Unraveling the spatial patterns of water conflicts in Africa through a multi-scale lens Open Access

Conflicts revolving around water have existed since the advent of human civilization. The first recorded major water conflict dates back to 2500 BC when the dispute over water between two cities in ancient Mesopotamia, Umma and Lagash, lasted for almost a century (Hatami & Gleick 1994). Globally, water conflicts tripled between 2000 and 2019 (Institute for Economics & Peace 2022), posing increasing threats to livelihoods in vulnerable areas (Chellaney 2013; AGRA 2022). Water conflicts are particularly prevalent in sub-Saharan Africa (SSA), where the frequent droughts have rendered water a strategic resource (Peña-Ramos et al. 2022). Over the past 20 years, water conflicts in SSA have increased in their spatial coverage, fatalities, frequency, and in the share relative to the total number of conflicts reported in the region¹.

Extreme climatic events, particularly droughts, are shown to be strongly correlated with conflict in Africa (Maystadt & Ecker 2014; Koubi 2019; Ide et al. 2021). Looking ahead, drought frequency and severity are projected to rise due to climate change (Elkouk et al. 2021). Understanding the mechanisms that run from drought to conflict is therefore critical for designing anticipatory policies to prevent and mitigate water conflicts in the face of future droughts. However, establishing causality remains a significant challenge in the literature due to the complex pathways linking drought to conflict and the difficulty in identifying the direct cause of each conflict event.

Moreover, research on water conflicts has predominantly focused on interstate water conflicts, which are often driven by sovereignty and the geopolitics surrounding transboundary water bodies. Conversely, the drivers and within-country patterns of intrastate water conflicts remain underexplored. Identifying regions most susceptible to intrastate water conflicts can help prioritize interventions in the most vulnerable areas, ensuring water security and peace – especially in resource-constrained settings. Yet this question remains insufficiently addressed in existing literature.

This study attempts to address this gap. We analyze past and present data on intrastate water conflicts and droughts to examine whether the influence of drought on intrastate water conflicts in Africa differs across locations within a country, and if so, how it shapes future water conflict patterns. We focus explicitly on intrastate, instead of interstate, water conflicts because the former has an immediate impact on communities and people's livelihoods. By so doing, we extend the literature in the following aspects.

First, this study advances the understanding of the causal influence of drought on conflicts by focusing exclusively and explicitly on water conflict events. Synthesizing the definition of intrastate water conflicts based on past literature (Makaya et al. 2020; Pacific Institute 2024b; United Nations 2024b), we distinguish three types of water conflicts: (1) conflicts related to the control, use, and access of water; (2) the attacking of water facilities; and (3) protests and demonstrations related to water cuts or shortages. These categories are derived from annotating a georeferenced dataset of water conflicts in 21 African countries (2010–2024). Compared to previous studies that included all types of conflicts, irrespective of their relevance to water issues (e.g., Soysa 2002), our focus on water conflicts offers a more nuanced modeling of water conflicts, which is more relevant for informing water policies.

Second, we reveal the spatially heterogeneous influence of drought on intrastate water conflicts. Previous studies have focused on the role of water scarcity in affecting total conflicts, offering mixed conclusions: while water scarcity is found to increase conflict frequency (Toset et al. 2000; Peña-Ramos et al. 2022), water abundance is also found to increase conflict frequency (Soysa 2002). However, such analyses are often performed at an aggregate level, implicitly assuming that the impact of water availability on conflicts is spatially homogeneous. Whether the role of water scarcity in affecting intrastate water conflicts differs across locations remains underexplored, especially when scarcity is driven by climate extremes. To this end, we examine the spatially differential impact of climate extremes on water conflicts to understand where water conflicts are more likely to manifest under climate extremes.

Finally, based on the results of our analysis, we postulate that different locations may represent varying strengths of water governance. We follow previous definition of water governance: ‘The range of political, institutional and administrative rules, practices and processes (formal and informal) through which decisions are taken and implemented, stakeholders can articulate their interests and have their concerns considered, and decision-makers are held accountable for water management’ (OECD 2015). Learning from the literature (Fukuyama 2013, 2016), especially on water governance (Woodhouse & Muller 2017), we find suggestive evidence that distance to the country border may partially capture this strength. Studies have also shown that weak governance can lead to water conflicts, even without a physical scarcity of water (Funder et al. 2004). However, conventional approaches to quantify governance strength often rely on the constructed index of government performance (Huther & Shah 2005), such as effectiveness (Kaufmann 2003) or corruption (Donchev & Ujhelyi 2014). We are not aware of any spatially explicit measurement of water governance at the pixel level. Conflicts have been used to measure governance stability (Mlachila et al. 2020), but this is not applicable in our context since water conflicts are the dependent variables.

