Resource Scarcity, Environmental Stress, and the Escalation of Armed Conflict

The case for a strong link was made most forcefully in 2013, when Hsiang, Burke and Miguel assembled the 60 most rigorous quantitative studies then available and found that each one-standard-deviation shift toward hotter temperatures or more extreme rainfall raised the frequency of intergroup conflict by a median of 14% and interpersonal violence by 4%. The Syrian civil war became the emblematic case: Kelley and colleagues showed that the drought preceding the 2011 uprising was the most severe in the instrumental record for the greater Fertile Crescent, made more than twice as likely by human-caused warming, and argued it had a catalytic effect on unrest in a country marked by poor governance and unsustainable agricultural policy.

Each claim drew sustained criticism. Selby and colleagues re-examined the Syrian case and concluded that there is no sound evidence that climate change was a factor in causing the civil war, and others questioned how robust pooled estimates are to the choice of data and model. In 2019 Mach and colleagues convened experts from across the disciplines to take stock. They agreed that climate has affected organised armed conflict within countries, but judged other drivers — low socioeconomic development and low state capabilities — to be substantially more influential, and named the mechanisms linking the two as a key uncertainty. Intensifying climate change, they concluded, is likely to raise future risks.

The most persuasive recent work looks for context rather than a universal effect. Von Uexkull and colleagues, studying ethnic groups across Asia and Africa from 1989 to 2014, found that drought under most conditions has little effect on the short-term risk of rebellion, but that for agriculturally dependent groups and politically excluded groups in very poor countries a local drought increases the likelihood of sustained violence. Harari and La Ferrara, working at the level of African grid cells, found that weather shocks in the growing season of the local crop raise conflict, with effects spilling into neighbouring areas. This dashboard tests the same chain with current data.

Most of the world's poorest farmers depend on rain rather than irrigation, so a dry year translates directly into a smaller harvest. To measure this, the analysis compares each country's cereal yield with its own average over the previous five years, which strips out the steady gains from better seed and fertiliser, and asks how that gap changes in years when national rainfall falls at least one standard deviation below its 1981–2010 average. Comparing each country only with itself, and controlling for global shocks common to every country in a year, a drought year lowers cereal yields by 5.9% (t = −6.1).

Crop failures — years in which yields fall at least 14% below the previous five-year average — become twice as common. They occur in about 6–8% of normal or wet years and 17% of years with rainfall more than 1.5 standard deviations below normal. Within the same country, a drought raises the probability of a crop failure by 7.9 percentage points (t = 6.0). The chart shows the full gradient: harvests rise with rainfall up to about one standard deviation above normal and then level off, while the failure rate falls almost steadily as rainfall increases.

National averages understate what happens on the ground. A country-level rainfall figure blends dry and wet regions, and a national yield figure blends irrigated and rain-fed farms, so the local shock to a herding community or a rain-fed village can be far larger. Irrigation dampens the effect — in the 26 countries where at least a tenth of farmland is irrigated, the estimated yield loss is about a fifth smaller, though the difference is imprecise. Section 8 returns to irrigation, drought-tolerant seed and insurance as ways of keeping the first link from closing.

To map the overlap, each country receives a scarcity exposure index: the average of its percentile ranks on water stress (the share of renewable freshwater withdrawn), the frequency of cereal crop failures since 2000, and the share of jobs in farming, which measures how directly scarcity reaches livelihoods. Fragility is measured by the World Bank's government effectiveness indicator, which rates the quality of public services and the capacity to implement policy. The chart places 159 countries on both axes and sizes each by deaths from organised violence in 2016–24, located by where each event happened.

With the default thresholds — exposure above the median and government effectiveness at or below −0.75 — 24 countries fall in the quadrant where scarcity meets fragility, among them Afghanistan, Burkina Faso, Chad, Haiti, Mali, Somalia, Sudan, Syria and Yemen. Eighteen of them, 75%, had state-based armed conflict in 2016–24; they hold 7% of the world's people and suffered 37% of all deaths from organised violence. Among 73 countries with low exposure and capable governments, 12 (16%) had armed conflict. Fragile states with lower scarcity exposure sit between the two, which is the signature of an accelerant: scarcity raises the risk most where institutions are already weak.

The World Bank's own lists show the same overlap. Its FY2027 conflict list names 24 countries where organised violence is geographically widespread, and its institutional fragility list 25, with 11 on both. The 24 conflict-list countries hold 13% of the people in this sample but 87% of deaths from organised violence in 2016–24 and 94% of deaths in communal conflicts; 40% of their jobs are in farming against 22% elsewhere, and undernourishment averages 22% against 8%. Move the thresholds to see how robust the pattern is, and pick a map layer to see each component.

