Dual Impacts of Space Heating Electrification and Climate Change Increase Uncertainties in Peak Load Behavior and Grid Capacity Requirements in Texas

Around 60% of households in Texas currently rely on electricity for space heating. As decarbonization efforts increase, non‐electrified households could adopt electric heat pumps, significantly increasing peak (highest) electricity demand in winter. Simultaneously, anthropogenic climate change is expected to increase temperatures, the potential for summer heat waves, and associated electricity demand for cooling. Uncertainty regarding the timing and magnitude of these concurrent changes raises questions about how they will jointly affect the seasonality of peak demand, firm capacity requirements, and grid reliability. This study investigates the net effects of residential space heating electrification and climate change on long‐term demand patterns and load shedding potential, using climate change projections, a predictive load model, and a direct current optimal power flow (DCOPF) model of the Texas grid. Results show that full electrification of residential space heating by replacing existing fossil fuel use with higher efficiency heat pumps could significantly improve reliability under hotter futures. Less efficient heat pumps may result in more severe winter peaking events and increased reliability risks. As heating electrification intensifies, system planners will need to balance the potential for greater resource adequacy risk caused by shifts in seasonal peaking behavior alongside the benefits (improved efficiency and reductions in emissions).


Introduction
Meeting 2050 economy-wide net zero emissions targets will require multiple simultaneous changes in energy infrastructure, including switching to low-carbon power sources, implementation of carbon capture utilization and storage (CCUS), improvements in energy efficiency, battery storage, and electrification of demand-side energy use (Horowitz, 2021;IEA, 2021;Williams et al., 2012).As of 2021, 13% of direct United States (U.S.) greenhouse gas emissions came from the residential and commercial sectors (U.S. EPA, 2023a), with one of the primary sources being the burning of fossil fuels for building heating (U.S. EPA, 2023b).Heating and cooling (air conditioning) jointly account for the largest annual electricity use in the residential sector (U.S. EIA, 2023).Compared to some other sectors, the building sector can be easier to electrify (Sugiyama, 2012), achieved in part via growth in the adoption of electric heat pumps (HPs) for space heating and cooling (Mai et al., 2018;Sugiyama, 2012).HPs are generally more energy efficient than other heating technologies (Sarbu, 2021;Vorushylo et al., 2018) (such as electric resistance heaters, furnaces, and boilers), and they have the added advantage of being dual-purpose (providing heating in the winter and cooling in the summer).This benefit could bring cost savings to consumers (Joe et al., 2021;Nerkar & Ngo, 2023).According to 2020 U.S. data, only about 36% of all space heating used in the residential sector is electric and 13% of all homes use HPs (U.S. EIA, 2020a).More widespread adoption of HPs could be fundamental to achieving deep decarbonization goals (Gaur et al., 2021;Walker et al., 2022).Yet, alongside the potential benefits of heating electrification via HPs, an open question is how they will affect demand dynamics and consequently, the ability of the power sector to reliably supply the electricity demand given the meteorological uncertainties.
At present, most U.S. regions experience peak electricity demand in the summer (Keskar et al., 2023).This is partly a result of large air conditioning demands in the summer, and the fact that winter heating demands are mostly met by direct combustion of fossil fuels (BDC, 2023; U.S. EIA, 2020b; Waite & Modi, 2020).While projected impacts vary by region and technology type (Fairey et al., 2004;Waite & Modi, 2020), several previous studies have shown that space heating electrification could be a major driver of changes in load shapes and peak electricity demand (Bistline et al., 2021), including shifts in the timing of peak demand occurrence from summer to winter (Mai et al., 2018;White et al., 2021a).One previous analysis (Waite & Modi, 2020) found that fully electrifying space heating with HPs could require a 70% increase in total U.S. generation capacity (increases also likely to vary by region).This degree of electrification could result in a seasonal shift in the regional peak electricity demand, with important implications for long-term planning in the electric sector (Bistline et al., 2021;Keskar et al., 2023;Vorushylo et al., 2018).For example, switching to winter peaking could mean less peak demand coinciding with peak solar generation, which could put more pressure on system operators to maintain existing firm capacity or build more.