We adopt a multi-scale approach that integrates data and results across multiple spatial levels. We first estimate interactions between drought, border distance, and water conflict (2010–2024), then extend findings to 2030–2040. In the second step, we incorporate country-level water governance and household-level perceived governance strength. We find that drought increases water conflict intensity (represented herein as the number of fatalities). The level of intensity increases proportionally as drought occurs nearer to the country border, after controlling for possible confounding factors. Future droughts may alter the locations of water conflicts. We further present findings in support of the postulate that there is an empirical linkage between within-country location and (perceived) governance strength. At the country level, stronger water governance may offset (partially) the influence of physical water scarcity on water conflicts.

Our results are expected to help better understand and visualize the interaction between spatial location and drought in affecting water conflicts in Africa, now and in the future. Understanding this interaction is critical to identifying where efforts are needed to strengthen domestic water governance and ensure sustainable water management in the light of increasing droughts. Different areas might face different exposures/harms and thus need tailored support. We conclude with implications on practical approaches that can be applied to alleviate water conflicts within a country, and possible directions for future studies.

Evidence shows that intrastate water conflicts may display different patterns across locations. This may be due to several factors. First, without any climate crisis, different areas may have different levels of provision of water infrastructure (Dangui & Jia 2022), public goods, dysfunctional governance, and sharing mechanisms (Susskind 2013). There are also well-documented socio-economic disparities in the development and autonomy of a country's remote areas compared to central areas (Zhang 2001; Erkut & Özgen 2003), ambiguous or insecure water/pasture tenure (Kugbega & Aboagye 2021), and lower compliance with policies (Ferrara 2002), which may affect water conflict patterns. Resource management in rural areas is often more informal, shaped by local customary practices and community-based governance. Remote regions also often lack equitable representation in national policymaking. Core regions in a country tend to control decisions that affects the remote regions (Wu 2016).

Climate extremes have been shown to exacerbate water conflicts (Almer et al. 2017; Pacific Institute 2024a). On the supply side, climate extremes such as temperature anomalies, droughts, and floods can disturb groundwater recharge (Gleick & Heberger 2014; Khan & Rodella 2021; Unfried et al. 2022), leading to the scarcity of renewable water mass (Moursi et al. 2017). Irrigation systems and water storage can buffer against water shortage, but they are also more prone to be targeted during conflicts (Khan & Rodella 2021). Ide et al. (2021) show that cuts of the public water supply are relevant predictors of water conflict. On the demand side, climate extremes can lead to an increase in water demand, which intensifies the competition for water from different sectors and contributes to widening the supply-demand gap of water (Bisung et al. 2014).

Combining the above two aspects, in the face of drought, remote areas may be more susceptible to water conflicts. First, climate extremes often exacerbate the fight for water (Giordano et al. 2019). Unclear water allocation rules in the face of drought, lack of coordination, and unclear actor responsibilities in the face of climate extremes may add up to more water crises and, in certain contexts, result in conflicts (Makaya et al. 2020). Some argue that it is rarely the lack of water as such that fuels conflict, but rather its governance and management (Detges 2016). Persistent grievances near the border that are deeply embedded into historic socio-economic structures, in combination with a triggering event like a drought, are hence driving such water conflicts, especially in the absence of conflict resolution mechanisms.

The dependent variables are two measurements of water conflicts in each pixel every month: (i) the total number of water conflicts and (ii) the total fatality in water conflicts. The original conflict data are obtained from the Armed Conflict Location & Event Data Project (ACLED) (Raleigh et al. 2023). ACLED data is an event-based dataset. The unit in ACLED is a conflict event occurring on a specific date at a specific location (longitude and latitude). The datasets also detail the event type, involved actors, and other characteristics of these incidents. The data are collected in real time from a wide range of local, national, and international sources, curated and cleaned according to well-developed methodological principles, and published on a weekly basis. More details can be found on the ACLED website².

ACLED includes a detailed description of the reason and nature of each conflict event, actor type, and their association. This allows us to identify and annotate water conflicts. We first exclude conflicts that are between two state actors (interstate). Next, we manually select conflict events based on the detailed description of each event. We identify three major types of water conflicts: (i) resource-based water conflicts; (ii) destruction or targeting of water infrastructure; and (iii) protests or uprisings in response to water use restrictions or changes in water rights.

