to the
assessment of the Sustainable Development Goal 6, Target 6.5 “Water resources management”, and in particular
to indicator 6.5.2 “Transboundary cooperation”.
https://doi.org/10.1016/j.gloenvcha.2018.07.001
Received 5 March 2018; Received in revised form 27 June 2018; Accepted 1 July 2018
⁎
Corresponding author at: European Commission, DG Joint Research Centre, Directorate D –Sustainable Resources, Unit D.02 Water and Marine Resources, Via E.
Fermi, 2749-21027, Ispra, VA, Italy.
E-mail address: fabio.farinosi@gmail.com (F. Farinosi).
Global Environmental Change 52 (2018) 286–313
0959-3780/ © 2018 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY license
(http://creativecommons.org/licenses/BY/4.0/).
T
1. Introduction
Future availability of freshwater for human consumption under a
changing world represents one of the main concerns of the current poli-
tical debate. Water crises have been placed among the major risk factors
for the coming decades by the Global Risks Perception Surveys conducted
by the World Economic Forum between 2015 and 2017 (WEF, 2017,
2016). Increasing demographic pressure, environmental degradation, and
climate change impacts on water spatio-temporal distribution represent
the largest determinants of current and future water related issues. Al-
though it is intuitive that water stress is likely to increase the competition
over water (Malthus, 1798), it is not completely clear how the combi-
nations of factors influencing water demand and availability alone could
lead to such different outcomes in different watersheds spread around the
planet. Evidence shows that the consequences of comparable levels of
physical water stress have been handled unevenly in different geo-
graphical areas and historical contexts (Wolf et al., 2003). Socioeconomic
and cultural characteristics (Wolf, 2009), jointly with topographic factors
(Beck et al., 2014;Gleditsch et al., 2006;Munia et al., 2016), were
identified as the drivers more likely influencing hydro-political dynamics.
Resource scarcity is likely to increase tensions, especially when associated
with socio-cultural stressors (Sirin, 2011), but, on the other hand, the lack
of a vital resource as water is also likely to boost cooperation between
actors sharing the same freshwater sources (Bernauer et al., 2012b;Wolf,
2009,2007;Wolf et al., 2003). The literature hardly identified common
features between countries involved in water issues: similar levels of
tension over water arose between countries independently of their climate
zone, population size, territorial extension, level of democracy (Wolf,
2009). Moreover, the same international water issue frequently resulted
in episodes of conflict and cooperation at the same time (Gerlak and
Zawahri, 2009;Kalbhenn and Bernauer, 2012;Wolf, 2009;Wolf et al.,
2003;Yoffe et al., 2004;Zeitoun et al., 2011;Zeitoun and Mirumachi,
2008). Although several cases of tensions, mostly non-violent, were also
recorded, the literature shows that water related issues are more likely to
be resolved with cooperation between the countries sharing the trans-
boundary basins (De Stefano et al., 2010b;Wolf, 2009,2007;Wolf et al.,
2003;Yoffeet al., 2004,2003). Analyzing historical events, Böhmelt et al.
(2014) concluded that physical availability and water demand compo-
nents are only part of the aspects to be considered for the analysis of water
related issues. The literature about political science, geopolitics, and di-
plomacy showed thatalso socioeconomic factors, jointly with institutional
capacity, legal framework, and cultural background influence the diplo-
matic interactions between countries or actors sharing resources
(Bernauer et al., 2012b;Wolf, 2009;Zeitoun et al., 2011)(
1
).
The goal of this study is to design an empirically based index aimed
at analyzing and mapping the interactions between biophysical and
socioeconomic factors linked to water issues at global scale. This was
done analyzing water availability and demand, as well as socio-
economic, institutional, legal, and cultural context: factors that are
likely to influence transboundary water issues. Final goal is to provide
the policy maker with an instrument able to capture historical and
current determinants of water related issues, but also the possibility to
construct scenarios and simulate sets of policy options. The hereby
presented index was calculated by applying a machine learning model
on data layers at detailed spatial resolution for the assessment of water
related issues and their determinants in the interactions between
countries in transboundary basins.
1.1. Assessing the factors influencing water cross-border issues
1.1.1. From water conflict and cooperation events to water interactions
Political debate at the highest level had often expressed the concern
for an increasing number of violent conflicts related to water resources
use and appropriation, in particular in the cases of transboundary ba-
sins. Such concern brought to the inclusion in Agenda 2030 of a specific
indicator on “Proportion of transboundary basin area with an opera-
tional arrangement for water cooperation”
2
(6.5.2), together with
“Degree of integrated water resources management implementation”
(6.5.1), for the assessment of Target 6.5 “Water resources manage-
ment”. Nevertheless, the analytic evidence of the correlation between
violent conflicts and climatic factors is not completely clear (Buhaug,
2010;Kallis and Zografos, 2014;Zeitoun and Mirumachi, 2008), and
thus the need emerges for methods oriented to pursue a scientifically
sound and quantitative assessment of available information, as the one
proposed herein.
The literature found a strong correlation between temperature
(Burke et al., 2009), or drought events (Couttenier and Soubeyran,
2014), and civil war episodes in Africa. Buhaug (2010) firmly contested
these findings and found the conflicts to be explained by structural and
contextual conditions, such as: exclusion of ethnical groups from the
political context, poor economic management, and geopolitical dy-
namics. Hsiang et al. (2011) proposed a meta-analysis based on 60
studies focusing on 45 historical conflicts on a global scale concluding
that temperature and rainfall variability are significantly connected to
violent events. Water related issues follow different dynamics respect to
civil conflicts: historical water crises were often resolved with more or
less satisfactory, formal or informal, agreements between the parties
(De Stefano et al., 2010b). Water conflicts in history are, in fact, per-
ipheral events and none of them reached a formal declaration of war
(Böhmelt et al., 2014;Kalbhenn and Bernauer, 2012;Katz, 2011;Wolf,
1998,2007,Yoffe et al., 2004,2003). The fact that water war episodes
were not recorded in the past does not imply that this could not happen
in the future (Kallis and Zografos, 2014). Water related disputes were
sometimes identified as igniting factors exacerbating international is-
sues of different nature (Wolf, 2009). On the other hand, cooperation
over transboundary basins often resulted in a benefit multiplier op-
portunity, associated with lower costs, increasing benefits and possi-
bility for cooperation beyond water (Sadoffand Grey, 2002). In the
analysis of historical hydro-political events, research points out that
certain degrees of conflict and cooperation coexists in the same water
related event (Gerlak and Zawahri, 2009;Kalbhenn and Bernauer,
2012;Wolf, 2009;Wolf et al., 2003;Yoffe et al., 2004;Zeitoun et al.,
2011;Zeitoun and Mirumachi, 2008). For this reason, some authors (in
particular Zeitoun and Mirumachi, 2008) claimed it would be more
appropriate to analyze the transboundary water interactions, conflict
and cooperation dynamics within the same water issue, regardless of
their nature (Kallis and Zografos, 2014;Watson, 2015;Zeitoun and
Mirumachi, 2008). In the proposed study, this approach was adopted
focusing on the historical water interactions, rather than on the specific
conflict or cooperation events linked with each of the water related
transboundary issues, and use this as an indicator of the hydro-political
risk, not intended as conflict risk, but rather risk of experiencing water
related issues. As specified in Kalbhenn and Bernauer (2012), each
water case underlying the interactions is defined as a water manage-
ment issue that manifests in multiple interrelated interactions. For in-
stance, the construction of a dam could represent a water case, while
the protests of the downstream countries, of the affected stakeholders,
the negotiations, and a possible international agreement would re-
present a series of events (conflict and cooperation) related to the
specific case of the construction of our dam. Following Wolf et al. (2003
and 2009), conflictive and cooperative events were defined water in-
teractions. In this paper, we will refer to the water interactions irre-
spectively of their specific nature and to more generic water issues or
cases, defined as the water management aspects determining the in-
terconnected water interactions, as for Wolf et al. (2003) and Yoffe
1
An overview about this topic is provided, among others, by the Correlates of
War Project (http://www.correlatesofwar.org/)
2
http://www.sdg6monitoring.org/indicators/target-65/indicators652/
F. Farinosi et al. Global Environmental Change 52 (2018) 286–313
287
et al. (2003). The likelihood of having water interactions is an indicator
of the complexity of the underlying water issue that, if not properly and
promptly addressed by the actors involved, could eventually increase
the hydro-political risk. Therefore, our index should then be interpreted
as a measure of the magnitude of water issues between specific actors in
a specific basin. The rationale behind this is the following: if there are
interactions about shared water resources in a specific basin, both in the
case of tension or cooperation events, there is a water allocation/
management/quality issue. Therefore, the fact that a water manage-
ment issue leads to cooperative or conflictive behaviors is unrelated to
the nature of the water issue itself. It attains more to the political,
cultural, institutional, and socioeconomic conditions of the actors in-
volved. The presence of a water issue is in itself an indicator of risk: it is
anecessary condition for having water interactions and a not sufficient
condition for either conflict or cooperation over water or both. Some
water interactions end up being conflictive, some others cooperative,
but all imply the existence of a water issue. This study focuses on the
analysis of the probability of having water issues, their intensity, and
their determinants: the necessary conditions for ensuing water tensions
or cooperation. The analysis of the factors that makes the water issue
being managed with a more confrontational or cooperative approach by
the actors involved falls outside the scope of this research.
1.1.2. Determinants of cross-border water issues
Economic, statistic and game theory approaches have been used to
analyze international dynamics over transboundary waters (Dinar,
2004). Some studies analyzed the dynamics of conflict and cooperation,
here defined water interactions (Bernauer and Böhmelt, 2014;Böhmelt
et al., 2014;Brochmann and Gleditsch, 2012;De Stefano et al., 2012,
2010b;Wolf, 2007;Wolf et al., 2003); other focused on the likelihood
of reaching bi- or multi-lateral agreements between countries (Dinar
et al., 2011;Espey and Towfique, 2004;Zawahri and Mitchell, 2011);
additional analyses used the existence of treaties and River Basin Or-
ganizations (RBO’s) as proxy to quantify the institutional resilience
toward potential hydro-political tensions (De Stefano et al., 2017,2012;
Petersen-Perlman, 2016). The likelihood of cooperating and finally
reaching water agreements is influenced by time-invariant factors, as
for geographical and topographic characteristics, and time-varying
correlates, as for climatic variables and socioeconomic characteristics.
Quantitative analysis was used to find the causal relations leading to
conflicting or cooperative interactions and the formation of treaties.
Wolf (2003, 2007, and 2009) underlined the central role of the quality,
stability and strength of the institutions, highlighting the need for
stronger institutional frameworks to cope with future challenges
(Giordano and Wolf, 2003). Zawahri and Mitchell (2011) argued that
the formation of treaties is a by-product of state interest, transaction
costs, and distribution of power. Dinar et al. (2011) analyzed the main
reasons why some treaties would be more likely discussed in some
basins relative to others. They found that scarcity and cooperation
follow an inversed U-shaped curvilinear relation: cooperation is higher
when water scarcity is moderate, instead of very high/low (also in
Dinar et al., 2010). Extreme scarcity situations were found to be in-
hibiting factors (Dinar et al., 2011). Institutional stability and effective
past agreements oriented toward a fair and efficient water allocation
between riparians were found to be cooperation boosting factors (Dinar
et al., 2015). These and other studies (Beck et al., 2014;Brochmann and
Gleditsch, 2012;Espey and Towfique, 2004) found evidence of the in-
fluence of economic factors, trade dependency, virtual water trade,
presence of water infrastructures, quality of the institutions, govern-
ance, presence of supra-national authorities, cultural background, on
the bi- and multi-lateral relations between the countries facing alloca-
tion, management, and pollution problems over shared water. A large
part of these analyses highlighted the non-linear nature of the relations
between water interactions and correlated factors.
In this study, we propose for the first time the use of a machine
learning approach to quantitatively assess the linear and non-linear
relations between the hydro-political interactions recorded and the
time-varying and time-invariant biophysical, topographic, and socio-
economic explanatory variables. We aim at combining information at
transboundary river basin level with gridded data into an empirically
based data driven index. A similar objective was pursued in the AQU-
EDUCT Water Risk Atlas developed by the World Resources Institute
(WRI) (Gassert et al., 2014,2013). AQUEDUCT did not specifically refer
to hydro-political risk, but rather to a global database of 12 main in-
dicators about water quantity, quality, and regulatory framework, from
about 15000 basins from all over the world that, once aggregated,
formed a composite index defined as overall water risk (Reig et al.,
2013). Similar gridded approach was used to calculate the Global Water
Security Index (GWSI), an index based on information about water
availability, accessibility, quality and management, aggregated through
spatial Multi Criteria Analysis (Gain et al., 2016). Other examples exist
at basin level spatial resolution, such as the Transboundary Waters As-
sessment Programme (TWAP) project (UNEP-DHI and UNEP, 2016). The
hydro-political tension component in TWAP is part of the overall Gov-
ernance indicator. This is based on three sub indicators: 1) Legal Fra-
mework,2)Enabling Environment, and 3) Hydro-Political Tensions. The
first is based on the rationale that governance of transboundary basins
is driven by the existence of bi- or multi-lateral treaties regulating in-
teractions between the countries. Legal Framework is based on the
presence in the treaties of the following principles: (a) “equitable and
reasonable utilization; (b) not causing significant harm; (c) environmental
protection; (d) cooperation and information exchange; (e) notification,
consultation or negotiation; (f) consultation and peaceful settlement of dis-
putes”(quoted from UNEP-DHI and UNEP, 2016). The coverage of all
the legal principles by the previous treaties, jointly with the ratification
of the UN WC Convention and/or UNECE Water Convention by the
countries involved, is considered a factor reducing risk. The Enabling
Environment attains to the single countries’capability of planning,
regulating, managing, and governing water resources (UNEP-DHI and
UNEP, 2016). The level of Hydro-Political Tension is obtained combining
the institutional vulnerability with planned infrastructural develop-
ment, where institutional vulnerability is higher in case the riparian
countries did not specifically regulate in a formal treaty water alloca-
tion and management of flow variability, in case they did not agree on a
conflict resolution mechanism, and in case the basin is not admini-
strated by a RBO (UNEP-DHI and UNEP, 2016). The indicator was de-
signed assigning a score to specific sub-indicators, then aggregated and
ranked, following the methodology developed in existing literature (De
Stefano et al., 2012,2010b,2010a). It is based on information derived
from the water treaties database (International Freshwater Treaties
Database - IFTD) (De Stefano et al., 2010b) created within the Trans-
boundary Freshwater Disputes Database (TFDD) (Wolf et al., 2003). This
work was then further developed in De Stefano et al. (2017). In this
updated version, the current institutional resilience of the trans-
boundary basins was calculated as a function of existing treaties and
river basin managing institutions (RBO’s), similarly to the methodology
used in the TWAP project (De Stefano et al., 2012;UNEP-DHI and
UNEP, 2016). The hydro-political vulnerability of the basins was then
quantified putting in relation with the institutional resilience destabi-
lizing factors, such as planned infrastructural development, and the
exacerbating factors, such as low income, climate driven water varia-
bility, reservoir depletion, armed internal or international conflicts,
past water disputes through a multi-criteria analysis (De Stefano et al.,
2017). The results were produced at basin level: thirty-six river basins
were classified within the high and very high categories of hydro-po-
litical risk.
F. Farinosi et al. Global Environmental Change 52 (2018) 286–313
288
We propose a different, somewhat complementary, approach com-
bining the information at transboundary basin level with local scale
gridded data processed in an empirically based model designed to take
into account linear and non-linear combinations between biophysical
and socioeconomic stressors and international water interactions. In a
second step, rather than assigning scores and aggregating sub-in-
dicators in ranked relative risk categories, we used the model fit with
past observations to construct a baseline and future projected scenarios.
Similarly to other approaches described in this section, our index
combines information at country level with gridded data, but, unlike
previous approaches, our outcome variable is computed at gridded re-
solution. This makes the hereby proposed index spatially explicit and
completely data driven.
2. Methodology and data
The empirically based analysis was designed upon concepts derived
from political science and environmental economics, with a set of in-
dicators selected covering information about: river basin freshwater
availability; climate stress; human pressure on water resources; socio-
economic conditions, including institutional development and power
imbalance; and topographic characteristics. A tool derived from ma-
chine learning, the Random Forests regression algorithm (Breiman,
2001), was used to estimate the relations between the indicators from
each of the groups with observed water interactions. The relative im-
pact of each time-varying and time-invariant indicator was in this way
assessed and empirically estimated using the water related events da-
tabase International River Basin Conflict and Cooperation –IRCC
(Kalbhenn and Bernauer, 2012). The Random Forests regression model
was trained based on historical information covering an eleven years
period (1997–2007). Medium term mean (1997–2012) of the selected
indicators at high spatial resolution (0.25 degrees) was then used to
estimate the spatial distribution of the likelihood of experiencing hydro-
political interactions (baseline scenario). Future scenarios of 2050 and
2100 were calculated by using the multi-model mean of the daily
temperature and precipitation estimates from 5 GCM’s belonging to the
Coupled Model Intercomparison Project Phase 5 (CMIP5) (Taylor et al.,
2012), considering two different emission and radiative forcing sce-
narios, Representative Concentration Pathways (RCP) 4.5 and 8.5
(Meinshausen et al., 2011,2009), for the 15 years period before the
reference time (respectively, 2036–2050 and 2086–2100).
2.1. Data
Data about historical water interactions are the basis for hydro-
political studies. Two main global dyadic databases of historical water
related events are currently available: the Transboundary Freshwater
Dispute Database (TFDD) International Water Event Database (IWED)
developed by the Oregon State University with the Basins at Risk pro-
ject (Wolf et al., 2003;Yoffe et al., 2003,2004; and later updated in De
Stefano et al., 2010b)
3
, providing information about international
water basin interactions between 1948 and 2008; and the International
River Cooperation and Conflict database (IRCC), reporting water re-
lated issues between 1997 and 2007 (Kalbhenn and Bernauer, 2012).
Both databases are set up in the form of water related events at dyad-
basin level. Each national territorial unit in a specific river basin is
defined as a basin-country unit (BCU), each of the possible pairs of
BCU’s in the same basin are classified as a dyad. Although the temporal
coverage (11 years) is limited, the IRCC database was preferred in this
analysis for the higher number of non-neutral interactions reported
(4797 - IRCC vs 1985 - TFDD) (Kalbhenn and Bernauer, 2012), and for
the data collection methodology coded from a homogeneous set of in-
formation (Bernauer and Böhmelt, 2014). The dyadic characterization
of the database, with a geographical scale limited to bilateral country
interactions for each transboundary basin, represents a limiting factor
for a detailed spatial analysis of the biophysical and socioeconomic
drivers determining the national and international water related issues.
Moreover, due to the nature of the algorithms used for the creation of
the database - mining water coded events from international news da-
tasets - the event data are characterized by an uneven geographical
distribution of the observations. More details and alternative water
interactions databases are presented in the Annex A.
The hydro-meteorological information used in this analysis were
derived from the highly spatially detailed climate data from the Multi-
Source Weighted-Ensemble Precipitation (MSWEP) database (Beck
et al., 2017). We calculated a precipitation anomaly indicator based
only on variation in the temporal distribution of precipitation: the
Standardized Precipitation Index (SPI) (McKee et al., 1993). This cli-
mate proxy, measuring rainfall anomalies, is widely used for drought
quantification and monitoring (WMO, 2012), (details in Annex A).