In addition to changes brought on by the electrification of energy usage, climate change is simultaneously exerting a strong influence on energy demand (Gi et al., 2018;Yalew et al., 2020), and ignoring these compounded impacts could significantly reduce the quality of load forecasts and the adequacy of long-term system planning efforts (Lee & Dessler, 2022).There is a well-established literature on the impacts of climate change on energy demand (Isaac & van Vuuren, 2009) and on the bulk electric grid (Amonkar et al., 2023;Jaglom et al., 2014;Romitti & Sue Wing, 2022).Due to increasing ambient air temperatures, most studies project that climate change will cause a decrease in total heating demand, an increase in total cooling demand (Amonkar et al., 2023;Ranson et al., 2014;U.S. EPA, 2023c), and an increase in peak load (Auffhammer et al., 2017), though here again the magnitude of these impacts is likely to be region dependent.
The effects of climate change and space heating electrification on electricity demand have largely received separate attention in the literature.There is limited research on the joint, net effects of simultaneous residential space heating electrification (using HPs) and climate change on the seasonality of peak load events and the corresponding impacts on grid reliability.Furthermore, most studies that have investigated the effects of climate change or heating electrification on electricity demands have relied on aggregated, statistical estimates of building parameters and energy use, rather than using detailed, hourly resolved building stock models to generate load profiles (Amonkar et al., 2023;Bistline et al., 2021;Buonocore et al., 2022;Eyre & Baruah, 2015;Gi et al., 2018;Pineau et al., 2022;Skiles et al., 2023;Steinberg et al., 2017).
We identified one previous study (Tarroja et al., 2018) that considered the joint effects of climate change and heating electrification on electricity demands while also taking a more mechanistic modeling approach.However, that study assumed a more futuristic, renewable-dominated grid (making it difficult to experimentally isolate the impacts on reliability caused by HP adoption and climate change).In addition, that study also only explored a small range of climate change and technology scenarios, and projected outcomes for a single future weather year.Finally, in no case could we find an example of a previous study that uses future demand projections of HP adoption (augmenting the HP penetration to replace all un-electrified home space heating) and climate change to estimate reliability impacts (load shedding) using operational models of the power grid.
In this paper, we attempt to exhaustively quantify the net impacts of climate change and HP adoption on peak and total load, the seasonality of peaking, and system reliability.We focus on the Electric Reliability Council of Texas (ERCOT) grid as a case study due to growing concerns around its sensitivity not only to extreme heat but also to extreme cold (Shaffer et al., 2022;Skiles et al., 2023).ERCOT is the independent system operator for the majority of the Texas region, managing about 90% of the state's electric power grid, and servicing more than 26 million customers (ERCOT, n.d.).
To estimate demand, we fit four machine learning (ML) models to hourly system-wide residential load data that were generated using mechanistic building stock models in a previous study (White et al., 2021a).The data we are leveraging consist of hourly annual time series of ERCOT residential electricity demand for 4 technology scenarios: three 100% space heating electrification scenarios, each assuming a different Air Source Heat Pump (ASHP) efficiency, and one model of the base case (57% electrified space heating in 2016) (White et al., 2021a).The HP scenarios electrify the 40% of households that currently use fossil fuels for space heating while maintaining the existing 60% electrified residential