The first type of water conflict (resource-based) involves disputes over ownership, use, control, or access to water resources, often driven by scarcity, inequitable allocation, or competing priorities among stakeholders. An example of this ‘type 1’ water conflict is the following:

We define the second type of water conflicts (‘type 2’) as the conflicts induced by deliberate destruction, or targeting, of water infrastructure and facilities during the acts of aggression or warfare, including attacks on civilian water systems, or strategic damage to water infrastructure. According to this definition, water and water systems can be ‘casualties’ of violence for reasons other than control of, or access to, water resources. Water tanks, dams, and water utility equipment have all been attacked in recent incidents around the world. The destruction of water facilities often causes injuries or deaths for those involved in the destruction and can directly influence those residents who rely on those water facilities. For example:

The third type of water conflict encompasses demonstrations, protests, or uprisings in response to water use restrictions or changes in water rights. An example of a ‘type 3’ conflict is the following:

In this paper, we use threshold values in the Standardized Precipitation Index (SPI) to identify a drought event. The SPI is a widely used index recommended by the World Meteorological Organization (WMO) to reflect short- and medium-term soil moisture and surface water conditions. It represents the precipitation over specific time intervals for a given location, relative to the probability density distribution of its long-term precipitation record. It is calculated monthly for a moving window of N months of the accumulation period. For example, SPI for a moving window of six months accumulation period is denoted as SPI-6.

For understanding drought-related water scarcity, SPI-6 is often the most relevant. It aligns with key seasonal water resource management decisions, accounting for both short-term precipitation deficits and medium-term storage changes. In the main specification, following WMO guides (World Meteorological Organization 2012), we define mild drought as SPI-6 below −1, which represents a dry episode that is likely to cause surface water reduction and shortages, drying of wells, boreholes, or reservoirs, and water usage restrictions. We define severe drought as SPI-6 below −1.5, representing an extreme and exceptional dry period that likely causes groundwater and reservoir reduction. As a robustness check, we also construct SPI-3 and SPI-9. The is calculated at the pixel level from historical monthly precipitation data obtained from Climate Hazards Group InfraRed Precipitation with Station data (CHIRPS) version 3.0 (Funk et al. 2015).

For future droughts, we first calculate the average number of drought events per month in each pixel for the historic period. Then, we apply the same definition of drought to calculate the average number of drought events per month in each pixel for the future period 2030–2040, using projected precipitation data from the INMCM6.0 Global Climate Model (GCM), under the SSP 245 climate change pathway (Volodin et al. 2017). We chose this GCM because it reflects the average global warming. Based on the above GCM and scenario, we download the raw data from Climate Copernicus (European Union Earth Observation Programme 2024).

We conducted two auxiliary analyses to further understand the relationship between water management, drought, and water conflicts. More details will be introduced in the method section. First, at the country level, we use Integrated Water Resources Management (IWRM) to approximate the strength of water governance and explore its interaction with drought in affecting water conflicts. IWRM is a framework widely adopted across Africa to quantify and improve the quality of water management. The IWRM score at the country level is sourced from three waves of surveys conducted by the United Nations Environment Programme (UNEP) in 2017, 2020, and 2023 (United Nations 2024a) to measure the implementation of IWRM. In these surveys, the respective ministry responsible for water resource management in each country assesses the level of water governance and policy implementation. It contains four sub-indicators: enabling environment, institutions and participation, management instrument, and financing.

Table 1 summarizes the statistics of key variables. Unconditionally, pixels far from borders tend to show higher mean water storage and irrigated cropland. Conflict frequency and fatalities are also greater near borders compared to pixels far from the borders. Both population and population density are higher in pixels far from the border. On average, border regions experience greater short-term drought (SPI-3: −0.007) than areas beyond (0.121). However, medium- to long-term indices (SPI-6 and SPI-9) suggest relatively wetter conditions near the border.

Table 1

Summary statistics.

