Balancing Arguments About Electricity for States and Utilities
People like to talk about the generation resources and carbon emissions of states. But is that actually meaningful? For more useful analysis we should look at the physical electric grid itself.
According to federal data, Vermont, wouldn’t you know, has the lowest CO2 emissions rate among states — over five times lower than the next one, Washington.
But what does that really mean? Is electricity in Vermont “five times cleaner” than that in Washington? Is provisioning it five times less dependent on fossil fuels? No.
In this post I try to explain why one should be careful with such claims. It’s not that they’re wrong; it’s that they’re narrower statements about discrete electricity generators but purport to be wider statements about electricity provision and consumption.
Instead, I argue the physical unit of “balancing areas” should be used instead of the political unit of states when comparing emissions rates between places. And I’ll show how government agencies offer the authoritative data for such analysis.
EPA’s authoritative but limited data
At the end of January every year the US Environmental Protection Agency (EPA) releases the latest version of its annual Emissions & Generation Resource Integrated Database dataset — or “eGRID” for short. As “the preeminent source of emissions data for the electric power sector,” in the agency’s words, the data set synthesizes power plant data from two governmental sources: generation data from the US Energy Information Administration (EIA), which collects that data from the power sector, and emissions data from the EPA itself. It lags real time by over a year, unfortunately — I’ve come to realize just how much work goes into making the dataset accurate — so last January only saw the release of the 2022 data.
In short, eGRID is the authoritative dataset for emissions of all individual generators on the grid in the United States.
Some might use eGRID to answer questions like “what was the highest sulfur dioxide-emitting power plant in New York State?” The Long Island Power Authority’s Northport plant, whose gas-powered generators came online in the 60s and 70s.
Or, given the generation data, it can be used to determine emissions rates: “what was New York City’s generator with the worst nitrous oxide emissions per kWh of generation?” The on-site, oil-fired emergency generators at the St. Mary’s Hospital for Children, according to estimated emissions based on fuel type. But among EPA’s measured emissions data on select facilities, using on-site pollution control equipment, it’s Con Edison’s old Hudson Avenue plant, which the utility retired in November 2022 to comply with the city’s “peaker rule.”
EPA aggregates its generator data to the level of states, and that’s where a claim like Vermont having the cleanest electricity in the nation comes from. But one shouldn’t derive too much meaning from that statement.
Sets of generators don’t tell us enough about electricity service
If one wants to compare generators between states, the state-level data is fine. Vermont’s generators produce five times less CO2 per unit of energy as compared to Washington’s generators. But if one wants to compare the electricity provisioned and consumed on the grid, that state-level data doesn’t cut the mustard. Let’s look at two concrete examples of the application of electricity in those two states.1
For Vermont, consider a maple creemee machine in Burlington, which uses electricity from the grid to freeze the delicious ice cream and to extrude it onto a crispy cone. For Washington, consider Microsoft’s Columbia Data Center, which uses electricity from the grid to run and cool a whole lot of computers. These are consumers, not generators, of electricity, and so nothing about them is covered by eGRID.
The question at hand is how clean and dependent on fossil fuels they are. That’s a property of the electricity service that is provisioned for them. And because neither Vermont nor Washington is an island (unlike Hawaii!), electrons flow in and out of each state along the physical connections of the grid, regardless of state borders. That makes a statement about not depending on certain generation resources, like on fossil-fueled power, essentially impossible to prove. Maybe a coal plant across state lines really keeps the maple creemees flowing.
Where does the electrical resource dependency stop? How far away might a generator be before a consumer doesn’t at all depend on it? At the broadest possible level, “the grid” refers to an AC network collectively — and somewhat miraculously — maintained at a frequency of 60 Hz. We have three spanning the continental United States: the Eastern Interconnection, which covers Vermont, the Western Interconnection, which covers Washington, and the Texas Interconnection, which covers most of Texas. A butterfly flapping its wings might contribute to a typhoon across the globe, and a nuclear plant in Miami might contribute to the maple creemees across the Eastern Interconnection.
But there is also a finer granularity of analysis of the physical electricity system, one that more accurately captures the reliable provisioning of electricity within the grid — the balancing authority area.
Balancing authorities paint a clearer picture
According to NERC, the North American body that determines electric reliability standards, a balancing authority is “the responsible entity that integrates resource plans ahead of time, maintains demand and resource balance within a Balancing Authority Area” — each area “the collection of generation, transmission, and loads within the metered boundaries of the Balancing Authority” — “and supports interconnection frequency in real time.”
Basically, they’re the fundamental grid operators performing that miraculous work of keeping electricity stable at all times, each one controlling a discrete territory within the larger system, each one connected to others.