Temperature data were derived from the WATCH Forcing Data meth-
odology applied to ERA-Interim (WFDEI) dataset (Weedon et al., 2014).
Water availability was assessed using a modified version of the Falk-
enmark Water Stress Indicator (Falkenmark and Lannerstad, 2005),
considering also the water resources flowing from upstream, calculated
using the 0.1 degrees resolution LISFLOOD global hydrological model
(De Roo et al., in preparation). River basin topographic data used for
the analysis were mainly represented by the river flow accumulation,
proxy for the upstream/downstream relations, and the share of national
territory in the basin (Beck et al., 2014).
Gross Domestic Product (GDP) statistics were derived from
Gleditsch (2002). The Governance indicator was calculated as mean
value of the six indicators (voice and accountability; political stability
and absence of violence; government effectiveness; regulatory quality;
rule of law; control of corruption) of the Worldwide Governance In-
dicators (WGI) project (Kaufmann et al., 2010). Agriculture (share of
GDP) and rural population (share of the total) were derived from the
World Development Indicator database (World Bank, 2018.). Popula-
tion dynamics were derived from the Gridded Population of the World
(GPW, v4) database (CIESIN, 2015) downscaled by the EC Joint Re-
search Centre (Freire and Pesaresi, 2015). Political and military im-
portance of the countries was represented in the model through the
Composite Index of National Capability (CINC) derived from the Na-
tional Material Capabilities (NMC v5.0) database within the Correlates
of War project (CoW) (Singer et al., 1972)
4
. CINC is calculated as a
share of the world power as function of six variables, namely: total
population, urban population, iron and steel production, military ex-
penditure, military personnel, and primary energy consumption (Singer
et al., 1972). The information about past bi- or multi-lateral water
treaties were derived from the International Freshwater Treaty Data-
base –IFTD (Oregon State University, Transboundary Freshwater Dis-
pute Database TFDD)
5
(De Stefano et al., 2012).
The climate projections data used in this study belong to the NASA
Earth Exchange Global Daily Downscaled Projections (NASA NEX-
GDDP) dataset downscaled (0.25 degrees) and bias corrected using the
Bias-Correction Spatial Disaggregation (BCSD) methodology described
in Thrasher et al. (2012). Due to computational constraints, we selected
5 out of the 21 climate models included in the NASA NEX-GDDP (details
in Annex A), chosen on the basis of the structural differences among
them, as described in Knutti et al. (2013).
Population density for the years 2050 and 2100 were estimated
applying the country specific population growth rates estimated by the
World Population Prospects of the UN/DESA (UN/DESA, 2017) to the
population density used for the baseline scenarios (CIESIN, 2015;Freire
3
www.transboundarywaters.orst.edu;http://gis.nacse.org/tfdd/index.php
4
http://cow.dss.ucdavis.edu/data-sets/national-material-capabilities/
national-material-capabilities-v4-0
5
www.transboundarywaters.orst.edu
F. Farinosi et al. Global Environmental Change 52 (2018) 286–313
289
and Pesaresi, 2015). Main statistics and variable description are sum-
marized in Annex A (Table A1); further information about data sources
could be found in Table A2.
2.2. Methodology: random forests regression
Different methodologies have been used in literature to analyze
dyadic data. Most of them were not designed to capture the non-linear
interactions. For this reason, in this work we propose a different ap-
proach applying the Random Forest (RF) regression algorithm
(Breiman, 2001). RF is a Classification and Regression Tree (CART)
based tool that involves an ensemble of regression trees. These are
calculated on random subsets of data randomly split in base of specific
features of each of the independent variables (Liaw and Wiener, 2002;
Strobl et al., 2009;Welling et al., 2016). RF is based on the decision
trees learning approach popular for non-linear multi-variate classifica-
tion and regression (Breiman, 2001;Tin Kam Ho, 1998). In this study,
we will refer to the RF regression, which is slightly different from the
classification algorithm and is structured in four subsequent steps de-
scribed below (RF algorithm logical steps, calibration, and validation
procedures are summarized in Annex B).
RF Model training: the model was used to find the linear and non-
linear relations between the dependent variable, a logarithmic trans-
formation of the number of water interactions for each of the country-
dyad/basin combinations observed for the 11 years (1997–2007)
available data, and the 19 independent variables selected for the ana-
lysis (variable selection was performed optimizing the model perfor-
mance as described in Annex B). Out of the 19 variables used in the
final specification of the model (see Table A1 for details), the 2 re-
presenting the basin’s topography were time invariant, while the re-
maining 17, representing biophysical and socioeconomic factors were
time varying (one is the time trend control). The interpretation of the
modeling set up should be intended as the relation between a percent
variation of the objective variable (the measure of the intensity of water
issues), in response to the variation of absolute values of the in-
dependent variables. Since the relations, in the majority of the cases,
are non-linear, by manipulating an independent variable, the variation
of the objective variable could be positive or negative depending on the
values of the remaining set of independent factors.
Baseline: the RF model set up in the previous step was used to
construct a baseline (or reference) scenario of the likelihood of hydro-
political issues at grid-cell level (each cell has dimensions 0.25 × 0.25
degrees, approximately 27 × 27 km at the equator). In order to reduce
the bias derived by climate variability and possible temporary shocks in
the specific independent variables, the baseline scenario was calculated
by averaging the values of the independent variables for the period
1997–2012 at grid-cell level. Variables’values at grid-cell level are cell
specific (as for the 8 climatic variables
6
; population density
7
; and water
availability
8
) or the same for all the cells of a country (as for all the
socioeconomic variables
9
). The production of a baseline or reference
scenario results in the possibility to map the spatial distribution of the
likelihood of having water interactions, our index, at global level, upon
present conditions of the factors determining water interactions.
Projections: using a procedure similar to the one described for the
baseline, the model was used to map the variations on the objective
variable as a response to four possible future climate and population
scenarios. The future conditions are based on climate projections to the
years 2050 and 2100 based on two different degrees of climate change
(RCP 4.5 –moderate climate change scenario; and RCP 8.5 –severe
climate change scenario). In order to reduce the bias derived from the
specific climate modeling exercises, we averaged in a multi-model
mean climate projections from 5 GCMs downscaled and bias corrected.
Climate projections were combined with population growth scenarios
at grid-cell resolution, calculated applying to the baseline population
density (CIESIN, 2015;Freire and Pesaresi, 2015), country specific
population growth rates for the years 2050 and 2100 (UN/DESA, 2017).
Comparison of the future and baseline scenarios to assess the change
in the index caused by population and climate dynamics.
3. Modeling results
3.1. Random forest model results
The RF model was trained using the entire set of observations
(N= 11801). Each of the observations reports the logarithmic trans-
formation of the number of hydro-political interactions for a specific
dyad of countries (749 country dyadic combinations considered in the
final panel) in a specific river basin (260 transboundary basins in-
cluded) for a specific year (11 years). Of the final 11801 observations
considered, 10062 reported no water interactions, while 1739 at least 1
interaction in the combination BCU/year
10
. The overall RF model was
found to explain about 70% of the variation (pseudo R
2
, details avail-
able in Annex B). Variable importance estimates for the RF model
highlighted that socioeconomic variables play the most important role.
Population density was the variable that mostly influenced the cap-
ability of the model to capture the variation of the set of observations
taken into account in this analysis. Time trend control resulted to be the
second most important variable in capturing the variability of the data:
this is likely due to the data collection algorithm of the hydro-political
event dataset strongly influenced by the increasing of news availability
in the period under consideration, coincident with internet develop-
ment. The upstream/downstream dynamics (represented by the flow
accumulation), jointly with territorial (area difference) and power im-
balance (Composite Index of National Capability - CINC) follow the
population dynamics. Per capita water availability (Falkenmark Index)
was reported as the most important of the biophysical variables, while
variables associated with precipitation and temperature follow in the
mid-lower portion of the permutation-based variable importance
ranking (Fig. A3).
The performed analysis highlighted the non-linear nature of the
relations between certain variables and their impact on the hydro-po-
litical interactions (further details in Annex B; partial dependence plots
in Fig. A4). The model finds an increasing inverse U-shaped relation
between population density and water interactions: sparsely populated
areas were associated with a lower probability of having water issues;
the likelihood increases till reaching its maximum at about 100 people
km
−2
. Above this value the relation decreases remaining positive and
leveling to zero for values above 400 people km
−2
. Almost opposite
results are found for the Falkenmark Index, indicating per capita water
availability including the amount of resources flowing from upstream:
in the areas where the water availability is the lowest, increasing values
are associated with a marginal decrease in the likelihood of water is-
sues. The slightly negative relation is however non-linear: it is positive
in areas where relatively more water is available and almost negligible
in water abundant areas. Relative territorial supremacy on the basin
(difference in the national territory in the shared watershed) was found
to have an inverse U-shaped relation: the likelihood of water interac-
tions appears to be very low among actors occupying similar territorial
extensions of the shared river basin; similar conclusions could be drawn
for countries occupying the majority of the basin territorial extensions,
6
TOT_Precip,MIN_Precip,SPI_12,AVG_Temp,TempMAX,TempMin,Temp_delta,
Temp_seasonal_var in Table A1.
7
Pop_density in Table A1.
8
Falkenmark_upst in Table A1.
9
Rural_pop, GDP, Agriculture, Governance_ind, cinc_mean, IFTD_treaties in Table
A1.
10
With a maximum of 166 interactions between Hungary and Romania in
the Danube river basin in the year 2000.
F. Farinosi et al. Global Environmental Change 52 (2018) 286–313
290
while hydro-political interactions are found to be more likely in the
middle cases. Low to medium levels of national power (composite index
of national capability) were found to be associated with higher like-
lihood of experiencing water interactions. Very upstream and very
downstream countries are found to be more likely to get involved in
water interactions. Rural and agricultural dependent economies and, in
general, lower to middle income countries are more prone to experience
water issues.
3.2. Model findings discussion
Socioeconomic and water demand side factors are found relatively
more important in determining hydro-political interactions respect to
supply side factors like shocks in precipitation or other climatic vari-
ables. Similar findings were highlighted in Böhmelt et al. (2014) where
population pressure, agricultural productivity, and in general economic
development were identified as important determinants for the for-
mation of water disputes, mitigated into cooperative interactions in
case of solid institutions and stable political conditions. Population
dynamics were found important drivers in other studies. Brochmann
and Gleditsch (2012), among others, found that countries characterized
by very large or very small population are more likely to get involved in
conflicts over water. In our case, population density is a proxy of human
pressure over water resources, but population is also linked with the
power of a nation and its economic and socio-political capabilities. Very
low densely populated areas were found to be less likely to experience
water interactions, but in case of rural communities (> 50% of the
population living in rural areas) extremely dependent on agricultural
productivity (> 30% of the GDP) for their economic development, the
combination of the three factors was found likely to experience water
issues (a 3-D dependence plot available in Fig. A5).
Increasing population density, by increasing human pressure on a
limited set of resources, was found likely to increase the probability of
experiencing water related issues, but this relation was found to be not
linear. This could be explained considering the role of hydraulic in-
frastructures in mitigating water stress in densely populated areas
(McDonald et al., 2014), and the extreme consequences when the ca-
pacity of the water infrastructures is no longer sufficient to cope with
climatic variability and population growth. Similarly, in Dinar et al
(2011) increasing human pressure on water resources, determining
water scarcity, was found to have an inverse U-shaped relation with
cooperative hydro-political interactions, while extreme cases were
more associated with tensions. The inverse U-shaped relation between
per capita water availability and likelihood of hydro-political interac-
tions confirms also the conclusions of Dinar et al (2010), that found
cooperative water interactions more likely in situation of average water
availability. Territorial and power imbalance were found significant
drivers of hydro-political interactions in the main literature available
(Brochmann and Gleditsch, 2012;Gleditsch et al., 2006;Zawahri and
Mitchell, 2011). This study’sfindings about upstream/downstream
dynamics confirm the accurate study performed by Munia et al. (2016)
quantifying the increasing water stress in the downstream part of the
basins due to upstream uses, and its connection with increasing water
tensions. Our results found an increasing trend of water related inter-
actions over time. On the one hand, the institutional development
brought an increasing collaboration over water related international
issues (De Stefano et al., 2012;Dinar et al., 2015;Kalbhenn and
Bernauer, 2012;Wolf, 2009). On the other hand, the trend is (at least
partially) explained by the increasing coverage of the international
press industry of the local news about water issues. The way water
event datasets were developed, in fact, is strongly influenced by the
publication of news in the main western languages: this sector has been
radically influenced by the digital revolution. As noted in De Stefano
et al. (2010b), the scarce representation of some areas of the world in
the water related events datasets is mainly due to the fact that the
search was performed analyzing international and local news in Eng-
lish. This methodology proved to be rather unsuccessful in capturing
information published in local languages or news from area not com-
pletely covered, such as war zones or politically or technologically
isolated countries. For this reason, data about historical water related
events represent the main limitation of the studies in this specificfield.
4. Model application to calculate the likelihood of hydro-political
interactions under current and upcoming conditions
One of the main objectives of this study was to draw a spatially
explicit data driven index aimed to help the policy makers in mon-
itoring the dynamics of the factors identified as influential in de-
termining water related issues, and in identifying the areas where co-
operation over water is more needed to timely address criticalities that
could eventually lead to water disputes. In order to achieve this ob-
jective, we calculated the medium-term mean (1997–2012, when
available) of the selected indicators at the highest spatial resolution
allowed by data availability (0.25 degrees), and we used the estimated
RF model to draw the spatial distribution of the likelihood of hydro-
political interactions. Not all the variables were available at sub-
country resolution, in particular: 10 variables were available at grid-cell
level; 5 at country level; 3 at country/basin level (more details in Table
A1). The spatial distribution of the index within the country borders is
therefore driven by climatic, population, and water availability drivers:
an unavoidable simplification caused by the limited availability of data
at sub-country and gridded level, partially compensated by the fact that
the spatial distribution of some variables, as for the national capability
(CINC), can be considered fairly homogeneous at intrastate level. A
high likelihood of hydro-political interactions identifies the areas where
water issues are more probable to raise. Although this index does not
give information about the degree of cooperation or conflict associated
with the specific interaction, it identifies the areas of possible hydro-
political risk that would be best addressed through a cooperative action
(Fig. 1). The index was calculated at pixel level, the values attributed to
each specific basin is the average of all the pixels within its boundaries.
To ensure the comparability of the different variables and indicators,
the corresponding values were normalized across the transboundary
basins in a 0–1 range through a simple min-max normalization proce-
dure. High values of the likelihood of hydro-political interactions are
linked with a larger water stress, due to lack of water supply and/or
human pressure in a more vulnerable institutional and socioeconomic
context. The spatial distribution of the index highlights the areas where
it could be more likely to experience issues related to water resources.
High likelihood of water related issues could be determined by poten-
tial water scarcity in densely populated areas, as in the case of the Nile
Delta, one of the basins that reach an high average value of the index
(score 0.761). Socioeconomic, political conditions and distribution of
water resources determine the differences in the index for the Upper
Nile. A combination of low governance, high population density, phy-
sical water stress, and almost complete economic dependency on agri-
cultural activities, shaped the distribution of the index in the Ganges-
Brahmaputra (highest in our ranking, score 1.000), and Indus basins
(score 0.675). A different climatic area, more pronounced precipitation
stress, with a lower population density and lower economic dependency
on agricultural production characterized the results for the Euphrates-
Tigris river basin (score 0.592). Population density, high economic
dependence on agriculture, and human pressure on water resources
determine the distribution of the index on the lower Niger (score
0.447), in particular within the borders of Burkina Faso and Nigeria.
Population distribution and socioeconomic conditions shape the index
F. Farinosi et al. Global Environmental Change 52 (2018) 286–313
291
in the Congo basin (score 0.432), while a relatively good governance
level characterizes the Zambezi river basin, with hotspots in the most
populated areas, and increasing values towards the outlet of the basin
(overall score 0.431). Human pressure and relatively heterogeneous
socioeconomic conditions determine the need for water cooperation in
the Mekong basin (score 0.492). Despite the evident progresses made
after the EU integration, our results highlight high likelihood of water
related issues in specific portions of the Danube basin (score 0.499),
especially in the eastern and southern parts, where there is still need to
consolidate institutional development and the economic dependency on
agriculture still remains relevant. A complete list of the results for the
transboundary river basin is available in the Annex D (Table A3).
Some of the areas highlighted in the results shown above (and in
Table A3) are well known hotspots for hydro-political issues. Other
areas are scenarios of national or international political tensions not
directly related with water. Although, given the different nature of our
study focusing on water interactions as a measure of the magnitude of
water issues, a direct comparison with previous studies aiming to
identify basins at risk of future water tensions is not possible, the results
of the different approaches are aligned. De Stefano et al. (2017) com-
pared the basins at risk identified using their approach with the ones
highlighted in the two previous assessments (Bernauer and Böhmelt,
2014;Wolf et al., 2003). Of the 12 basins found to be more likely to
experience water issues in this study (Table A3), 10 are identified as
basin at risk in previous analyses, namely: Ganges/Brahmaputra (Wolf
et al., 2003), Pearl/Bei Jiang (De Stefano et al., 2017), Nile (Wolf et al.,
2003), Feni (or Fenney) (Bernauer and Böhmelt, 2014), Indus
(Bernauer and Böhmelt, 2014;Wolf et al., 2003), Colorado (Bernauer
and Böhmelt, 2014), Tarim (De Stefano et al., 2017), Shatt al-Arab -
Tigris/Euphrates (Bernauer and Böhmelt, 2014;Wolf et al., 2003), Hari
(Bernauer and Böhmelt, 2014), and Irrawaddy (De Stefano et al., 2017;
Wolf et al., 2003). Therefore, the probability of observing hydro-poli-
tical interactions is to some extent correlated with the hydro-political
risk analyses conducted in previous studies identifying basins at risk.
That supports the idea that the index proposed herein should be con-
sidered for systematic application in support to the assessment of the
SDG 6, in particular for what concerns the impacts of future potential
biophysical or socio-environmental changes on the likelihood of hydro-
political issues at global scale. The proposed index can also be used to
assess interlinkages with other SDGs, and in particular SDG 16 on
peace, justice and institutions. In order to achieve a global perspective,
our analysis was extended also outside the borders of the international
river basins initially included in the analytical framework (Fig. 2 and
Fig. A7 in the Annex). The results outside the boundaries of the inter-
national river basins and in the portions of them not or poorly re-
presented in the database of hydro-political events used to fit the RF
model, might be affected by certain degrees of error and, that for,
should to be considered purely indicative.
The evolution of the index under future climate and population
scenarios was estimated for the years 2050 and 2100 considering
changes in population density, by applying UN/DESA population
growth estimates to the 2015 data, and climate conditions, considering
the multi-model-mean of the projected precipitation and temperature
for the periods 2036–2050 and 2086–2100 (Fig. 3 –Additional details
in the Annex –Figs. A8 and A9). As mentioned above, population
density is among the top drivers determining the likelihood of hydro-
political interactions, while, conversely, climate factors are relatively
less important in terms of magnitude, but more relevant in terms of
impacted area extent. The reason for choosing the combination of cli-
mate and population dynamics as driver for change is motivated mainly
by data availability. When alternative scenarios of other important
variables and relevant dynamics, as for instance the institutional
Fig. 1. Likelihood of the occurrence of hydro-political interactions in the main transboundary river basins (from the top-left [normalized likelihood of hydro-political
issues, min = 0 and max =1]: Ganges-Brahmaputra [1.000], Nile [0.761], Indus [0.675], Euphrates-Tigris [0.592], Danube [0.499], Mekong [0.492], Aral Sea
[0.455], Niger [0.447], Congo [0.432], Zambezi [0.431], Senegal [0.372]). In the radar chart the normalized score of the main factors determining the likelihood in
the specific river basins. Not all the variables explicitly used for the model are represented in the radar chart: the non-included factors, however, are derived from the
climatic variables displayed.
