Just published: Analysis of pollution from power plants (particularly coal) and its impacts on people, crops, and climate in the United States. [“The downstream air pollution impacts of the transition from coal to natural gas in the United States,” Nature Sustainability, 6 January 2020. (DOI: 10.1038/s41893-019-0453-5)]
Where can I find the study?
Read/download the paper here.
Supplementary Information (all the big tables for you table geeks) is here.
I wrote a “Behind the Paper” post for Nature that describes a bit of the project background; check it out here.
Replication code for the paper can be found here. To run the replication, download the zipped compiled data file here (546 MB). [The above file contains a few additional files for plotting and is organized for use with the replication code, but is otherwise identical to the data that were published with the paper here.]
(Note: The data and code linked above reproduce the analyses, figures, tables, and main results from the paper without first building up the final datasets from their sources. The full underlying data are large (50 GB) and compilation is slow; if that’s your jam, data are in four parts here [AdminPlantsCropsMortality.zip Aerosols.zip SO2NO2O3.zip EPA AQ Report.zip] and the repository with instructions and code for replication/compilation is here. Data source information is in the paper.)
Frequently Asked Questions
Can you describe this project in one sentence? I used observations (i.e., measurements, not models) and statistics to show that shutting down old coal plants in recent years saved lives and crops nearby, and resulted in a cleaner atmosphere over the whole country.
That was brief. How about five sentences? All fossil-fuel burning electric power plants emit a mix of other pollutants along with carbon dioxide. This study measured changes in pollution nearby when old electric power units turned off (most of these were coal-fired) or when new units turned on (most of these were natural gas-fired). I found that on average an old coal unit shutting down resulted in cleaner air (both particulate matter and ozone), lower mortality rates (i.e., fewer deaths), and higher crop yields (corn, wheat, soybeans) nearby. The reduced pollution contributed to an overall clearer atmosphere, with more sunlight reaching the surface of the earth. Although this study provides some hard numbers for how detrimental coal-fired power plants have been, it also showed that newer natural gas plants are not perfect — although they were not associated with increased deaths or crop impacts in the same way that the coal-fired plants were, they did create higher pollution levels nearby when they came online and will require further study.
How did you do the study / what methods did you use? The study examined power plants in the continental United States from 2005-2016. This date range is in part a function of when some satellite observations of the atmosphere became available, but also captures a lot of the transition in the US electric power sector, when many coal-fired plants were decommissioned and many new natural gas-fired plants were brought online. I used publicly available data from U.S. federal agencies (CDC, USDA, NASA, EPA) and statistically examined changes in all locations in the US, comparing locations where a power plant turned off or on to other locations where there were no changes. I looked at pollution levels, both at the surface of the earth and through the whole atmosphere, as well as mortality and yields of major crops (corn, wheat, soybeans), before and after a plant turned on or off. I find that there were statistically significant changes (meaning these are not likely to be random) associated with coal-unit shutdowns across the country — near these locations, mortality rates dropped, crop yields rose, and the atmosphere as a whole was clearer. I used these changes to estimate the total impact over time of this energy sector transformation, in terms of lives, crop production, and regional radiative forcing (this is one way to think about earth’s energy balance, or the instantaneous difference between incoming sunlight and outgoing radiation at the top of the atmosphere).
What are the basic findings? On average, counties that had a coal-fired unit shut down either within the county or within about 25km of the county saw a 0.9% drop in local mortality rates after the shutdown. Counties producing corn, soybeans, or wheat saw an average increase in yields of 7.2%, 6.3%, and 4.0%, respectively, after a shutdown. Aggregating these numbers up (multiply the average effects by local population or crop area, and the number of years the unit was off in this study period) gives the totals in the paper — ~26,610 lives saved and ~570 million bushes of staple crop production. If we assume that plants left in operation would have the same effects if shut down, we can also estimate the lives and crops lost due to coal-fired plants still in operation: ~329,000 lives and 10.2 billion bushes. These are the the conservative estimates — the numbers when transport-related impacts out to 200km are included are higher: 38,000 lives and 4.8 billion bushes saved by shut-downs, and (using the same assumptions) 510,000 lives and 75 billion bushes over the 12 year period.
