The Bite Count

The mosquito is the deadliest animal to humans — not through its own venom, but as a vector for the pathogens it carries, which kill an estimated 725,000 to 1,000,000 people every year, more than any other creature. Yellow fever is one of its most lethal cargoes: a modelling study estimated 84,000–170,000 severe cases and 29,000–60,000 deaths in Africa in a single year, roughly 90% of the global burden (Garske et al., 2014). What makes yellow fever unusual is that we have a near-perfect vaccine, a well-characterised vector, and reliable satellite proxies for its habitat. That combination makes it possible to do something rare in infectious disease: attach a probability to the risk.

What the satellite measures

Two Sentinel-2 indices proxy mosquito habitat from orbit. NDVI — the Normalized Difference Vegetation Index — measures living green cover: denser vegetation gives adult mosquitoes resting sites and holds the humidity that larvae need. MNDWI — the Modified Normalized Difference Water Index — detects surface water: the rivers, pools and flooded ground where Aedes breeds. We queried real Sentinel-2 imagery through Google Earth Engine for three endemic zones and derived a composite Habitat Suitability Index (HSI), scored 0–10, weighting vegetation density and peak water presence equally. The HSI is our own construct — a way to compress two satellite signals into one comparable number, not an established epidemiological metric.

Three zones, measured from orbit

West Africa — Nigeria/Benin axis (2.5–5.5°E, 6.2–8.5°N): NDVI mean 0.517, MNDWI peak 0.522, HSI 6.4/10. This corridor runs from the Niger Delta wetlands to the Guinea forest edge — a mosaic of vegetation and seasonal standing water carrying Aedes aegypti alongside the sylvatic vectors Ae. africanus and Ae. luteocephalus. It is also where the hard field numbers come from: Nigeria's nationwide surveillance during the 2017–2020 outbreaks detected yellow fever virus in about a quarter of Aedes mosquito pools, with minimum infection rates ranging from 0.9 to 62.5 per 1,000 across species (Parasites & Vectors, 2024) — the empirical anchor for the model below.

Amazon Basin — Manaus area, Brazil (61–58°W, 4–1°S): NDVI mean 0.637, MNDWI peak 1.0, HSI 8.2/10 — the highest of the three. NDVI above 0.6 indicates near-continuous primary canopy; an MNDWI peak of 1.0 marks open water — the floodplain lakes and tributary channels that braid hundreds of kilometres of forest-water edge. This is ideal habitat: the sylvatic yellow-fever cycle here runs through Haemagogus and Sabethes mosquitoes biting by day in the canopy, with Aedes aegypti carrying the urban cycle in settlements. Suitability is highest precisely where forest meets standing water.

Congo Basin — Kinshasa area, DRC (14.5–17.5°E, 5–2°S): NDVI mean 0.504, MNDWI peak 0.827, HSI 7.1/10. Dense equatorial forest threaded by the Congo River and its tributaries. It is also one of the least entomologically surveyed endemic zones — a real intelligence gap for a region that records yellow fever every year. The satellite read matches West Africa on overall suitability, with lower spatial variability: more continuous forest, less fragmentation.

The probability model

Bites and infection follow a compound Bernoulli process. Let f = the fraction of biting Aedes carrying viable yellow fever virus, and p = the probability that one infectious bite transmits. The chance of at least one infection after n independent bites is P(n) = 1 − (1 − f·p)ⁿ. The model is deliberately simple: it treats each bite as an independent trial at constant risk and ignores acquired immunity, so read the numbers as orders of magnitude, not clinical predictions. Its value is that every input is stated and swappable.

Inter-epidemic baseline: take f ≈ 0.002 — the low end of the measured Nigerian range — and p ≈ 0.6, mid-range for Aedes vector competence in the laboratory, giving an effective per-bite risk f·p ≈ 0.0012. The model then yields 88 bites for a 10% cumulative chance of infection, 578 bites for 50%, and 1,917 bites for 90%. In a rural West African endemic area at peak season, where biting can reach tens per night, 578 bites accumulate in roughly two to three weeks of unprotected outdoor exposure.