.	Pixels within 50 km from the border .		Pixels beyond 50 km from the border .
	Mean .	Standard Dev. .	Mean .	Standard Dev. .
SPI-3 months	−0.007	0.86	0.121	0.92
SPI-6 months	0.027	0.85	−0.038	0.88
SPI-9 months	0.014	0.94	−0.037	0.91
Total conflict (count)	183	59	160	61
Average fatalities (deaths)	2.89	5.01	1.89	6.22
(count)	1.14	0.33	1.10	0.41
Average (deaths)	1.81	6.75	1.69	3.88
Irrigated crop land (hectares)	560	1,211	670	1,462
Ground water storage (depth in mm)	16,460	10,705	22,203	13,676
Population density (people/km2)	54.09	40.12	78.68	61.24
Population (people)	48,681	29,546	70,812	38,974
Total observations (pixel-month)	167		272

.	Pixels within 50 km from the border .		Pixels beyond 50 km from the border .
	Mean .	Standard Dev. .	Mean .	Standard Dev. .
SPI-3 months	−0.007	0.86	0.121	0.92
SPI-6 months	0.027	0.85	−0.038	0.88
SPI-9 months	0.014	0.94	−0.037	0.91
Total conflict (count)	183	59	160	61
Average fatalities (deaths)	2.89	5.01	1.89	6.22
(count)	1.14	0.33	1.10	0.41
Average (deaths)	1.81	6.75	1.69	3.88
Irrigated crop land (hectares)	560	1,211	670	1,462
Ground water storage (depth in mm)	16,460	10,705	22,203	13,676
Population density (people/km2)	54.09	40.12	78.68	61.24
Population (people)	48,681	29,546	70,812	38,974
Total observations (pixel-month)	167		272

View Large

We consider two dependent variables. In the first set of equations, the dependent variable represents the frequency of water conflicts in pixel i during period t, where t takes the value from the year 2010 to 2024. In the second set of equations, denotes the number of fatalities attributed to water conflicts. Each equation set considers two drought measurements, namely severe and mild. measures the distance of pixel i to its country border. The dummy variable takes value 1 if a drought occurs in pixel i during period t, and in case of no drought. In the main specification, we use SPI-6 to define drought following the reasons in Section 3.3. We consider two drought variables: mild drought, where SPI-6 is smaller than negative 1, and severe drought where SPI-6 is smaller than negative 1.5. In total, we estimate four equations: two conflict outcomes (frequency and fatalities), over two drought measurements (mild and severe drought). represents the vector of control variables, which include month fixed effects to account for possible seasonality patterns that may be related to increase competition over water, as well as fixed effects at administrative level 1 (the first sub-national administrative level in a country) to control for time-invariant features such as water sharing rules that correlate with both the independent variables and conflicts. Vector also includes key time-variant confounders such as total conflicts excluding water conflicts, population, population density, irrigation, and underground water storage. The error term is denoted by ⁠. The model is estimated to have random effects at the pixel level.

Once Equation (1) is estimated, we combine the results with the future drought described in Section 3.3 to identify future water conflict spots (2030–2040). For each pixel, the estimated increase in drought frequency is multiplied by the estimated marginal effect of drought on water conflicts in the historic period ⁠. The resulting change in drought frequency between the historic and the future period in each pixel is calculated as the difference in the monthly average number of drought events. We measure this change as an average over multiple years because we are interested in the long-term trend of drought frequencies, rather than in the change between specific years, which may be subject to more uncertainty. This calculation is performed only for pixels that experienced at least one mild drought and at least one fatal water conflict over the period 2010 − 2024. We assume that the state border will remain constant over time, and that the marginal influence of drought on water conflicts will follow a similar trajectory in the future as it has in the past.

The variable is derived from the Integrated Water Management Score for country i in year t. The score we use in the regression is the summed score from all the Water Management modules, which range from 1 to 100, where a higher score indicates a higher level of water management, as indicated by the officials. Although this score is self-reported, we consider government officials' self-reported evaluations of a country's water management level to be a plausible proxy of the quality of water governance in that country, as these officials possess insider knowledge of institutional performance, policy implementation, and resource allocation. These assessments reflect expertise and direct involvement, making them reliable indicators of governance capacity and water management efficiency at the national level. The variable measures the frequency of drought events that occur in country i during year t, and in case of no drought event. represents the vector of control variables which are the same as the pixel-level analysis, including country fixed effects and year fixed effects. The error term is denoted by ⁠. We consider two drought measurements, namely severe and mild. In total, we estimate four equations: two conflict outcomes (frequency and fatalities), times two drought measurements (mild and severe drought). All equations have the same specification.

At the pixel level, we used a representative georeferenced survey to examine whether the distance to country border and the distance to the capital reflect the differences in residents’ perceived governance strength. Although we do not have specific measurements of governance strength related to water, we expect the perception reflects overall governance strength that includes that for managing water and other natural resources. The panel data on perceived governance strength are obtained from the World Value Survey over 2017–2022 for Nigeria and Kenya. We selected questions asking people's attitude on different dimensions of governance factors, for example, people should obey their rulers; their satisfaction with the political system performance, if the respondent feel close to his or her village, town, region, and country, and the extent to which the respondent felt unsafe from crime in his or her own home.