F. Farinosi et al. Global Environmental Change 52 (2018) 286–313
292
development, will be available, these could be taken into consideration
as well.
Changes in population density are expected to exacerbate the an-
thropogenic pressure on water resources, the availability of which is
strongly impacted by changes in climate. The combination of these two
factors is estimated to impact negatively on the overall hydro-political
risk. The likelihood of water related issues is expected to increase
globally, with gridded values averaging between +36.4% in the case of
moderate climate change (RCP 4.5) and +37.1% in the case of the
more pessimistic climate scenario (RCP 8.5) for the year 2050, and
respectively between +39.3 and +46.8% for the year 2100.
Aggregating the results for the main transboundary river basins,
excluding the areas of the globe not falling in transboundary basins, the
likelihood of experiencing hydro-political interactions was calculated to
increase on average between 74.9% (2050 RCP 4.5) to 95.3% (2100
RCP 8.5), especially in sub-Saharan Africa, South America, Southern
North America, Southern and Eastern Europe, Central and Southern
Asia. Table 1 presents the main statistics for the global projections, and
the results for the transboundary basins most represented in the original
IRCC database that were found to be likely of experiencing more hydro-
political interactions in the future. The convergence of the increasing
trends in population density and temperature, jointly with decreasing
precipitation is the combination that most influences the future in-
creasing hydro-political risk, as for instance in the case of Southern
Fig. 2. Global distribution of the current likelihood of hydro-political issues among the main transboundary basins (transboundary basin borders in black, non-
transboundary areas shaded).
Fig. 3. Change in the likelihood of hydro-political issues considering the four future climate change and population scenarios respect to the baseline presented in
Fig. 2.
F. Farinosi et al. Global Environmental Change 52 (2018) 286–313
293
Europe, Central Asia, and Middle East (Figs. A8 and A9). Increasing
population and temperature were found to be dominant respect to in-
creasing precipitation, as in the case of some tropical areas in sub-Sa-
haran Africa and South-East Asia, in some cases due to the seasonal
distribution of the rainfall. Harsher climate conditions were found to
offset the benefits derived by decreasing population density, as in the
case of North-Eastern China in the second half of the 21st Century.
Only a handful of transboundary basins are expected to benefitor
not being impacted by the global climatic and population changes: one
in Central Asia, Chuy Basin (from -8% 2050 RCP 4.5 to -37% 2100 RCP
4.5); two in the North of the Scandinavian peninsula: Tuloma Basin
(between Russia and Finland, from -3% 2050 both RCP’s to +3% 2100
RCP 8.5), and Näätämö basin (at the border between Finland and
Norway, -3.2% 2050 RCP 8.5 to +1% 2100 RCP 8.5); and two in
Ireland: Bann Basin (-13.4% 2100 RCP 4.5 to +2% 2100 RCP 8.5), and
Flurry Basin (-17.4% 2100 RCP 4.5 to -1.6% 2100 RCP 8.5). All these
basins are characterized by low population density and, the ones in the
northern latitudes, abundant water availability. A detailed list of the
projected population and climatic variables, and the estimated results
in terms of hydro-political risks are available in the Annex (Figs. A8,
A9, and Table A3, respectively).
The increasing pressure that future climate and population dy-
namics are expected to pose upon the already problematic basins,
especially in the Sahelian and Sub-Saharan Africa, Central, South and
South-Eastern Asia, should be carefully monitored in order to avoid
eventual hydro-political turmoil. In particular, the institutional and
governance capacity of the national and supranational institutions
(RBO’s) should be enhanced in order to minimize the vulnerability of
the specific biophysical and socioeconomic basin-systems to the in-
creasing pressure. This aspect could significantly increase the capability
of the river systems to deal with the increasing magnitude of change.
5. Conclusion
In this paper, we presented an innovative analysis of the past hydro-
political issues in international river basins and their determinants
through the application of the Random Forest regression algorithm. Our
analysis had two main goals: highlighting the factors that are more
relevant in determining the hydro-political interactions, capturing also
the non-linear relations between the main drivers; and producing a tool
able to map and monitor the evolution of the hydro-political risk over
space and time, under specific socioeconomic and biophysical sce-
narios. We did that by designing an empirically estimated, data-driven,
and spatially explicit global index of the magnitude of hydro-political
issues. The factors that were found to be more relevant in determining
hydro-political interactions were mainly represented by, respectively:
population density, water availability (quantified through the
Falkenmark index), upstream/downstream dynamics (represented by
the flow accumulation), with territorial (area difference) and power
imbalance (Composite Index of National Capability –CINC), and cli-
matic conditions. Current climatic and socioeconomic conditions were
used to design a baseline scenario of the distribution of the likelihood of
hydro-political interactions. This output allows to map the spatial dis-
tribution of the areas within the basins where water management issues
are more likely to rise under current conditions. Among the basins
found to be more likely to experience water issues in this study, some
were already identified as basin at risk in previous analyses, namely:
Ganges/Brahmaputra, Pearl/Bei Jiang, Nile, Feni (or Fenney), Indus,
Colorado, Tarim, Shatt al-Arab - Tigris/Euphrates, Hari, and Irrawaddy.
The hereby proposed index adds the possibility to identify the most
critical areas within the basin boundaries. The baseline scenario was
then compared to four distinct climate and population density projec-
tions, designed by combining the most updated bias corrected and
spatially detailed climate and the most recent estimates of the future
population changes. The results of this work allow the identification of
the areas where water interactions are more likely to arise under pre-
sent and upcoming conditions, and cooperation over water should be
pursued to avoid possible hydro-political tensions. Future demographic
and climatic conditions are expected to heavily increase the probability
of experiencing water management issues in already stressed basins,
such as the Nile, the Indus, the Colorado, the Feni, the Irrawaddy, the
Orange, and the Okavango.
One of the characteristics of the analysis presented is that we chose
not to make a distinction between past episodes of cooperation and
dispute over water, using them collectively as water interactions, a
measure of the magnitude of the associated water issue. This was mo-
tivated by the fact that water disputes had virtually never ended in
violent conflicts, at least in the most recent historical eras, and by the
consideration that the classification of positive (cooperative) and ne-
gative (conflictive) interactions in the event databases has often been
arbitrary and ambiguous. Our focus was then more oriented towards
understanding the preconditions increasing the likelihood of
Table 1
Summary of estimated change of the likelihood of experiencing hydro-political interactions under four future projected scenarios. Data are presented aggregated per
geographic areas or river basins. Values are presented as average (minimum and maximum variation). A more comprehensive table is presented in the Annex (Table
A3).
2050 RCP 4.5 2050 RCP 8.5 2100 RCP 4.5 2100 RCP 8.5
Avg % change
(Min / Max)
St. Dev Avg % change
(Min / Max)
St. Dev Avg % change
(Min / Max)
St. Dev Avg % change
(Min / Max)
St. Dev
Globe 36.4 (-72/+5944) 56.3 37.1 (-71.5/+5861) 56.9 39.3 (-76.5/+5120) 60.9 46.8 (-69.9/+5235) 67.5
Transboundary basins
(all)
74.9 (-61/+5944) 66.7 76.2 (-61/+5861) 97.3 80.7 (-66/+5120) 72.1 95.3 (-57/+5235) 79.4
Lake Chad 77.2 (+12.8/+439) 48 76.7 (+11.7/+439) 48.3 85.7 (+10.9/+567) 67.3 78.4 (-4.2/+557) 64.4
Congo 70.9 (-13.7/+547) 71.6 71 (-12/+546) 70.9 78.1 (-4.6/+601) 75.7 83.3 (-3.6/+514) 72.7
Niger 64.1 (-2.3/+346) 50.6 62.3 (-3.6/+333) 47.9 76 (-3/+378) 68.8 66 (-8.1/+339) 59.5
Nile 43.3 (-35.5/+599) 70.5 43.1 (-35.5/+607) 69.9 45.2 (-30.1/+697) 85 42.4 (-32.5/+734) 84.2
Zambezi 38.9 (-6.4/+321) 34.2 38.5 (-5.8/+312) 33.8 48.4 (-7.7/+418) 47.4 47.1 (-10.2/+342) 43.7
Senegal 36.7 (-5.7/+234) 48.7 36.4 (-5.9/+235) 48.5 45.4 (-5.8/+265) 59.4 41 (-5.1/+247) 51.7
Aral Sea 33.2 (-11.2/+249) 30.1 34.2 (-12.9/+252) 31.1 35.6 (-12.1/+259) 32.1 41.9 (-16.4/+292) 37.4
Euphrates-Tigris 23.1 (-2.5/+349) 46.4 23.5 (-2.3/+364) 48.4 26.5 (-3.5/+446) 50.4 32.5 (-5.3/+563) 59.1
Danube 24.2 (-55/+510) 82.7 26 (-55.5/+518) 84.7 19.2 (-66.1/+555) 88.3 34.7 (-52.3/+651) 110
Indus 12.3 (-27/+169) 26.5 12.5 (-32.5/+168) 26.8 15.3 (-34.9/+224) 30.3 19.1 (-32.3/+262) 35
F. Farinosi et al. Global Environmental Change 52 (2018) 286–313
294
experiencing hydro-political interactions due to emerging water man-
agement issues. More than being exhaustive, our approach tends to
boost the interest in the hydro-political field of study, by offering a new
perspective through the application of a methodology that had never
been considered before in this kind of analyses, dealing with aspects
that are different by the only institutional resilience, and by exploring
the possibility of creating a spatially explicit interactive tool able to
assist stakeholders and policy makers in dealing with water related is-
sues in different socioeconomic and climatic contexts through the
analysis of what-if scenarios. Future studies could further develop the
instrument by integrating updated socioeconomic, biophysical, and
demographic projections.
The difficulties and the limitations encountered in this process were
multiple. Beside the logical constraints that every global analysis has, as
the other studies in this field, this work is affected by limitations in data
availability. Water events database are extremely hard and expensive to
collect and to manage. Data collection is mostly conducted through the
application of mining algorithms operating in the news databases
available only in the most widely spoken western languages. For this
reason, the available datasets are necessarily biased and incomplete.
Their temporal coverage is very limited, only eleven years in our case,
and the sub-national geographic characterizations of the specific water
related events is, in the majority of the cases, not considered. These
particular factors make very difficult to apply the existing datasets for
the development of spatially explicit interactive decision making tools.
As stated above, the index presented in this paper could be applied
for the Agenda 2030 monitoring activities and in particular for Target
6.5 –Water Resources Management, where the only indicator regarding
hydro-political dynamics used is the 6.5.2 Proportion of transboundary
basin area with an operational arrangement for water cooperation. This is
an indicator capturing mainly the institutional resilience in trans-
boundary basins, with no consideration for the other determining fac-
tors specifically analyzed in this study. Therefore, the use of the pro-
posed index could provide a substantial contribution to move from the
mere recording of facts, to the understanding of phenomena the me-
chanisms behind them, which are prerequisites for identification of
effective sustainability policies.
As noted already in previous global analyses (Bernauer and
Böhmelt, 2014;De Stefano et al., 2017;Yoffe et al., 2003), the results of
this study should be intended to be an indicator of the areas that might
require closer investigation under present and possible upcoming sce-
narios. We recommend to further explore the development of this
analysis in regional or sub-regional contexts where more detailed data
is available. Future research will be focused in specific transnational
river basins in developing countries where potential water stress ex-
acerbated by climate change and variability, rapid population growth,
and unsustainable development could be further destabilizing factors
for the already tumultuous political context.
Author attribution
F. Farinosi, G. Bidoglio, A. Reynaud, and C. Carmona-Moreno de-
signed the study; F. Farinosi and G. Ceccherini developed the modeling
framework; F. Farinosi processed data, coded the methodology, and
performed the analysis; F. Farinosi, C. Giupponi, A. Reynaud, A. De
Roo, G. Bidoglio, and C. Carmona-Moreno discussed the results; F.
Farinosi with comments from the co-authors wrote the manuscript.
Conflict of interest
The authors declare no conflict of interest.
Funding
Arnaud Reynaud gratefully acknowledges the financial support of
the Research Chair "Finance Durable et Investissement Responsable"
and the Research Chair Amundi.
Acknowledgments
Climate projection scenarios used were from the NEX-GDDP dataset,
prepared by the Climate Analytics Group and NASA Ames Research
Center using the NASA Earth Exchange, and distributed by the NASA
Center for Climate Simulation (NCCS). The authors would like to thank
Stefano Barchiesi (JRC) and Mehmet Pinar (Edge Hill University) for
the valuable comments provided during the preparation of the manu-
script, Tatevik Hovhannissian and Ilenia Babetto (WFP) for proof-
reading the final product.
Annex A Data specification and source
IRCC and TFDD water events databases
The Transboundary Freshwater Dispute Database (TFDD) International Water Event Database (IWED) (Wolf et al., 2003;Yoffe et al., 2003,2004;
De Stefano et al., 2010b) provides information about international water basin interactions between 1948 and 2008; the International River Co-
operation and Conflict database (IRCC), reports water related issues between 1997 and 2007 (Kalbhenn and Bernauer, 2012). Both databases are set
up in the form of water related events at dyad-basin level. Each national territorial unit in a specific river basin is defined as a basin-country unit
(BCU), each of the possible pairs of BCU’s in the same basin are classified as a dyad. Water related events (or interactions) are classified in the basis of
a scale assigning a score representing the intensity of the issue, and its nature (conflict/cooperation): +6 most cooperative, -6 most confrontational
in the IRCC case (Kalbhenn and Bernauer, 2012); -7/+7 in the TFDD case (Yoffe et al., 2004;Yoffe and Larson, 2002). The interactions related to the
same water issue involving two or more BCU’s are clustered in a specific“case”(multiple events could be attributed to each water case), representing
the hydro-political issue determining the interactions between the countries sharing a watershed. Although the temporal coverage (11 years) is
limited, the IRCC database was preferred in this analysis for the higher number of non-neutral interactions reported.
Due to the nature of the algorithms used for the creation of the database, mining water coded events from international news datasets, the event
data are characterized by an uneven geographical distribution of the observations. About 4800 of the 5881 observed events refer to the most
represented international river basins, namely Danube, Nile, Zambezi, Mekong, Euphrates/Tigris, Ganges/Brahmaputra, Aral Sea, Elbe/Labe
(Kalbhenn and Bernauer, 2012). In total, the IRCC dataset counts 15965 entries (5881 events and 10084 combinations of dyad countries-basin-year
with no events), it presents data about 262 transboundary basins, 760 dyads countries, and 1279 combinations basin/dyads (respectively 261, 725,
and 1249 for the TFDD dataset) (Kalbhenn and Bernauer, 2012). Due to data limitation, 11 dyads and 2 minor basins (total of 43 observations, 21 of
which non-zero)
11
were excluded.
11
Due to data limitation, the observations including Brunei Darussalam (a total of 43 - 21 with interactions in the Mekong river for joint management in the
context of the ASEAN political talks, and 22 without) were excluded. After removing these observations, the IRCC panel remained with 15,922 entries (5,860 events
vs 10,062 dyad-basin-year combinations with no events).
F. Farinosi et al. Global Environmental Change 52 (2018) 286–313
295
Table A1
Descriptive statistics of the data used in the analysis.
Group / Sub-group Name Description Abbreviation Unit Mean SD Min Max Spatial resolution Time varying?
Biophysical
Falkenmark index Per capita water available including upstream flow Falkenmark_upst m
3
year
−1
52,816 80,549 146 904,414 Grid-cell Y
Precipitation Total precipitation TOT_Precip mm year
−1
1,028 596 31 4,971 Grid-cell Y
Minimum precipitation MIN_Precip mm year
−1
764 607 0 4,910 Grid-cell Y
Standard Precipitation Index SPI_12 –−2.5 1.66 −10.8 1.7 Grid-cell Y
Temperature Average temperature AVG_Temp °C 16.36 8.86 −12.81 30.02 Grid-cell Y
Maximum temperature TempMAX °C 20.01 8.44 −10.51 33.68 Grid-cell Y
Minimum Temperature TempMin °C 9.95 10.68 −26.39 28.79 Grid-cell Y
Temperature variation Temp_delta °C 3.21 3.71 0 25.69 Grid-cell Y
Seasonal variability Temp_seasonal_var Stand. dev. 2.93 2.51 0.03 15.85 Grid-cell Y
Socioeconomic
Population Population density Pop_density People sqkm
−1
67.62 89.30 0 1,433 Grid-cell Y
Share of rural population Rural_pop % 49.42 19.52 Country Y
GDP Per capita Gross Domestic Product GDP 2005 USD 8,659 10,013 313 59,384 Country Y
Agricultural dependency of the economy Agricultural share of the GDP Agriculture % 17.60 12.92 0 61.96 Country Y
Institutional development and quality Worldwide Governance Indicators (WGI) Governance_ind Normalized value (range: -2.5/+2.5) −0.274 0.83 −1.939 1.910 Country Y
Power imbalance Composite Index of National Capability cinc_mean Normalized value (range: 0/1) 0.010 0.021 0 0.136 Country Y
Previous collaboration Number of bi-or multi-lateral water treaties IFTD_treaties number 17.84 15.61 1 95 Country/
basin
Y
Topography
Flow accumulation % of the flow accumulated in the country flow_acc % 0.436 0.31 0 1 Country/
basin
N
Territorial imbalance Difference of national territory in the basin area_diff‘000 km
2
190.8 455.4 0 3,739 Country/
basin
N
Others
Time trend Time trend variable year –––––Y
Projected Scenarios - Climate
Precipitation Total precipitation 2050 rcp 4.5 TOT_Precip_2050_45 mm year
−1
723 691 0 7,359 Grid-cell Y
Total precipitation 2050 rcp 8.5 TOT_Precip_2050_85 mm year
−1
730 695 0 7,566 Grid-cell Y
Total precipitation 2100 rcp 4.5 TOT_Precip_2100_45 mm year
−1
745 715 0 7,780 Grid-cell Y
Total precipitation 2100 rcp 8.5 TOT_Precip_2100_85 mm year
−1
794 765 0 8,879 Grid-cell Y
Minimum precipitation 2050 rcp 4.5 MIN_Precip_2050_45 mm year
−1
9 17 0 252 Grid-cell Y
Minimum precipitation 2050 rcp 8.5 MIN_Precip_2050_85 mm year
−1
9 17 0 274 Grid-cell Y
Minimum precipitation 2100 rcp 4.5 MIN_Precip_2100_45 mm year
−1
9 18 0 278 Grid-cell Y
Minimum precipitation 2100 rcp 8.5 MIN_Precip_2100_85 mm year
−1
9 19 0 351 Grid-cell Y
(continued on next page)
F. Farinosi et al. Global Environmental Change 52 (2018) 286–313
296
Table A1 (continued)
Group / Sub-group Name Description Abbreviation Unit Mean SD Min Max Spatial resolution Time varying?