Under the same model, a confirmed outbreak collapses the count to 23 bites for a 50% chance of infection — a total an unprotected adult can reach in a single evening outdoors at the peak of the season.

Outbreak phase: vector infection prevalence climbs. Nigeria's own surveillance recorded minimum infection rates up to 62.5 per 1,000 (≈6%) in the most-infected species, so an outbreak f ≈ 0.05 is conservative; with p ≈ 0.6 that gives f·p ≈ 0.03. Now the model returns 4 bites for 10%, 23 bites for 50%, and 76 bites for 90%. These are not abstractions: across the Americas in 2025, PAHO confirmed 346 yellow-fever cases and 143 deaths in seven countries — a 41% case-fatality rate — with the virus reaching São Paulo state and other areas outside the historical risk map.

Re-running the scan: Nairobi, and a second disease

A reader posed the sharper question: bitten 20, 30, 40, 50, 60 times over Nairobi, how likely is malaria — or yellow fever? We re-ran the same pipeline over the city (36.65–37.05°E, 1.15–1.45°S) on 16 June 2026: seven low-cloud Sentinel-2 scenes, NDVI mean 0.388, MNDWI mean −0.37 with isolated peaks to 0.51. Moderate canopy, little standing water. On the habitat index alone the city reads borderline-live. It isn't — and the reason is the one variable a vegetation index can't see.

Altitude. Nairobi sits at ~1,795 m; June nights fall toward 10 °C — below the thermal floor for Plasmodium sporogony, the stage where the parasite matures inside the mosquito's salivary glands. The cold simply stops it. No sporozoite-positive Anopheles has been found in the city since the 1920s; it is classified low-risk, effectively non-transmitting. The malaria recorded in Nairobi clinics is overwhelmingly imported from the Lake Victoria basin and the coast — not acquired from a local bite.

The same compound Bernoulli model carries over, now with p = q·s·b: q the share of your bites from a competent vector, s its infection (sporozoite) rate, b the transmission efficiency per infectious bite. Western-Kenya field surveys put sporozoite rates at 2–13% in the endemic lowlands and at zero in Nairobi. That gives a per-bite p of roughly 0.005–0.03 in endemic lowland Kenya, against about 0.000005 — five in a million — inside the city.

Run the bite counts that prompted this. In Nairobi: at 60 bites the malaria probability is still about 0.03%, and yellow fever stays near 0% — the virus does not circulate in the city at all. Move the same bites to endemic lowland Kenya and the curve straightens: 20 bites → 10–46%, 60 bites → 26% (low season), 60% (moderate), 84% (intense holoendemic). For yellow fever during an active West Africa or Amazon outbreak (f·p ≈ 0.03), 60 bites → 84%, matching the outbreak figure above.

The danger is in where you are bitten, not how many times. One infectious bite in Kisumu outweighs sixty in Nairobi.

The disease that earns its name

Yellow fever is named for the jaundice of its severe form. About 85% of symptomatic infections resolve after the acute phase — sudden fever, severe headache, myalgia, and Faget's sign, the paradoxical slowing of the pulse as fever climbs. After a brief remission, roughly 15% enter the toxic phase: jaundice, haemorrhage (the 'vomito negro', black vomit of degraded blood), simultaneous liver and kidney failure, and shock. WHO estimates that about half of those who enter the toxic phase die within 7 to 10 days. There is no antiviral; care is supportive. Reported case fatality rises steeply when outbreaks are caught late — 41% across the Americas in 2025 — almost entirely among the unvaccinated.

The only number that resets the probability to zero

The 17D yellow-fever vaccine reaches ≥99% seroconversion within 30 days of a single dose, and in 2013 WHO confirmed that protection is lifelong — no booster required. Every strand above converges on one line: vaccinate before entering high-HSI territory. The mosquito does not keep your calendar; the virus does not read your budget. The satellite only shows where both of them live. Intelligence, harvested.