space heating (primarily electric resistance) (White et al., 2021a).All four scenarios use the historical 2016 weather year as the meteorological forcing.ASHPs are the most common HP type (U.S. DOE, n.d.) and can be a direct alternative to buildings that currently use fossil fuels (Kelly & Cockroft, 2011;Sarbu, 2021).They are more cost competitive in areas with milder climates, and their performance declines under colder ambient temperatures (Deetjen et al., 2021;Kaufman et al., 2019).The HP efficiencies modeled in this paper include the Standard Efficiency, High Efficiency, and Ultra-High Efficiency, with Seasonal Energy Efficiency Ratings (SEER) values of 15.0, 22.0, 29.3, and Heating Seasonal Performance Factors (HSPF) of 8.5, 10.0, and 14.0, respectively.For reference, effective in 2023, the U.S. Department of Energy residential standards require a minimum SEER of 14 and 15 in the Northern and Southern U.S. respectively and a minimum HP HSPF of 8.8 (U.S. EIA, 2019).SEER and HSPF are measures of cooling and heating performance respectively.We also fit a fifth ML model to hourly non-residential load data for 2016.This was obtained by deducting the base case simulated residential demand (White et al., 2021a) from ERCOT's reported 2016 historical total load.We then force the 5 ML models with climate change data (Figure 1) to predict changes in aggregated ERCOT electricity consumption, both residential and total, under climate change conditions.For the climate change simulations, we use an open-source Thermodynamic Global Warming (TGW) meteorological data set (Jones  (Burleyson et al., 2023).Top: ERCOT maximum hourly annual population-weighted temperature under historical and future climate change scenarios RCP 8.5 Hotter and RCP 4.5 Cooler.Bottom: ERCOT minimum hourly annual population-weighted temperature under historical and future climate change scenarios RCP 8.5 Hotter and RCP 4.5 Cooler.Representative Concentration Pathways (RCPs) 4.5 and 8.5 are projections of plausible future climate conditions under different levels of anthropogenic activity.Hotter and Cooler are distinctions of High-and Low-sensitivity models respectively under the same RCP (Jones et al., 2023).Shared Socioeconomic Pathways (SSPs) track different socioeconomic scenarios.Historical U.S. population projections are based on the U.S. census; future populations are consistent with the SSPs (Jiang et al., 2020;Zoraghein & O'Neill, 2020).Also see Table 1.
Earth's Future 10. 1029/2024EF004443 et al., 2022, 2023)).The future TGW data comprise four Representative Concentration Pathways (RCP) scenarios, each based on RCP 4.5 or RCP 8.5, paired with "Hotter" or "Cooler" model sensitivities, and spanning 80 years (2020-2099).The ML models took these 320 future simulation years of meteorology, and given the four electrification scenarios, were used to predict 1,280 divergent future simulation years of hourly load for the ERCOT service territory.The load data that come out of the ML models are then passed through an open-source direct current optimal power flow (DC OPF) production cost model (Akdemir et al., 2024) for ERCOT to simulate hourly operations of the ERCOT bulk power system.We use the DC OPF model to assess the impacts on grid reliability (load shedding) under the different futures/pathways.Results are reported in terms of seasonality and magnitude of peak load, electricity demand, and impact on loss of load frequency.See the Methods section (Figure 2) for an in-depth discussion of model workflows.
Our findings shed light on the potential long-term impacts of full residential electrification of space heating under climate change.Specifically, there are clear benefits to HP adoption in terms of decarbonization of building energy use and reducing air conditioning loads, especially under the hotter climate scenarios.HPs help promote system reliability under hotter climates, but they could also increase the grid's vulnerability to acute cold events.These dynamics are complex and depend on joint climate and technology adoption scenarios.Although this study is based on ERCOT, our approach is transferable and can easily be used across other U.S. regions.This type of analysis could be used to inform integrated resource planning efforts as the U.S. grid decarbonizes, expands, and continues to experience stress due to extreme conditions (Busby et al., 2021;Dumas et al., 2019;Stone et al., 2021).