Temperature Average temperature 2050 rcp 4.5 AVG_Temp_2050_45 °C 11.86 13.70 −25.31 32.67 Grid-cell Y
Average temperature 2050 rcp 8.5 AVG_Temp_2050_85 °C 12.22 13.69 −25.16 33.01 Grid-cell Y
Average temperature 2100 rcp 4.5 AVG_Temp_2100_45 °C 12.90 13.49 −23.56 33.75 Grid-cell Y
Average temperature 2100 rcp 8.5 AVG_Temp_2100_85 °C 15.76 12.86 −18.57 36.03 Grid-cell Y
Maximum temperature 2050 rcp 4.5 TempMAX_2050_45 °C 20.01 8.44 −10.51 33.68 Grid-cell Y
Maximum temperature 2050 rcp 8.5 TempMAX_2050_85 °C 29.72 8.96 −5.61 48.56 Grid-cell Y
Maximum temperature 2100 rcp 4.5 TempMAX_2100_45 °C 30.72 8.97 −4.99 49.86 Grid-cell Y
Maximum temperature 2100 rcp 8.5 TempMAX_2100_85 °C 33.54 8.89 −2.39 53.37 Grid-cell Y
Minimum temperature 2050 rcp 4.5 TempMin_2050_45 °C −5.86 19.44 −53.79 27.47 Grid-cell Y
Minimum temperature 2050 rcp 8.5 TempMin_2050_85 °C −5.46 19.33 −53.53 27.76 Grid-cell Y
Minimum temperature 2100 rcp 4.5 TempMin_2100_45 °C −4.59 18.97 −52.00 28.16 Grid-cell Y
Minimum temperature 2100 rcp 8.5 TempMin_2100_85 °C −1.28 17.65 −47.29 29.69 Grid-cell Y
Temperature variation 2050 rcp 4.5 Temp_delta_2050_45 °C 11.29 2.91 0.48 21.23 Grid-cell Y
Temperature variation 2050 rcp 8.5 Temp_delta_2050_85 °C 11.23 2.91 0.46 21.38 Grid-cell Y
Temperature variation 2100 rcp 4.5 Temp_delta_2100_45 °C 11.27 2.97 0.47 21.51 Grid-cell Y
Temperature variation 2100 rcp 8.5 Temp_delta_2100_85 °C 11.06 3.04 .47 21.74 Grid-cell Y
Seasonal variability 2050 rcp 4.5 Temp_s_var_2050_45 Stand. dev. 8.66 5.26 0.33 22.77 Grid-cell Y
Seasonal variability 2050 rcp 8.5 Temp_s_var_2050_85 Stand. dev. 8.66 5.24 0.33 22.95 Grid-cell Y
Seasonal variability 2100 rcp 4.5 Temp_s_var_2100_45 Stand. dev. 8.57 5.16 0.33 22.64 Grid-cell Y
Seasonal variability 2100 rcp 8.5 Temp_s_var_2100_85 Stand. dev. 8.45 4.85 0.36 21.75 Grid-cell Y
Projected Scenarios - Population
Population Population density 2050 Pop_density_2050 People sqkm
−1
50 269 0 22,622 Grid-cell Y
Population density 2100 Pop_density_2100 People sqkm
−1
55 315 0 29,243 Grid-cell Y
Table A2
Main data used for the analysis.
Variable Spatial coverage / resolution Temporal coverage /
resolution
Database Reference URL
Hydro-political interaction Global /Basin-Country_Unit
(BCU)
1997-2007 International River Cooperation and
Conflict IRCC
Kalbhenn and
Bernauer, 2012
http://www.ib.ethz.ch/data.html
World river basins Global / Spatial polygon –International River Dataset Beck et al., 2014 http://ir-s01.ethz.ch/
Climate forcing Global / 0.25 degrees 1979 –2015 / 3 hours Multi-Source Weighted-Ensemble
Precipitation (MSWEP)
Beck et al., 2017 http://www.gloh2o.org/
Climate forcing Global / 0.25 degrees 1979 –2015 / 3 hours WATCH Forcing Data methodology applied
to ERA-Interim (WFDEI)
Weedon et al., 2014 ftp://rfdata:forceDATA@ftp.iiasa.ac.at
Gross Domestic Product (GDP) Global / Country 1950-2011 / year Expanded GDP data –Version 6.0 Gleditsch, 2002 http://privatewww.essex.ac.uk/∼ksg/exptradegdp.html
Governance indicator Global / Country 1996-2015 / year Worldwide Governance Indicators (WGI) Kaufmann et al., 2010 http://info.worldbank.org/governance/wgi/#home
Agriculture (% GDP) and
Rural population (% of
the total)
Global / Country 1960-2016 / year World Development Indicator (WDI)
database
World Bank 2017
(accessed in February
2017)
http://data.worldbank.org/data-catalog/world-development-indicators
Population Global / 1 km Multitemporal (1975,
1990, 2000, 2015)
GHS population grid, derived from the
Gridded Population of the World (GPW, v4)
Freire and Pesaresi,
2015 –CIESIN, 2015
http://data.europa.eu/89h/jrc-ghsl-ghs_pop_gpw4_globe_r2015a
http://ghsl.jrc.ec.europa.eu/ghs_pop.php
Composite Index of National
Capability (CINC)
Global / Country 1816-2007 / year National Material Capability (NMC)
version 5.0 –Correlates of War (CoW)
Singer et al., 1972 http://www.correlatesofwar.org/data-sets/national-material-capabilities
International Freshwater
Treaty Database –IFTD
Global / BCU 1820 - 2007 International Freshwater Treaty Database –
IFTD - Transboundary Freshwater Dispute
Database TFDD
De Stefano et al., 2012 http://www.transboundarywaters.orst.edu/database/interfreshtreatdata.
html
Climate future scenarios Global / 0.25 degrees 1950 –2100 / day NASA Earth Exchange Global Daily
Downscaled Projections (NEX-GDDP)
dataset
Thrasher et al., 2012 https://nex.nasa.gov/nex/projects/1356/
https://dataserver.nccs.nasa.gov/thredds/catalog/bypass/NEX-GDDP/
catalog.html
Population trends Global / Country 1950 –2100 / 5 years United Nations - Department of Economic
and Social Affairs, Population Division
UN/DESA, 2017 https://esa.un.org/unpd/wpp/
F. Farinosi et al. Global Environmental Change 52 (2018) 286–313
297
Fig. A1. 10-k cross validation: comparison between Generalized Linear Model (GLM), Boosted Decision Trees (BDT), and Support Vector Machine (SVM). Lower
values of RMSE and higher values of R-squared are preferred.
Fig. A2. Parameter calibration –top: left: ntree vs MSE; right: mtry vs Out of Bag Error OOB-MSE; bottom: NodeSize vs MSE (left), and NodeSize vs Pseudo R2 (right).
Fig. A3. Permutation-based variable importance calculated running the model 100 times: importance expressed as percent increase of MSE (left) - Standard deviation
of the permutation-based importance measure (right).
F. Farinosi et al. Global Environmental Change 52 (2018) 286–313
298
Fig. A4. Partial Dependence plots. Y-axis indicates the cross validated contributions due to the explanatory variable (change in likelihood of hydro-political in-
teractions due to changes in variable), X-axis indicates variable value. Colors indicate the mutual interactions between variables, while the coefficient of de-
termination on top of the boxes indicates the goodness of fit of the trend line (Welling et al., 2016).
F. Farinosi et al. Global Environmental Change 52 (2018) 286–313
299
Alternative water events databases
Alternative event datasets are currently available, namely: the Water and Conflict Chronology (Gleick, 1998;Gleick et al., 1994)
12
,a
collection of historical conflicts where water was the object of the dispute, a side political goal, a military tool, or object of terroristic attacks;
the Water-Related Intrastate Conflict and Cooperation (WARICC) (Bernauer et al., 2012a), providing information about water related events
in 35 countries in the Mediterranean, Middle-East, and Sahel areas; and the Issue Correlate of War –River Claims Data Set (ICOW-River),
collecting river management issues data about 82 dyads in 36riversmainlyintheWesternHemisphereandMiddleEast(Hensel et al., 2008,
2006). The mentioned alternative databases were not taken into consideration in this analysis due to respectively the bias potentially rising
from the un-homogeneous data collection methodology (in the case of Water and Conflict Chronology), and the limited geographical coverage
(in the cases of WARICC and ICOW-River). Moreover the WARICC dataset is a collection of domestic water tension/cooperation events, which
only in part overlaps the domain of the transboundary interactions object of the analysis hereby presented. However, in this study, the
WARICC dataset was still considered as a source of information about the spatial distribution of water management critical hotspots within
the selected basins.
Standardized precipitation Index
The index is a measure expressed in standard deviation units of the variation of the precipitation of a specific number of months respect to the
long run average (WMO, 2012). The number of months based on which the SPI could be calculated usually varies between 3 and 48 months. Shorter
time scales SPI is considered a good indicator of variations of soil moisture, while on longer scales (up to 24 months), it could be associated with
groundwater or reservoir levels variation (WMO, 2012). That for, a shorter SPI (3 months) is often utilized to detect meteorological droughts; a
medium SPI (6 months) is usually associated with agricultural droughts; a longer SPI (12–24 months) is associated with hydrological droughts. SPI
was calculated using the R package SPEI (Beguería and Vicente-Serrano, 2014).
Climate modeling outputs - projections used in the study
Climate projections data used in this study belong to the NASA Earth Exchange Global Daily Downscaled Projections (NASA NEX-GDDP) dataset
downscaled (0.25 degrees) and bias corrected using the Bias-Correction Spatial Disaggregation (BCSD) methodology described in Thrasher et al.
(2012). NEX-GDDP includes all the 21 GCM’s built in support of the Fifth Assessment Report of the Intergovernmental Panel on Climate Change
(IPCC AR5). Due to computational constraints, we selected 5 out of the 21 climate models, chosen in base of the structural differences among them,
as described in Knutti et al. (2013). In particular, in this study the following GCM’s were chosen: the National Center for Atmospheric Research
(NCAR) Community Climate System Model (CCSM) version 4.0 (NCAR CESM, 2011); the National Oceanic and Atmospheric Administration
(NOAA) Geophysical Fluid Dynamics Laboratory (GFDL) Coupled Model (CM) version 3.0 (Donner et al., 2011); the Institute for Numerical
Mathematics (INM) Climate Model (CM) version 4.0 (Volodin et al., 2013,2010); the Institut Pierre-Simon Laplace (IPSL) Climate Model (CM)
version 5.0 A –Medium Resolution –(Dufresne et al., 2013); and the Model for Interdisciplinary Research on Climate (MIROC) version 5.0
(Watanabe et al., 2010).
Fig. A5. 3-Dimensional partial dependency plot of the interactions between rural population and agricultural economic dependency and their combined impacton
hydro-political interactions. Colors indicate the mutual interactions between variables, while the coefficient of determination on top of the boxes indicates the
goodness of fit of the (grey) regression plane (Welling et al., 2016).
12
http://worldwater.org/water-conflict/
F. Farinosi et al. Global Environmental Change 52 (2018) 286–313
300
Fig. A6. Comparison between water events reported in the WARICC database between 1997 and 2009 (Bernauer et al., 2012a) and the likelihood of hydro-political
interactions presented in this study (shaded areas are not considered in the WARICC database).
F. Farinosi et al. Global Environmental Change 52 (2018) 286–313
301
Fig. A7. Input Data Baseline Scenario.
F. Farinosi et al. Global Environmental Change 52 (2018) 286–313
302
Fig. A8. Input data projected scenarios.
Fig. A9. Baseline and projected scenarios results.
F. Farinosi et al. Global Environmental Change 52 (2018) 286–313
303
Annex B Random Forest regression algorithm
RF is a Classification and Regression Tree (CART) based tool that involves an ensemble of regression trees (Breiman, 2001). The estimated trees,
each of them being a regression model fit using a subset of the input data and a portion of the independent variables, are then averaged in order to
reduce the bias typical of the bagging (bootstrap aggregating) techniques (Breiman, 1994). Each of the tree is a noisy but virtually unbiased
regression model: the aggregation procedure allows to reduce the variance (Hastie et al., 2009).
The RF regression algorithm procedure could be summarized in few steps:
A A random subset of the training data is drawn (each tree is trained using about 63%, ∼2/3, of the initial observations);
B For each of the bootstrapped subsets, a tree is grown by recursively repeating three actions: select a random subset of the independent variables
(m<p, being pthe total number of the independent variables); calculate the best variable/split among the mvariables; generate two sub-nodes.
This iteration is repeated until the minimum number of observation per node is reached. Each tree is tested against the remaining observations
(about 37%) and the Out-of-the-Bag error (OOB –the mean prediction error on each training subset) is calculated (specifically, in the case of RF
regression, the Mean Squared Error –MSE (Eq. 1))
C Aggregate the generated trees in an ensemble and calculation of the overall MSE (Hastie et al., 2009).
The model is tuned through the calibration of three parameters: number of trees (ntree); the number of variables per node (mtry); and the
minimum number of observations for the final node, often called leaf (nodesize)(Li et al., 2016;Malekipirbazari and Aksakalli, 2015). RF error is
determined by mainly two factors: correlation between the trees; and strength of the individual tree. The mnumber of randomly selected in-
dependent variables, by default chosen as a third of the whole number of regressors (p), is crucial for this. In fact, a larger m(mtry) value would
increase the correlation between the trees, while increasing the trees strength, and vice-versa (Breiman, 2001). Increasing the number of trees
(ntrees) would stabilize the model and reduce the overall error, until reaching a point in which the correlation between the trees would start to rise,
consequently decreasing the overall model performance. Given its characteristics, RF is almost insensitive to tuning in the size of the final node
(nodesize), and consequently to the length of the tree (Segal, 2004). RF is particularly effective in capturing complex non-linear relations; the model
can handle a certain degree of multicollinearity between the dependent variables; it is almost completely immune from overfitting; it is insensitive to
outliers, and does not require data pre-processing (Hastie et al., 2009). Moreover, RF is not sensitive to monotonic transformations of the in-
dependent variables; at the same there is no need to perform a feature selection: RF automatically ignores the variables that do not ensure a good
split. This model was successfully applied in many fields of study in which the traditional statistical analysis is afflicted by the problem of multi-
collinearity and the independent variables are characterized by high covariance, as for instance: genomics (Chen and Ishwaran, 2012), remote
sensing (Jing et al., 2016;Rasquinha and Sankaran, 2016;Vogels et al., 2017), public health (Loidl et al., 2016), hydrology (Li et al., 2016;Mohr
et al., 2017;Núñez et al., 2016), agriculture (Jeong et al., 2016), and ecological indicators (Pourtaghi et al., 2016). To our best knowledge, the
assessment presented in this paper is the first application of a RF approach to a dyadic dataset in the context of international water interactions.
Since it is performed internally while estimating the OOB error, RF does not need cross-validation. However, as in previous studies (Jeong et al.,
2016;Li et al., 2016;Malekipirbazari and Aksakalli, 2015), we performed a 10 fold cross validation to test the performance of the RF model in
comparison with alternative algorithms, namely: Generalized Linear Regression (GLM), Boosted Decision Trees (BDT), and Support Vector Machine
(SVM). The model performance was compared calculating the coefficient of determination R
2
(Eq. 2) and the Root Mean Squared Error RMSE (Eq. 3).
∑
=−
=
M
SE y y n(ˆ)/
i
n
ii
1
2
(1)
=−
∑−
∑−
=
=
R
yy
yy
1(ˆ)
()
i
n
ii
i
n
ii
21
2
1
2(2)
∑
=−
=
R
MSE y y n(ˆ)/
i
n
ii
1
2(3)
Being y
i
the observed values,
y
ˆi
the modeled, and
y
i
the observation mean.
The final calibrated RF model was trained using the entire set of observations (N=11,801). Model performance was estimated calculating MSE
(Eq. 1) and pseudo R
2
(Eq. 4), a measure of the variation explained by the model (Kvålseth, 1985;Seber and Lee, 2003).
=−
p
seudo R MSE
Var y
1()
2
(4)
Where
M
S
E
represents the Mean Standard Error (Eq. 1), and
V
ar y(
)
represents the variance of the observed values.
An additional important feature of RF is the possibility to quantify the relative importance of each of the explanatory variables by estimating the
MSE variations when a specific independent variable is permuted (Breiman, 2001;Hastie et al., 2009). Given the random nature of the model,
variable relative importance is rather volatile. Although, more stable values could be achieved if the number of trees is sufficiently high, variable
importance is likely to vary within a certain range especially in case of correlated variables (Altmann et al., 2010). That for, variable ranking could
virtually experience permutations every time a RF model is performed (Hastie et al., 2009;Strobl et al., 2007). In order to avoid this problem, our
final tuned model was run 100 times recursively and the variable relative importance estimates were presented in a boxplot form (Fig. A3). Variable
interactions and final results of the model were presented using 2-dimensional and 3-dimensional partial dependency plots (Hastie et al., 2009;
Welling et al., 2016). The analytical experiment was performed using the statistical software Rin combination with the packages: randomForest (Liaw
and Wiener, 2002), forestfloor (Welling et al., 2016), caret (Kuhn, 2008), and varSelRF (Diaz-Uriarte, 2014).
In order to validate our choice in terms of methodological approach, we tested the RF model performance in comparison with alternative
statistical approaches, one linear model and two other algorithms derived by machine learning. In the 10-fold cross validation, the RF model
outperformed the alternative methods by minimizing the error (mean RMSE = 0.218) and maximizing the coefficient of determination (mean
R
2
=0.679). The Generalized Linear Model (GLM) was the least performant, followed by the Boosted Decision Trees (BDT), and Support Vector
Machine (SVM) (Fig. A1). In order to ensure comparability, all the statistical approaches were tuned with a multiple-steps procedure and the best
performant parameters were selected for the final assessment.
F. Farinosi et al. Global Environmental Change 52 (2018) 286–313
304
RF parameter calibration was performed running the model recursively minimizing the Mean Squared Error (MSE) and maximizing the variation
explained (pseudo R
2
). Increasing the number of trees rapidly stabilized (ntree∼150) the MSE below 0.05: the performance marginal gain for values
above this ntree was found almost completely negligible. However, the final setting of this parameter was set to 500 to give more robustness to the
model. The number of variables randomly selected per split (mtry), by default set at 6 (∼p/3), was found to be optimal at a value of 8. While the
minimum number of observations for the final node (nodesize) value that minimized the error was found to be the default one (2 observations) (Fig.
A2). This confirms the theoretical literature indicating that the selected model handles comfortably fully grown trees without overfitting.
Annex C Comparison with the spatial distribution of the water interaction recorded in the WARICC database
Our analysis highlights the areas of the transboundary basins where water interactions are more likely to rise. As an empirical validation of our
results, we compared the spatial distribution of our index with geospatial data about water interactions. The only dataset of historical water related
events available at fine (sub-dyadic) resolution is the Water-Related Intrastate Conflict and Cooperation (WARICC) dataset (Bernauer et al., 2012a),
classifying water events recorded in 35 countries in the Mediterranean, the Middle East, and the Sahel in the period 1997–2009. A comparison
between the spatial distribution of the water events classified in the WARICC dataset and the here presented likelihood of hydro-political interactions
is presented in Fig. A6. It should be noted that our index was structured studying the transboundary water interactions, while the WARICC database
is a repository of domestic conflict and cooperation events over water. The spatial distribution of the events corresponds to the highest values of our
index for the majority of the areas under consideration, especially within the boundaries of the main transboundary river basins. The correspondence
of the presented index and the events outside the boundaries of the main river basins, especially in the Adriatic portion of the Balkans, is less evident.