We conduct robustness checks using SPI-3 and SPI-9. SPI-3 measures precipitation anomalies over three months, capturing short-term fluctuations in rainfall, particularly for monitoring seasonal variability, agricultural droughts, and rapid changes in soil moisture. This index is sensitive to temporary dry spells but may not fully reflect deeper hydrological deficits. SPI-9 examines precipitation trends over nine months, reflecting prolonged dry or wet periods. This index is relevant for hydrological and groundwater droughts, as it captures cumulative precipitation deficits affecting river flows, aquifers, and reservoir storage. Moreover, in the main specification, we use the distance of a pixel's centroid to the country's border. In the robustness check, we use the pixel's distance to a country's capital.

We conduct two sets of placebo tests. Note that robustness checks and placebo tests serve distinct validation purposes in our context. Robustness checks assess the stability of results by varying model specification on the key variable (distance). In contrast, placebo tests detect spurious effects by applying the methodology to scenarios where no treatment effect should exist, ensuring the credibility of empirical findings. First, for each pixel, we randomly change the timing of drought and re-estimate the regression equations. The placebo test fails if the results using the actual timing of drought are similar to the random timing of drought. Second, for each pixel that experienced drought, we randomly change the location of the pixel to generate an artificial distance to the border and re-estimate Equation (1). If the impact of distance to the country border on affecting water conflicts was an artifact, we anticipate seeing the actual results similar to the pattern based on this placebo test.

Pixel-level regression results are shown in Table 2. Holding all variables constant, neither stand-alone drought nor distance to borders is significantly associated with a higher frequency of water conflicts. However, both factors are significantly linked to higher fatalities, particularly during severe droughts, which nearly double the number of fatalities compared to months without drought. For every 10 km closer to the country border during a severe drought, water conflict fatalities increase by 0.17, representing an approximate 0.7% rise in deaths on average. The signs of control variables are consistent with expectations. Population size positively affects water conflict frequency and fatalities under both drought levels. Higher underground water storage is associated with more fatalities in water conflicts. This may be driven by increasing competition over extraction rights and strategic control, leading to more violent disputes (Gleick & Iceland 2018). Larger irrigated cropland areas likely reflect investments in water infrastructure and cooperative resource management, reducing competition and mitigating conflict severity. Figure S1, Supplementary material, visualizes the estimated increase in water conflict fatalities after drought, as a function of distance to the country border.

Table 2

Pixel level regression results (SPI-6).

Dependent variable .	Total number of water conflicts .		Total fatalities in water conflicts .
	Mild drought (<− 1) .	Severe drought (<− 1.5) .	Mild drought (<− 1) .	Severe drought (<− 1.5) .
(km)	−0.0001 (0.0001)	0.0002 (0.0002)	−0.008** (0.001)	−0.005*** (0.001)
	0.032 (0.137)	0.193 (0.331)	1.661*** (0.104)	4.289*** (0.368)
	−0.0004 (0.003)	0.005 (0.004)	−0.008** (0.005)	−0.017* (0.003)
Total conflicts	0.0003 (0.0002)	0.0002 (0.0002)	0.0001 (0.0001)	0.0001 (0.0002)
Irrigated crop land (hectare)	−0.0001 (0.0001)	−0.0001 (0.0001)	−0.0001** (0.00004)	−0.0001** (0.00005)
Underground water storage	−0.0001 (0.0002)	−0.0001 (0.0002)	0.0001* (0.00002)	0.0001 (0.0001)
	0.043* (0.016)	0.043* (0.017)	0.051* (0.018)	0.052* (0.020)
	−0.003 (0.009)	−0.003 (0.005)	−0.002 (0.011)	−0.002 (0.013)
District fixed effect	Yes	Yes	Yes	Yes
Month fixed effect	Yes	Yes	Yes	Yes
F-statistics	0.23	0.71	131.422	15.965
N. of Obs.	436	436	436	436

