Differential effects of climate change on average and peak demand for heating and cooling across the contiguous USA

Extreme weather events pose an operational risk to infrastructure systems and the humans who depend on them and are a major cause of power outages and energy price spikes across the United States. Hot (cold) temperatures create a demand for cooling (heating), which in turn drive demand for energy. For example, Winter Storm Uri, which caused cascading failures through interconnected and interdependent infrastructure systems as well as loss of human life in Texas in 2021, was caused not only by supply-side failures of the energy system but also by unanticipated surges in demand for heating. Similarly in August 2020, an extreme heat wave in California caused surging demand for cooling, leading the grid operator to institute rolling blackouts. This problem is not limited to the electricity sector. Severe winter weather in New England can lead to scarcity-driven spikes in wholesale prices of electricity and natural gas. At present, peak electric load events across the contiguous United States (CONUS) occur during the summer months and when high temperatures lead to demand for electricity to power air-conditioning. A large fraction of energy demand for heating during the winter is met by gas or oil furnaces. However, modeled pathways to deep decarbonization typically require electrification of sectors including building heating, which may lead to peak demands for electricity during winter cold spells. Because winter temperatures are typically farther from a thermal comfort level than summer temperatures, electrification of space-heating will change the seasonality of electricity demand, with large portions of the United States projected to become winter peaking systems. Thus, a key question is how climate variability and change will affect peak demands for heating and cooling in an electrified future. Theory and climate models indicate anthropogenic climate change drives robust increases in surface temperatures globally; if variability were unchanged, heating demand would decrease and cooling demand would increase. However, warming trends are accompanied by changes in the severity and duration of extreme events such as heat waves, while shifts in cold extremes remain uncertain. In this paper, the authors present a retrospective, spatially explicit analysis of trends in heating and cooling demand using temperature-based proxies over 72 years (1950–2021) across the CONUS, quantifying changes to annual average and annual maximum (peak) thermal demands and exploring implications for grid reliability and economics. They identify a north–south divide in emergent patterns, especially for the ratio of average to peak demands and the relative importance of peak cooling versus peak heating. To aggregate findings to decision-relevant scales, trends are estimated for major electric grid systems, highlighting Florida and the Midcontinent Independent System Operator (MISO) as archetypes of hot and cold regions.

Data: Hourly 2-meter temperature from ERA5 reanalysis at 0.5°×0.5° over CONUS (3267 grid points), 1950–2021. Population data from Gridded Population of the World v4 (GPWv4), using 2020 population, aggregated to the ERA5 grid. Electric grid sub-regions derived from EPA eGRID shapefiles, modified to better align with grid operators (e.g., merging NYC, Long Island, and New York State into NYISO). Regions analyzed include AZ/NM, CAISO, ERCOT, Florida, Wisconsin (Rural), MISO, ISO-NE, Northwest, NYISO, PJM West, Michigan, PJM East, Colorado, Kansas, Oklahoma, Arkansas/Louisiana, Missouri, Southeast, Tennessee Valley, and Carolinas. Inferred thermal demand definitions: Using a comfort threshold of 65°F (18.33°C), hourly inferred heating demand HD_i,t = max(65 − T_i,t, 0); cooling demand CD_i,t = max(T_i,t − 65, 0); total thermal demand TTD_i,t = |T_i,t − 65|. Results are robust to reasonable alternative thresholds (e.g., 68°F). Population-weighted aggregation: For each grid region and hour t, compute population-weighted inferred heating, cooling, and total thermal demand by summing grid-cell values multiplied by the cell’s fraction of 2020 population within the region. This fixes population weights to isolate temperature-driven trends. Peak events: Define annual peak inferred demand intensities as the annual maxima of moving-window averages over specified durations (mainly 72 hours; additional analyses for 6–336 hours). For cooling, annual cycle is January–December (1950–2021); for heating and total thermal, annual cycle is September–August (1951–2021) to maintain winter season continuity. Trend detection: Use the nonparametric Mann–Kendall (MK) test (two-sided, α=5%) to assess monotonic trends and Theil–Sen estimator to quantify slope (robust to non-normality and outliers). Field significance: Employ a bootstrap field significance test that resamples the entire spatial field by time (1000 samples) to account for spatial correlation; compare observed count of significant grid cells to the (1−α) percentile of bootstrap distribution to reject the null of no field-wide trend. Load factor: Define annual thermal load factor as the ratio of annual mean thermal demand to annual peak thermal demand (peak based on 72-hour window). Analyses conducted at grid-cell scale and aggregated to grid sub-regions (population-weighted) for decision relevance.

