Given the numerous heatwaves plaguing Europe in recent years, our research began with the question: what happens to the grid during extreme summer heat? To our surprise, we found that existing research only partially answered this question, which motivated our work.
Our paper in Nature Energy shows how heatwaves can simultaneously increase demand, reduce generation availability and limit how much electricity transmission lines can carry. These effects interact: even when electricity is available elsewhere, the network may struggle to deliver it where it is needed.
While individual aspects have been separately studied in the past, our investigation is among the first to simultaneously investigate their coupling and properly evaluate their joint effect in a single framework.
We used scenario-based stress tests to examine where these combined pressures can expose capacity bottlenecks in European national grids. Although we cannot precisely forecast when or where blackouts will occur, the results do show where grids are vulnerable and the conditions that expose those weak points. We’ve released our model and data as an open toolkit called HEAT-Analysis, so that operators can run the same analysis on their own grids.
How heat affects temperate Europe
The project was born in 2023 from a reading group that discussed Cambridge alumnus Simon Sharpe’s Five Times Faster, a book about putting extreme (rather than average) weather at the heart of climate thinking. It led us to wonder what extreme heat might do to power grids. We set about building a modelling framework to capture these interacting effects.
We found that a heatwave hits the grid in three ways at the same time. Demand rises, because everyone tries to cool themselves using power-hungry air conditioners. Meanwhile, generation falls because thermal power plants struggle to shed heat, solar panels lose efficiency and, in many cases, wind output drops too. The transmission lines themselves also lose capacity because they can’t dissipate heat to the already-hot air fast enough.
That last effect is the least familiar. Due to line resistance, a high-voltage line warms up as it carries current. It must stay below a temperature limit (typically around 90°C) or it can sag towards trees and the ground. On a cool, breezy day the line sheds heat easily even with a high load. On a hot, still day it can’t shed heat as easily, so capacity must be reduced to keep it from exceeding its limit, sometimes by as much as half of the line’s rated value.
Conditions also change along a line’s length, further complicating the picture. For example, one line in northwestern Spain crosses eight weather regions along its 130-kilometre length. A single hotspot can reduce capacity along the entire line, so we modelled lines section by section.
Another consideration is that heat changes the line’s resistance and, in turn, how much electricity flows.
Capturing these effects together across a national grid quickly became computationally unmanageable, especially when repeated over hundreds of weather scenarios.
Our solution was to separate the calculation into two simpler problems: finding power flows with fixed temperatures, then updating temperatures using those flows. Alternating between them preserved the feedback and closely matched the full calculation at about half the computing cost, making the study possible.
Another challenge was getting valid data about real grids. Grid assets and operations are sensitive information and not always publicly available. We therefore turned to the open-source PyPSA-Eur model of the European network, combining it with weather reanalysis and demand data. We know this leaves out some parts of the grid, but it’s the best we have. We expect individual grid operators to know enough about their own grid to use our open-source tools to find their true hotspots.
What the model revealed
Three results stood out. First, standard methods are inadequate. Because they don’t track conductor temperatures, they let lines run hotter than their limits and so overstate how much power the grid can move, missing potential problems.
The second was how limited the benefits of storage were in the configurations we tested. Even when we let every storage site in the system start fully charged when a heatwave arrived, load shedding (the controlled power cuts operators use to protect the rest of the grid) fell only marginally. Extra energy can’t make up for higher demand and lost generation if lines can’t deliver it.
The third was that neighbours can’t always help. In our simulations, France gained substantial relief when linked to less-stressed neighbours. Spain gained almost none, because heatwaves often span large regions and leave the grids next door short of spare capacity too.
Across the eight European grids we studied, our simulations indicated load shedding to be likely in France, Spain and Italy, while Germany, the UK, the Netherlands, Belgium and Portugal maintained supply in those scenarios. When we added modest annual demand growth of 1% to 3% (as expected from electric vehicles, electrified heating and cooling, and data centres), load shedding rose in step.
Why are there so few blackouts in practice?
Our scenarios cover 2026–2030, but they are not forecasts of the exact heatwaves we will experience. A real heatwave will differ in its intensity, location and timing, creating a different combination of electricity demand, available generation and transmission capacity. Not every combination produces a shortage.
Even under the same weather conditions, the grid represented in our model is not an exact copy of the real network. Gaps in public data, recent upgrades and operating options can mean that the model exaggerates some bottlenecks or misses others.
The results are therefore best understood as a stress test to identify potential weak points and the conditions that expose them, rather than predict when and where the next blackout will occur.
What can operators do?
Operators can use our open-source HEAT-Analysis tools with their own grid data to find vulnerabilities or test proposed upgrades. When bottlenecks are found, there are ways to respond, from reconductoring vulnerable lines (restringing existing towers with conductors that tolerate more heat) to placing storage near bottlenecks.
Our model was developed for Europe because its grid data are unusually open, but the same approach could be applied in China, India, Texas or across Africa, where some grids already plan around summer peaks and others may soon need to.
Next, we want to use machine learning to make the analysis fast enough to run over millions of scenarios rather than hundreds, and to extend it to other extremes such as storms and drought. The aim is to help planners assess demand, generation and transmission together under extreme weather, and use the resulting bottlenecks to guide further investigation and the testing of possible upgrades.
Keshav and Minghua conceived the study. Enming collected the data, developed the algorithm and ran the simulations, and all three of us developed the model and interpreted the results.
Header photo: High-voltage lines on a hazy summer day. Credit: Birte Fritsch / Flickr (CC by 2.0)