On the climate side, reduced local pollution from coal plants resulted in cleaner skies overall (fewer aerosols in the atmosphere), and this changed earth’s energy balance (radiative forcing at the top of the atmosphere). Over the whole country, IRF increased by 0.5 W/m2 (meaning warmer), with larger increases in the eastern/mid-Atlantic regions. Why is this? Aerosols produced by coal combustion also block some incoming sunlight from reaching the surface of the earth (this is part of why they impact crop growth). This aerosol-based cooling effect actually “masks” some of the warming of the atmosphere caused by greenhouse gases; as the air has cleared in recent years with coal shutdowns, the full extent of warming becomes apparent. (Importantly, aerosols don’t undo or cancel out greenhouse gas warming; it’s two different mechanisms at play: I think of this like wearing a hat on an extra hot sunny day to keep your face cool. Your hat doesn’t change the warming, it changes what your face is feeling.) So this analysis shows how coal-based combustion in recent years actually kept us from fully feeling the impacts of global warming. It’s important to realize that clearing the air in the U.S. will in the short term speed up warming by removing the aerosol-based cooling effects.
Why do this study in this way? This type of study is important (IMHO) for two reasons: First, because it is empirical (i.e., it relies on observations and measurements as opposed to modeled data*), it tells the story of what actually happened recently in the US. We can see who the winners and losers of this feedstock transition have been (in terms of bearing the burden of pollution-related impacts). This analysis shows that there is a lot of heterogeneity in the impacts from stack emissions — it’s not just some nebulous pollution burden shared by everyone. Second, there has been a LOT of incredible work that explores how changes in pollution concentrations matter for both human and plant health, but they aren’t necessarily linked back to the tractable technology or policy changes that caused the variations in pollution concentrations. At the end of the day, if we want to understand the consequences of our decisions as a society, we have to be able to connect policies and technologies to the emissions changes they cause, then connect those emissions to the pollution that results (there is a lot of chemical and physical change between the mix of pollutants emitted from combustion of various fossil fuels and the ambient pollution that results), and finally link those pollution changes directly to outcomes we care about. This study makes the full connection — it’s the difference between saying “pollution causes X” and “the pollution from that power plant caused X.”
* Important: I am not implying that there is anything wrong with modeling: we learn a ton from model-based studies, and they are the right (and sometimes only!) choice in many cases (e.g., thinking about future scenarios). But to tell a similar story here using a modeling approach requires getting emissions, atmospheric transport and chemistry, and impact functions all correct, and recent studies have shown that this can be difficult to achieve with any kind of granularity. All that said, I think the right approach is to get at these questions and see how well different types of studies agree.
How do these numbers agree (or not) with previous studies? Overall these numbers agree broadly with other studies, when converted to impacts-per-concentration change. They’re higher than other empirical studies, which may be due to (a) having a cleaner causal identification strategy, (b) measuring the full impacts of *all* pollution related to plant emissions, or (c) differential toxicity of different types of (e.g.) particulate pollution or different aerosol/ozone mixes (in the case of crops). It is a bit trickier to compare to climate modeling studies, because those are dynamic, but the radiative forcing changes I estimate are in line with the overall literature (for example, as compiled in the IPCC 5th Assessment Report).
What are the main sources of uncertainty in this study? [Note: Uncertainty here means precision — e.g., I find that mortality rates dropped by 0.9% after a shutdown, with a window around that estimate of 0.1% to 1.7%. What creates that range / why isn’t it just one number?] There are a few sources of uncertainty here which mean that the likely impacts fall within the ranges noted in the paper. One is that these estimates are for an average unit shut-down, but units vary in sizes, technologies, and more. A second is that different types of populations are exposed in different locations — there may be more vulnerable populations in some areas compared to others. There is also simple measurement error involved (here again, I mean precision) everywhere. Reported county crop yields are averages, and the pollution estimates here are aggregated up to yearly values, which may hide some of the variation in exposures that matters. I provide estimates based on overall power in a set of tables that accompany the paper (Supplementary Information linked at the top). Finally, when aggregating total statistics, I combine the average effects with local population and crop numbers. This means that the exact impact when looking at any one location may be slightly off (a bit too high in some places, a bit too low in others).