A naive test compares conflict after dry years with conflict after normal years. In fragile states — government effectiveness of −1 or below in the previous year — armed conflict was active in 42% of years following a drought against 31% following a normal year, and 50% following a severe drought. New wars broke out in 8.9% of peaceful years after a drought against 6.7% after a normal year, and communal violence was recorded in 17% of years after a drought against 12%. The gaps are much smaller in capable states. Taken at face value, these numbers would make drought a powerful trigger of war in fragile states.

But the countries that suffer most droughts are also, for many other reasons, the countries most prone to conflict. When each country is compared only with itself, with common global shocks removed, the picture changes. In fragile states the effect of an ordinary drought on active conflict the next year shrinks to about one percentage point, and on new wars to about two, neither statistically distinguishable from zero; for communal violence the estimate even turns negative. Most of the raw gap, in other words, is about which places are dry, not which years are dry.

Severe droughts — rainfall 1.5 standard deviations or more below normal — are the exception. Across all countries they raise the probability that a new armed conflict starts the following year by 2.5 percentage points (t = 2.2) on a base of 4.0%, an increase of about 60%, and the probability of active conflict by 2.7 points. The estimate in fragile states alone is larger (5.7 points) but imprecise. Scarcity, on this evidence, can tip a fragile peace into war when the shock is extreme; an ordinary dry year mostly does not.

National averages hide the places where drought bites. The local test divides 107 low- and middle-income countries into 6,500 one-degree grid cells, roughly 110 kilometres square, and uses gridded monthly rainfall from CRU TS 4.09 to measure how each cell's main rainy season — its wettest four months on average — compared with normal in each year from 1990 to 2024. It counts violent events in the same cell from the UCDP Georeferenced Event Dataset and compares each cell with itself while removing everything that happened to its country in that year, so that the estimate reflects only the local drought. Seasons without station observations are excluded.

The results do not fit a simple story. In the season the rains fail, state-based fighting in fragile countries falls by about 2 percentage points on a base of 12.5% (t = −2.9): armies and rebels may fight less when resources to seize are scarce, or the timing of campaigns may follow the seasons rather than the harvest. The year after, the balance shifts: non-state violence in fragile-state cells rises by about 0.5 points on a base of 3.6% (about 15%, t = 1.9) and communal violence by about 0.5 points on 1.6% (about 31%, t = 1.4). After severe local droughts, non-state violence rises by about 30%.

These later effects sit at the edge of statistical significance, so they are suggestive rather than settled. But they line up with the expert judgement and the actor-level evidence in section 1: drought matters most for violence between herders and farmers, between ethnic militias and between communities competing for land and water, in places where the state is too weak to mediate — the violence that dominates parts of Nigeria, South Sudan, Sudan, Ethiopia and the Democratic Republic of the Congo. In capable states the effects the following year are close to zero.

If drought's effect on war is modest and conditional, war's effect on food is not. Fighting destroys crops and stores, blocks markets and aid, drives farmers off their land and collapses the currencies that pay for imported food. In 2024 the average share of people undernourished was 16% in countries with active state-based conflict against 8% in countries at peace, a gap that has persisted since 2001. The 2026 Global Report on Food Crises mid-year update projects that about 327,000 people will face catastrophe — the most extreme phase of acute food insecurity — across five countries and territories in 2026, and identifies a risk of famine in Somalia, South Sudan and Sudan, and in the Gaza Strip through April 2026. Every one of them is a conflict zone.

That loop is what makes the combination of scarcity and fragility so dangerous. A drought that would cause hardship in a stable country can, in a country already at war, become a famine, because the systems that would absorb it — markets, roads, social protection, humanitarian access — have been broken by fighting. And hunger can feed the next round of conflict by weakening the state and enlarging the pool of people with little to lose. The update notes that conflict, weather extremes, rising food and input prices, a strengthening El Niño and declining aid could reverse recent gains; Nigeria recorded the largest deterioration, with nearly 5 million more people facing crisis or worse than in 2025.

Water itself is increasingly a target and a weapon. The Pacific Institute's Water Conflict Chronology, updated in November 2025, records more than 2,700 events across 100 countries in which water was a trigger, a weapon or a casualty of violence. Protecting water and food systems in war, and restoring them after it, is therefore part of breaking the chain, not only a humanitarian afterthought.