However, there is the possibility that the high concentration of the WARICC reported events in the Balkan area could be linked with the civil conflicts
that followed the collapse of the Socialist Federal Republic of Yugoslavia (1992, with the civil war spanning a decade between 1991 and 2001).
Similar bias could affect the WARICC event data at the border between Eritrea and Ethiopia involved in a territorial dispute between 1998 and 2000.
Annex D Results
Table A3
Normalized likelihood of hydro-political interaction, future scenarios, and main determining factors per river basin.
Basin_ID National
Power
(CINC)
Governance Economic
Wealth
Economic
dependence
on
agriculture
International
treaties in
the basin
Population
Density
Population in
Rural Areas
Water
availability
Precipitation
stress
Yearly
precipitation
131 0.914 0.369 0.370 0.737 0.625 0.942 0.794 0.234 0.514 0.672
135 1.000 0.330 0.495 0.602 0.623 0.865 0.649 0.195 0.629 0.702
57 0.984 0.858 0.942 0.061 0.999 0.506 0.160 0.440 0.527 0.504
130 0.516 0.129 0.262 0.882 0.420 0.675 0.806 0.394 0.518 0.442
248 0.776 0.482 0.588 0.384 0.693 0.674 0.125 0.626 0.533 0.638
141 0.847 0.338 0.294 0.736 0.600 0.963 0.821 0.297 0.490 0.886
114 0.837 0.239 0.341 0.765 0.472 0.832 0.756 0.242 0.621 0.469
82 0.985 0.854 0.941 0.062 1.000 0.416 0.160 0.425 0.529 0.252
84 0.999 0.327 0.494 0.610 0.624 0.400 0.647 0.208 0.527 0.034
104 0.647 0.191 0.593 0.491 0.412 0.619 0.307 0.330 0.634 0.283
110 0.513 0.101 0.558 0.715 0.363 0.495 0.585 0.734 0.422 0.181
132 0.735 0.050 0.393 0.938 0.280 0.661 0.802 0.529 0.475 0.745
44 0.630 0.623 0.697 0.318 0.667 0.779 0.385 0.170 0.286 0.429
79 0.847 0.272 0.669 0.400 0.688 0.704 0.254 0.629 0.407 0.541
245 0.255 0.337 0.347 0.664 0.535 0.583 0.861 0.420 0.246 0.483
51 0.967 0.871 0.936 0.069 0.976 0.391 0.160 0.579 0.367 0.509
123 0.345 0.332 0.596 0.385 0.262 0.932 0.074 0.011 0.523 0.518
242 0.545 0.578 0.539 0.352 0.666 0.438 0.518 0.571 0.212 0.253
233 0.305 0.506 0.552 0.416 0.460 0.180 0.600 0.665 0.036 0.380
143 0.929 0.329 0.432 0.678 0.535 0.816 0.740 0.188 0.654 0.847
107 0.647 0.177 0.607 0.575 0.476 0.562 0.403 0.665 0.390 0.348
120 0.519 0.000 0.273 0.819 0.236 0.497 0.828 0.404 0.538 0.118
244 0.560 0.498 0.473 0.565 0.682 0.599 0.568 0.393 0.267 0.498
59 0.568 0.584 0.719 0.471 0.588 0.720 0.421 0.621 0.434 0.504
246 0.546 0.469 0.502 0.435 0.675 0.622 0.610 0.279 0.408 0.493
122 0.836 0.301 0.435 0.793 0.464 0.671 0.781 0.741 0.496 0.710
118 0.547 0.413 0.402 0.640 0.161 0.751 0.475 0.043 0.553 0.250
35 0.939 0.306 0.605 0.521 0.658 0.551 0.420 0.546 0.660 0.427
237 0.339 0.409 0.426 0.520 0.472 0.345 0.717 0.736 0.000 0.368
112 0.408 0.020 0.418 0.802 0.241 0.460 0.772 0.646 0.598 0.309
171 0.550 0.206 0.029 0.968 0.411 0.700 0.973 0.248 0.639 0.427
64 0.486 0.209 0.532 0.683 0.587 0.519 0.608 0.551 0.442 0.265
106 0.750 0.453 0.734 0.720 0.119 0.999 0.239 0.173 0.414 0.683
121 0.527 0.384 0.547 0.687 0.379 0.800 0.270 0.062 0.606 0.043
139 0.497 0.268 0.187 0.865 0.544 0.598 0.757 0.295 0.482 0.425
164 0.323 0.365 0.300 0.590 0.413 0.900 0.488 0.307 0.429 0.727
241 0.525 0.485 0.522 0.537 0.631 0.598 0.557 0.481 0.318 0.380
239 0.367 0.148 0.454 0.717 0.420 0.572 0.759 0.569 0.162 0.441
158 0.693 0.450 0.664 0.279 0.876 0.906 0.207 0.297 0.429 0.798
133 0.691 0.209 0.615 0.476 0.521 0.476 0.323 0.268 0.546 0.087
228 0.504 0.027 0.184 0.758 0.324 0.630 0.692 0.798 0.368 0.648
188 0.487 0.055 0.087 0.830 0.329 0.486 0.690 0.787 0.464 0.688
(continued on next page)
F. Farinosi et al. Global Environmental Change 52 (2018) 286–313
305
Table A3 (continued)
Basin_ID National
Power
(CINC)
Governance Economic
Wealth
Economic
dependence
on
agriculture
International
treaties in
the basin
Population
Density
Population in
Rural Areas
Water
availability
Precipitation
stress
Yearly
precipitation
231 0.374 0.281 0.312 0.688 0.428 0.520 0.752 0.617 0.123 0.550
95 0.934 0.193 0.409 0.851 0.509 0.717 0.532 0.205 0.769 0.522
119 0.345 0.332 0.596 0.385 0.262 0.907 0.074 0.154 0.292 0.474
203 0.594 0.525 0.600 0.529 0.317 0.794 0.520 0.484 0.244 0.882
240 0.367 0.362 0.123 0.809 0.527 0.485 0.795 0.558 0.231 0.576
108 0.632 0.293 0.525 0.711 0.495 0.807 0.420 0.175 0.365 0.363
155 0.484 0.149 0.137 0.839 0.290 0.643 0.899 0.257 0.343 0.330
234 0.444 0.169 0.394 0.497 0.366 0.465 0.722 0.699 0.052 0.487
111 0.923 0.669 0.855 0.181 0.951 0.524 0.182 0.379 0.601 0.295
124 0.894 0.613 0.820 0.210 0.934 0.545 0.188 0.362 0.426 0.228
185 0.509 0.227 0.129 0.915 0.393 0.667 0.941 0.245 0.387 0.563
138 0.798 0.161 0.370 0.892 0.475 0.491 0.810 0.478 0.449 0.839
56 0.394 0.457 0.413 0.762 0.320 0.147 0.382 0.594 0.776 0.085
58 0.881 0.910 0.915 0.094 0.882 0.551 0.158 0.744 0.577 0.568
86 0.843 0.274 0.665 0.416 0.680 0.556 0.263 0.619 0.258 0.570
116 0.668 0.161 0.573 0.594 0.480 0.364 0.436 0.321 0.590 0.208
125 0.929 0.203 0.463 0.804 0.499 0.552 0.720 0.451 0.396 0.649
19 0.853 0.268 0.676 0.373 0.700 0.605 0.241 0.629 0.500 0.450
12 0.832 0.274 0.664 0.393 0.692 0.373 0.287 0.616 0.468 0.401
2 0.841 0.287 0.661 0.445 0.680 0.206 0.255 0.626 0.426 0.415
117 0.572 0.256 0.454 0.546 0.207 0.823 0.389 0.000 0.562 0.252
173 0.586 0.295 0.314 0.812 0.359 0.909 0.874 0.479 0.402 0.733
115 0.657 0.323 0.563 0.688 0.498 0.695 0.406 0.154 0.237 0.537
223 0.338 0.294 0.288 0.898 0.161 0.359 0.997 0.792 0.181 0.908
140 0.802 0.300 0.265 0.736 0.552 0.966 0.828 0.344 0.527 0.891
136 0.938 0.328 0.438 0.674 0.545 0.791 0.731 0.417 0.534 0.684
92 0.578 0.507 0.654 0.529 0.487 0.654 0.286 0.680 0.357 0.450
215 0.587 0.333 0.553 0.513 0.090 0.658 0.242 0.721 0.219 0.871
152 0.457 0.134 0.160 0.790 0.243 0.561 0.877 0.276 0.333 0.173
229 0.319 0.294 0.286 0.900 0.161 0.187 1.000 0.792 0.248 0.938
243 0.579 0.418 0.654 0.466 0.627 0.000 0.029 0.720 0.602 0.000
148 0.261 0.352 0.197 0.839 0.554 0.443 0.655 0.403 0.281 0.351
113 0.465 0.387 0.521 0.536 0.217 0.736 0.359 0.006 0.202 0.375
49 0.842 0.273 0.662 0.409 0.688 0.591 0.250 0.609 0.423 0.395
161 0.329 0.396 0.194 0.889 0.405 0.623 0.741 0.273 0.439 0.546
159 0.345 0.308 0.431 0.608 0.490 0.924 0.580 0.317 0.375 0.812
144 0.559 0.266 0.316 0.843 0.400 0.687 0.836 0.831 0.651 0.715
80 0.963 0.241 0.444 0.786 0.556 0.683 0.571 0.311 0.800 0.449
145 0.589 0.287 0.323 0.813 0.397 0.742 0.835 0.805 0.711 0.725
154 0.131 0.457 0.512 0.647 0.116 0.393 0.587 0.715 0.627 0.756
129 0.763 0.480 0.703 0.268 0.888 0.275 0.203 0.304 0.601 0.367
24 0.174 0.743 0.710 0.302 0.333 0.436 0.300 0.516 0.432 0.471
182 0.486 0.119 0.132 0.948 0.340 0.528 0.891 0.265 0.577 0.361
162 0.299 0.380 0.220 0.596 0.327 0.788 0.464 0.344 0.414 0.735
55 0.692 0.409 0.515 0.702 0.482 0.126 0.347 0.603 0.640 0.204
235 0.694 0.433 0.591 0.499 0.642 0.551 0.147 0.750 0.522 0.626
163 0.288 0.364 0.211 0.718 0.546 0.540 0.654 0.430 0.390 0.510
249 0.512 0.801 0.661 0.341 0.262 0.573 0.065 0.580 0.533 0.731
48 0.637 0.678 0.725 0.253 0.689 0.766 0.379 0.109 0.300 0.424
66 0.777 0.294 0.585 0.522 0.630 0.840 0.289 0.536 0.420 0.404
232 0.398 0.372 0.068 0.838 0.521 0.448 0.825 0.510 0.354 0.595
146 0.354 0.379 0.560 0.448 0.363 0.889 0.323 0.119 0.168 0.740
77 0.337 0.544 0.712 0.385 0.262 0.601 0.459 0.621 0.397 0.531
214 0.157 0.329 0.613 0.383 0.165 0.201 0.163 0.823 0.508 0.744
254 0.512 0.801 0.661 0.341 0.262 0.298 0.065 0.586 0.366 0.702
166 0.282 0.359 0.271 0.616 0.278 0.652 0.497 0.431 0.155 0.716
178 0.574 0.247 0.584 0.417 0.213 0.366 0.119 0.743 0.276 0.824
109 0.691 0.209 0.615 0.476 0.521 0.599 0.323 0.279 0.821 0.519
127 0.555 0.365 0.420 0.614 0.176 0.442 0.448 0.035 0.639 0.112
99 0.844 0.265 0.667 0.394 0.684 0.531 0.260 0.601 0.261 0.539
76 0.965 0.305 0.588 0.526 0.659 0.602 0.483 0.538 0.757 0.439
169 0.217 0.244 0.150 0.897 0.406 0.603 0.667 0.504 0.392 0.618
184 0.579 0.281 0.572 0.459 0.171 0.650 0.167 0.737 0.123 0.816
206 0.726 0.436 0.557 0.458 0.637 0.280 0.183 0.653 0.407 0.818
205 0.655 0.557 0.668 0.461 0.000 0.114 0.339 0.917 0.339 0.834
238 0.425 0.481 0.505 0.578 0.413 0.131 0.279 0.682 0.418 0.225
194 0.395 0.340 0.412 0.750 0.091 0.103 0.623 0.952 0.344 0.799
30 0.794 0.929 0.905 0.105 0.823 0.304 0.157 0.785 0.420 0.390
105 0.660 0.775 0.832 0.257 0.734 0.517 0.236 0.212 0.332 0.347
137 0.455 0.146 0.163 0.942 0.463 0.588 0.796 0.622 0.618 0.451
227 0.477 0.337 0.512 0.540 0.552 0.618 0.303 0.591 0.154 0.517
251 0.579 0.418 0.654 0.466 0.627 0.000 0.029 0.580 0.511 0.927
10 0.936 0.889 0.927 0.081 0.938 0.021 0.159 0.674 0.638 0.394
128 0.554 0.370 0.418 0.617 0.175 0.362 0.451 0.035 0.659 0.102
40 0.813 0.401 0.687 0.341 0.693 0.613 0.292 0.482 0.383 0.466
247 0.668 0.599 0.612 0.467 0.646 0.385 0.045 0.879 0.471 0.668
(continued on next page)
F. Farinosi et al. Global Environmental Change 52 (2018) 286–313
306
Table A3 (continued)
Basin_ID National
Power
(CINC)
Governance Economic
Wealth
Economic
dependence
on
agriculture
International
treaties in
the basin
Population
Density
Population in
Rural Areas
Water
availability
Precipitation
stress
Yearly
precipitation
172 0.633 0.328 0.336 0.742 0.389 0.964 0.835 0.390 0.404 0.823
68 0.821 0.281 0.636 0.457 0.668 0.391 0.265 0.487 0.388 0.402
175 0.613 0.198 0.474 0.849 0.307 0.676 0.768 0.539 0.417 0.917
134 0.696 0.202 0.414 0.745 0.360 0.451 0.659 0.174 0.586 0.049
179 0.424 0.378 0.192 0.830 0.479 0.814 0.673 0.385 0.427 0.619
33 0.679 0.940 0.898 0.111 0.780 0.307 0.157 0.804 0.336 0.580
94 0.435 0.521 0.629 0.532 0.473 0.280 0.279 0.746 0.503 0.431
258 0.553 0.591 0.657 0.416 0.520 0.290 0.045 0.635 0.459 0.258
15 0.825 0.428 0.738 0.349 0.699 0.491 0.215 0.612 0.515 0.471
43 0.787 0.467 0.693 0.324 0.690 0.582 0.318 0.102 0.304 0.507
31 0.261 0.642 0.684 0.349 0.413 0.612 0.319 0.467 0.467 0.463
253 0.546 0.632 0.658 0.403 0.486 0.066 0.049 0.612 0.281 0.801
126 0.570 0.271 0.449 0.556 0.203 0.269 0.397 0.010 0.573 0.006
189 0.185 0.510 0.633 0.391 0.217 0.541 0.371 0.586 0.255 0.981
93 0.586 0.299 0.548 0.621 0.328 0.678 0.418 0.354 0.424 0.465
252 0.561 0.539 0.657 0.432 0.557 0.090 0.041 0.650 0.384 0.814
36 0.736 0.278 0.592 0.543 0.596 0.646 0.285 0.445 0.417 0.438
261 0.550 0.609 0.658 0.410 0.506 0.308 0.047 0.634 0.053 0.249
256 0.549 0.616 0.658 0.408 0.500 0.023 0.048 0.638 0.499 0.569
7 0.894 0.906 0.917 0.091 0.894 0.000 0.158 0.731 0.595 0.255
62 0.969 0.869 0.937 0.068 0.979 0.266 0.160 0.566 0.280 0.833
69 0.736 0.686 0.870 0.190 0.724 0.875 0.311 0.129 0.610 0.585
41 0.773 0.283 0.642 0.422 0.678 0.433 0.363 0.594 0.500 0.286
236 0.444 0.374 0.456 0.580 0.514 0.442 0.304 0.691 0.300 0.404
63 0.975 0.359 0.480 0.644 0.582 0.258 0.586 0.402 0.408 0.172
250 0.556 0.571 0.657 0.422 0.535 0.371 0.044 0.654 0.527 0.865
90 0.472 0.536 0.700 0.500 0.490 0.700 0.298 0.629 0.293 0.449
218 0.097 0.236 0.662 0.251 0.090 0.191 0.403 0.763 0.641 0.847
61 0.668 0.324 0.488 0.625 0.536 0.738 0.391 0.319 0.407 0.429
262 0.556 0.571 0.657 0.422 0.535 0.044 0.044 0.624 0.119 0.253
222 0.167 0.304 0.590 0.373 0.146 0.153 0.191 0.855 0.347 0.730
102 0.657 0.776 0.830 0.256 0.735 0.723 0.243 0.219 0.396 0.439
81 0.292 0.373 0.557 0.555 0.033 0.601 0.648 0.515 0.286 0.664
201 0.595 0.174 0.199 0.848 0.498 0.777 0.655 0.322 0.359 0.817
42 0.650 0.906 0.907 0.094 0.675 0.552 0.309 0.256 0.673 0.595
255 0.532 0.705 0.659 0.377 0.413 0.060 0.056 0.612 0.395 0.523
221 0.706 0.286 0.406 0.648 0.161 0.261 0.600 0.423 0.336 0.882
151 0.639 0.397 0.603 0.446 0.790 0.645 0.345 0.309 0.460 0.765
165 0.271 0.318 0.364 0.642 0.262 0.850 0.564 0.431 0.315 0.655
97 0.496 0.585 0.730 0.461 0.543 0.504 0.257 0.659 0.231 0.488
8 0.841 0.344 0.708 0.362 0.700 0.440 0.228 0.622 0.386 0.458
220 0.435 0.302 0.177 0.846 0.431 0.526 0.889 0.293 0.448 0.506
170 0.231 0.155 0.150 0.842 0.514 0.499 0.735 0.460 0.433 0.724
216 0.517 0.300 0.524 0.566 0.403 0.606 0.331 0.625 0.285 0.855
20 0.925 0.894 0.924 0.084 0.925 0.053 0.159 0.748 0.388 0.673
225 0.454 0.308 0.507 0.569 0.531 0.593 0.347 0.583 0.229 0.613
