Public health impacts of an imminent Red Sea oil spill

The Safer, a single-hulled oil tanker moored off Yemen’s Red Sea coast, has been abandoned since 2015 and contains 1.1 million barrels of oil—over four times the amount spilled by the Exxon Valdez. Ongoing conflict and blockade have impeded maintenance, inspections, and any long-term solution to remove the oil, increasing the likelihood of a catastrophic spill. Yemen is acutely vulnerable due to heavy reliance on Red Sea ports (Hudaydah and Salif) through which most humanitarian aid and fuel enter, and due to its fragile health, water, and food systems. Oil contamination could also disrupt desalination plants across the region and devastate Yemen’s fisheries, a critical livelihood and food source. The research question is to quantify the immediate public health impacts—on fuel, clean water, food, fisheries, and air-pollution-related health outcomes—of a potential spill from the Safer, to inform urgency and scope of preventive interventions.

Major oil spills have well-documented environmental, economic, and health consequences, including increased cardiorespiratory hospitalizations and a spectrum of psychiatric, neurological, dermatological, and haematological effects. Prior analyses warned of threats to the Red Sea’s unique ecosystems, including resilient coral reefs, and highlighted likely seasonal currents directing spill trajectories north in winter and south in summer. However, immediate public health impacts specific to the Safer scenario—particularly disruptions to humanitarian aid, fuel, water, and food supplies—had not been comprehensively quantified. This work builds on and extends earlier single-simulation trajectories by conducting thousands of simulations incorporating oil properties to characterize the range of possible spill trajectories and downstream health-relevant disruptions.

The study models immediate public health impacts via integrated spill, oil fate, air pollution, and supply disruption analyses.

• Data sources: ERA5 surface wind fields (hourly, 0.25°), HYCOM currents, sea temperature and salinity (3-hourly, 1/12°), NOAA Oil Library properties for Marib Light crude, World Food Programme fuel prices, UNVIM fuel import data, multiple sources for desalination plant locations and capacities, Yemen port food aid flows (2019), and gridded fish yield (2016) for the Red Sea and Gulf of Aden. Simulated datasets are available at DOI:10.7910/DVN/XPESIR.
• Spill trajectory modelling: Used NOAA’s GNOME via pyGNOME to simulate transport and weathering processes (advection, diffusion, entrainment, emulsification, evaporation, spreading, shoreline interaction, dissolution). Conducted 1,000 Monte Carlo simulations for summer (Jun–Aug) and winter (Dec–Feb), each with 3-week timelines and 1,000 particles per run representing surface oil concentration on ~100 m grid (lat–lon rounded to 0.001°). Also included 1,000 “uncertainty” particles per simulation representing extreme weather assumptions; convex hull of all uncertainty particles across simulations delineated the approximate 90% trajectory region. Outputs reported as relative surface concentration percentiles to account for unknown spill volume; bilinear interpolation used between grid points. Sensitivity analyses varied spill duration (24 h vs. 7 d release), seasons (spring, autumn), grid resolution (≈1.1 km), and particle counts (10,000).
• Oil fate modelling: Employed NOAA’s ADIOS to estimate weathering over six days under two scenarios: evaporation only and an optimistic clean-up scenario (immediate response; skimming 14 barrels/hour at 100% efficiency; in situ burning 70,000 m² at 50% efficiency; 15% sprayed with dispersants at 20% efficiency; 29°C; 42% salinity). Conducted 1,000 Monte Carlo runs varying time/date; summarized mean and 95% uncertainty intervals.
• Air pollution modelling: Used NOAA HYSPLIT to simulate atmospheric transport/dispersion of pollutants emitted at sea level for 144 h, with daily runs every 3 h for Jan–Mar (winter) and Jun–Aug (summer). Scenarios: fast release (24 h) and slow release (72 h), with and without combustion. Each simulation emitted 2,500 particles; average concentrations computed on ~100 m grid with bilinear interpolation. Assumed a full spill of 150,000 metric tons. Converted emitted oil to PM2.5 using oil-to-PM conversion factors (Middlebrook et al.) for evaporation and, where relevant, combustion; divided by emission duration and multiplied by HYSPLIT concentration fields. Estimated population-weighted average increased risk (IR) of cardiovascular and respiratory hospitalizations using concentration–response from Burnett et al., averaging risk over exposed person-days (population exposed to ≥10 µg/m³ PM2.5 times exposure duration). Propagated uncertainty via 1,000 Monte Carlo draws for PM conversion rate and IR parameters assuming normal distributions derived from published point estimates and CIs; truncated negative lower bounds to zero. Additional analyses used alternative health functions (Wei et al. for respiratory hospitalizations; Kloog et al. for short-term mortality).
• Supply disruptions: Fuel disruption estimated from mean (and 95% intervals) of monthly Red Sea fuel imports Jan–May 2020 (n=5). Fuel price shock inferred from Hudaydah median fuel prices (diesel/petrol/gas) comparing 15 Oct 2017 vs 15 Nov 2017 during full port closures. Desalination disruption estimated by intersecting simulated spill extents with plant locations/capacities; converted to population-equivalent daily water use by country-specific per capita use. Food aid disruption estimated by applying 2019 port shares (Hudaydah, Aden) to average monthly targeted beneficiaries (WFP situation reports, Mar 2020–Feb 2021, n=10) with uncertainty via quantile interpolation. Fisheries impact calculated as fraction of Yemen’s gridded fish yield within cells with at least the 10th percentile of simulated surface oil concentration (sensitivity: no threshold; 20th percentile).

