Cheatsheets
Mortality accounting, not energy-team rhetoric

Deaths per Terawatt-Hour: the Energy Risk Table, With Receipts.

Public fear ranks energy sources by vivid disasters. The denominator ranks them by statistical deaths per unit of useful electricity. Those are different arguments.

Unit: deaths per TWh of electricity Scope: global reference values, accident + air-pollution mortality

Deaths per TWh, log scale central values from OWID 2021 synthesis; fossil values likely conservative globally

Axis: 0.01 -> 0.1 -> 1 -> 10 -> 100 deaths/TWh
One city-year = 1 TWh for ~150,000 EU residents
Highest central value
32.72

Brown coal deaths/TWh in OWID's 2021 synthesis.

Coal vs nuclear
~820x

24.62 divided by 0.03. The exact ratio shifts; the order-of-magnitude gap does not.

Fukushima answer
0-1 / 2,351

Radiation-attributed deaths versus Fukushima Prefecture's disaster-related deaths as of May 1, 2026.

Bottom cluster
0.019-0.04

Solar, nuclear, wind, and hydro excluding Banqiao overlap at chart precision.

Quick Reference: the Ranking Table fundamentals

A terawatt-hour is 1,000,000,000 kWh. OWID uses the helpful translation that this is about the annual electricity use of 150,000 EU residents, using 2021 EU per-capita electricity consumption of about 6,400 kWh. The table below keeps every bar's value in text, because screenshots travel farther than footnotes.

Rank Source Deaths/TWh Dominant death mechanism One statistical death per... Receipts Gotcha
1 Brown coal 32.72 Combustion air pollution; lignite has lower energy density and high emissions per kWh. 0.031 TWh; ~11 city-days OWID 2021 Not every table separates lignite from hard coal; when merged, use the coal row.
2 Coal 24.62 Fine particulate pollution and other combustion products; mining and transport accidents are smaller than the chronic air-pollution burden. 0.041 TWh; ~15 city-days OWID 2022 update Markandya 2007 OWID says fossil values are likely conservative globally because the underlying rows use older European-control estimates.
3 Oil 18.43 Combustion pollution plus extraction, refinery, and transport accidents. 0.054 TWh; ~20 city-days OWID 2021 Markandya 2007 Oil is a small electricity source in many rich-country grids, but per-unit mortality is still high.
4 Biomass 4.63 Combustion pollution from burning organic material; not zero-carbon or zero-pollution at the stack. 0.216 TWh; ~79 city-days OWID 2021 Biomass can be renewable in a carbon-cycle sense and still harmful in a human-health sense.
5 Gas 2.821 Lower particulate burden than coal, but still combustion pollution plus explosions, extraction, and pipeline accidents. 0.354 TWh; ~129 city-days OWID 2021 Markandya 2007 Gas looks clean only relative to coal; it is not in the wind/solar/nuclear safety cluster.
6 Hydropower, including Banqiao 1.3 Rare dam-failure tail risk; OWID's all-history row is dominated by the 1975 Banqiao failure. 0.769 TWh; ~281 city-days OWID hydro method Sovacool 2016 If you omit Banqiao without saying so, you understate hydro's historical tail risk.
7 Hydropower, excluding Banqiao ~0.04 Routine hydropower accidents, excluding the dominant outlier. 25 TWh; ~25 city-years OWID 2022 update This row is useful for normal-operation intuition, not for erasing dam-collapse risk.
8 Wind 0.035 Supply-chain, construction, maintenance, helicopter, fire, and offshore-work accidents. 28.6 TWh; ~29 city-years OWID 2021 Sovacool 2016 Wind is not literally risk-free; the risk is mostly occupational and construction-related.
9 Nuclear, central accounting 0.03 Severe accidents plus very low routine-operation mortality in commercial power generation. 33.3 TWh; ~33 city-years OWID method UNSCEAR Do not sell this as "nuclear killed nobody." The honest claim is lower deaths per unit than fossil fuels.
10 Solar 0.019 Construction, manufacturing, electrical, and rooftop-installation accidents. 52.6 TWh; ~53 city-years OWID 2021 Sovacool 2016 Rooftop work is the small-print risk; utility-scale and rooftop mixes differ.

How to cite the ranking without overclaiming

Use the central table for the base-rate correction, then add: "The low-carbon sources overlap at the bottom; fossil rows are probably conservative globally; hydro changes sharply depending on whether Banqiao is included."

Why Intuition Inverts working knowledge

People remember named accidents. The death ledger is mostly unnamed chronic exposure. That is the whole inversion.

Air pollution is the invisible bulk

Definition: air-pollution mortality counts premature deaths attributable to exposure, mainly fine particulate matter, using exposure-response epidemiology rather than death certificates that say "coal plant."