Annual mean demand trends: • Across the CONUS, annual mean inferred cooling demand increased while annual mean inferred heating demand decreased over 1950–2021. • Total thermal demand generally decreased across most of CONUS because winter heating magnitudes typically exceed summer cooling, but southern hot regions (parts of Florida, Arizona, Texas, Southern California) show increases where rising cooling outweighs falling heating. • Largest increases in mean cooling demand occur in central Colorado (~1%/year); largest decreases in mean heating demand occur in southern Florida (below −1%/year). • Field significance tests reject the hypothesis of no trend for cooling, heating, and total thermal demand. Aggregated grid entities: • Florida: Average cooling exceeds heating; total thermal demand exhibits a statistically significant increasing trend. • MISO (Midcontinent): Average heating dominates; warming-driven decreases in heating lead to a net decrease in total thermal demand, representative of northern grids. Peak demand (72-hour events unless noted): • Peak cooling demand intensity increased across most of CONUS; median trend 0.16%/year (range −0.25 to 1.77%/year). Largest trends in central Colorado (>1%/year). Statistically significant increases found in most of the western US, New England, New York, Florida, Louisiana, Pennsylvania, and large parts of Texas, Virginia, and North Carolina; small decreases in the Dakotas. Patterns persist across other event durations. • Peak heating demand intensity decreased across almost all of CONUS; median −0.1%/year (range −0.41 to 0.03%/year). Few areas with significant increases; significant decreases mostly in Southern California and parts of the southwest and southeast (areas where winter heating is relatively low). Patterns consistent across durations. • Peak total thermal demand intensity shows a median −0.1%/year (range −0.37 to 0.20%/year) with decreases over most of CONUS, driven by decreases in peak heating intensity, but increases in southernmost Florida, Texas, Arizona, and California where peak cooling increases outweigh peak heating. • No systematic shifts in the seasonality or day-of-year occurrence of peak heating and cooling events were found. Variability and regional contrasts: • Peak heating exhibits much larger inter-annual and decadal variability than peak cooling. In Florida, peak cooling increasingly dominates after 2010, but rare cold outbreaks generate exceptional peak heating years. In MISO, peak total thermal events are exclusively winter heating peaks with increasing inter-annual variability after ~1980. Load factors (mean/peak): • Thermal load factor trends are heterogeneous; median 0.01%/year with range (−0.17, 0.42)%/year. • Spatially coherent positive trends occur in the Southeastern US (statistically significant in southernmost regions). Florida shows increasing load factors with large decadal variability; MISO shows a mild, generally decreasing trend (notably a decrease in the 1950s). • Northern CONUS and some western mountain regions show decreasing load factor trends (significant in parts of California and the Great Lakes), indicating mean thermal load is decreasing faster than peak, implying lower utilization. In some areas of Southern California and Arizona, decreasing load factors are driven by peak increasing faster than mean. • Field significance tests generally reject the null of no trend for peak cooling, peak heating, and peak total thermal demand intensity.

The study addresses how climate variability and change differentially affect average and peak heating and cooling demands relevant to an electrified future. While average temperatures have risen, leading to broadly decreasing annual mean heating demand and increasing cooling demand, the behavior of extremes is more nuanced. Peak cooling demand has risen robustly across many populated and western/northeastern regions, consistent with more intense/longer heat events. In contrast, peak heating demand exhibits smaller, less robust decreases and pronounced inter-annual variability, reflecting the long left tail of the temperature distribution where rare severe cold snaps dominate extremes. These findings imply that even as mean heating needs decline, electrification of space heating could shift or intensify winter peak electricity requirements in many regions. Operators must plan for rare but severe cold events to ensure reliability, especially in northern systems such as MISO where peak events are dominated by winter heating and variability has increased since about 1980. The north–south divide in utilization trends (load factors) has implications for system economics: southern grids (e.g., Florida) may see higher utilization (mean growing faster than peak), while northern grids face declining utilization (mean declining faster than peak), which can raise average costs. High variability in winter peaks underscores the value of improved seasonal forecasting, operational flexibility (e.g., scheduling maintenance), and adequate winter reserve capacity to mitigate risks exemplified by events like Texas’s 2021 Winter Storm Uri.

Significant climate-driven changes in thermal loads have occurred across the CONUS. Average winter heating demand is decreasing while average summer cooling demand is increasing. Peak loads display less consistency: peak cooling loads are increasing in many population-dense regions, whereas peak heating loads are relatively unchanged or modestly decreasing, with high inter-annual variability. Capacity utilization trends diverge geographically: in northern regions where heating dominates, average heating demand is decreasing faster than peak, producing declining load factors; in southernmost regions where cooling dominates, average cooling demand is increasing faster than peak, producing rising load factors. If heating electrification proceeds and these trends persist, northern regions may face increasing costs to maintain reliability due to lower utilization, whereas southernmost regions may see decreasing costs. The results reflect temperature-driven changes only and are a precursor to future-climate projections using models. Extrapolation beyond CONUS, particularly to rapidly air-conditioning-adopting cities in the global South, is complicated by differing technology adoption trends. Overall, planning for winter reliability under electrification requires accounting for rare, severe cold snaps and leveraging improved seasonal forecasts and operational strategies.

The analysis uses temperature-based proxies for inferred heating and cooling demand with a fixed comfort threshold (primarily 65°F) and does not incorporate additional nonlinear drivers of peak loads such as humidity (especially relevant for cooling), building characteristics and envelope performance, occupancy, demographics, installed heating/cooling technologies and capacities, or price elasticity. Population weights are fixed to 2020 to isolate temperature effects; changes in population distribution over time are not modeled. Place-specific nonlinear responses and technology differences limit spatial generalization of more complex demand models across CONUS. Results focus on temperature-induced changes and may not capture all determinants of electricity demand. Extrapolation outside CONUS is limited, particularly in regions with rapidly increasing air-conditioning adoption. Trend detection is based on nonparametric methods applied to reanalysis data and is subject to uncertainties in reanalysis and observational inputs (e.g., potential issues noted in high-elevation Colorado stations).