Frequency
1 in 1,710
turbines / year
Upper research estimate. Public logs capture closer to one fire in ten of those events.
Environmental assessment · interactive figures
Wind, fire, and the environmental ledger
Wind turbines do not generate electricity by burning fuel. They do, occasionally, catch fire — usually in the nacelle, usually beyond the reach of hoses. This study pulls the published frequencies, soil-burn measurements, wildfire case reports, and the other side of the ledger (avoided air pollution and carbon) into one set of interactive figures.
Loss given fire
90%
total or near-total
Once the nacelle is alight, height and oil load usually finish the machine.
Air-quality dividend
231
deaths avoided, US 2014
One year of RPS-linked wind, 95% CI 146–318 — orders of magnitude above fire casualties.
Carbon payback
< 2 yr
versus gas turbines
A typical onshore farm offsets its 30-year manufacturing footprint in 1.5–1.7 years.
01 · Incidence
How often do turbines burn?
Public logs and insurance notes undercount. Accident researchers who reconstructed 30 years of the global fleet put the true rate about ten times higher than headlines — still rare per machine, material across hundreds of thousands of towers.
Figure 1
Reported series follows CWIF / media compilations (≈12/year through 2012, rising toward the mid-20s). Estimated series scales the Imperial College–Edinburgh–SP finding that ~91% of fires never appear in public logs, anchored at 117 events around 2011–12. Later years are an indicative continuation, not a new census.
Per-turbine rate
1 / 1,710–2,000
per year, research vs conservative
Share of accidents
2nd
after blade failure (~19% vs fire ~15%)
Unreported share
~91%
if the 117 vs 12 reconstruction holds
02 · Mechanism
Lightning first, then electrics
Reviews rank ignition sources rather than publishing a complete global census. The shares below are a synthesis of that ranking (lightning, electrical, mechanical, maintenance) totaling 100% so the figure is readable — treat them as relative weights, not a single registry extract.
Figure 2
Oil-soaked acoustic foam, gearbox and transformer oil, and a glass-reinforced nacelle shell turn a small electrical event into a total loss.
Figure 3
Fire is not the most common accident. It is the most expensive common one: 90% of nacelle fires destroy the machine.
03 · The nacelle
What is actually burning
A nacelle is a sealed room the size of a small bus, packed with a gearbox, hydraulics, a transformer, and foam lining that becomes kindling once it is soaked with oil. Firefighters on the ground hold the perimeter. They almost never put water on the fire.
Figure 4 · interactive
1,300 L
Range 200–2,000 litres · reference case 1,300 L
Burning area
150 m²
worst-case pool, no soak-in
Vs football pitch
2.1%
association pitch ≈ 7,140 m²
Scaled from the Earthdoc trial: 1,300 L of turbine oil produced a ~150 m² burning pool on sandy soil (absorption ignored — worst-case). Drag the slider to resize the inventory.
04 · Ground
Heat and oil go into the soil
The 2022 field burn found a 22-minute free-burning oil fire. Surface soil hit 130 °C immediately after, then 70 °C six minutes later. Oil coked the top 5 cm and showed up heterogeneously to 30 cm. Around turbines that sit in cropland, that is a topsoil-replacement problem, not a scorch mark.
Figure 5
Flame temperature is the midpoint of the 680–710 °C range. The remaining points are measured soil temperatures from the sandy-loam experiment.
05 · Escape
The fire that leaves the tower
A nacelle fire is usually a single-machine loss plus spot grass fires. The large 2026 Greek burn was not a nacelle event — investigators tied it to defective overhead collector lines. That distinction matters: electrical infrastructure, not the rotor, is the wildfire pathway with real acreage.
Figure 6
Log scale is the honest default: Attica–Boeotia (collector line) is three orders of magnitude above typical nacelle debris fires. The 80,000-hectare figure sometimes attached to Lake Bonney (2006) is not used here; contemporaneous reporting describes a local grass fire from falling debris.