The first ledger shows the 24 countries on the World Bank's FY2027 conflict list, one row each, with every year from 1990 to 2024 coloured by its national rainfall anomaly and a red dot sized by the deaths from organised violence that year. Read across a row and the pattern of section 4 becomes visible: in most of these countries violence, once begun, continues through wet and dry years alike. Somalia's droughts of 2016 and 2021–2022 came amid continuous war with al-Shabaab, but the violence neither started nor stopped with the rains; in the central Sahel the dry year of 2021 coincided with the spread of jihadist and communal violence, but so did wetter years. Just as often the worst violence comes in ordinary or wet years.

The second ledger lets you choose any country and see three rows together: rainfall, cereal yields against the previous five years, and the timeline of violent deaths. Crop failures often follow droughts, as section 2 shows, but some of the largest yield collapses come during wars rather than droughts, when fighting stops farming. The national data hide important differences inside countries, which is why the local estimates in section 5 matter.

Rainfall data deserve caution. National ERA5 averages come from a reanalysis that blends models and observations, and in places with few stations they can drift: the South Sudan series, for example, shows drought in most years since 2008, a pattern that the station-based CRU data do not show and that is likely an artefact. Cereal yield statistics for countries at war are often estimates rather than measurements. The ledgers are best used to explore patterns and ask questions, not to date the causes of particular wars.

The analysis points to a division of labour. The first link — rain to harvest — is strong and predictable, so it can be weakened with known tools: irrigation and water storage, drought-tolerant seed, soil and water conservation, and index insurance that pays out when rains fail. The third link — harvest to hunger and lost income — can be cut by shock-responsive social protection that scales up automatically when a drought is forecast or declared, by anticipatory cash released before a crisis peaks, and by grain reserves and open trade that keep food prices from spiking.

The fourth link is where most of the danger lies, and it is the hardest to fix. Scarcity turns into violence where land and water rights are insecure, where herders and farmers have no shared rules or mediators, where groups are excluded from power and where the state cannot deliver help. Building inclusive institutions, recognising customary tenure, protecting pastoral corridors and investing in local dispute resolution are slow work, but they are what separate a drought that causes hardship from one that causes killing. Climate adaptation in fragile states therefore needs to be conflict-sensitive — designed so that new wells, dams and irrigation schemes reduce rather than create competition.

The final defences are early warning and protection. Severe droughts, which do raise the risk of new war, can often be anticipated months ahead from seasonal forecasts; combining climate forecasts with conflict early-warning systems allows mediation, peacekeeping and humanitarian resources to be positioned before tensions peak. And because conflict feeds hunger more reliably than drought feeds conflict, protecting food and water systems in war — and restoring them after — is one of the most effective ways to stop the loop from closing. Limiting warming slows the drying trends that make severe droughts more common in places such as the eastern Mediterranean and keeps the first link from tightening further.

The country panel covers 159 countries with at least 1 million people in 2020, from 1989 to 2024 (5,724 country-years). Annual national rainfall comes from ERA5 via the World Bank Climate Change Knowledge Portal; drought is a z-score of −1 or below relative to 1981–2010, severe drought −1.5 or below. Cereal yields are FAO data via World Bank WDI; a crop failure is a yield at least 15 log points (about 14%) below the previous five-year mean. Fragility is WGI government effectiveness of −1 or below in the previous year, available from 1996. Armed conflict (incidence and onset, at least 25 battle deaths a year) comes from the UCDP/PRIO Armed Conflict Dataset 26.1, non-state and communal conflict from the UCDP Non-State Conflict Dataset 26.1 (communal = informally organised identity groups), and deaths by country and year from the UCDP Georeferenced Event Dataset 26.1, located by event; events in Gaza and the West Bank are assigned to Palestine.

National estimates are linear probability models of each outcome in the following year on drought, with country and year fixed effects and standard errors clustered by country; onset models include only countries at peace in the drought year, and incidence and communal models control for the current state. The local panel divides land into 1° cells, assigns each to a country, and for 107 low- and middle-income countries measures rainfall in each cell's climatological wettest four months using CRU TS 4.09 monthly precipitation, keeping only seasons with station observations in every month (207,404 cell-years in 6,496 cells). Models include cell and country-year fixed effects and violence in the cell in the previous year, with standard errors clustered by country.

The scarcity exposure index averages each country's percentile ranks on water stress (SDG 6.4.2), crop-failure frequency 2000–2024 and agricultural employment share. Limitations: fixed-effects estimates capture short-run triggers, not slow structural effects; rainfall data are weakest in conflict zones, biasing estimates toward zero (CRU misses Somalia's 2016–2017 and 2020–2023 droughts; ERA5 likely overstates South Sudan's); yield statistics in war are often estimates; annual and seasonal totals miss timing within seasons and heat. Related reports: the geography of survivability under climate stress and the network of interdependent risks.