Dependent variable .	Total number of water conflicts .		Total fatalities in water conflicts .
	Mild drought (<− 1) .	Severe drought (<− 1.5) .	Mild drought (<− 1) .	Severe drought (<− 1.5) .
(km)	−0.0001 (0.0001)	0.0002 (0.0002)	−0.008** (0.001)	−0.005*** (0.001)
	0.032 (0.137)	0.193 (0.331)	1.661*** (0.104)	4.289*** (0.368)
	−0.0004 (0.003)	0.005 (0.004)	−0.008** (0.005)	−0.017* (0.003)
Total conflicts	0.0003 (0.0002)	0.0002 (0.0002)	0.0001 (0.0001)	0.0001 (0.0002)
Irrigated crop land (hectare)	−0.0001 (0.0001)	−0.0001 (0.0001)	−0.0001** (0.00004)	−0.0001** (0.00005)
Underground water storage	−0.0001 (0.0002)	−0.0001 (0.0002)	0.0001* (0.00002)	0.0001 (0.0001)
	0.043* (0.016)	0.043* (0.017)	0.051* (0.018)	0.052* (0.020)
	−0.003 (0.009)	−0.003 (0.005)	−0.002 (0.011)	−0.002 (0.013)
District fixed effect	Yes	Yes	Yes	Yes
Month fixed effect	Yes	Yes	Yes	Yes
F-statistics	0.23	0.71	131.422	15.965
N. of Obs.	436	436	436	436

Notes: *** p < 0.01, ** p < 0.05, * p < 0.1. Robust standard errors are reported in the parenthesis.

View Large

Our analysis shows that areas close to a country's borders tend to face more fatal water conflicts after drought. This effect cannot be explained by differences in overall conflicts, irrigation, population, and underground water storage, which we explicitly controlled for in the model. Our postulation is that this may be partially due to overall weaker policy enforcement and management.

To examine this postulate, at the country level, we use Integrated Water Resources Management (IWRM), a framework widely adopted across Africa, aiming to quantify and improve the quality of the management of water. Table 3 presents the panel regression results at the country level. When there is no drought, there is no significant relationship between water conflicts and stronger water governances – as proxied by distance to border – indicating that the presence of robust water management structures alone does not necessarily reduce conflict.

However, drought conditions are strongly associated with water conflicts, both in terms of frequency and intensity (fatalities). More importantly, the interaction terms show that the number of water conflicts and related fatalities is significantly lower in countries that report higher water management scores during droughts. Specifically, countries with stronger water management scores experience about 250 fewer conflicts and 112 fewer fatalities when facing drought conditions annually. This suggests that water management by itself may not mitigate water conflict, but can have a significant impact on the face of droughts.

Table S4, Supplementary material, shows the individual scores that are significant in affecting water conflict with respect to drought, defined over SPI-6 being smaller than negative 1.5. The results indicate the interaction effects between water management measures and drought on water conflicts and their associated fatalities. In the first table, the interaction term between management score and drought shows significant negative effects on both the total number of water conflicts and fatalities. This suggests that higher water management scores mitigate the detrimental effects of drought on water conflicts. In the second table, a similar interaction term with the financing score demonstrates a significant reduction in the number of water conflicts, but its effect on fatalities is statistically insignificant. This indicates that while financing efforts during drought periods may reduce conflict occurrence, their effectiveness in mitigating fatalities remains limited. The coefficients for drought alone are positive and significant, highlighting its substantial contribution to both conflicts and fatalities in the absence of strong water management or financing strategies.

Figure S2, Supplementary material, shows that respondents located further away from the country border are significantly more likely to report feeling unsafe, and distant from their country. The analysis reveals notable differences in perceptions of governance strength among individuals, based on their proximity to capitals and country borders. Satisfaction with the political system is higher among individuals living near the capital, indicating stronger government presence and governance efficacy in central areas. The trust in police is also lower (0.6) for respondents. For respondents who are within 50 km of the country border, 9.8% of the interviewed households feel extremely unsafe, whereas 4.8% of those who lived within 50 to 100 km from the country border feel unsafe. The inverse trust of police and army is both significantly lower for respondents who are near the country border, compared to those who are closer to the capital. Individuals near the capital report higher levels of obedience to rulers, suggesting greater trust in or compliance with centralized authority. In contrast, perceptions of safety reveal disparities linked to both capital and border proximity. Those living far from the capital and near borders also feel less safe in their homes, reflecting weaker enforcement or security provisions in these regions. Interestingly, perceptions of closeness to community and national entities differ by location. While individuals near the capital and borders feel a stronger connection to their country and region, those far from the capital or near borders exhibit stronger ties to their immediate localities, such as villages or towns. This suggests that localized social cohesion may act as a substitute for weaker national governance in more remote or border regions.