89 0.676 0.772 0.841 0.260 0.726 0.599 0.190 0.176 0.471 0.477
259 0.512 0.801 0.661 0.341 0.262 0.478 0.065 0.580 0.057 0.254
263 0.553 0.588 0.657 0.417 0.523 0.197 0.045 0.623 0.294 0.302
230 0.656 0.289 0.372 0.760 0.161 0.347 0.738 0.651 0.270 0.800
71 0.623 0.334 0.453 0.663 0.481 0.512 0.457 0.278 0.444 0.344
212 0.000 0.428 0.436 0.665 0.058 0.008 0.506 1.000 0.278 0.823
183 0.185 0.510 0.633 0.391 0.217 0.573 0.371 0.586 0.135 0.958
70 0.635 0.331 0.461 0.655 0.495 0.639 0.441 0.331 0.354 0.364
150 0.687 0.443 0.657 0.307 0.867 0.318 0.224 0.308 0.664 0.735
4 0.438 0.962 1.000 0.126 0.596 0.343 0.192 0.700 0.530 0.479
74 0.914 0.898 0.922 0.087 0.914 0.270 0.158 0.709 0.294 0.644
257 0.532 0.705 0.659 0.377 0.413 0.012 0.056 0.624 0.112 0.340
204 0.453 0.187 0.229 0.853 0.373 0.352 0.853 0.255 0.368 0.314
149 0.564 0.403 0.557 0.550 0.662 0.269 0.458 0.463 0.535 0.693
22 0.794 0.439 0.687 0.352 0.615 0.433 0.264 0.597 0.503 0.475
181 0.190 0.614 0.605 0.494 0.217 0.382 0.355 0.549 0.107 0.963
72 0.527 0.739 0.809 0.211 0.420 0.715 0.499 0.357 0.728 0.650
264 0.579 0.418 0.654 0.466 0.627 0.000 0.029 0.666 0.575 0.340
260 0.512 0.801 0.661 0.341 0.262 0.090 0.065 0.580 0.022 0.304
91 0.346 0.346 0.559 0.690 0.207 0.708 0.518 0.468 0.380 0.650
67 0.875 0.912 0.914 0.095 0.877 0.327 0.158 0.751 0.385 0.631
50 0.744 0.838 0.853 0.100 0.800 0.815 0.240 0.195 0.456 0.464
27 0.228 0.639 0.679 0.323 0.385 0.478 0.311 0.527 0.465 0.480
98 0.712 0.433 0.667 0.552 0.490 0.371 0.318 0.208 0.432 0.434
174 0.363 0.221 0.216 0.810 0.467 0.546 0.643 0.284 0.473 0.573
147 0.300 0.186 0.339 0.920 0.363 0.822 0.550 0.130 0.111 0.673
177 0.243 0.170 0.141 0.905 0.454 0.561 0.740 0.460 0.541 0.774
219 0.126 0.328 0.640 0.346 0.147 0.166 0.185 0.805 0.496 0.807
88 0.346 0.414 0.610 0.567 0.210 0.685 0.417 0.333 0.327 0.490
16 0.442 0.962 0.998 0.126 0.594 0.424 0.190 0.699 0.411 0.593
(continued on next page)
F. Farinosi et al. Global Environmental Change 52 (2018) 286–313
307
Table A3 (continued)
Basin_ID National
Power
(CINC)
Governance Economic
Wealth
Economic
dependence
on
agriculture
International
treaties in
the basin
Population
Density
Population in
Rural Areas
Water
availability
Precipitation
stress
Yearly
precipitation
28 0.262 0.643 0.684 0.350 0.414 0.547 0.319 0.477 0.356 0.492
25 0.800 0.928 0.905 0.104 0.826 0.000 0.157 0.804 0.380 0.837
157 0.339 0.309 0.425 0.612 0.474 0.782 0.578 0.341 0.523 0.652
199 0.373 0.419 0.263 0.876 0.307 0.681 0.581 0.344 0.541 0.663
45 0.760 0.902 0.878 0.000 0.883 0.619 0.159 0.181 0.739 0.540
18 0.760 0.933 0.902 0.107 0.807 0.003 0.157 0.790 0.579 0.648
85 0.676 0.771 0.840 0.262 0.727 0.503 0.192 0.166 0.426 0.673
39 0.677 0.905 0.902 0.080 0.724 0.592 0.284 0.349 0.594 0.628
168 0.234 0.424 0.439 0.670 0.179 0.694 0.432 0.549 0.205 0.778
209 0.541 0.576 0.634 0.532 0.090 0.094 0.338 0.463 0.053 0.928
21 0.707 0.938 0.899 0.110 0.787 0.000 0.157 0.804 0.371 0.669
187 0.237 0.168 0.126 0.929 0.411 0.679 0.724 0.486 0.487 0.814
32 0.243 0.634 0.679 0.333 0.397 0.438 0.314 0.479 0.406 0.490
52 0.758 0.904 0.893 0.046 0.847 0.903 0.231 0.256 0.506 0.565
193 0.538 0.237 0.563 0.517 0.226 0.080 0.204 0.836 0.465 0.810
3 0.451 0.978 0.969 0.163 0.630 0.056 0.170 0.658 0.279 0.410
65 0.745 0.833 0.871 0.150 0.942 0.735 0.199 0.328 0.556 0.586
26 0.232 0.629 0.676 0.325 0.389 0.358 0.312 0.529 0.432 0.470
23 0.699 0.939 0.899 0.110 0.785 0.003 0.157 0.803 0.262 0.679
176 0.233 0.182 0.125 0.960 0.350 0.590 0.727 0.481 0.510 0.770
87 0.655 0.777 0.829 0.256 0.736 0.601 0.249 0.215 0.408 0.466
29 0.719 0.363 0.663 0.454 0.558 0.500 0.269 0.536 0.495 0.479
60 0.754 0.821 0.865 0.158 0.951 0.866 0.196 0.333 0.316 0.490
96 0.657 0.776 0.830 0.256 0.735 0.601 0.244 0.282 0.569 0.661
75 0.754 0.821 0.865 0.158 0.951 0.639 0.196 0.331 0.484 0.509
186 0.213 0.248 0.129 0.890 0.305 0.691 0.725 0.317 0.402 0.633
78 0.754 0.821 0.865 0.158 0.951 0.349 0.196 0.116 0.456 0.610
100 0.701 0.425 0.656 0.565 0.471 0.539 0.334 0.241 0.524 0.646
34 0.468 0.444 0.668 0.482 0.476 0.617 0.295 0.405 0.437 0.458
200 0.185 0.510 0.633 0.391 0.217 0.097 0.371 0.729 0.085 1.000
83 0.673 0.772 0.838 0.261 0.729 0.602 0.201 0.175 0.519 0.650
9 0.540 0.989 0.884 0.168 0.558 0.141 0.108 0.535 0.423 0.455
103 0.408 0.437 0.643 0.705 0.442 0.605 0.489 0.412 0.320 0.688
17 0.549 0.973 0.913 0.127 0.474 0.493 0.117 0.568 0.358 0.519
160 0.253 0.329 0.327 0.701 0.185 0.542 0.486 0.524 0.592 0.749
5 0.466 1.000 0.917 0.207 0.671 0.045 0.140 0.586 0.314 0.444
6 0.600 0.943 0.880 0.240 0.687 0.081 0.138 0.571 0.316 0.428
180 0.365 0.171 0.225 0.780 0.487 0.642 0.606 0.323 0.618 0.662
190 0.224 0.123 0.070 0.917 0.401 0.582 0.664 0.519 0.507 0.786
226 0.441 0.328 0.154 0.860 0.474 0.507 0.875 0.321 1.000 0.494
198 0.234 0.113 0.017 0.970 0.183 0.534 0.593 0.538 0.698 0.775
11 0.565 0.974 0.870 0.242 0.694 0.134 0.128 0.549 0.438 0.443
208 0.626 0.473 0.577 0.579 0.118 0.256 0.431 0.441 0.182 0.919
191 0.202 0.111 0.000 0.968 0.236 0.524 0.617 0.560 0.469 0.841
37 0.760 0.902 0.878 0.000 0.883 0.786 0.159 0.181 0.653 0.568
14 0.567 0.973 0.870 0.242 0.694 0.245 0.128 0.546 0.570 0.445
202 0.373 0.339 0.253 0.848 0.374 0.661 0.582 0.325 0.531 0.661
197 0.206 0.113 0.016 0.959 0.279 0.600 0.626 0.559 0.569 0.764
38 0.665 0.968 0.888 0.105 0.633 0.747 0.134 0.219 0.448 0.531
156 0.345 0.308 0.431 0.608 0.490 0.580 0.580 0.327 0.671 0.853
53 0.660 0.844 0.868 0.083 0.804 1.000 0.015 0.199 0.328 0.516
167 0.248 0.332 0.316 0.715 0.161 0.568 0.466 0.554 0.198 0.729
195 0.201 0.138 0.023 1.000 0.090 0.524 0.637 0.546 0.438 0.864
47 0.760 0.902 0.878 0.000 0.883 0.742 0.159 0.181 0.706 0.564
207 0.582 0.179 0.202 0.840 0.495 0.265 0.644 0.357 0.515 0.906
54 0.648 0.846 0.868 0.076 0.787 0.824 0.000 0.297 0.286 0.497
196 0.324 0.146 0.157 0.880 0.390 0.542 0.597 0.473 0.700 0.746
Basin_ID Climatic zone
(Temperature)
Likelihood
hydro-polit
issues_norm
likelihood_bline %
change_2050_RC-
P4.5_vs_bline
%
change_2050_RC-
P8.5_vs_bline
%
change_2100_RC-
P4.5_vs_bline
%
change_2100_RC-
P8.5_vs_bline
Name of the
Basin
no.
131 0.726 1.000 0.535 7.177 7.143 6.657 7.740 Ganges/
Brahmaputra
1
135 0.764 0.999 0.535 9.735 10.721 7.268 9.612 Pearl 2
57 0.552 0.799 0.436 37.895 38.335 39.870 42.001 Mississippi 3
130 0.933 0.761 0.418 43.399 43.142 45.220 42.469 Nile 4
248 0.715 0.756 0.415 −8.030 −8.225 −37.233 −30.656 Chuy 5
141 0.911 0.679 0.377 30.565 30.362 31.508 31.298 Feni 6
114 0.623 0.675 0.375 12.325 12.571 15.398 19.152 Indus 7
82 0.570 0.670 0.373 19.048 20.118 21.261 25.488 Colorado 8
84 0.478 0.658 0.367 12.355 13.503 14.840 18.797 Tarim 9
104 0.770 0.592 0.335 23.191 23.550 26.545 32.532 Shatt
al-Arab - tigris/
Euphrates
10
(continued on next page)
F. Farinosi et al. Global Environmental Change 52 (2018) 286–313
308
Table A3 (continued)
Basin_ID Climatic zone
(Temperature)
Likelihood
hydro-polit
issues_norm
likelihood_bline %
change_2050_RC-
P4.5_vs_bline
%
change_2050_RC-
P8.5_vs_bline
%
change_2100_RC-
P4.5_vs_bline
%
change_2100_RC-
P8.5_vs_bline
Name of the
Basin
no.
110 0.682 0.577 0.327 16.011 16.935 15.378 17.544 Hari 11
132 0.827 0.573 0.326 26.150 26.687 26.387 28.578 Irrawaddy 12
44 0.492 0.566 0.322 13.420 14.818 3.903 21.293 Vistula 13
79 0.465 0.565 0.321 14.538 15.129 16.204 24.305 Terek 14
245 0.832 0.564 0.321 49.606 48.309 61.302 55.689 Umbeluzi 15
51 0.447 0.543 0.311 22.426 23.327 26.039 33.005 Columbia 16
123 0.702 0.538 0.308 24.587 25.240 21.536 33.438 Wadi Al Izziyah 17
242 0.750 0.522 0.300 33.894 33.924 35.468 39.138 Orange 18
233 0.834 0.521 0.300 22.035 22.276 23.819 24.930 Okavango 19
143 0.835 0.515 0.297 39.760 39.426 33.798 34.072 Ka Long 20
107 0.655 0.506 0.293 28.520 29.178 29.114 30.587 Atrek 21
120 0.713 0.505 0.292 30.626 31.593 32.292 40.272 Sistan Basin 22
244 0.796 0.503 0.291 34.967 34.728 39.035 40.384 Komati 23
59 0.524 0.499 0.289 24.236 26.075 19.280 34.751 Danube 24
246 0.766 0.492 0.286 44.019 43.416 46.850 47.083 Maputo 25
122 0.818 0.492 0.285 25.191 25.722 23.442 24.132 Mekong 26
118 0.636 0.486 0.283 40.140 39.405 38.553 47.336 Bon Naima 27
35 0.273 0.486 0.283 22.040 22.141 22.208 30.770 Amur 28
237 0.830 0.474 0.277 32.909 33.264 40.085 39.081 Etosha pan 29
112 0.647 0.463 0.271 27.397 28.805 26.423 40.916 Morghab 30
171 0.866 0.456 0.268 48.466 48.526 53.271 52.514 Lake Abbe 31
64 0.516 0.455 0.268 33.255 34.251 35.684 41.908 Aral Sea 32
106 0.533 0.453 0.266 39.803 40.756 43.653 51.858 Han 33
121 0.744 0.452 0.266 34.974 35.191 36.876 38.270 Jordan/Dead
Sea
34
139 0.975 0.447 0.263 64.156 62.392 76.026 66.048 Niger 35
164 0.827 0.439 0.260 42.833 39.332 38.346 38.480 Paz 36
241 0.816 0.438 0.259 37.244 36.743 39.585 41.529 Limpopo 37
239 0.807 0.438 0.259 32.701 33.190 37.219 33.807 Save 38
158 0.854 0.435 0.258 36.891 36.078 35.299 40.257 Coatan 39
133 0.895 0.433 0.256 21.881 20.925 19.092 19.746 Bahu Kalat 40
228 0.896 0.432 0.256 76.422 72.838 89.122 86.112 Chiloango 41
188 0.882 0.432 0.256 70.974 71.093 78.190 83.396 Congo 42
231 0.839 0.431 0.256 38.997 38.543 48.453 47.166 Zambezi 43
95 0.391 0.431 0.256 49.424 49.336 54.487 63.095 Yalu 44
119 0.741 0.431 0.255 41.130 42.978 42.289 50.967 Nahr al-Kabir
al-Janoubi
45
203 0.949 0.430 0.255 56.644 56.191 58.877 63.133 Golok 46
240 0.842 0.429 0.255 44.165 45.605 57.977 57.144 Buzi 47
108 0.672 0.427 0.254 43.776 45.252 50.121 62.190 Orontes 48
155 0.918 0.425 0.253 44.890 45.321 55.501 54.110 Mareb 49
234 0.768 0.421 0.251 29.415 28.827 38.116 32.273 Cunene 50
111 0.716 0.419 0.250 35.540 36.125 36.227 39.494 Grande 51
124 0.628 0.417 0.249 44.239 43.658 43.027 42.764 Tijuana 52
185 0.867 0.416 0.248 50.010 50.492 58.676 59.113 Lake Turkana 53
138 0.808 0.415 0.248 27.036 27.538 26.590 32.981 Kaladan 54
56 0.212 0.413 0.247 12.058 12.288 13.367 23.324 Khyargas Nuur 55
58 0.403 0.412 0.246 30.829 32.246 37.467 55.786 Saint Lawrence 56
86 0.365 0.410 0.245 6.540 8.538 11.796 30.443 Sulak 57
116 0.723 0.410 0.245 36.506 41.215 37.430 49.433 Kowl-e
Namakzar
58
125 0.600 0.405 0.243 11.027 11.374 9.907 12.980 Salween 59
19 0.399 0.402 0.242 14.911 17.514 16.818 38.846 Volga 60
12 0.301 0.400 0.241 20.203 20.882 22.444 34.406 Ob 61
2 0.167 0.399 0.240 15.028 15.150 16.938 21.381 Yenisei 62
117 0.664 0.398 0.239 60.379 60.820 61.724 69.363 Tafna 63
173 0.973 0.397 0.239 52.008 53.182 54.810 54.593 Vam Co 64
115 0.686 0.393 0.237 60.662 60.731 60.745 73.029 Nahr El Kebir 65
223 0.913 0.391 0.236 66.842 69.560 72.281 73.869 Sepik 66
140 0.904 0.381 0.231 64.706 65.589 66.616 64.365 Karnaphuli 67
136 0.796 0.380 0.231 44.530 45.087 45.591 53.619 Red 68
92 0.583 0.380 0.230 50.216 53.315 44.838 59.882 Maritsa 69
215 0.765 0.377 0.229 40.394 41.311 39.038 47.912 Patia 70
152 0.951 0.377 0.229 43.335 44.161 54.466 57.474 Baraka 71
229 0.910 0.374 0.228 69.094 72.969 78.131 77.008 Fly 72
243 0.393 0.372 0.227 38.994 37.835 58.107 73.794 Laguna de Tara 73
148 1.000 0.372 0.227 36.799 36.410 45.407 41.005 Senegal 74
113 0.699 0.371 0.226 63.119 63.409 62.729 66.377 Medjerda 75
49 0.490 0.370 0.226 46.478 51.039 50.766 67.073 Don 76
161 0.977 0.369 0.225 81.633 79.409 96.946 93.278 Volta 77
159 0.779 0.367 0.224 52.838 52.184 51.782 49.953 Suchiate 78
144 0.851 0.366 0.224 44.895 46.094 45.062 49.497 Ma 79
80 0.356 0.354 0.218 51.331 52.762 52.802 69.604 Tumen 80
145 0.862 0.353 0.217 59.430 60.639 60.561 61.692 Ca 81
154 0.899 0.350 0.216 53.080 54.425 55.682 65.549 Belize 82
129 0.756 0.350 0.216 51.042 50.632 50.584 49.151 Yaqui 83
24 0.445 0.350 0.216 26.681 32.136 27.052 45.597 Parnu 84
(continued on next page)
F. Farinosi et al. Global Environmental Change 52 (2018) 286–313
309
Table A3 (continued)
Basin_ID Climatic zone
(Temperature)
Likelihood
hydro-polit
issues_norm
likelihood_bline %
change_2050_RC-
P4.5_vs_bline
%
change_2050_RC-
P8.5_vs_bline
%
change_2100_RC-
P4.5_vs_bline
%
change_2100_RC-
P8.5_vs_bline
Name of the
Basin
no.
