Impacts of long-term temperature change and variability on electricity investments

The paper investigates how long-term changes in temperature and its spatial-temporal variability affect electricity demand profiles, installed capacity, and capital investments in the United States. Prior work often focused on annual electricity demand or did not integrate socioeconomic evolution and multisectoral interactions. The authors aim to provide a comprehensive assessment that captures subannual (monthly, day/night, and peak) dynamics, incorporates socioeconomic drivers (population, income), and reflects competition among fuels and sectors. The study addresses key gaps by linking temperature-driven changes in heating and cooling service demands to subannual load profiles and consequent capacity expansion and investment needs, highlighting the importance of planning for peak loads rather than only annual totals.

Previous studies have shown temperature affects building energy needs and electricity systems globally and within countries. However, three main limitations recur: (1) multisector models typically use annual demands and miss subannual peak impacts; (2) econometric/empirical studies often exclude evolving socioeconomics, technology, prices, and infrastructure; and (3) power-sector tools may not capture multisectoral interactions and broader market dynamics. The literature indicates potential nontrivial investment impacts from warming, but a comprehensive, subannual, multisectoral, socioeconomic-integrated assessment at state-level granularity for the U.S. has been lacking. This study builds on and extends prior work by endogenizing subannual load responses to temperature and embedding them within a multisector modeling framework with state-level detail.

Scenarios and framework:

• Core scenarios: RCP 8.5 with SSP2-style socioeconomic and technology assumptions, run in two variants: (a) without temperature impacts (constant climate), and (b) with temperature impacts affecting subannual loads and investment needs. Sensitivity cases adopt SSP3 (low population/economic growth) and SSP5 (high population/economic growth) while similarly toggling temperature impacts.
• Modeling tool: Modified GCAM-USA (v4.1) with state-level detail, improved power-sector representation, and fully endogenized temperature impacts on buildings’ heating and cooling service demands, subannual electricity load profiles, and capacity/investment requirements. The model is myopic, dynamic-recursive, partial equilibrium.

Subannual load representation and dispatch:

• Each model year is divided into 25 time segments: day and night for each month (24 total) plus one "super-peak" segment representing the top 10 load hours. Segments respect chronology; night segments prohibit dispatch of solar-based technologies. Segment durations reflect state-specific daylight hours.
• Electricity demand includes buildings, transport, and industry. Buildings’ heating/cooling demands respond endogenously to temperature in each segment; other sectoral loads follow fixed shapes.
• Capacity is dispatched via a linear least-cost routine (fuel, O&M, variable costs). Total capacity must be at least 15% above the super-peak load (reserve margin), consistent with NERC and other U.S. planning models.
• New capacity choices across four investment segments are determined via a nonlinear logit approach based on levelized capital, O&M, and fuel costs; fuel costs are endogenous from supply curves.

Socioeconomic drivers and multisector interactions:

• Exogenous annual GDP and population by state drive floorspace (buildings) and service demands in other sectors; these in turn affect energy prices and competition across fuels (electricity, gas, liquids, biomass, hydrogen). The model tracks capital stock vintages and retirements.

Electricity trade:

• Within-grid-region trade is unconstrained with common prices; interregional trade across 15 grid regions uses a nonlinear logit-based formulation calibrated to historical net trade and implicit transmission capability. Future trade responds to relative regional price differentials, with increasing marginal costs to expand trade.

Temperature, HDH/CDH, and climate data processing:

• Hourly, gridded (20 km) temperatures from RESM (WRF atmosphere + CLM land) forced by CESM-CCSM4 (CMIP5) under RCP 8.5 for 2005–2100. Historical downscaling for 1975–2004. Bias-corrected via BCSD (quantile mapping vs. NLDAS-2), with trend-preserving correction applied to future residuals after linear trend removal.
• Heating Degree Hours (HDH) and Cooling Degree Hours (CDH) use a 65°F (18°C) threshold, computed per hour and aggregated to each time segment. HDH/CDH are population-weighted and aggregated to the state level and fed into GCAM-USA.
• Buildings’ heating and cooling energy demand per unit floorspace in each segment uses calibrated functional forms dependent on HDH/CDH, thermal conductance, internal gains, surface-to-floor ratio, income, and service prices. Calibration constants match historical final-year energy use and are held fixed forward.

Evaluation and outputs:

• The model computes monthly day/night electricity loads, peak/mean loads, technology-specific capacity additions, and generation by segment. RCP 8.5 with and without temperature impacts are compared to quantify temperature-induced changes in generation, installed capacity, and cumulative capital investments (2015 USD, undiscounted). Sensitivities under SSP3 and SSP5 quantify dependence on socioeconomic trajectories.

