Europe's warming rate
Since the 1980s, Europe has warmed at more than twice the global average rate.
WMO climate assessment
Extreme heat can increase electricity demand, change power-plant output and limit how much power transmission lines can carry. HEAT-Analysis brings these effects together in grid models built from public data to examine where projected heatwaves could expose bottlenecks under the represented infrastructure configurations.
AI-generated editorial illustration, shown for context rather than as a record of an outage.
Why this matters
Recent European heatwaves have brought widespread heat stress and a substantial health burden. Electricity networks must also operate through these conditions.
Since the 1980s, Europe has warmed at more than twice the global average rate.
WMO climate assessmentOn 17 July 2024, one fifth of European land reached at least strong heat stress.
Copernicus ESOTC 2024A peer-reviewed study estimated this number across 35 European countries in summer 2022.
Nature Medicine studyFor electricity systems, heat can raise demand just as generation and transmission capacity become constrained.
The problem
Electricity supply depends on both available generation and the network's ability to deliver it. A heat-related constraint on a transmission line can therefore matter even when power is available elsewhere.
More air conditioning increases the amount of electricity that must reach homes and businesses.
Some thermal power plants lose available output as the air warms, while wind and solar production respond to the accompanying weather.
Air temperature, sunshine and wind alter conductor heating and therefore the current that a route can carry within the modelled temperature limit.
Weather to demand
Cooling demand rises as conditions become hotter. We calibrate each country's response using historical electricity use and weather, separating base demand from heating, cooling and workday effects.
The daily estimate is converted to an hourly profile using historical demand ratios from hot days. Future growth changes the base-demand component, while the calibrated weather response is retained.
Daily demand = base demand + heating + cooling + workday effect
Heating and cooling terms respond to BAIT below and above their respective thresholds. The reference model's separate annual trend term is not used.
Base demand grows from its 2025 level at a compound annual rate of 1% in the baseline and up to 3% in the sensitivity cases. This growth applies only to base demand, not to the weather-dependent components.
Weather to available power
Higher temperatures can reduce the output available from conventional generators. We use technology-specific relationships that can be refined with plant-specific data.
Wind and solar availability follow the accompanying weather through Atlite profiles. These limits describe how much power could be produced, not how much the grid ultimately dispatches.
Available output = nominal capacity × availability factor × temperature derating
Availability and derating are applied according to generator type. Where matching future weather-driven profiles are unavailable, historical mean availability derived from PyPSA is used. The OPF then chooses generation within these limits.
OCGT: open-cycle gas turbine. CCGT: combined-cycle gas turbine.
Weather to line capacity
Current and sunlight heat a conductor, while convection and radiation remove heat. Wind, air temperature and sunshine therefore change the current it can carry without exceeding its temperature limit.
We calculate this balance along individual line segments. Local weather conditions determine the thermal current limit, which is also capped at the line's nominal rating.
Convection + radiation = solar heating + electrical heating
Building on the IEEE 738 heat balance, we calculate temperatures segment by segment to capture weather variations along a line. The most restrictive segment sets the line's thermal current limit.
We also approximate thermal interactions between conductors in a bundle by multiplying convective and radiative cooling by 0.8. This baseline choice is assessed alongside alternative conductor settings and should be adapted to the actual line configuration, using local weather and conductor-bundle data.
Electrical heating depends on current squared and temperature-dependent resistance. The OPF and heat-balance calculations exchange current and temperature to update line resistance.
The framework
For each projected heatwave hour, the model tests whether the represented grid can meet demand under the corresponding weather conditions.
For each projected heatwave hour, we combine the weather-dependent inputs with the represented network, generation and storage configuration. We solve an independent grid-operation problem within the modelled operating limits to determine generation, power flows and any unmet demand, recorded as load shedding.