Lake Bonney, South Australia · Jan 2006
NacelleFalling debris ignited grass beneath a Vestas turbine. Extent of the grass fire was local; later claims of tens of thousands of hectares are not corroborated.
Pubnico Point, Nova Scotia · 2019
NacelleTwo blades and nacelle damaged; burned out in about two hours.
Clements Gap, South Australia · 2023–24
Nacelle + grassDestroyed a turbine and burned about 30 hectares of farmland.
Mavro Spithari, Greece · 27 Jun 2026
Collector lineShort circuit on the farm’s private overhead line; ~12 acres.
Pubnico Point, Nova Scotia · 21 Jul 2026
NacelleFull loss of hub, nacelle and blades on a 78 m V80. Spot fires in grass only.
Attica–Boeotia, Greece · 31 Jul 2026
Collector line~35,000 acres, ~200 homes. Investigators tied ignition to defective overhead lines serving a new wind farm.
Greater Changhua 4, Taiwan · 8 Aug 2026
Offshore nacelleDetected, tripped, and extinguished by the operator. No injuries, farm otherwise online.
06 · The other column
What the fleet avoids while a few machines burn
A nacelle fire is photogenic. The air-quality paper is not. One year of US wind tied to renewable portfolio standards was estimated to avoid 231 premature deaths and 32 million tonnes of CO₂ — a public-health dividend that swamps the fire record.
Figure 7
Air quality alone US$2.0 billion. Total with carbon and fuel savings ~US$5.0 billion. Targeting the dirtiest thermal plants first could raise health benefits to US$8.4 billion — without closing exposure gaps by itself.
Figure 8
Each bar is the percentage of US power-sector emissions avoided that year by RPS-linked wind (right axis conceptually 1.3–1.6%). Absolute tonnes are in the tooltips.
Figure 9
10.8 gCO₂eq/kWh in a New Zealand onshore case. Manufacturing dominates. Recycling blades could trim that toward 9.7 g.
Twenty-nine percent of those US health benefits accrued to racial/ethnic minority populations and 32% to low-income groups — below a later 40% disadvantaged-community target. Wind improved exposure gaps in some states and worsened them in others.
07 · Siting
Where you put the farm changes the carbon math
Life-cycle studies that skip land-use change understate forest sites. In one Chinese comparison, land-use change was 37.9% of a forest farm’s life-cycle emissions versus 1.2% on desert. Inner Mongolia’s clustered build-out also lifted nighttime land-surface temperature by 0.23 °C and cut net primary productivity.
Figure 10
1.2 km² in 1990 to 10,755 km² in 2020. Nighttime warming 0.23 °C; NPP −12.37 gC/m², largest over forest.
Figure 11
Forest siting is a different climate project than desert siting. Filter the ecosystem to isolate a class.
Figure 12
China-wide modelling, summer 2015–2018: PM₂.₅ rose 4.39 µg/m³ in the northeast and fell 3.27 µg/m³ in the southeast as mesoscale winds shifted.
08 · Collision
The quiet, continuous kill
Fires are rare pulses. Blade strike and barotrauma are the chronic ecological cost. US-scale reviews put annual fatalities near 679,000 birds and 888,000 bats, with enormous site-to-site variance.
Figure 13
Per-turbine literature ranges run 0–125 birds and 0–287 bats per year. Collision risk tracks season, weather, turbine layout, and species — not a single national rate.
09 · Scale it
A fleet risk model
Set the global (or regional) turbine count, pick a fire rate, and pick a unit cost. This is an expected-value sketch — it does not forecast a specific farm, and it does not price wildfire escaped from collector lines.
Turbine count
Machines in the modelled fleet370,000
Expected fires / year
216.4
5.85 per 10,000 machines
Expected total losses
194.7
90% of fires destroy the nacelle
Expected annual cost
$974M
at US$4.5M each, downtime included
Even at the high rate, a 370,000-machine fleet implies on the order of a couple of hundred fires a year worldwide — against a technology whose US air-quality paper alone counted 231 avoided deaths in a single year.
