Case study
Targeting Hyper-Polluting Power Plants
RASEI, University of Colorado (NSF-funded) · 2019–2021 · Data engineer
The world's most-polluting 5% of power plants produced 73% of the electricity sector's CO2 in 2018. The data work behind that finding, published in Environmental Research Letters in 2021.
Context
The question
Burning fossil fuels for electricity is the largest single source of CO2 by sector, and sector-wide policies have proved hard to put in place. Some researchers had argued that a country could cut a large share of its electricity emissions by acting on a small group of extreme plants, but only two studies had tested that across countries, and both used 2009 data, from the middle of a financial crisis and before the Paris Agreement. The project asked three things of the world's fossil-fueled fleet in 2018: where the biggest emitters are, how unevenly emissions are spread across each country's plants and whether that changed since 2009, and how much each country would cut if only its worst 5% changed. Before any of that could be asked, the plant-level picture had to exist: one file covering 29,078 plants in 221 countries, with emissions estimated where plants do not report them and coordinates found where none were recorded.
My part
What I did
I was the project's data engineer and one of the paper's three authors. The plant file itself, the 2018 edition of the Carbon Monitoring for Action database, was built by its original architect, Kevin Ummel, with two models that estimate output and emissions for the plants that do not report them. My part was to make that file usable as a map and as a base for policy analysis.
- Coordinates. 13% of the plants, 3,813 of 29,265, had no location: the World Electric Power Plant Database records none, and India and Australia do not report them. Merging the coordinates from the original CARMA file covered 54.8%. For the rest I wrote a forward geocoder in Python, with Pandas and GeoPy, that placed plants from their country, region and city names and took coverage to 97.9%.
- Matching. I reconciled the plant records across CARMA, the World Electric Power Plant Database and the World Resources Institute's Global Power Plant Database, where the same plant appears under different names and identifiers, so that one record could carry all three.
- Policies. I matched each plant to the decommissions, emission controls and climate and energy policies that apply to it: the layer for the larger dataset the paper announces, and for the wider project's question, the effect of climate policy on power plants.
Model-filled values are estimates, not measurements, and geocoded locations are derived. Both should be read as layers on top of the source records.
Finding
What came out
The world's most-polluting 5% of plants produced 73% of electricity-sector CO2 in 2018, 14.6 times their share of the fleet. The ten worst plants all burn coal, in Poland, India, South Korea, Taiwan, China, Germany and Japan; the worst of them, Bełchatów in Poland, emitted more CO2 in 2018 than the whole of Switzerland. The ten were also 28% to 76% more carbon-intensive than the other fossil-fuel plants in their own countries: they were not just big, they were dirty. Between 2009 and 2018 the spread widened in most of the ten largest emitting countries. How concentrated emissions are varies a great deal by country, and with it what targeting the worst plants can do: in China the top 5% produced 24.5% of electricity CO2; in the United States, Japan, South Korea, Germany and Australia, between 75% and 90%. The table gives the paper's estimates of what would follow if only the worst 5% of plants changed.
From the paper
If only the worst 5% of plants changed
How much electricity-sector CO2 would have fallen in 2018, by country, under four changes to the top 5% of emitters. The bars share one scale.
| Country | Brought to their country's average | Brought to the world average | Coal and oil switched to gas | Carbon capture fitted |
|---|---|---|---|---|
| China | −3.8% | −9.6% | −11.1% | −20.7% |
| United States | −29.3% | −29.5% | −41.7% | −63.7% |
| India | −8.4% | −20.4% | −23.6% | −44.2% |
| Japan | −39.2% | −35.2% | −43.1% | −76.1% |
| Russia | −20.9% | −17.4% | −34.1% | −37.6% |
| South Korea | −40.0% | −32.7% | −39.3% | −76.0% |
| Germany | −55.9% | −35.1% | −43.7% | −75.9% |
| South Africa | −8.6% | −11.6% | −13.3% | −25.2% |
| Australia | −28.3% | −34.9% | −42.9% | −75.5% |
| Indonesia | −9.0% | −28.0% | −33.4% | −60.5% |
| World | −16.8% | −25.4% | −29.5% | −48.9% |
Since then
What it changed
The paper's argument is that policies aimed at a small set of extreme plants belong alongside sector-wide and economy-wide instruments, and that no single version of it fits every country: where the fleet is more even, as in China, the target has to widen from the top 5% toward the top 10%. It was published open access in Environmental Research Letters in August 2021 and has been cited 38 times since (Crossref, September 2026). The paper announces the next step: the same plants with their policies, decommissions and emission controls matched, as a larger dataset for the question of whether policy changes what plants emit.
Sources
Where the record is
- The paper. Grant, D., Zelinka, D. and Mitova, S. (2021). Reducing CO2 emissions by targeting the world's hyper-polluting power plants. Environmental Research Letters 16(9), 094022. Open access under a Creative Commons Attribution licence, with supplementary material on the journal's site.
- The data. The 2018 CARMA file draws on plant-level emissions reports from the United States, the European Union, Australia, Canada and India, the World Electric Power Plant Database from S&P Global Platts, and country production data from the International Energy Agency. The paper's data statement: the dataset is available from the corresponding author on reasonable request. It is not public, so the numbers on this page can be checked against the paper, not re-derived from the file.
- The funding. National Science Foundation award 1824106.
- My publications. The full list is on ResearchGate.