Concrete example: WHO's October 2024 fact sheet estimates ambient outdoor air pollution caused 4.2 million premature deaths worldwide in 2019; combined ambient and household air pollution is associated with 6.7 million deaths annually.

Gotcha: these are statistical deaths, not imaginary deaths. The uncertainty is in attribution and dose-response shape, not whether PM2.5 damages hearts and lungs.

Dread risk beats base rates

Definition: dread risk is the human tendency to overweight involuntary, catastrophic, unfamiliar, contamination-like hazards relative to routine distributed hazards.

Concrete example: Fukushima and Chernobyl dominate nuclear fear because they are named events with images. Coal mortality is a background flow spread across millions of homes, hospitals, and death certificates.

When not to use it: dread risk does not prove a hazard is fake. It explains why perceived risk and measured deaths/TWh diverge.

Denominator blindness

Definition: denominator blindness is comparing incidents without dividing by the amount of service produced.

Concrete example: "a dam failure killed tens of thousands" and "hydro usually has low routine deaths" can both be true. Deaths/TWh asks how much electricity was produced for the historical death record.

Gotcha: per-TWh is not the whole policy spreadsheet. It is the mortality column, not cost, speed, land use, dispatchability, or political feasibility.

The arithmetic that changes the conversation

At 24.62 deaths/TWh, coal statistically produces one death for every 0.041 TWh. At 0.03 deaths/TWh, nuclear produces one death for every 33.3 TWh under OWID's central accident accounting. A 1 TWh/year city gets one coal-linked statistical death about every 15 days; the same city on nuclear reaches one statistical death after about 33 years.

The Accidents, Honestly edge cases

This is where the page earns trust. The correct answer is neither "nothing happened" nor "every modeled cancer is a confirmed corpse." It is an accounting split.

Chernobyl, 1986

Purpose of the block: separate confirmed acute effects from modeled long-term cancer projections.

Acute radiation syndrome134 high-dose workers; 28 died in first 3 months Other early deaths2 workers in explosion; common shorthand: 30 early deaths Thyroid cancer>6,000 diagnosed by 2005 in exposed youth; ~15 fatal cases documented Modeled projectionWHO/Forum: up to ~4,000 eventual radiation-related deaths

UNSCEAR says 28 ARS deaths occurred in the first three months and another 19 ARS survivors died later from causes not necessarily linked to radiation. WHO's 2005 Chernobyl Forum release gives the often-cited "up to 4,000" eventual projection for higher-exposed groups.

Gotcha: long-term estimates diverge because low-dose cancer models extrapolate small risk over very large populations. That is not the same evidentiary status as the acute toll.

Fukushima Daiichi, 2011

Purpose of the block: answer the two different death questions separately.

Radiation deaths0 documented population deaths; 1 worker compensation case often counted Detectable public effectsUNSCEAR: no documented resident health effects directly attributable to radiation Disaster-related deaths2,351 in Fukushima Prefecture as of May 1, 2026 Forced evacuation>160,000 residents initially forced to evacuate

The uncomfortable lesson is that response design can be a health hazard. Evacuation saved exposure, but prolonged displacement, interrupted care, and stress killed people too, especially older and medically fragile residents.

Gotcha: saying "Fukushima killed nobody" hides evacuation harm. Saying "Fukushima killed thousands by radiation" is also wrong.

Banqiao Dam, 1975

Definition: Banqiao is the dominant hydropower tail event in OWID's all-history hydro row.

OWID accident count~171,000 deaths attributed to Banqiao Hydro total used~176,000 hydro accident deaths, 1965-2021 Hydro denominator138,175 TWh cumulative hydro electricity, 1965-2021 Result1.3 deaths/TWh with Banqiao; ~0.04 without

Banqiao death counts are reported as ranges because 1975 disaster records mix immediate flood deaths, famine, disease, downstream failures, and politically constrained reporting. The table uses OWID's accident-accounting value rather than pretending the archive is cleaner than it is.

Gotcha: the honest hydro row is a pair. Excluding Banqiao explains normal operation; including it respects historical tail risk.

The fossil ledger

Definition: fossil death accounting is mostly routine exposure, not a single disaster list.

Air pollutionWHO: 4.2M ambient deaths in 2019 Current global contextState of Global Air 2025 covers 2023 health impacts Power-sector caveatNot all air-pollution deaths are from electricity Acute eventsMining, explosions, refinery, oil-train, and pipeline deaths are extra, smaller rows

Gotcha: if your intuition requires a named event, fossil fuels look safer than they are.

Wind and solar's small print

Definition: wind and solar deaths are low but not zero; the residual risk is mostly occupational and construction-related rather than routine public exposure.