We conduct a battery of robustness checks. Table 4 compares the main coefficients of interest across specifications with different SPI timescales. Across all models, the negative and significant coefficients on distance to country border (ranging from −0.002 to −0.008) indicate that water conflicts tend to be more severe closer to borders. The coefficients on drought are positive and substantial, and SPI-6 exhibits the strongest and most significant effect. SPI-3 shows a weaker but similar effect, while SPI-9 retains the same directional effect but is no longer significant, suggesting that long-term drought trends may allow for adaptation measures, mitigating immediate water conflict risks. The full results are presented in the Supplementary material.

Table 4

Comparing results with different SPI accumulation periods.

.	SPI-3 .		SPI-6 .		SPI-9 .
	Mild (⁠⁠) .	Severe (⁠⁠) .	Mild (⁠⁠) .	Severe (⁠⁠) .	Mild (⁠⁠) .	Severe (⁠⁠) .
Distance to country border (km)	−0.002** (0.001)	−0.004*** (−0.002)	−0.008** (0.002)	−0.005*** (0.001)	−0.003** (0.001)	−0.005*** (0.001)
	0.399* (0.104)	2.263* (0.008)	1.661*** (0.104)	4.289*** (0.368)	0.866*** (0.104)	3.179*** (0.312)
{Drought = 1} × {Distance to country border (km)}	−0.001* (0.005)	−0.003* (0.001)	−0.008** (0.005)	−0.017* (0.003)	0.0004 (0.0005)	−0.0005 (0.0003)

.	SPI-3 .		SPI-6 .		SPI-9 .
	Mild (⁠⁠) .	Severe (⁠⁠) .	Mild (⁠⁠) .	Severe (⁠⁠) .	Mild (⁠⁠) .	Severe (⁠⁠) .
Distance to country border (km)	−0.002** (0.001)	−0.004*** (−0.002)	−0.008** (0.002)	−0.005*** (0.001)	−0.003** (0.001)	−0.005*** (0.001)
	0.399* (0.104)	2.263* (0.008)	1.661*** (0.104)	4.289*** (0.368)	0.866*** (0.104)	3.179*** (0.312)
{Drought = 1} × {Distance to country border (km)}	−0.001* (0.005)	−0.003* (0.001)	−0.008** (0.005)	−0.017* (0.003)	0.0004 (0.0005)	−0.0005 (0.0003)

Notes: *** p < 0.01, ** p < 0.05, * p < 0.1. Robust standard errors are reported in the parenthesis.

View Large

Next, instead of distance to border, we use the distance to the capital as the ‘inverse’ of remoteness. Drought is again positively associated with more water conflicts, especially under severe drought. For total fatalities in water conflicts, distance to the capital is not significant, but drought presence significantly increases fatalities, particularly under severe drought. Regarding country-level results, Table S2, Supplementary material, shows the results using individual water management scores. Drought always worsens the water conflict condition; at the country level, most of the individual water measurements do not show. This alone does not affect conflict significantly, but the interaction with drought is significant. To test if the observed interaction of remoteness and drought on water conflicts is due to overfitting or randomness, we conducted two sets of falsification tests described in the method section. If the pattern still holds in these artificial scenarios, it suggests that our original findings in Tables 2 and 3 might be spurious or due to some confounding factors. Figure S3 in the Supplementary material presents two falsification tests where (i) the timing of drought is randomly generated for each pixel (ii) the location of climate extremes is randomly generated. The results show a substantially less significant effect compared to the real results (in solid vertical line), validating that our results are less likely due to coincidence or spurious pattern.

In Africa, water conflicts have recently increased in frequency, intensity, and spatial coverage, partially due to worsening climatic conditions (especially droughts) that trigger more water scarcity (Gleick & Iceland 2018). Understanding both the mechanism that runs from drought to water conflict and the spatial pattern of this mechanism is critical for informing policies that aim to mitigate water conflicts.

Using a multi-scale approach, our analysis offers three key findings. First, we find robust evidence of the impact of drought on water conflicts. Between 2010 and 2024, drought is associated with an increase in water conflict fatalities, which is consistent with previous literature (Ide et al. 2021). The impact of drought on conflicts depends highly on where the drought occurs. During both mild and severe droughts, there is an increase in water conflict fatalities for every kilometer closer to the country border. Remote areas, on average, appear to experience more water conflicts. This may stem from geographical isolation and logistical challenges that hinder the central government from maintaining a strong presence, and/or limit the governance capacity of local authorities. The spatial pattern of water conflicts suggests that water policies should be tailored to different regions, considering the local governance capacity, vulnerability to drought and conflicts, and the prevalent type of water conflicts.