182 0.910 0.348 0.215 51.620 52.139 60.391 61.988 Jubba 85
162 0.862 0.344 0.213 65.255 65.715 64.319 67.353 Lempa 86
55 0.215 0.344 0.213 27.674 27.868 28.422 40.177 Uvs Nuur 87
235 0.816 0.340 0.211 61.130 61.532 61.541 65.097 La Plata 88
163 0.987 0.340 0.211 45.265 45.516 53.507 50.611 Gambia 89
249 0.564 0.339 0.210 62.603 65.476 66.910 77.887 Itata 90
48 0.511 0.336 0.209 26.985 29.252 25.959 49.819 Oder 91
66 0.529 0.336 0.209 66.922 70.718 68.196 79.323 Mius 92
232 0.868 0.335 0.208 70.740 70.465 74.673 74.437 Ruvuma 93
146 0.868 0.335 0.208 39.982 38.825 37.597 46.008 Dajabon 94
77 0.630 0.332 0.207 40.074 38.096 30.458 44.491 Krka 95
214 0.901 0.332 0.207 61.174 63.429 67.555 74.938 Ogooue 96
254 0.428 0.330 0.206 16.922 14.697 26.705 71.237 Aysen 97
166 0.848 0.329 0.205 66.726 69.933 67.155 70.093 Goascoran 98
178 0.917 0.329 0.205 54.367 55.484 56.799 61.101 Orinoco 99
109 0.609 0.325 0.203 148.591 149.671 139.922 145.642 Astara 100
127 0.759 0.324 0.203 45.242 44.446 44.398 40.608 Daoura 101
99 0.350 0.318 0.200 21.566 23.694 28.921 47.519 Samur 102
76 0.379 0.318 0.200 55.322 57.039 58.810 85.333 Razdolnaya 103
169 0.969 0.316 0.199 67.153 69.001 71.816 66.659 Geba 104
184 0.885 0.315 0.198 62.539 62.080 61.424 66.345 Catatumbo 105
206 0.905 0.314 0.198 69.531 71.949 74.245 79.205 Amazon 106
205 0.922 0.314 0.198 83.329 81.464 82.615 97.463 Maroni 107
238 0.464 0.314 0.198 34.925 36.805 38.625 49.510 Salar de
Coipasa
108
194 0.918 0.314 0.198 90.354 93.346 95.117 94.494 Essequibo 109
30 0.344 0.313 0.198 43.446 42.903 47.988 71.267 Nelson 110
105 0.681 0.309 0.196 55.291 56.500 55.106 57.243 Guadiana 111
137 0.959 0.308 0.195 77.261 76.721 85.745 78.406 Lake Chad 112
227 0.763 0.300 0.191 78.424 79.710 79.433 78.167 Chira 113
251 0.395 0.298 0.190 29.415 27.369 29.453 67.673 Reremo 114
10 0.161 0.288 0.186 12.881 12.881 12.410 13.457 Yukon 115
128 0.782 0.286 0.184 45.867 44.195 42.345 36.756 Draa 116
40 0.477 0.284 0.183 35.546 38.240 38.994 54.089 Pregolya 117
247 0.724 0.284 0.183 86.172 87.663 88.132 90.378 Lagoa Mirim 118
172 0.916 0.278 0.180 89.899 91.832 93.084 95.622 Saigon 119
68 0.538 0.274 0.179 103.114 105.280 100.864 114.788 Elancik 120
175 0.935 0.271 0.177 99.318 98.860 100.004 111.787 Kraburi 121
134 0.896 0.270 0.177 39.833 39.831 43.966 50.476 Dasht 122
179 0.961 0.269 0.176 90.197 89.592 98.651 94.495 Oueme 123
33 0.363 0.266 0.174 15.388 17.901 22.999 50.747 Fraser 124
94 0.591 0.265 0.174 86.589 86.945 94.064 101.435 Veleka 125
258 0.454 0.260 0.171 46.802 48.582 50.719 59.267 Gallegos 126
15 0.392 0.259 0.171 1.093 5.871 6.675 39.432 Neva 127
43 0.476 0.256 0.170 37.708 38.605 42.487 63.132 Prokhladnaya 128
31 0.463 0.254 0.169 36.106 38.221 41.699 57.345 Lielupe 129
253 0.452 0.254 0.169 33.442 33.457 38.158 90.112 Palena 130
126 0.788 0.254 0.169 43.824 43.493 42.999 48.813 Guir 131
189 0.683 0.250 0.167 55.589 56.549 56.286 64.103 Chiriqui Viejo 132
93 0.519 0.249 0.166 94.790 98.658 102.868 123.806 Kura 133
252 0.460 0.249 0.166 47.315 46.236 57.048 93.187 Yelcho 134
36 0.487 0.248 0.166 77.638 82.214 80.318 115.561 Dnieper 135
261 0.444 0.243 0.163 45.337 52.485 58.706 59.204 Chorrillo Gama 136
256 0.422 0.243 0.163 38.084 37.882 41.435 69.739 Pascua 137
7 0.000 0.240 0.162 41.525 42.947 43.800 37.571 Firth 138
62 0.391 0.240 0.162 22.857 25.265 29.616 67.280 Skagit 139
69 0.533 0.238 0.161 41.556 41.986 45.016 64.626 Po 140
41 0.428 0.238 0.161 75.811 80.973 85.096 115.802 Ural 141
236 0.473 0.236 0.160 59.399 61.091 63.889 74.703 Lake Poopo 142
63 0.351 0.235 0.159 70.909 72.570 77.699 103.299 Ulungur Lake 143
250 0.441 0.235 0.159 66.619 64.372 72.727 111.168 Puelo 144
90 0.547 0.229 0.157 94.276 100.482 104.004 136.217 Struma 145
218 0.898 0.228 0.156 120.453 117.838 118.462 132.059 Muni 146
61 0.515 0.226 0.155 90.221 101.880 97.406 133.495 Dniester 147
262 0.430 0.226 0.155 31.452 33.336 40.908 54.445 Carmen Sylva 148
222 0.905 0.220 0.152 101.254 101.497 103.809 117.353 Nyanga 149
102 0.640 0.220 0.152 90.515 91.062 90.375 96.143 Tagus 150
81 0.532 0.217 0.151 71.715 74.709 72.156 108.972 Neretva 151
201 0.935 0.213 0.149 122.078 121.610 135.796 134.107 Cross 152
42 0.505 0.213 0.148 18.451 18.818 19.793 23.536 Erne 153
255 0.429 0.211 0.147 28.357 24.239 33.961 57.870 Baker 154
221 0.908 0.210 0.147 122.709 126.033 128.722 138.947 Mamberamo 155
151 0.881 0.209 0.147 101.391 102.178 105.692 115.989 Grijalva 156
165 0.863 0.208 0.146 96.493 98.875 97.382 105.066 Choluteca 157
97 0.515 0.208 0.146 99.893 117.166 123.834 150.927 Nestos 158
8 0.285 0.207 0.146 −3.351 −3.359 −3.072 3.904 Tuloma 159
220 0.802 0.206 0.145 101.007 94.896 111.627 109.755 Lake Natron 160
(continued on next page)
F. Farinosi et al. Global Environmental Change 52 (2018) 286–313
310
Table A3 (continued)
Basin_ID Climatic zone
(Temperature)
Likelihood
hydro-polit
issues_norm
likelihood_bline %
change_2050_RC-
P4.5_vs_bline
%
change_2050_RC-
P8.5_vs_bline
%
change_2100_RC-
P4.5_vs_bline
%
change_2100_RC-
P8.5_vs_bline
Name of the
Basin
no.
170 0.946 0.204 0.144 112.089 111.106 117.249 108.119 Corubal 161
216 0.725 0.204 0.144 68.638 69.153 70.800 72.413 Mira 162
20 0.275 0.202 0.143 10.037 10.291 13.495 34.315 Chilkat 163
225 0.819 0.195 0.140 130.927 130.265 126.865 124.303 Tumbes 164
89 0.591 0.191 0.138 79.343 81.642 84.993 106.493 Ebro 165
259 0.450 0.191 0.138 32.763 30.245 35.212 38.374 Cullen 166
263 0.428 0.188 0.136 43.748 46.420 50.323 63.073 Grande 167
230 0.948 0.188 0.136 127.423 143.191 158.373 154.442 Maro 168
71 0.560 0.187 0.136 142.102 149.462 125.675 181.356 Sarata 169
212 0.926 0.185 0.135 174.396 182.228 195.985 195.665 Courantyne 170
183 0.733 0.185 0.135 91.256 90.034 86.362 121.519 Changuinola 171
70 0.555 0.184 0.135 143.751 154.680 144.589 190.962 Kogilnik 172
150 0.937 0.184 0.134 121.309 123.991 126.775 135.037 Candelaria 173
4 0.314 0.184 0.134 4.914 5.518 8.702 15.068 Jacobselv 174
74 0.429 0.182 0.134 113.693 120.834 127.937 159.299 St. Croix 175
257 0.415 0.180 0.132 56.967 58.024 58.052 71.070 Serrano 176
204 0.965 0.179 0.132 73.013 71.153 85.044 83.953 Lotagipi
Swamp
177
149 0.920 0.178 0.131 119.900 120.154 122.313 137.571 Hondo 178
22 0.436 0.178 0.131 33.026 41.514 47.018 88.089 Narva 179
181 0.762 0.177 0.131 90.168 94.537 95.792 111.275 Sixaola 180
72 0.507 0.176 0.130 54.230 56.041 65.278 92.729 Soca 181
264 0.384 0.176 0.130 62.659 65.489 65.602 75.172 Azopardo 182
260 0.446 0.173 0.129 30.289 31.493 37.774 46.175 San Martin 183
91 0.523 0.169 0.127 115.140 122.155 125.039 178.758 Drin 184
67 0.394 0.167 0.126 82.972 89.305 102.765 171.721 St. John 185
50 0.516 0.167 0.126 47.848 53.256 56.612 94.477 Elbe 186
27 0.443 0.164 0.124 20.937 27.914 26.188 79.302 Gauja 187
98 0.570 0.162 0.123 176.007 172.150 185.032 187.255 Rezovo 188
174 0.955 0.157 0.121 141.517 144.171 165.421 164.383 Komoe 189
147 0.867 0.157 0.121 120.411 115.799 118.560 131.013 Artibonite 190
177 0.948 0.156 0.120 137.920 139.242 152.438 154.596 Great Scarcies 191
219 0.896 0.155 0.120 173.706 167.770 173.505 187.674 Komo 192
88 0.556 0.152 0.119 153.223 162.162 162.986 217.157 Vardar 193
16 0.337 0.151 0.118 16.796 18.906 26.135 40.091 Glomma 194
28 0.465 0.149 0.117 26.519 29.577 29.132 67.981 Venta 195
25 0.311 0.149 0.117 18.827 18.859 19.131 31.272 Whiting 196
157 0.828 0.147 0.116 107.254 108.978 114.931 131.645 Motagua 197
199 0.950 0.147 0.116 138.030 137.874 160.241 160.190 Tano 198
45 0.499 0.146 0.115 20.030 17.133 10.005 16.141 Castletown 199
18 0.212 0.143 0.114 31.921 31.729 30.558 37.024 Alsek 200
85 0.604 0.143 0.114 80.148 79.947 88.327 144.245 Bidasoa 201
39 0.495 0.140 0.113 13.628 14.706 17.044 24.157 Foyle 202
168 0.926 0.139 0.112 171.857 171.636 179.760 192.135 San Juan 203
209 0.858 0.136 0.111 155.412 158.018 179.530 211.183 Bangau 204
21 0.278 0.135 0.110 23.876 21.825 20.978 42.618 Taku 205
187 0.910 0.134 0.110 206.005 203.225 219.225 215.247 Moa 206
32 0.476 0.124 0.105 23.132 26.922 34.758 72.400 Bartuva 207
52 0.518 0.124 0.105 65.772 68.285 74.209 108.764 Rhine 208
193 0.922 0.123 0.105 210.320 210.304 230.299 245.266 Amakuru 209
3 0.241 0.123 0.104 6.523 4.455 2.202 2.831 Tana 210
65 0.522 0.121 0.103 73.025 77.041 87.550 126.725 Rhone 211
26 0.449 0.121 0.103 36.497 42.554 48.109 106.947 Salaca 212
23 0.264 0.120 0.103 39.127 37.569 35.044 54.081 Stikine 213
176 0.933 0.117 0.102 195.890 194.008 214.415 211.983 Little Scarcies 214
87 0.585 0.113 0.099 155.668 158.431 161.268 185.709 Douro 215
29 0.443 0.113 0.099 73.402 86.984 85.046 140.055 Daugava 216
60 0.567 0.112 0.099 107.133 106.555 113.580 167.160 Seine 217
96 0.570 0.111 0.099 111.822 115.256 136.297 229.442 Lima 218
75 0.594 0.111 0.099 141.552 148.393 151.884 178.245 Garonne 219
186 0.946 0.106 0.096 212.271 221.510 247.575 251.098 Mono 220
78 0.411 0.104 0.095 8.357 15.140 18.320 97.793 Roya 221
100 0.437 0.102 0.094 161.495 173.102 193.599 264.468 Coruh 222
34 0.469 0.102 0.094 90.623 95.664 99.557 153.148 Neman 223
200 0.883 0.100 0.093 231.996 231.870 230.575 289.672 Jurado 224
83 0.568 0.100 0.093 108.232 108.621 117.314 186.477 Minho 225
9 0.263 0.097 0.092 17.455 13.514 10.072 23.592 Torne 226
103 0.584 0.097 0.091 208.022 216.359 230.081 254.334 Vjose 227
17 0.412 0.096 0.091 44.189 53.186 65.902 110.718 Gota alv 228
160 0.866 0.093 0.090 193.171 193.212 195.294 205.530 Coco 229
5 0.271 0.093 0.090 −2.885 −3.258 −3.179 0.967 Naatamo 230
6 0.271 0.091 0.088 13.682 13.299 10.993 23.334 Paatsjoki 231
180 0.917 0.091 0.088 232.020 231.195 268.440 268.096 Sassandra 232
190 0.898 0.084 0.085 310.934 308.310 311.350 299.824 Saint Paul 233
226 0.884 0.082 0.084 192.595 190.027 207.611 215.557 Umba 234
198 0.918 0.081 0.084 324.308 316.243 308.188 327.404 Cestos 235
11 0.295 0.080 0.083 15.783 15.669 15.313 51.772 Kemijoki 236
(continued on next page)
F. Farinosi et al. Global Environmental Change 52 (2018) 286–313
311
References
Altmann, A., Tolosi, L., Sander, O., Lengauer, T., 2010. Permutation importance: a cor-
rected feature importance measure. Bioinformatics 26, 1340–1347. https://doi.org/
10.1093/bioinformatics/btq134.
Beck, L., Bernauer, T., Siegfried, T., Böhmelt, T., 2014. Implications of hydro-political
dependency for international water cooperation and conflict: insights from new data.
Polit. Geogr. 42, 23–33. https://doi.org/10.1016/j.polgeo.2014.05.004.
Beck, H.E., van Dijk, A.I.J.M., Levizzani, V., Schellekens, J., Miralles, D.G., Martens, B., de
Roo, A., 2017. MSWEP: 3-hourly 0.25° global gridded precipitation (1979-2015) by
merging gauge, satellite, and reanalysis data. Hydrol. Earth Syst. Sci. 21, 589–615.
https://doi.org/10.5194/hess-21-589-2017.
Beguería, S., Vicente-Serrano, S.M., 2014. Calculation of the Standardised Precipitation-
Evapotranspiration Index.
Bernauer, T., Böhmelt, T., 2014. Basins at risk: predicting international river basin con-
flict and cooperation. Glob. Environ. Polit. 14, 116–138. https://doi.org/10.1162/
GLEP_a_00260.
Bernauer, T., Böhmelt, T., Buhaug, H., Gleditsch, N.P., Tribaldos, T., Weibust, E.B.,
Wischnath, G., 2012a. Water-related intrastate conflict and cooperation (WARICC): a
new event dataset. Int. Interact. 38, 529–545. https://doi.org/10.1080/03050629.
2012.697428.
Bernauer, T., Böhmelt, T., Koubi, V., 2012b. Environmental changes and violent conflict.
Environ. Res. Lett. 7, 15601.
Böhmelt, T., Bernauer, T., Buhaug, H., Gleditsch, N.P., Tribaldos, T., Wischnath, G., 2014.
Demand, supply, and restraint: determinants of domestic water conflict and co-
operation. Glob. Environ. Chang. 29, 337–348. https://doi.org/10.1016/j.gloenvcha.
2013.11.018.
Breiman, L., 1994. Bagging Preddictors. Berkeley, CA, USA.
Breiman, L., 2001. Random Forest. Mach. Learn. 45, 5–32. https://doi.org/10.1023/
A:1010933404324.
Brochmann, M., Gleditsch, N.P., 2012. Shared rivers and conflict –a reconsideration.
Polit. Geogr. 31, 519–527. https://doi.org/10.1016/j.polgeo.2012.11.001.
Buhaug, H., 2010. Climate not to blame for African civil wars. Proc. Natl. Acad. Sci. 107,
16477–16482. https://doi.org/10.1073/pnas.1005739107.
Burke, M.B., Miguel, E., Satyanath, S., Dykema, J.A., Lobell, D.B., 2009. Warming in-
creases the risk of civil war in Africa. Proc. Natl. Acad. Sci. 106, 20670–20674.
https://doi.org/10.1073/pnas.0907998106.
Chen, X., Ishwaran, H., 2012. Random forests for genomic data analysis. Genomics 99,
323–329. https://doi.org/10.1016/j.ygeno.2012.04.003.
CIESIN, 2015. Gridded Population of the World (GPW), v4. New York, NY, USA. https://
doi.org/10.7927/H4SF2T42.
Couttenier, M., Soubeyran, R., 2014. Drought and civil war in sub-saharan africa. Econ. J.
124, 201–244. https://doi.org/10.1111/ecoj.12042.
De Stefano, L., Duncan, J., Dinar, S., Stahl, K., Strzepek, K., Wolf, A.T., 2010a. Mapping
the Resilience of International River Basins to Future Climate Change-Induced Water
Variability. Washington DC, USA.
De Stefano, L., Edwards, P., de Silva, L., Wolf, A.T., 2010b. Tracking cooperation and
conflict in international basins: historic and recent trends. Water Policy 12, 871.
https://doi.org/10.2166/wp.2010.137.
De Stefano, L., Duncan, J., Dinar, S., Stahl, K., Strzepek, K.M., Wolf, A.T., 2012. Climate
change and the institutional resilience of international river basins. J. Peace Res. 49,
193–209. https://doi.org/10.1177/0022343311427416.
De Stefano, L., Petersen-Perlman, J.D., Sproles, E.A., Eynard, J., Wolf, A.T., 2017.
Assessment of transboundary river basins for potential hydro-political tensions. Glob.
Environ. Chang. 45, 35–46. https://doi.org/10.1016/j.gloenvcha.2017.04.008.
Diaz-Uriarte, R., 2014. varSelRF: Variable Selection Using Random Forests.
Dinar, A., 2004. Exploring transboundary water conflict and cooperation. Water Resour.
Res. 40https://doi.org/10.1029/2003WR002598. n/a-n/a.
Dinar, A., Blankespoor, B., Dinar, S., Kurukulasuriya, P., 2010. Does precipitation and
runoffvariability affect treaty cooperation between states sharing international bi-
lateral rivers? Ecol. Econ. 69, 2568–2581. https://doi.org/10.1016/j.ecolecon.2010.
07.036.
Dinar, S., Dinar, A., Kurukulasuriya, P., 2011. Scarcity and cooperation along interna-
tional rivers: an empirical assessment of bilateral treaties1. Int. Stud. Q. 55, 809–833.
https://doi.org/10.1111/j.1468-2478.2011.00671.x.
Dinar, S., Katz, D., De Stefano, L., Blankespoor, B., 2015. Climate change, conflict, and
cooperation: global analysis of the effectiveness of international river treaties in
addressing water variability. Polit. Geogr. 45, 55–66. https://doi.org/10.1016/j.
polgeo.2014.08.003.
Donner, L.J., Wyman, B.L., Hemler, R.S., Horowitz, L.W., Ming, Y., Zhao, M., Golaz, J.-C.,
Ginoux, P., Lin, S.-J., Schwarzkopf, M.D., Austin, J., Alaka, G., Cooke, W.F.,
Delworth, T.L., Freidenreich, S.M., Gordon, C.T., Griffies, S.M., Held, I.M., Hurlin,
W.J., Klein, S.A., Knutson, T.R., Langenhorst, A.R., Lee, H.-C., Lin, Y., Magi, B.I.,
Malyshev, S.L., Milly, P.C.D., Naik, V., Nath, M.J., Pincus, R., Ploshay, J.J.,
Ramaswamy, V., Seman, C.J., Shevliakova, E., Sirutis, J.J., Stern, W.F., Stouffer, R.J.,
Wilson, R.J., Winton, M., Wittenberg, A.T., Zeng, F., 2011. The dynamical core,
physical parameterizations, and basic simulation characteristics of the atmospheric
component AM3 of the GFDL global coupled model CM3. J. Clim. 24, 3484–3519.
https://doi.org/10.1175/2011JCLI3955.1.