Methodology
The experimental design is summarized in Figure 2. It first outlines the principal modeling tools and approaches used by the authors of White et al. (2021a) to generate the residential energy demand data using building stock models.This previously generated data (shaded gray in Figure 2) was central to the models developed in this technical analysis.Using these data, we develop, train, validate, and utilize predictive ML models of electricity  2021a), together with the 2016 weather data were used to fit 4 ML models under four scenarios of space heating electrification: Base, Standard Efficiency, High Efficiency, and Ultra-High Efficiency HP adoption.A fifth ML model was trained based on non-residential demand data and 2016 weather data.The 5 ML models were then used to predict electricity demand using future climate data.The output from those models was then passed through a direct current optimal power flow model for the ERCOT grid to simulate reliability impacts.
Earth's Future 10.1029/2024EF004443 demand under concurrent climate change, and technology adoption (heating electrification).These demand data are then passed through a calibrated power system operations model for ERCOT (GO ERCOT).We focus on electrification of the residential heating sector both due to data availability and because it can be a disproportionate driver of the system's peak demand (White et al., 2021a).

Previous Building Energy Data
The authors of White et al. (2021a) used the National Renewable Energy Lab's (NREL) ReStock (NREL, n.d.) analysis tool to generate a large spatially diverse building stock sample that is statistically representative of existing households in ERCOT.The EnergyPlus (Crawley et al., 2001) model took this data (building dimensions and characteristics), alongside hourly meteorological parameters, and used it to simulate hourly energy demand.
EnergyPlus is an integrated Physics-based dynamic building energy model that estimates energy usage under different physical and thermal attributes (Tarroja et al., 2018;White et al., 2021a).The year 2016 was selected for their analysis because it had a relatively hot summer and cold winter.Electric resistance heating was assumed as an auxiliary heat source for the Standard and High Efficiency HPs when the winter outdoor temperatures dip very low, with the compressor lockout temperature set to 17.8°C.Currently, most HPs come with auxiliary heating (electric resistance or backup burner) to generate supplemental heat when demand exceeds the compressor's capacity (Kaufman et al., 2019).The Ultra-High Efficiency HP in the experiment did not utilize any auxiliary heating.Details of their modeling approach can be found in White et al. (2021a).
From this previous study, we took 4 hourly "8760s", that is, hourly time series of residential load, one for each of the four different space heating electrification scenarios (base case, and HP adoption under the Standard Efficiency, High Efficiency, and Ultra-High Efficiency HP technologies), all assuming a 2016 weather year.Again, the HP adoption scenarios represent 100% electrified residential space heating for ERCOT (keeping the existing electrified residential space heating (60% primarily resistance heating) and electrifying the remaining 40% (fossil fuel heating) with HPs).

Machine Learning Prediction of ERCOT Demand
We then trained 4 separate, predictive ML models of 2016 residential demand in ERCOT.Features included: dayof-week (weekday/weekend), hour-of-day, and spatially resolved hourly time series of meteorological variables (Burleyson et al., 2023).The 4 ML models are labeled ML Model 1 to 4 in Figure 2, one for each space heating electrification scenario.
Hourly values of the 2016 non-residential load were also estimated by first subtracting the predicted base scenario hourly residential load from the actual, observed 2016 hourly load reported by ERCOT, and then training a fifth ML model to predict this difference (shown as ML Model 5 in Figure 2).Details of the data sources, predictor variables, and methods used in generating and evaluating the models can be seen in Table S1 of the Supporting Information S1.The performance metrics of the selected ML models on test data are shown in Table S2 of the Supporting Information S1.

Integrated Models of Climate, Socioeconomic Change & HPs Energy Demand
The ML demand models (trained on the full 2016 meteorology data set) were then used to predict hourly residential load for the 4 HP technology scenarios and non-residential load under four future 80-year projections of climate (2020-2099).These future climate projections span representative concentration pathways (RCPs) 4.5 and 8.5 and include two scenarios of climate model uncertainty for each RCP (cooler and hotter climate sensitivity) (Jones et al., 2022(Jones et al., , 2023)).The hourly meteorological variables used were population-weighted using population projections consistent with the Shared Socioeconomic Pathway SSP3 (Burleyson et al., 2023).Collectively, the combinatorial experiment entails the simulation of 1,280 (4 × 1 × 80 × 4) individual operating years as shown in Table 1.

Grid Operations (GO) ERCOT Model
The hourly load profiles for all 1,280 simulation years (80 future climate years, 4 heating electrification scenarios, 4 climate scenarios), along with hourly values of variable renewable energy production, were passed through the GO ERCOT model.