Second, at the country level, our results support the above postulate that there is significant interplay between drought and water governance (proxied by distance to borders) in affecting water conflicts. Strong water governance may partially offset the negative impacts of physical water scarcity on water conflicts. Options to strengthen water governance can include enhancing the effectiveness of administrative systems and ensuring comprehensive implementation of IWRM principles. An example is the establishment of inclusive water management processes through the engagement of diverse stakeholders. Given the spatial specificity of water conflicts and their driving factors revealed herein, these mitigating solutions should also be more locally centered, by prioritizing areas that have low water governance capacity.

A third key finding is that in the future, we anticipate an increase in water conflict fatalities in the majority of pixels. However, some pixels that are currently peaceful may experience a substantial increase in water conflicts, suggesting that future drought may exacerbate water conflict risks in vulnerable, drought-prone regions that are currently not the hotspots of water conflicts. Since climate change is expected to alter the spatial distribution of drought, it is important to identify areas that are currently water secure but may face water scarcity in the future and design interventions that can either prevent water-related conflicts or mitigate their impact on local communities, for example, enhancing water storage and safeguarding critical water facilities from being damaged during conflicts.

From a methodological perspective, our approach implies that, to understand water conflicts, constructing a specific dataset, together with evidence across spatial scales, can help offer a more comprehensive picture. First, a key challenge in analyzing the relationship between drought and water conflicts is endogeneity. Instead of relying primarily on econometric identification (Sarsons 2015), we focus exclusively on water conflicts to strengthen the attribution of drought as an exogenous driver. By explicitly narrowing the scope to water conflicts, we also mitigate confounding influences from broader socio-political unrest. Our approach suggests an alternative way to address endogeneity: rather than attempting to model the entire causal chain within a single framework, future research can separately analyze different subcategories of events. Second, a multi-scale approach may address the limitation of any single-scale approach.

Still, there are a few empirical caveats in this paper that need attention. First, our datasets for water conflicts are annotated from existing datasets of ACLED, which may contain possible measurement errors (Eck 2012). For example, conflicts lasting multiple days are recorded as separate incidents, and the location of the conflict may not be accurate (Parker & Vadheim 2017). To mitigate the influence of potential error, all conflict data used in our analysis are aggregated from point to pixel of 30 by 30 km and from daily to monthly. Moreover, our manual selection of the water conflict events relies on the detailed description of the conflict in ACLED dataset. In cases where the event descriptions are brief or vague, it is possible that we may miss out on a water conflict event. However, it is very unlikely that we mistakenly annotate a conflict event that is not related to water as a water conflict. Thus, our count of water conflicts and the associated estimates can be interpreted as a lower bound. Future studies can benefit from a more comprehensive dataset on conflicts.

Second, we do not model explicitly the possible spatial spillover of conflicts, since locations that experience water conflicts are far apart and are thus less likely to experience any substantial spatial autoregressive process. We use Moran's I based on a distance weight matrix to show that both conflict frequency and conflict fatality display very weak spatial autocorrelation (less than 0.001 for both conflict frequency and fatality). However, it is still possible that a specific water conflict event may spread and trigger water conflicts in nearby areas. This may lead to counting one single conflict as multiple events.

The third caveat is related to the quantification of water governance strength. As pointed out by Fukuyama (2013), there is an overall lack of empirical measures of the quality of governance. At the country level, we use self-reported water management scores to measure governance strength. These scores may reflect the officials' perceptions rather than reality: officials might overstate their country's water management status to present a positive image. If we assume no overstatement, more water conflicts might indeed drive the need for more robust policies, leading to a situation where water conflict levels and policy implementation ratings rise together. At the pixel level, however, quantifying the strength of governance remains a challenge. Although we use the distance to borders as a quantitative proxy, future research can rely on additional geospatial data or other spatially explicit information, like satellite imagery on water infrastructure, to objectively measure different aspects of water governance and related outcomes.

Our study provides a starting point for a discussion on how to address these, largely methodological, issues. Therefore, we believe that it can motivate further policy research on unraveling the complex spatial relationship between conflicts, droughts and governance at the local level. Improving our understanding of this topic is key for designing interventions that can promote peace and ultimately enhance peoples' welfare, especially in the face of climate extremes.

This work was carried out under the CGIAR Science Program on Policy Innovations. We would like to thank all funders who supported this research through their contributions to the CGIAR Trust Fund (www.cgiar.org/funders).

All relevant data are available from an online repository or repositories: https://dataverse.harvard.edu/dataset.xhtml?persistentId=doi%3A10.7910%2FDVN%2F7MKDJZ&version=1.0.

The authors declare there is no conflict.