Designing a sector-coupled European energy system robust to 60 years of historical weather data

TL;DR

The paper tackles the challenge that long-horizon infrastructure planning for a highly renewable, sector-coupled Europe is vulnerable to interannual weather variability, which is often ignored when using a single weather year. Using the open PyPSA-Eur model, the authors perform a greenfield joint capacity and dispatch optimization over 62 design years to achieve net-zero $CO_2$ emissions, then fix capacities and dispatch them across the remaining 61 weather years to assess robustness. They find that total system costs vary by about $\pm 10\%$ across design years, with compound weather events driving more robust and cost-effective capacity layouts, and that CO$_2$-emitting backup generation can be a cost-efficient robustness measure with average emissions well below 1% of 1990 levels. The study demonstrates that a sector-coupled European energy system can be designed to withstand six decades of historical weather variability, informing transmission planning, backup capacity, and carbon-management strategies for a resilient decarbonization path.

Abstract

As energy systems transform to rely on renewable energy and electrification, they encounter stronger year-to-year variability in energy supply and demand. However, most infrastructure planning is based on a single weather year, resulting in a lack of robustness. In this paper, we optimize energy infrastructure for a European energy system designed for net-zero CO$_2$ emissions in 62 different weather years. Subsequently, we fix the capacity layouts and simulate their operation in every weather year, to evaluate resource adequacy and CO$_2$ emissions abatement. We show that interannual weather variability causes variation of $\pm$10\% in total system cost. The most expensive capacity layout obtains the lowest net CO$_2$ emissions but not the highest resource adequacy. Instead, capacity layouts designed with years including compound weather events result in a more robust and cost-effective design. Deploying CO$_2$-emitting backup generation is a cost-effective robustness measure, which only increase CO$_2$ emissions marginally as the average CO$_2$ emissions remain less than 1\% of 1990 levels. Our findings highlight how extreme weather years drive investments in robustness measures, making them compatible with all weather conditions within six decades of historical weather data.

Figures (73)

• Figure 1: Total annual system costs for the capacity optimization using as input weather years from 1960 to 2021. The figure shows (a) the Europe-aggregate system costs split by technology, (b-f) the total system costs correlated with the Europe-aggregate weather-dependent variables for the corresponding years, and (g) violin plot of the total system cost with annotations of the 25% (q1), 50% (Med.), and 75% (q3) quantiles. In a-f, the solid lines depict the linear regression, while the coefficient of determination r and p-values are shown in the lower left corner. See Supplemental Fig. \ref{['sfig:co2_emissions_price']} for the distribution of the CO$_2$ emissions prices derived from the capacity optimization.
• Figure 2: European aggregate robustness metrics summarized by (a) loss of load aggregated for every coherent hours with unserved energy (sequential unserved energy) for Europe in all operational years averaged for all capacity layouts, (b) peak loss of load across all 3 hour time steps and (c) net CO$_2$ emissions for all operational years (red markers) run for the different capacity layouts (design year). The black lines in (b) and (c) indicate the average of all operational years, and gray annotations indicate the operational year causing the highest peak loss of load and net CO$_2$ emissions. See Supplemental Fig. \ref{['sfig:unserved_energy_avg']} for the loss of load in every 3 hour timestep, Supplemental Fig. \ref{['sfig:CO2_emissions_t']} for the net-CO$_2$ emissions, and Supplemental Fig. \ref{['sfig:dispatch_optimization_all']} for distribution plots of cumulative and sequential unserved energy.
• Figure 3: CO$_2$ emissions split by technology for every operational year, averaged over all capacity layouts. For the CO$_2$ emissions in every capacity layout, see Supplemental Fig. \ref{['sfig:CO2_emissions']}.
• Figure 4: Wind resources in the operational year (1968) causing the highest peak loss of load. See Supplemental Fig. \ref{['sfig:wind_December']} for the nodal wind resources in the loss of load event occuring in December. See Supplemental Figs. \ref{['sfig:unserved_energy_and_renwable_droughts_EU_1966']}-\ref{['sfig:unserved_energy_and_renwable_droughts_EU_2010']} for the same depiction with wind and solar PV resources, hydro inflow, and heating demand in 1966, 1968, 1972, 1996, and 2010.
• Figure 5: Geographical distribution of the annual unserved energy obtained with different capacity layouts (design years) simulated with all operational years. The top-left subfigure depicts the average unserved energy in all capacity layouts for every operational years. Furthermore, we show the two capacity layouts obtaining the lowest peak loss of load (1966 and 1968), and the most expensive system (2010). The annotations indicate the number of hours with average electricity demand (average hourly load; av.h.l.) uncovered per year in every country. See Supplemental Figs. \ref{['sfig:map_unserved_energy_all_designs']} and \ref{['sfig:map_unserved_energy_all_operations']} for all capacity layouts and all operational years.