• Under SSP2/RCP 8.5, mean temperature changes increase annual electricity generation by about 5% nationally in 2100 (0.5–8% across states) relative to a constant-climate baseline. However, installed capacity rises by 14% and cumulative capital investments by 16% (ranges across states: capacity +3–20%; investments +3–22%).
• Temperature-induced cumulative capital investments total roughly USD 1 trillion nationally (undiscounted 2015 USD) over 2015–2100, averaging about USD 10 billion per year; national installed capacity requirements are driven by peak load increases rather than mean demand.
• Subannual impacts: The super-peak segment load increases by approximately 150 GW due to higher cooling loads, necessitating significant additional capacity. Cooling accounts for a substantial share of peak loads (e.g., 34% of super-peak in 2015 and 26% in 2100).
• Dispatch dynamics: Added coal, nuclear, and renewable capacities (low variable costs) are dispatched across all segments, displacing some gas generation in lower-demand segments.
• Fuel-specific responses: Largest capacity increases occur in gas, followed by coal and solar. Largest investment increases occur in gas, then solar, wind, and coal.
• Spatial heterogeneity: Highest temperature-induced increases in 2100 installed capacity and cumulative investments are concentrated in California (≈18 GW; ≈USD 71B), Illinois (≈13.6 GW; ≈USD 67B), Pennsylvania (≈13.8 GW; ≈USD 57B), and Texas (≈13.2 GW; ≈USD 39B). Patterns reflect differing socioeconomic growth, regional fuel prices, existing stock and turnover, and renewable resource quality.
• Electricity trade as flexibility: Capacity and investment hotspots do not map one-to-one to peak CDH changes due to interstate and interregional trade. Example: Texas electricity exports in 2100 rise by 128% with temperature impacts, indicating capacity added in TX helps meet demand elsewhere.
• Sensitivity to socioeconomic pathways: Temperature-induced investments vary strongly with population/economic growth. Relative to SSP2, SSP5 increases temperature-induced investments by roughly 34–65% across states; SSP3 reduces them by ~27–36%. Nationally under SSPs, temperature-induced changes (SSP2/SSP3/SSP5) are: installed capacity +241/+125/+404 GW; cumulative investments ≈USD 993/669/1480 billion (undiscounted).

Findings demonstrate that focusing solely on annual electricity demand understates the economic impact of warming on the power sector. Peak load growth from higher temperatures drives disproportionate increases in required capacity and capital investments. Integrating subannual load dynamics, regional heterogeneity, socioeconomic evolution, and multisector interactions is essential for robust long-term capacity expansion planning. Electricity trade can mitigate localized impacts by shifting where capacity is added, decoupling capacity build locations from local peak temperature changes. The results also show strong sensitivity to socioeconomic pathways, implying that future demographic and economic trajectories will significantly influence the scale and composition of temperature-induced investments and capacity expansion.

This study advances electric sector planning by endogenizing temperature-driven, subannual load dynamics within a multisector, state-level framework (GCAM-USA), linking heating/cooling service demands to capacity and investment outcomes. Under SSP2/RCP 8.5, temperature changes modestly raise annual generation but substantially increase installed capacity and cumulative investments, with notable spatial heterogeneity and a central role for trade. The work highlights the need to plan for peak loads, consider socioeconomic uncertainty, and account for multisectoral interactions. Future research directions include: exploring alternative climate scenarios (e.g., RCP 4.5) and their system-wide transformations; incorporating state-level policies (e.g., RPS, SB100, CAFÉ) and electrification trends; assessing different reserve margins; evaluating expanded transmission and alternative trade patterns; quantifying roles for storage, load-levelling, and demand response; and examining EV penetration and charging strategies on temperature-induced peak impacts.

• Climate scenario scope: Core analysis uses a single climate model realization (CESM-CCSM4) under RCP 8.5; results may differ with warmer/cooler ensemble members or alternative RCPs.
• Impact scope: Focuses on temperature effects on buildings’ heating/cooling services and subannual electricity load profiles; does not quantify climate impacts on energy resources, agricultural yields, water availability, or other sectors.
• Policy exclusion: Does not include state-level policies (e.g., SB100, RPS) or transportation policies (e.g., CAFÉ), which could alter technology mix and investment patterns.
• Modeling assumptions: Fixed 15% reserve margin; myopic, dynamic-recursive decision-making; fixed load profiles for transport and industry; unconstrained intra-regional trade and calibrated interregional trade; absence of explicit storage and demand response modeling.
• Uncertainty analysis: No full uncertainty exploration across combined SSP-RCP spaces; sensitivity shown only for population/economic growth (SSP3/SSP5).
• Economic evaluation: Reported capital investments are undiscounted (with discounted values in supplemental materials); results sensitive to discount rate assumptions.