We start with historical weather records (ERA5) for selected heat events in 2019, 2022 and 2024. For temperature, we take the difference between each historical heatwave and its reference weather, then add it to the future reference climate. We use bias-corrected projections for 2026–2030 (C3S/CORDEX, under the RCP 4.5 climate scenario). This carries the historical event's hourly and spatial temperature pattern into a future climate setting. Changes in solar irradiance and wind speed are also transferred to their future reference conditions.
The scenarios depend on the historical events and reference climate data; they do not predict how future heatwave dynamics will change. Other climate models or weather scenarios can also supply inputs to the grid analysis.
Each hour is analysed independently, with a specified initial storage charge rather than a charging and discharging schedule across hours. The weather-dependent inputs feed into an alternating-current optimal power flow (AC-OPF) calculation. Demand that cannot be met within the operating constraints is recorded as load shedding. The Spain example shows one set of inputs and its solved network state.
Power flows and conductor temperatures depend on one another. We alternate between solving grid operation at fixed temperatures and updating segment temperatures and line resistances using the resulting currents. Separating these calculations retains their feedback while reducing the cost of analysing many heatwave scenarios.
The European analysis uses two updates. Compared with the same procedure continued until the results stabilise, this gives average errors below 1% for both load-shedding and line-temperature metrics. The Spanish method comparison shows the differences in results and running time.
Explore one solved hour
Start with one projected heatwave hour: 13:00 on 24 July 2030 in the represented Spanish grid. The map brings together the network configuration, weather conditions and solved grid operation.
Illustrative 2030 snapshot under the represented grid configuration; not a forecast of an outage. Weather fields come from the bundled Spain 2030 worked example. Solved results match Extended Data Fig. 2 Source Data, and installed capacities come from the SI Spanish grid example in the figure source packages.
Spain: comparing methods
The map shows one solved hour. Across a wider set of heatwave conditions, how much does the modelling approach change the bottlenecks we identify? We compare four approaches on the Spanish grid using the same 480 projected heatwave hours.
AC-OPF uses fixed transmission limits, while Quad-OPF approximates their temperature dependence. Iter-OPF is the study's two-update method; TD-OPF continues the temperature-dependent calculation until the results stabilise, providing a convergence reference.
In this comparison, AC-OPF and Quad-OPF yield less load shedding but allow some lines to exceed the modelled 90°C thermal limit. Iter-OPF captures the additional constraints and closely matches the TD-OPF reference, with a shorter running time. The important difference lies in local thermal bottlenecks, even though average line temperatures are similar.
What the simulations show
The Spanish case shows why local thermal limits matter. Extending the analysis to eight country-level grid representations reveals how stress varies with weather, network structure, generation, demand and storage. Each grid is evaluated over 480 projected heatwave hours.
The temperature distributions show that the represented grids experience different heat conditions across the projected scenarios.
The French, Spanish and Italian grid models show greater load shedding, while several other represented systems remain less affected under the tested conditions.
The available line-capacity distributions vary across grids, reflecting differences in weather exposure and the represented transmission networks.
What the results mean
Heat can constrain electricity supply through demand, generation and transmission at the same time. Considering these effects together reveals local bottlenecks that simpler grid models can miss, while cross-border connections provide relief only where neighbouring systems and connecting lines retain spare capacity.
The findings describe the represented grids under the tested heatwave conditions, not predictions of national outages. They depend on the public-data infrastructure baseline and the climate, demand and operational assumptions. Future investment and changes in grid operation could alter these outcomes.
More detailed grid and cost data would support comparisons of adaptation options, including demand response, transmission reinforcement, targeted storage and cross-border support. Alternative climate scenarios and multi-period modelling could extend the assessment to other weather conditions and storage operation over time.
Reproducible examples
Three short examples follow the same Spanish 2030 heatwave hour, from weather inputs through grid operation. The final notebook runs a single OPF using the bundled inputs.
Notebook 01
Inspect hourly air temperature, wind and sunlight across the selected 2030 day.
View on GitHubNotebook 02
See the modelled demand profile, generator availability and weather-dependent line limits.
View on GitHubNotebook 03
Run one temperature-dependent OPF and compare its result with the saved map.
View on GitHub