10 · Methods
What this is, and is not
Figures are compiled from the papers and incident reports listed below. Where a study published a range, the midpoint or both bounds are shown. Cause shares and the post-2012 ‘estimated’ fire series are syntheses, labelled as such in the captions. This is not a new meta-analysis and does not replace site-level environmental impact assessment.
01 · Imperial / Edinburgh / SP Technical (2014–2023)
30-year accident review (~200,000 turbines). Media logged ~12 fires/year; researchers estimated ~117/year (~91% unreported). Fire is the second-largest accident class after blade failure; ~90% of nacelle fires are a total or near-total loss.
02 · Caithness Windfarm Information Forum (2012)
200 reported fire incidents 1995–2012 (avg 11.7/year). Fire ~12.5–15% of reported accidents; blade failure ~19%.
03 · GCube Underwriting (2015–2025)
Average insured cost of a turbine fire ~US$4.5 million (replacement + downtime). Later industry notes cite US$7–8 million.
04 · Earthdoc soil-fire experiment (2022)
Field burn of ~1,300 L turbine oil on sandy soil. Combustion area ~150 m². Surface 680–710 °C; 81 °C at 20 cm; oil coking to 5 cm and impregnation to ~30 cm.
05 · US wind & air quality (2011–2017) (2024)
Wind associated with 2014 RPS: US$2.0 billion air-quality health benefits, 231 premature deaths avoided (95% CI 146–318), 32 Mt CO₂, 51 kt SO₂, 25 kt NOₓ avoided. Total monetized benefits ~US$5.0 billion.
06 · Onshore LCA payback (NZ case) (2024)
10.8 gCO₂eq/kWh; greenhouse-gas payback 1.5–1.7 years versus combined-cycle gas; energy payback 0.4–0.5 years. Manufacturing dominates the carbon footprint.
07 · Inner Mongolia remote-sensing study (2024)
Wind-farm area 1.2 km² (1990) to 10,755 km² (2020). Nighttime land-surface temperature +0.23 °C; net primary productivity down 12.37 gC/m², strongest over forest.
08 · China plant biomass (2,404 farms) (2023)
NDVI and productivity decline within 1–10 km of farms. Average forest carbon-sink loss 12,034 t per farm (~US$1.81 million).
09 · China land-use LCA (2025)
Land-use change is 37.9% of life-cycle emissions on forest sites, 4.3% on grassland, 1.2% on desert. Biomass carbon loss 243.88 t C/turbine (forest), 9.95 (grass), ~5 (desert).
10 · Onshore wind environmental review (2024)
US annual collision estimates: ~679,000 birds and ~888,000 bats. Fatality ranges 0–125 birds and 0–287 bats per turbine per year.
11 · China mesoscale air-quality modelling (2024)
Farms do not add emissions but redistribute pollutants via circulation changes: PM₂.₅ +4.39 µg/m³ in the northeast, −3.27 µg/m³ in the southeast (2015–2018).
12 · Attica–Boeotia wildfire investigation (2026)
July 2026 fire of ~35,000 acres / ~200 homes attributed to defective overhead collector lines from a newly commissioned wind farm. A smaller short-circuit fire (~12 acres) had occurred on the same line on 27 June.
13 · Pubnico Point, Nova Scotia (2026)
21 July 2026 nacelle fire on a 2005 Vestas V80 farm (17 turbines, 30.6 MW, 78 m hub). Fire crews could not reach the nacelle; they held the perimeter. Spot grass fires only. Prior nacelle fire at the same site in 2019.
14 · South Australia turbine-fire record (2006–2024)
Clements Gap nacelle fire also ignited ~30 ha of grassland. Earlier losses at Cathedral Rocks, Starfish Hill, Lake Bonney (2006, falling debris started a grass fire).
Suppression systems, vegetation buffers, and underground collector lines are the practical mitigations the same literature keeps returning to.