• Spill trajectories: Most simulations move towards Yemen’s northwest coast; seasonal variation shows northward movement in winter and southeastward along Yemen’s coast in summer, with wide uncertainty allowing either direction across seasons. Estimated 6–10 days for oil to reach Yemen’s western coastline.
• Oil fate and cleanup: About 51% (95% UI 46–54%) of oil evaporates within 24 h; heavier fractions persist on water. Modeled clean-up (skimming, in situ burns, dispersants) removes negligible oil in first six days. Under optimistic clean-up conditions, by day 6, 39.7% of oil remains floating versus 38.2% with evaporation alone.
• Port and infrastructure exposure: Within two weeks, Hudaydah and Salif ports likely experience high surface oil concentrations (≈90th percentile relative to exposed area). By three weeks, spill may reach Aden and desalination plants and ports in Eritrea and Saudi Arabia; unmitigated spread could impede Gulf of Aden passage.
• Fuel disruption: Each month of Red Sea port closure disrupts delivery of 200,000 (180,000–250,000) metric tons of fuel, ≈38% of Yemen’s national requirement. Historical precedent: during Nov 2017 full port closures, Hudaydah fuel prices rose 72% in one month.
• Clean water: Potential desalination disruption totals 77,000 m³/day (summer) and 362,000 m³/day (winter), equivalent to daily water use of approximately 1.0–1.9 million people regionally. Additional severe water access losses in Yemen expected due to fuel shortages powering pumps/trucking, echoing prior crises.
• Food aid: If Red Sea ports close within two weeks, food aid disrupted for an estimated 5.7 (3.7–8.1) million people; if Aden also closes, 8.4 (5.4–11.9) million people.
• Fisheries: Yemen’s Red Sea fisheries threatened at 66.5–85.2% within one week and 93–100% within three weeks (season-dependent); in summer, up to 2.6% of Gulf of Aden fisheries threatened within three weeks.
• Air pollution and health: Population-weighted average increased risk of cardiovascular and respiratory hospitalizations ranges from 5.8% (0–7.5%) across 11.3 (0–27.2) million person-days for a slow-release winter leak to 31.2% (6.5–50.6%) across 19.5 (0.4–24.2) million person-days for a fast-release summer leak. With combustion, IR ranges from 6.7% (5.2–7.9%) across 22.3 (1.2–41.8) million person-days (slow-release winter) to 42.0% (21.9–61.4%) across 22.7 (17.0–26.0) million person-days (fast-release summer). PM2.5 can reach ≈1,600 µg/m³ near the oil, corresponding to ≈530% (460–590%) increased hospitalization risk for directly exposed individuals (e.g., clean-up workers). Seasonal winds drive pollution east into Yemen in summer and westward offshore in winter; some winter simulations show no Yemen exposure.

Findings indicate that a spill from the Safer would have catastrophic immediate public health consequences, especially for Yemen. Disruption of fuel imports would jeopardize hospital operations, water pumping, sewage treatment, solid waste collection, and electrical grids, exacerbating risks of water-borne diseases. Contamination and precautionary shutdowns of desalination plants would intensify the regional water crisis. Interruptions to food aid and closure of fisheries would worsen food insecurity and famine risk. Medical supply imports could also be impeded, straining an already under-resourced health-care system. Clean-up is likely slow and logistically challenging given conflict, sea mines, and port closures; premature reopening risks remobilizing oil. Air pollution health effects, while moderate at the population level relative to supply disruptions, could severely affect clean-up workers and further burden health services. Beyond immediate health impacts, long-term ecological damage (including to resilient Red Sea coral reefs) and global trade disruptions through the Bab el-Mandeb Strait are plausible. Monte Carlo simulations confirm seasonal tendencies but underscore substantial trajectory uncertainty in all seasons, reinforcing that catastrophic impacts are likely regardless of spill timing and that all regional actors would bear consequences. Overall, results directly answer the research question by quantifying disruptions to fuel, water, food, fisheries, and air-pollution-related health risks, emphasizing the urgency of preventive offloading of the oil.

An imminent spill from the Safer poses an extreme, preventable public health risk, particularly to Yemen, by jeopardizing access to fuel, clean water, food aid, and fisheries, and by increasing air-pollution-related hospitalizations. While trajectory and impact uncertainties exist, simulations consistently show severe disruptions across seasons, and clean-up would be slow under optimistic assumptions. The disaster remains avertable through urgently implementing a long-term solution to offload and secure the oil. Future work should refine estimates with real-time environmental data, assess long-term ecological and economic impacts, evaluate mitigation and response strategies (including protective measures for clean-up workers), and improve modeling of port operations and desalination plant resilience under spill conditions.

Key limitations include: (1) Spill simulations rely on historical environmental data; actual spills may occur under extreme conditions not fully captured. (2) Variable data quality across inputs (e.g., desalination plant inventories, population grids) may introduce measurement error. (3) Analyses assume a full spill and do not model partial leaks; durations of port/desalination closures are not predicted due to uncertain clean-up timelines and responses. (4) Clean-up modeling is optimistic and based on Deepwater Horizon recovery rates; containment booms and wave turbulence effects are simplified; tides are not explicitly modeled (assumed implicit in currents). (5) Uncertainty regions may over/underestimate affected areas due to wind/current patterns and lack of full wave turbulence modeling; ports are considered disrupted only when directly oiled, a conservative assumption. (6) Air pollution estimates depend on oil-to-PM conversion factors and concentration–response functions from other populations; volatile organic compounds and non-cardiorespiratory health outcomes are not modeled; emission source is assumed at a fixed site rather than distributed along the slick, likely biasing exposure inland downward. (7) Stored supplies could buffer disruptions, but given ongoing shortages, buffers are likely minimal; price spikes could still render supplies inaccessible.