When the Rains Don’t Come: Drought Risk in the Horn of Africa

1Background: two seasons, one margin of error

The eastern Horn of Africa depends on two rainy seasons, the longer March to May (MAM) rains and the shorter October to December (OND) rains. Both are critical for replenishing groundwater and pasture and for supporting crop and livestock production.^[6] The region is dominated by arid and semi-arid lands, characterized by low and irregular rainfall and periodic droughts.^[7]

That leaves little margin. Households engaged in rain-fed agriculture, agropastoralism and pastoralism are among the most severely affected when rains fail. They are forced to spend limited assets buffering losses and are consequently pushed further into poverty.^[2]

This brief asks three questions. What has changed in the region’s climate? How do those changes become human harm? And what does the evidence say about the value of acting before a crisis is declared?

2The rainfall record is not a simple decline

The IPCC reports that over equatorial east Africa the short rains have shown a long-term wetting trend from the 1960s until the present. In contrast, the long rains experienced a long-term drying trend between 1986 and 2007, with rainfall declines in each month of the season and a shortening of the wet season.^[1] Since around 2000 the long rains have shown a significant relationship with the El Niño-Southern Oscillation (ENSO), as multiple droughts have occurred during recent La Niña events.^[1]

The same chapter reports that since 2005 drought frequency has doubled, from once every 6 years to once every 3, and has become more severe in the long and summer rainfall seasons than in the short rainfall season.^[1] A review of east African drought studies is consistent with that seasonal split, reporting a decrease in drought frequency in the boreal autumn, but cautions that the underlying studies rest on sparse, short-term regional climate data and are subject to uncertainties.^[8]

World Weather Attribution reached a similar picture for the 2020–2023 event. Using two observation-based rainfall products, it found a trend toward less rainfall in the long rains but not in the short rains, which show the opposite, a wettening. There is no trend when the two seasons are combined over the 24 months. The drought years also saw consecutive La Niña conditions, and below-average short rains correlate strongly with La Niña, whereas the long rains do not.^[2]

The pattern is therefore not simply more failed seasons. It is a region whose two seasons are moving in different directions and whose worst sequences coincide with La Niña-like Pacific conditions.

3Heat turns a dry spell into a drought

Rainfall is only half of the water balance. World Weather Attribution combined its rainfall assessment with potential evapotranspiration, the amount of water that evaporates from soil and plants because of higher temperatures, over the 24 months of the drought. It found that the combination of low rainfall and high evapotranspiration as unusual as the recent conditions would not have led to drought at all in a 1.2°C cooler world. In today’s climate the same event is classified as an exceptional drought (D4), with major crop and pasture losses and widespread water shortages. The authors attribute the change in severity primarily to the strong increase in evaporative demand caused by higher temperatures, and give a conservative estimate that such droughts have become about 100 times more likely.^[2]

Pacific variability still sets the timing. A multi-agency alert issued in late 2021 reported that between 1950 and 1998 only 28 percent of OND La Niñas were followed by below-normal MAM rains in the eastern Horn, whereas since 1999 they were followed 78 percent of the time (Figure 1). The alert states that climate change has increased the frequency of poor MAM rains during La Niña-like seasons, and adds that these links have been used to anticipate dry seasons in 2016/17 and 2020/21.^[5]

Two features matter for planning. The sequences are partly predictable, which is what makes early action possible. And their severity is amplified by heat, which makes the consequences of a given rainfall deficit larger than they were in the past.

4The 2020 to 2023 drought and its toll

In the southern Horn, covering parts of southern Ethiopia, southern Somalia and eastern Kenya, the short rains were below average in 2020, 2021 and 2022, and the long rains in 2021 and 2022. World Weather Attribution reports that the drought turned underlying conditions into acute food insecurity for over 4 million inhabitants of that area.^[2] By November 2021, a multi-agency alert reported that Kenya and Somalia had declared drought emergencies and that rainfall totals and vegetation imagery suggested lowest-on-record values in many areas.^[5]

The mortality estimates are among the most striking evidence of how drought becomes harm. A report by Somalia’s Federal Ministry of Health, WHO and UNICEF, carried out by the London School of Hygiene and Tropical Medicine and Imperial College London, estimated that 43,000 excess deaths might have occurred in Somalia in 2022 relative to the 2017–2018 drought, with half of them among children under 5. The highest death rates were estimated in south-central Somalia, especially around Bay, Bakool and Banadir. The estimate rests on a statistical model in which the crude death rate rose from 0.33 to 0.38 deaths per 10,000 person-days over 2022. The release also noted that, although famine had been averted for the time being, the crisis was already more severe than the 2017–2018 drought crisis.^[3]

Weather alone does not explain the outcome. World Weather Attribution found that state fragility and conflict, as well as the length of the drought, played a significant role in worsening outcomes, especially for people in Somalia.^[2] Displacement follows the same logic. During the 2016–2017 Somalia drought, 926,000 newly displaced people were reported, with around 40% of drought-related displacements accommodated in Mogadishu, Baidoa and Kismayo, adding to overcrowding in urban areas.^[1]