What are the shortcomings of this method? There are a few shortcomings to this analysis: (1) As noted above, this method/study design calculates average impacts, the total numbers are estimates and the values for each individual location might not be spot-on. (2) It’s not really possible to separate out the impacts of (for example) aerosols versus other pollutants on people and plants in this analysis — there just isn’t enough independent variation. To do that, you’d need to have places where plants shut down and had very different changes on different types of pollution to be able to tease apart the individual impacts. But across the study area, these things are strongly related: When a plant closes, most of the related pollution drops. (The one exception here is ozone which has complicated and non-linear behavior as you move further away from a plant, and varies a lot based on background pollutant levels.) So these are estimates for all power-plant related emissions, not individual pollutants. (3) Regional climate impacts — Here I calculated changes in the instantaneous top of the atmosphere radiative forcing, but there are all sorts of dynamic responses of the climate that are not included here. Aerosols change the temperature structure throughout the atmosphere and that changes circulation dynamics. In addition, aerosols have a bunch of other impacts (for example on cloud formation and precipitation) that are not included. (4) Finally, it may not be straightforward to translate these findings to other contexts. This analysis tells the story of what happened recently in the US, but the effects may not be same when thinking about the future or other locations. If fuels and technologies differ, the impacts would be expected to differ as well.
* One other note: this is only a study of what happens downstream from power plants, post-combustion. There are certainly important environmental and human impacts of upstream activities associated with different feedstocks, and life-cycle impacts to consider as well. Recent studies have, for example, highlighted large fugitive methane emissions from natural gas extraction and transport that would certainly impact both local pollution (near leaks) and the overall greenhouse gas / climate impacts of natural gas. Although they do not produce either greenhouse gas or short-lived pollutant emissions in the same ways, nuclear, solar, and wind all have their own environmental, human, and land/resource use issues that would play into a full cradle-to-cradle comparison.
How could this study be useful? Power plants are politically hot topics. In recent years there has been a lot of rhetoric about the local economic benefits (e.g., jobs) that come with having a local coal-fired power plant. (There have also been political, health, environmental justice, and indigenous rights discussions surrounding natural gas infrastructure.) The point is that a lot of economic considerations, as well as some political and social considerations, tend to go into energy policies and decisions. However, in an ideal world, stakeholders would weigh all costs and benefits of a decision to (e.g.) build or retire an electric power generation unit, including impacts on health, agriculture, and climate. Unfortunately it’s often tricky to pin down those numbers for inclusion in the decision-making process. This study gives some hard figures that communities might be able to use in thinking through energy priorities. Perhaps most important, this analysis shows that there’s a lot of variation in who was and still is paying some of these costs of coal (in terms of lives and crops) — there is a lot of potential benefit yet to be gained from further retirements.
What’s next? A personal interest of mine is understanding pollution impacts on crops, and there’s a lot more here to be understood about sunlight effects (i.e., what aerosol pollution does to the light environment for crop growth) versus toxic exposures like NOx and Ozone. I’m looking forward to digging into the crop data more. There’s also an opportunity to look at other human outcomes besides death. Mortality is like the tip of the iceberg in terms of impacts on humans — there’s lots of evidence that pollution impacts health in lots of non-mortal ways, from reduced productivity to acute hospitalizations to long-term conditions. Finally, I’m excited to learn more about the dynamic response of the climate to these kinds of changes. Pollution’s impacts on climate are really important to understand from a planning perspective. Decision-makers need to know what to expect in terms of regional climate from policies they might pursue to clean up the air so they can be ready and make complementary adaptation decisions.
[Last updated 10 January 2020]