Wind0.035 deaths/TWh; maintenance, offshore work, helicopter, fire, and construction accidents dominate Solar0.019 deaths/TWh; rooftop falls, electrical work, manufacturing, and installation dominate MechanismSmall, distributed worker-safety risk instead of chronic air pollution or single-site catastrophe Policy lessonTreat low-carbon construction safety as real safety work, not as a rounding error

Gotcha: "near zero" is not "zero." Counting installation falls honestly is part of why the table is credible.

Chernobyl high-end sensitivity

If you force a conservative sensitivity check by adding 4,000 modeled Chernobyl deaths to OWID's central nuclear accident count, then divide by OWID's 96,876 TWh cumulative nuclear generation through 2021, the added term is about 0.041 deaths/TWh. Nuclear moves from 0.03 to roughly 0.07, still below gas by about 40x and coal by about 350x. The ranking changes inside the bottom cluster, not between fossil and low-carbon sources.

Objections, Steel-Manned advanced

A strong table survives strong objections. These are the test points to check before using the chart as a receipt in an argument.

The nuclear numbers exclude waste.

Best version of the objection: long-lived waste creates deep-time risk that cannot be measured by historical deaths/TWh.

Answer: correct that historical deaths/TWh is record-based, not a prophecy. But stored civilian nuclear waste has caused approximately no public deaths to date in the record used for this table, and waste is engineered as a concentrated, monitored hazard rather than a dispersed inhalation stream. The right caveat is "record-based, not infinite-future proof," not "therefore coal is safer."

LNT means all low-dose radiation numbers are guesses.

Best version of the objection: linear-no-threshold modeling extrapolates below directly observable risk and can turn tiny individual risks into large population counts.

Answer: yes. That is why Chernobyl has confirmed deaths, attributable disease, and modeled projections as separate bins. The high-end sensitivity check above deliberately grants the projection and still leaves the fossil versus nuclear gap enormous.

Per-TWh ignores catastrophe potential.

Best version of the objection: a table of realized deaths can miss fat-tail risks that have not occurred yet or were narrowly avoided.

Answer: true. Per-TWh is a realized-mortality metric. Use it for the record, then add separate columns for worst-case envelope, governance, proliferation, plant age, and design. The same discipline exposes hydro's Banqiao tail instead of hiding it.

Air-pollution deaths are statistical fiction.

Best version of the objection: no doctor writes "coal electricity" as the cause of death on a certificate.

Answer: epidemiology routinely attributes deaths to risk factors: smoking, hypertension, heat, occupational exposure, and PM2.5. To make coal rank near nuclear, the fossil air-pollution burden would need to be wrong by hundreds of times, not by the usual uncertainty band.

The real comparison is system-level.

Best version of the objection: electricity systems need reliability, capacity, transmission, storage, fuel security, cost control, and build speed.

Answer: agreed. This page is the mortality column. It does not settle the grid-design spreadsheet. It prevents one common mistake: ranking energy sources by fear while ignoring deaths per unit of service.

Context Blocks working knowledge

Nuclear deaths averted

Kharecha and Hansen estimated in 2013 that historical nuclear power prevented about 1.84 million air-pollution deaths and 64 GtCO2-eq emissions by displacing fossil generation.

Gotcha: this is a counterfactual model, not a body count. Cite it as an estimate.

No banana-dose theater

This page avoids cute radiation comparisons. Dose analogies can be useful, but they often distract from the accounting question: deaths per unit of electricity.

When to use dose tables: emergency sheltering, exposure-rate decisions, and radiological preparedness, not this ranking table.

Trend line

Energy production generally gets safer as safety engineering, pollution controls, occupational rules, and monitoring improve. The table is not a license to freeze old plants or old practices in place.

Gotcha: a cleaner future grid is not just a fuel swap; it is better engineering plus better siting plus better emergency response.

Common Mistakes anti-patterns

MistakeCiting the table as exact point truth.
CorrectionSay the low-carbon bottom cluster overlaps and that fossil values are likely conservative globally.
MistakeComparing every reactor to Chernobyl's RBMK design.
CorrectionUse Chernobyl as a real accident in the record, not as a generic model of the modern fleet.
MistakeUsing mortality as an energy-policy argument-ender.
CorrectionMortality is one column. Cost, speed, land, reliability, security, and emissions are separate columns.
MistakeQuoting Fukushima's radiation toll without the evacuation toll.
CorrectionKeep the split visible: radiation-attributed deaths are 0-1; disaster-related evacuee deaths are over 2,000.
MistakeForgetting hydro's two-row honesty.
CorrectionShow hydro with Banqiao for historical tail risk and without Banqiao for normal-operation intuition.
MistakeTreating air-pollution deaths as unreal because they are statistical.
CorrectionRisk-factor attribution is how public health measures smoking, heat, PM2.5, and occupational hazards.

Source Map

These links are visible by design. The page is useful only if each controversial number has a receipt.