Dufresne, J.-L., Foujols, M.-A., Denvil, S., Caubel, A., Marti, O., Aumont, O., Balkanski, Y.,
Bekki, S., Bellenger, H., Benshila, R., Bony, S., Bopp, L., Braconnot, P., Brockmann, P.,
Cadule, P., Cheruy, F., Codron, F., Cozic, A., Cugnet, D., de Noblet, N., Duvel, J.-P.,
Ethé, C., Fairhead, L., Fichefet, T., Flavoni, S., Friedlingstein, P., Grandpeix, J.-Y.,
Guez, L., Guilyardi, E., Hauglustaine, D., Hourdin, F., Idelkadi, A., Ghattas, J.,
Joussaume, S., Kageyama, M., Krinner, G., Labetoulle, S., Lahellec, A., Lefebvre, M.-
P., Lefevre, F., Levy, C., Li, Z.X., Lloyd, J., Lott, F., Madec, G., Mancip, M., Marchand,
M., Masson, S., Meurdesoif, Y., Mignot, J., Musat, I., Parouty, S., Polcher, J., Rio, C.,
Schulz, M., Swingedouw, D., Szopa, S., Talandier, C., Terray, P., Viovy, N., Vuichard,
N., 2013. Climate change projections using the IPSL-CM5 earth system model: from
CMIP3 to CMIP5. Clim. Dyn. 40, 2123–2165. https://doi.org/10.1007/s00382-012-
1636-1.
Espey, M., Towfique, B., 2004. International bilateral water treaty formation. Water
Resour. Res. 40https://doi.org/10.1029/2003WR002534. n/a-n/a.
Falkenmark, M., Lannerstad, M., 2005. Consumptive water use to feed humanity - curing
a blind spot. Hydrol. Earth Syst. Sci. 9, 15–28. https://doi.org/10.5194/hess-9-15-
2005.
Freire, S., Pesaresi, M., 2015. GHS Population Grid, Derived from GPW4, Multitemporal
(1975, 1990, 2000, 2015).
Gain, A.K., Giupponi, C., Wada, Y., 2016. Measuring global water security towards sus-
tainable development goals. Environ. Res. Lett. 11, 124015. https://doi.org/10.
1088/1748-9326/11/12/124015.
Gassert, F., Landis, M., Luck, M., Reig, P., Shiao, T., 2013. AQUEDUCT Global Maps 2.0.
Gassert, F., Landis, M., Luck, M., Reig, P., Shiao, T., 2014. Aqueduct Global Maps 2.1.
Washington DC, USA.
Gerlak, A., Zawahri, N., 2009. Navigating international river disputes to avert conflict.
Int. Negot. 14, 211–227. https://doi.org/10.1163/157180609X432806.
Giordano, M.A., Wolf, A.T., 2003. Sharing waters: Post-Rio international water man-
agement. Nat. Resour. Forum 27, 163–171. https://doi.org/10.1111/1477-8947.
00051.
Gleditsch, K.S., 2002. Expanded trade and GDP data. J. Conflict Resolut. 46, 712–724.
https://doi.org/10.1177/0022002702046005006.
Gleditsch, N.P., Furlong, K., Hegre, H., Lacina, B., Owen, T., 2006. Conflicts over shared
rivers: Resource scarcity or fuzzy boundaries? Polit. Geogr. 25, 361–382. https://doi.
org/10.1016/j.polgeo.2006.02.004.
Gleick, P.H., 1998. Water and conflict. The World’s Water 1998-1999. pp. 105–135.
Gleick, P.H., Yolles, P., Hatami, H., 1994. Water, war & peace in the Middle East.
Environment 36 6-.
Hastie, T., Tibshirani, R., Friedman, J., 2009. The Elements of Statistical Learning.
Springer Series in Statistics, Springer New York, New York, NY. https://doi.org/10.
1007/978-0-387-84858-7.
Hensel, P.R., McLaughlin Mitchell, S., Sowers, T.E., 2006. Conflict management of ri-
parian disputes. Polit. Geogr. 25, 383–411. https://doi.org/10.1016/j.polgeo.2005.
11.001.
Hensel, P.R., Mitchell, S.M., Sowers, T.E.I., Thyne, C.L., 2008. Bones of contention:
comparing territorial, maritime, and river issues. J. Conflict Resolut. 52, 117–143.
Table A3 (continued)
Basin_ID Climatic zone
(Temperature)
Likelihood
hydro-polit
issues_norm
likelihood_bline %
change_2050_RC-
P4.5_vs_bline
%
change_2050_RC-
P8.5_vs_bline
%
change_2100_RC-
P4.5_vs_bline
%
change_2100_RC-
P8.5_vs_bline
Name of the
Basin
no.
208 0.896 0.077 0.082 271.152 272.996 278.096 325.800 Sembakung 237
191 0.902 0.074 0.080 317.685 315.511 317.048 313.667 Lofa 238
37 0.513 0.074 0.080 −13.186 −11.370 −13.433 2.049 Bann 239
14 0.344 0.069 0.078 21.673 28.563 33.546 144.646 Oulujoki 240
202 0.950 0.066 0.076 229.932 228.168 260.177 259.918 Bia 241
197 0.911 0.066 0.076 355.824 352.873 341.956 351.122 Saint John 242
38 0.511 0.058 0.072 37.020 38.330 54.346 113.547 Vida 243
156 0.910 0.045 0.066 264.224 265.564 278.304 310.664 Sarstun 244
53 0.554 0.044 0.065 99.300 104.848 123.365 190.222 Scheldt 245
167 0.916 0.042 0.065 364.270 361.837 339.826 352.057 Negro 246
195 0.915 0.036 0.062 411.293 404.598 422.166 459.592 Mano 247
47 0.511 0.025 0.056 −12.873 −10.599 −17.494 −1.670 Flurry 248
207 0.935 0.020 0.054 412.128 407.433 470.367 462.026 Akpa Yafi249
54 0.553 0.001 0.044 60.077 80.453 111.309 260.706 Yser 250
196 0.911 0.000 0.044 546.906 535.628 538.646 589.893 Cavalla 251
F. Farinosi et al. Global Environmental Change 52 (2018) 286–313
312
Hsiang, S.M., Meng, K.C., Cane, M.A., 2011. Civil conflicts are associated with the global
climate. Nature 476, 438–441. https://doi.org/10.1038/nature10311.
Jeong, J.H., Resop, J.P., Mueller, N.D., Fleisher, D.H., Yun, K., Butler, E.E., Timlin, D.J.,
Shim, K.-M., Gerber, J.S., Reddy, V.R., Kim, S.-H., 2016. Random forests for global
and regional crop yield predictions. PLoS One 11, e0156571. https://doi.org/10.
1371/journal.pone.0156571.
Jing, W., Yang, Y., Yue, X., Zhao, X., 2016. A comparison of different regression algo-
rithms for downscaling monthly satellite-based precipitation over North China.
Remote Sens. (Basel) 8, 835. https://doi.org/10.3390/rs8100835.
Kalbhenn, A., Bernauer, T., 2012. International Water cooperation and conflict: a new
event dataset. Ssrn Electron. J. https://doi.org/10.2139/ssrn.2176609.
Kallis, G., Zografos, C., 2014. Hydro-climatic change, conflict and security. Clim. Change
123, 69–82. https://doi.org/10.1007/s10584-013-0893-2.
Katz, D., 2011. Hydro-political hyperbole: examining incentives for overemphasizing the
risks of water wars. Glob. Environ. Polit. 11, 12–35. https://doi.org/10.1162/GLEP_
a_00041.
Kaufmann, D., Kraay, A., Mastruzzi, M., 2010. The Worldwide Governance Indicators:
Methodology and Analytical Issues.
Knutti, R., Masson, D., Gettelman, A., 2013. Climate model genealogy: generation CMIP5
and how we got there. Geophys. Res. Lett. 40, 1194–1199. https://doi.org/10.1002/
grl.50256.
Kuhn, M., 2008. Caret Package. J. Stat. Softw. 28.
Kvålseth, T.O., 1985. Cautionary note about R 2. Am. Stat. 39, 279–285. https://doi.org/
10.1080/00031305.1985.10479448.
Li, B., Yang, G., Wan, R., Dai, X., Zhang, Y., 2016. Comparison of random forests and
other statistical methods for the prediction of lake water level: a case study of the
Poyang Lake in China. Hydrol. Res. 47, 69–83. https://doi.org/10.2166/nh.2016.
264.
Liaw, A., Wiener, M., 2002. Classification and regression by randomForest. R News.
Loidl, V., Oberhauser, C., Ballert, C., Coenen, M., Cieza, A., Sabariego, C., 2016. Which
environmental factors have the highest impact on the performance of people ex-
periencing difficulties in capacity? Int. J. Environ. Res. Public Health 13, 416.
https://doi.org/10.3390/ijerph13040416.
Malekipirbazari, M., Aksakalli, V., 2015. Risk assessment in social lending via random
forests. Expert Syst. Appl. 42, 4621–4631. https://doi.org/10.1016/j.eswa.2015.02.
001.
Malthus, T., 1798. Essay on the Principle of Population.
McDonald, R.I., Weber, K., Padowski, J., Flörke, M., Schneider, C., Green, P.A., Gleeson,
T., Eckman, S., Lehner, B., Balk, D., Boucher, T., Grill, G., Montgomery, M., 2014.
Water on an urban planet: urbanization and the reach of urban water infrastructure.
Glob. Environ. Chang. 27, 96–105. https://doi.org/10.1016/j.gloenvcha.2014.04.
022.
McKee, T.B., Doesken, N.J., Kleist, J., 1993. The relationship of drought frequency and
duration to time scales. Prepr. 8th Conf. Appl. Climatol. 179–184.
Meinshausen, M., Meinshausen, N., Hare, W., Raper, S.C.B., Frieler, K., Knutti, R., Frame,
D.J., Allen, M.R., 2009. Greenhouse-gas emission targets for limiting global warming
to 2 degrees C. Nature 458, 1158–1162. https://doi.org/10.1038/nature08017.
Meinshausen, M., Smith, S.J., Calvin, K., Daniel, J.S., Kainuma, M.L.T., Lamarque, J.-F.,
Matsumoto, K., Montzka, S.A., Raper, S.C.B., Riahi, K., Thomson, A., Velders, G.J.M.,
Vuuren, D.P.P., 2011. The RCP greenhouse gas concentrations and their extensions
from 1765 to 2300. Clim. Change 109, 213–241. https://doi.org/10.1007/s10584-
011-0156-z.
Mohr, C.H., Manga, M., Wang, C.-Y., Korup, O., 2017. Regional changes in streamflow
after a megathrust earthquake. Earth Planet. Sci. Lett. 458, 418–428. https://doi.org/
10.1016/j.epsl.2016.11.013.
Munia, H., Guillaume, J.H.A., Mirumachi, N., Porkka, M., Wada, Y., Kummu, M., 2016.
Water stress in global transboundary river basins: significance of upstream water use
on downstream stress. Environ. Res. Lett. 11, 014002. https://doi.org/10.1088/
1748-9326/11/1/014002.
NCAR CESM, 2011. Community Climate System Model/Climate and Global Dynamics
Division/National Center for Atmospheric Research/University Corporation for
Atmospheric Research, and Data Support Section/Computational and Information
Systems Laboratory/National Center for Atm. Boulder, CO - USA. https://doi.org/10.
5065/D6TH8JP5.
Núñez, J., Hallack-Alegría, M., Cadena, M., 2016. Resolving regional frequency analysis
of precipitation at large and complex scales using a bottom-up approach: the Latin
America and the Caribbean drought atlas. J. Hydrol. 538, 515–538. https://doi.org/
10.1016/j.jhydrol.2016.04.025.
Petersen-Perlman, J.D., 2016. Projecting River Basin resilience in the Zambezi River Basin
through global analyses and basin realities. Water Resour. Manag. 30, 1987–2003.
https://doi.org/10.1007/s11269-016-1264-3.
Pourtaghi, Z.S., Pourghasemi, H.R., Aretano, R., Semeraro, T., 2016. Investigation of
general indicators influencing on forest fire and its susceptibility modeling using
different data mining techniques. Ecol. Indic. 64, 72–84. https://doi.org/10.1016/j.
ecolind.2015.12.030.
Rasquinha, D.N., Sankaran, M., 2016. Modelling biome shifts in the Indian subcontinent
under scenarios of future climate change. Curr. Sci. 111, 147. https://doi.org/10.
18520/cs/v111/i1/147-156.
Reig, P., Shiao, T., Gassert, F., 2013. Aqueduct Water Risk Framework. Washington DC,
USA.
Sadoff, C.W., Grey, D., 2002. Beyond the river: the benefits of cooperation on interna-
tional rivers. Water Policy 4, 389–403. https://doi.org/10.1016/S1366-7017(02)
00035-1.
Seber, G.A.F., Lee, A.J., 2003. Linear Regression Analysis. John Wiley & Sons, Inc,
Hoboken, NJ, USA. https://doi.org/10.1002/9780471722199.
Segal, M.R., 2004. Machine Learning Benchmarks and Random Forest Regression.
Singer, J.D., Bremer, S., Stuckey, J., 1972. Capability distribution, uncertainty, and major
power wars, 1820-1965. In: Russett, B.M. (Ed.), Peace, War, and Numbers. Sage,
Beverly Hills, CA, pp. 19–48.
Sirin, C.V., 2011. Scarcity-induced domestic conflict: examining the interactive effects of
environmental scarcity and ‘Ethnic’population pressures. Civ. Wars 13, 122–140.
https://doi.org/10.1080/13698249.2011.576141.
Strobl, C., Boulesteix, A.-L., Zeileis, A., Hothorn, T., 2007. Bias in random forest variabe
importance measures. BMC Bioinf. 8, 25. https://doi.org/10.1186/1471-2105-8-25.
Strobl, C., Malley, J., Tutz, G., 2009. An introduction to recursive partitioning: rationale,
application, and characteristics of classification and regression trees, bagging, and
random forests. Psychol. Methods 14, 323–348. https://doi.org/10.1037/a0016973.
Taylor, K.E., Stouffer, R.J., Meehl, G.A., 2012. An overview of CMIP5 and the experiment
design. Bull. Am. Meteorol. Soc. 93, 485–498. https://doi.org/10.1175/BAMS-D-11-
00094.1.
Thrasher, B., Maurer, E.P., McKellar, C., Duffy, P.B., 2012. Technical note: bias correcting
climate model simulated daily temperature extremes with quantile mapping. Hydrol.
Earth Syst. Sci. 16, 3309–3314. https://doi.org/10.5194/hess-16-3309-2012.
Tin Kam Ho, 1998. The random subspace method for constructing decision forests. IEEE
Trans. Pattern Anal. Mach. Intell. 20, 832–844. https://doi.org/10.1109/34.709601.
UN/DESA, 2017. World Population Prospects: The 2017 Revision, Key Findings and
Advance Tables (Working Paper No. ESA/P/WP/248).
UNEP-DHI, UNEP, 2016. Transboundary River Basins: Status and Trends. Nairobi,
Kenya.
Vogels, M.F.A., de Jong, S.M., Sterk, G., Addink, E.A., 2017. Agricultural cropland
mapping using black-and-white aerial photography, object-based image analysis and
random forests. Int. J. Appl. Earth Obs. Geoinf. 54, 114–123. https://doi.org/10.
1016/j.jag.2016.09.003.
Volodin, E.M., Dianskii, N.A., Gusev, A.V., 2010. Simulating present-day climate with the
INMCM4.0 coupled model of the atmospheric and oceanic general circulations. Izv.
Atmos. Ocean. Phys. 46, 414–431. https://doi.org/10.1134/S000143381004002X.
Volodin, E.M., Diansky, N.A., Gusev, A.V., 2013. Simulation and prediction of climate
changes in the 19th to 21st centuries with the institute of numerical mathematics,
Russian academy of sciences, model of the earth’s climate system. Izv. Atmos. Ocean.
Phys. 49, 347–366. https://doi.org/10.1134/S0001433813040105.
Watanabe, M., Suzuki, T., O’ishi, R., Komuro, Y., Watanabe, S., Emori, S., Takemura, T.,
Chikira, M., Ogura, T., Sekiguchi, M., Takata, K., Yamazaki, D., Yokohata, T.,
Nozawa, T., Hasumi, H., Tatebe, H., Kimoto, M., 2010. Improved climate simulation
by MIROC5: mean states, variability, and climate sensitivity. J. Clim. 23, 6312–6335.
https://doi.org/10.1175/2010JCLI3679.1.
Watson, J.E., 2015. Beyond Cooperation: Environmental Justice in Transboundary Water
Management (PhD Dissertation). Oregon State University.
Weedon, G.P., Balsamo, G., Bellouin, N., Gomes, S., Best, M.J., Viterbo, P., 2014. The
WFDEI meteorological forcing data set: WATCH forcing data methodology applied to
ERA-interim reanalysis data. Water Resour. Res. 50, 7505–7514. https://doi.org/10.
1002/2014WR015638.
WEF, 2016. World Economic Forum - The Global Risks Report 2016. Geneva,
Switzerland.
WEF, 2017. World Economic Forum - The Global Risks Report 2017. Geneva,
Switzerland.
Welling, S.H., Refsgaard, H.H.F., Brockhoff, P.B., Clemmensen, L.H., 2016. Forest Floor
Visualizations of Random Forests. ArXiv e-prints.
WMO, 2012. Standardized Precipitation Index - User Guide - WMO-No. 1090.
Wolf, A., 1998. Conflict and cooperation along international waterways. Water Policy 1,
251–265. https://doi.org/10.1016/S1366-7017(98)00019-1.
Wolf, A.T., 2007. Shared waters: conflict and cooperation. Annu. Rev. Environ. Resour.
32, 241–269. https://doi.org/10.1146/annurev.energy.32.041006.101434.
Wolf, A.T., 2009. A long term view of water and international security 1. J. Contemp.
Water Res. Educ. 142, 67–75. https://doi.org/10.1111/j.1936-704X.2009.00056.x.
Wolf, A.T., Yoffe, S.B., Giordano, M., 2003. International waters: identifying basins at
risk. Water Policy 5 29 LP-60.
World Bank, 2018. WDI - World Development Indicators Databank. n.d..
Yoffe, S.B., Larson, K., 2002. Chapter 2: Water events database methodology. In: Basin at
Risk: Conflict and Cooperation Over International Freshwater Resources (PhD
Dissertation). Oregon State University.
Yoffe, S., Wolf, A.T., Giordano, M., 2003. Conflict and cooperation over international
freshwater resources: indicators of basins at risk. J. Am. Water Resour. Assoc. 39,
1109–1126. https://doi.org/10.1111/j.1752-1688.2003.tb03696.x.
Yoffe, S., Fiske, G., Giordano, M., Giordano, M., Larson, K., Stahl, K., Wolf, A.T., 2004.
Geography of international water conflict and cooperation: data sets and applica-
tions. Water Resour. Res. 40https://doi.org/10.1029/2003WR002530. n/a-n/a.
Zawahri, N.A., Mitchell, S.M., 2011. Fragmented governance of international rivers: ne-
gotiating bilateral versus multilateral treaties1. Int. Stud. Q. 55, 835–858. https://
doi.org/10.1111/j.1468-2478.2011.00673.x.
Zeitoun, M., Mirumachi, N., 2008. Transboundary water interaction I: reconsidering
conflict and cooperation. Int. Environ. Agreements Polit. Law Econ. 8, 297–316.
https://doi.org/10.1007/s10784-008-9083-5.
Zeitoun, M., Mirumachi, N., Warner, J., 2011. Transboundary water interaction II: the
influence of ‘soft’power. Int. Environ. Agreements Polit. Law Econ. 11, 159–178.
https://doi.org/10.1007/s10784-010-9134-6.
F. Farinosi et al. Global Environmental Change 52 (2018) 286–313
313