In traditional areas of the power system, the utilities that send you monthly bills act as BAs, like in parts of Washington, but in restructured areas with liberalized power markets, larger entities called RTOs/ISOs do that work, like in Vermont. Generators and even some large consumers are centrally dispatched by the BA in order to keep electricity balanced and stable.2 In some places there are multiple BAs covering a state, like Washington, but in other places a state is entirely covered by a single BA that spans other states too, like Vermont.
Back to our two examples. The electricity serving maple creemees in Burlington is the result of balancing supply and demand within the entire BA — ISO New England, or ISO-NE for short — whose territory spans not just Vermont but Connecticut, Maine, Massachusetts, New Hampshire, and Rhode Island too. The electrons powering the machine can’t be traced to any particular generators, within Vermont or without; they are just as much the result of a Vermont solar farm as a Massachusetts gas plant. It’s ISO-NE’s job to keep those electrons stable at all times, and to ensure that grid resources are planned, installed, and retired in such a way that they can perform that job over the years.
In this way, balancing the grid rests on the interplay of all the resources. One can’t say the solar farm (or a nuclear plant for that matter) in the ISO-NE footprint doesn’t depend on the availability of a gas plant. On the contrary, ISO-NE might very well be relying on such a gas plant to balance the fluctuations of solar output and demand. It could even be relying on resources in a neighboring BA; we’ll look at such interchange with neighbors later.
The Columbia Data Center, on the other hand, is part of a much smaller BA, the Grant County Public Utility District, or Grant PUD for short. There, the region’s abundant water resources provide bountiful carbon-free hydroelectric power, and unlike wind or solar resources, the hydropower can be ramped up and down as needed for balancing.3 Because of its limited territory, Grant PUD has a far lighter responsibility, balancing far less supply and demand, than does ISO-NE. It’s also not operating its own auctions to transact for power and incentivize new resources.
But if ISO-NE is so large, might there be constraints on the grid that effectively isolate certain areas? For example, if there was only one small transmission line connecting utility systems in Vermont to those in other states within ISO-NE, technically the respective systems would be within the same BA but congestion on that line would turn Vermont into a de facto island. (In that case, the state-level analysis might appear more accurate after all.) Unfortunately, we simply don’t have standardized public data on the power system at a finer granularity than BAs,4 so we have to settle for BAs as the analytical sweet spot. Their responsibility for balancing makes that sweet spot sweeter.
Maple creemees vs. data centers, round two
We’ve seen that eGRID synthesizes EPA emissions data with EIA generation data, and that its aggregation at the state level doesn’t help us understand the “dirtiness” of electricity on the grid in a given state. Thankfully, eGRID also aggregates data at the BA level too.
What then are the relative CO2 emissions rates of the ISO-NE BA, which plans and balances resources encompassing the Vermont maple creemee machines, and the Grant PUD BA, which does the same for a much tinier footprint that contains the Washington data center? As shown in the table below, the emissions rates between the two are markedly different.
The BA-level numbers are so different from the naive state-level ones that they tell a completely different story: the maple creemees’ electricity isn’t five times cleaner than the data center’s; the data center’s electricity is, err, infinitely cleaner than the maple creemees’.
Another result of the BA analysis is that the provisioning of electricity within Vermont is actually substantially dirtier — that is, much more physically dependent on fossil fuel generation — than the Green Mountain State’s residents and politicians would have the rest of us believe.
As a caveat, however, Vermont and Massachusetts both have a considerable amount of behind-the-meter solar generation (i.e. rooftop solar) which is not accounted for in the eGRID dataset. If it were, CO2 rates for generation in both states and the whole ISO-NE would be a bit lower. That’s the flip side to generation resources not on the grid: our public data doesn’t know about it.
What balancing area analysis reveals about the grid
Zooming out to the whole country, let’s take a look at all the BAs, or at least those of substantial size, ranked by CO2 emissions rate. The chart below, drawn from eGRID, shows the 20 BAs with at least 10 GW of installed capacity. To make it more interesting, I’ve also added columns showing the percentage of different kinds of generation in each BA.
Two interesting results from the chart above are worth highlighting.
BAs with lower emissions rates aren’t simply those with higher percentage of wind and solar generation. Of the top 10, only two have more than 10% wind and solar on the grid, while some BAs with substantially higher percentage of wind and solar generation also have substantially higher CO2 emissions rates. The deregulated market covering most of Texas, ERCOT, is a prime example of the latter.
BAs for vertically integrated monopoly utilities sometimes have lower emissions rates — sometimes far lower — than BAs for restructured markets that have more wind and solar generation. Look at Salt River Project, the two Duke territories, Florida Power & Light, and the Tennessee Valley Authority for example; all are cleaner than the BAs for the competitive markets of ERCOT, PJM, MISO, and SPP. Advocates of such competitive markets would have you believe this simply cannot be true.
