Which US states look most exposed to losing power, and why that is not the same problem in every state. I built the score from public data.
Tobyn Smith / built from open EIA, EPA and NOAA data
Use the arrow keys, the marks along the bottom, or a swipe on a phone. The outage figures and the plant capacity are real, while peak demand is still a sample.
Keeping the power on is one of the bigger infrastructure questions in US policy, and a few pressures are hitting the grid at the same time.
More frequent and more severe storms, heat and cold snaps are behind most of the large outages.
A lot of the wires and substations were built decades ago and are being asked to carry more.
Coal is retiring, gas and renewables are growing, and that shifts where the weak spots sit.
Most of the data for this is already public, but it is split across agencies and formats. I pulled a few of those sources into one score per state so they can be compared.
Three public inputs, from two federal agencies, with no paywall and no login.
The industry reliability measures, SAIDI and SAIFI, for how long and how often customers lose power.
Generating capacity by plant and fuel, from EPA's eGRID, which shows how much a state leans on one fuel or one big station.
Installed capacity set against peak demand, so you can see how thin the margin is on a bad day.
The outage figures are the real EIA-861 numbers for 2023, for every state, and plant capacity is real too, from EPA eGRID2023. Peak demand still comes from a built-in sample, which is why the project runs without a data key.
Every state gets a score from 0 to 100, where a higher score means it looks more likely to struggle to keep the power on, and the number comes from three pieces.
How often the power goes out and for how long, from the SAIDI and SAIFI measures.
How much of a state's power leans on one fuel type or a single very large plant.
Whether there is much spare capacity above peak demand, and how varied the supply is.
I put the three on the same scale and then combine them. The weights sit in one settings file so they are easy to change, and the next two slides let you do that.
Darker means more exposed. Hover a state for its score and what is pushing it up.
Hover or tap a state to see its score.
How much each part counts is a choice I made, not a fact. If you drag the weights, the ranking re-sorts, and the map on the previous slide re-shades as you do it.
| # | State | Score |
|---|
Showing the top 15. At the default 45 / 30 / 25 this matches the pipeline output, and if you push the weight onto one part, states like West Virginia climb or slide. Hover a row to find that state on the map.
A high score is not one problem. It depends which piece is doing the work, and that changes by state.
Mississippi takes 84 per cent of its capacity from gas, and West Virginia 83 per cent from coal. They are the two most concentrated supplies in the country, and they lead the table on that alone. If either fuel has a bad year, there is not much else to fall back on.
Maine has the least concentrated fuel mix of any state and still comes third, because its customers lost thirty-one hours of power in 2023. That is a lines and restoration problem, not a plant problem.
Spreading a fuel mix, hardening lines and adding spare capacity are three different jobs, with different costs and different timescales. A rank on its own does not say which job it is.
The label under each state on the map is just the short version of that.
The limits matter more than the ranking if you are going to use this for policy.
State-level figures blur a lot. A state can look fine overall and still have fragile pockets inside it.
Outage figures and plant capacity are both real, but peak demand is not, so the spare-capacity part is still a worked example, and I have labelled the parts where they appear.
The score says nothing about the transmission network that moves power around, which probably matters as much as the plants do.
The top and the bottom hold. The middle moves around, and a precise place in it is not stable.
The index leans hardest on outages, so I checked that data against a source collected by someone else. NOAA's Storm Events Database is compiled by the National Weather Service, and it has nothing to do with what utilities file to the EIA.
The EIA publishes outage minutes twice, counting major storm days and excluding them. The gap between the two is the part storms account for. If the weather story is right, that gap should track how much damaging weather a state actually had.
Across the fifty places that report both figures, everywhere bar Hawaii, the rank correlation is 0.21 for how many damaging storms a state had and 0.24 for how much damage they did. Texas and Georgia record the most storm events in the country and lose few minutes. Maine records a fraction as many and lost the most of anyone.
Storms obviously matter, or the two outage figures would not sit so far apart, but how much bad weather arrives does not decide who suffers most. What the grid is like when it arrives, and how fast crews restore it, look like they matter more.
It is one year of data, and NOAA event counts partly reflect how densely a state is observed.
This is a US project, but the question is not only a US one. Looking at it from outside changes what counts as a normal year.
In much of western Europe a typical customer loses power for minutes a year, while in the US the median state lost four hours in 2023 once major storms are counted, and Maine, the worst, lost thirty-one. Countries measure this differently, so the figures do not line up neatly, but the gap is too wide to be a measurement artefact.
Countries that perform well tend to bury more of their lines and pay for it in their bills. The US runs long overhead lines through forest and hurricane country, which is cheaper right up until the weather arrives.
Nothing in the scoring is US-specific. With capacity, reliability and demand data, the same three components would run on German Länder or Indian states, and the hard part abroad is getting hold of the data.
The US publishes enough of this that someone else could rebuild the score. Not many countries do.
Georgia is where I live, so I looked at one state more closely than the ranking allows. I set the score next to Georgia's figures from the EIA, EPA eGRID and NOAA, and there is a map of every power plant of 50 MW and up that you can filter by fuel.
On the real data Georgia is mid-table, nineteenth of fifty-one, which is much lower than the sample had it. Its 343 minutes off supply in 2023 sit close to the national middle, and the fuel mix is broad enough that concentration is not what pushes it up.
The analysis runs in Python. It pulls the public data, or falls back to a built-in sample, then cleans
and joins it, scores each state, and writes the charts and a results table. The map is drawn in R with
the usmap package, and the same results export as a GeoPackage so they open in QGIS.
Built with pandas, matplotlib, ggplot2 and geopandas. With Python installed, from the project folder, it is two commands:
# install the libraries, then run the pipeline
pip install -r requirements.txt
python pipeline.pyFull setup notes, including the optional R map and how to plug in a live EIA data key, are in the project README.
Data: EIA forms 860 and 861, EPA eGRID, NOAA storm records. Code and the full write-up are on GitHub.