5Recurrence in 2025 and 2026, and the turn toward floods

FEWS NET reported in December 2025 that the eastern Horn was experiencing one of the driest October to December rainy seasons on record, driven by La Niña conditions and a historic strong negative Indian Ocean Dipole, less than three years after the drought that struck from late 2020 to early 2023. It estimated that 20–25 million people across Somalia, Ethiopia and Kenya were in need of humanitarian food assistance, with drought the main driver for 50–55 percent of them. In Somalia, rangeland conditions were already worse than in the poor 2016/17 and 2021/22 OND seasons, and non-irrigated harvests in Bay and Bakool were projected to reach less than 10 percent of average.^[6]

The hazard then changed. A joint Government of Somalia and United Nations release of 23 September 2026 describes the below-average 2026 Gu rains, and warns that heavy flooding associated with the anticipated El Niño is expected to worsen conditions between October and December. The latest Integrated Food Security Phase Classification (IPC) analysis projects that 5.4 million people will face crisis levels of hunger or worse in that period, including 1.5 million at emergency levels or worse. It estimates that 1.92 million children under five will need treatment for acute malnutrition between July 2026 and June 2027, including 433,000 with severe acute malnutrition, and it identifies a risk of famine among internally displaced people in Baidoa, based on data collected in April 2026, with a follow-up assessment planned for October 2026.^[4]

Somalia’s disaster management agency estimates that 2.5 to 3.5 million people could be exposed to flooding, with 1 million likely to be displaced from areas where they farm and graze livestock.^[4] The same households that lost herds and harvests to dry seasons are now exposed to the opposite hazard. The release asks for food, cash and nutrition supplies to be pre-positioned in flood-prone rural areas before flooding restricts road access in October.^[4]

6The economics of acting early

Humanitarian aid has historically arrived late, well into the peak of a crisis, and the case for acting earlier is partly financial. A study by Courtenay Cabot Venton for the USAID Center for Resilience modeled costs and avoided losses over 15 years (2000–2015) for a population of 15 million people in Kenya, Ethiopia and Somalia, using actual production and price data that mostly reflect repeated drought cycles. The counterfactual is a late humanitarian response (Table 1).^[7]

When intervention costs are offset against avoided humanitarian aid and avoided income and livestock losses, the chapter reports net benefits of US$3.6 for every US$1 spent on resilience programming. A separate analysis for Ethiopia, covering 8.7 million people, estimated that an early response would save about US$965 million in response costs alone, and found average benefits of US$3.3 for every US$1 invested in Ethiopia’s Somali region and US$2.4 in Tigray.^[7,9]

These are modeled results, and the author is careful about what they do and do not show. The resilience package is unspecified: the analysis assumes households receive interventions that raise income by a set amount and does not measure what it costs to graduate a household to a resilient status.^[7] The IPCC offers independent, qualitative support for the direction of the finding, assessing with high confidence that early warning systems based on targeted climate services can be effective for disaster risk reduction, social protection programs and managing risks to health and food systems.^[1]

The cost of delay is also institutional. World Weather Attribution notes that the length of the 2020–2023 drought raises serious questions about how long a drought the government drought management systems and the international aid infrastructure should be prepared to handle.^[2]

7Implications for disaster risk management

Analysis. This section is Disaster Science’s interpretation of the evidence above. It is not drawn from the cited sources.

Plan for variability, not a single trend. A region whose short rains are wetting while its long rains dry, and which can move from record dry seasons to flood warnings within a year, needs risk plans that treat drought and flood as parts of one system. Triggers, budgets and stockpiles should be designed for both.

Use the forecastable part. Because La Niña-linked sequences are partly predictable, forecasts should be wired to pre-agreed actions and funding, so that money moves on a trigger rather than after a declaration.

Make heat part of the drought definition. Rainfall indices alone understate the severity of a hot season. Drought monitoring that includes evaporative demand, vegetation and water-point status gives earlier and more accurate signals of agricultural drought.

Track mortality, not only food insecurity. The 2022 estimates came from a retrospective study. Routine mortality and nutrition surveillance, which the 2026 release calls for in the form of seasonal assessments and monitoring, would make the human cost visible while there is still time to act.^[4]

Build the layers together. The modeled returns in Table 1 come from combining early response, predictable safety net transfers and income-raising measures. Each layer is more valuable when the others are in place before the shock.

8Uncertainty and limitations

Trends. Rainfall trends in the region depend on the season, the period and the dataset. The review cited above stresses sparse, short-term data,^[8] and the “doubling” of drought frequency is an IPCC-reported figure drawn from that literature, not a result of this brief.^[1]

Attribution. The “100 times more likely” figure is a conservative estimate for a specific event, defined by rainfall and evaporation over 24 months in the southern Horn.^[2] This brief cites the 2023 report. A peer-reviewed version later appeared in Weather and Climate Extremes (2025) and has since been the subject of a corrigendum (2026); readers who need the final numbers should consult the journal version and the corrigendum.

The Pacific statistic. The 28% and 78% shares come from a single multi-agency alert, which does not report the underlying number of events in the text we reviewed.^[5]

Mortality and models. The Somalia mortality figures are modeled estimates (“might have occurred”), not counts.^[3] The economic returns are model outputs that depend on assumptions about response timing and household behavior.^[7]

Projections. FEWS NET and IPC figures are projections. A “risk of famine” is not a famine classification, and the 2026 release states that a follow-up assessment is planned to confirm the level of risk.^[6,4]

Scope. This brief synthesizes published assessments, agency alerts and peer-reviewed studies. It presents no original field data and is an editorial synthesis that has not undergone peer review.