Why is the CO2 emissions rate of a BA not exactly correlated with the percentage of wind and solar in it? Because the big carbon-free power sources of hydro and nuclear dominate in the cleaner BAs.
Coveting thy neighbor’s resources
Earlier I said neither Vermont nor Washington was an island, but the same is true for the corresponding BAs. On top of managing the resources within its area, each one also manages imports and exports of power with its neighboring BAs along the physical transmission lines between them. That’s what makes the BAs add up to a whole Interconnection — a grid.
So far I’ve argued that one should look at the generators inside a BA rather than merely those inside a state to understand the “dirtiness” of electricity. Going further, one should also look at the BA’s reliance on imports. A hypothetical BA that consists entirely of intermittent wind and solar generation would have no supply-side means of balancing that power to meet demand — but it might rely on “dispatchable” imports from a neighbor to balance.
The EPA eGRID dataset doesn’t tell us anything about power interchange between BAs. Instead, the EIA provides public data on generation and interchange for all BAs in the country. There are two key ways that EIA’s BA data differs from eGRID.
EIA’s BA data is provided on an hourly basis, not annually, and it lags real-time by roughly a day or so, not more than a year like eGRID. That’s because…
EIA’s BA data does not include any generator-level data, whether generation or emissions. Instead, it only offers the total amount of generation and estimated CO2 emissions by fuel type (for wind, solar, gas, etc.), within each BA, along with demand and imports and exports with neighbors.
For example, consider the following chart showing ISO-NE’s interchanges over the first week of June this year. There’s a clear pattern of daytime and afternoon exports (positive values) to New York ISO, the BA serving New York State, followed by evening imports (negative values) from the Hydro-Quebec, the (public power) utility and BA serving Quebec.5
Putting it all together with flow tracing
With the addition of imports and exports of power, the BA-level analysis can move us even closer to an understanding of “dirtiness” of different areas of the grid. If ISO-NE had relatively clean generation internally but relied heavily on imports from a coal-heavy neighbor for balancing, for example, then for an understanding of electricity consumption in ISO-NE we should take into account the emissions rates of those imports.
That’s exactly what the “flow tracing” emissions accounting method accomplishes. It was first applied to the level of BAs in work published in a 2019 PNAS article by Stanford researcher Jacques de Chalendar and his colleagues. Since then it has been adopted by the popular Electricity Maps app and even the EIA itself. Electricity Maps has a nice blog post about the method, but the gist is that you process the entire networked system of BAs alongside the hourly generation and interchange data to come up with a CO2 emissions rate or “carbon intensity” of electricity consumption, not just generation, for each BA.6
de Chalendar’s website hosts visualizations of the emissions rates of various BAs. Below is a chart showing the weekly emissions rate of ISO-NE over the past several years, with the weekly average emissions rate of generation, of imports, and of overall consumption (“Demand”) shown in the respective lines.
The clear takeaway is that imports, like Quebecois hydropower, have always been far cleaner than local generation inside ISO-NE.
Returning to the original example, now equipped with BA-level consumption data that takes imports into account, we see slightly dirtier CO2 emissions rates for both maple creemees and the data center:
It’s not perfect, but it’s much closer to reality than state-level aggregation.
When it comes to carbon intensity of electricity, a region’s generators do not tell the whole story of that region’s electricity as a consumed service. That’s because that service is the result of the work of a balancing authority using all the resources at its disposal, including interchange with its neighbors.
“Did you know Vermont’s electricity is the cleanest in the nation?” Hopefully now you’re better equipped to understand — and rebut — a misleading claim like that.
Since we care about grid electricity in this analysis, we’re not considering the possibility that any local generation and/or storage is at play, like a rooftop solar system, or a portable diesel generator.
The RTOs/ISOs additionally operate the markets in which power is bought and sold. These are the entities highlighted by analyst and author
in her book Shorting the Grid. Like the distinction between an RTO/ISO’s dual role as grid operator and as market operator, she distinguishes between the “physical grid” and the “policy grid.”Typically, though, the operations of hydropower facilities come with many strings attached, since the operation of those dams come with state and federal constraints around, say, fish populations or river navigability.
The EIA data collection for balancing authorities was first announced in 2013. You can find the justification for it in the Federal Register here.
For a closer look at those diurnal import patterns in ISO-NE, see this chart. As an exercise to the reader, look up the generation data for the BA to see how natural gas and solar generation follow the same rhythm, with nuclear as a horizontal line.
The flow tracing analysis rests on a method of determining the carbon emissions generated inside each BA. But the EIA data does not offer measured emissions data; instead it merely offers generation by generation type — 5 MWh of gas generation, 10 MWh of solar generation, etc. The emissions data is then derived from “emission factors” for each generation type — “870 lb/MWh for gas generation, 0 lb/MWh for solar generation, etc.