Model Design
GO ERCOT is an open-source framework for instantiating, calibrating, and testing reduced network, DC OPF production cost models of the ERCOT grid.It is part of the GO family of modeling software (GO WEST (Akdemir et al., 2024), GO EAST, and GO ERCOT) for the 3 grid interconnections of the conterminous US that can be used to simulate the detailed hourly operations of the grid.The models' potential applications are wide in scope, including exploring the behavior of the grid during periods of acute stress and under long-term uncertainty in climate, technology, and/or population changes.With the aim of balancing model fidelity and computational time, GO ERCOT customizes the creation of models by searching over a range of parameters such as: the number of nodes; mathematical formulation (linear programming (LP) or mixed integer linear programming (MILP)); transmission line limits; and the shape of operating reserve demand curves (ORDC) (ERCOT, 2022).GO's ability to build and calibrate reproducible models, developed based on open-source data, and with inherent flexibility in model parameterization, allows researchers to tailor model versions to the experimental design and available computational resources.GO ERCOT makes use of synthetic grid data sets from Texas A&M University (TAMU) (Birchfield et al., 2016;Texas A&M University Engineering, 2017).The data sets contain information on the full 2,000-node topology consisting of highvoltage transmission, load centers, and generation assets in the ERCOT system.Based on user-selected parameters, GO relocates assets (generation, transmission, and load), creating an equivalent, reduced (or "backbone") representation of the grid (Shi et al., 2012).Reducing network complexity reduces the number of decision variables, constraints, and runtime.GO disaggregates time series of electricity demand, solar, and wind generation from the balancing authority (BA) level to the nodal level using pre-determined nodal weights.Figure 3 shows the simplified network (a 150-node topology reduced from 2000 nodes) used in this paper.
Outputs of the DC OPF production cost models produced by GO ERCOT include hourly: locational marginal prices (LMPs) at every node; power generation at each plant; loss of load at every node; power flows on every transmission line; and voltage angles at every node.A more detailed description of the GO model scripts' functionalities can be found in Akdemir et al. (2024).The version of GO ERCOT used for this experiment was calibrated and validated using 2019 data.Additional details regarding the calibration of the model version used in this study can be found in Supporting Information S1.

Results and Discussion
The goal of our experiment is to assess the impact of transitioning from the base scenario of about 60% home space heating electrification to 100% electrified space heating via replacement of fossil fuel heating (40% of current residential heating) with HPs.This transition is analyzed in terms of the magnitude and seasonality of peak load, the total electricity demand, and changes in the reliability of the ERCOT system (quantified by loss of load).The experiment also sought to investigate the sensitivity of these results to different HP technologies, and future climate scenarios.

Broad Adoption of HPs Would Cause a Sporadic Increase in Winter Peaking Years
Figures 4 and 5 show the seasonal peaking patterns of the ERCOT residential load and total load respectively, for each of the future climate years and heating electrification scenarios.The color of each cell indicates the season of the year with the highest hourly load occurrence.In general, we find that the season in which peak load occurs is strongly influenced by the adoption of HP technologies and future climate pathways.It is also influenced by whether we consider residential load only or total ERCOT load, and by our definition of "peak" load.For example, whether peak load was categorized as: the highest hourly annual load ("Single Hour"); the mode of the season corresponding to the peak 0.5% hours of the year ("Top 0.5%"); or the top 1% hours of the year ("Top 1%").We also see differences in peaking patterns depending on the future time period in question, that is, medium term (2020-2059) and long term (2060-2099).

Residential Load Only
The top panel of Figure 4 shows results for residential load only under the assumption that the highest "Single Hour" sets peak load.Here, we find that the base scenario remains mostly summer peaking across all climate scenarios and 80 future years.When the system switches to 40% HP/60% mainly resistance heating (via either Standard, High, or Ultra-High Efficiency HPs), there is a noticeable shift from summer to winter peaking that occurs from 2020 to around the mid-2060s.Then, from the mid-2060s onwards, there is a gradual reversal back to summer peaking, as warming from climate change increases and overpowers the increase in winter demands due to HPs.The degree to which the system reverts back to a summer peaking system (i.e., the "redness" of the second half of the 21st century) is strongly influenced by the climate change scenario, with the hotter RCP 8.5 scenario showing the most consistent summer peaking behavior.The middle and bottom panels of Figure 4 show how the results can change depending on the definition of peak load.In the middle panel, we define peak load as the "Top 0.5%" of hours in a year, and in the bottom panel, it is the "Top 1%" of hours in the year.In both these panels, we see that the base scenario is generally summer peaking.
While the HP adoption scenarios do result in more frequent winter peaking relative to the baseline scenario (at least up until the mid-2060s), the effect is muted for these alternate definitions of peak load.As with the single peak hour definition, the effects of climate change intensify after the 2060s, and the system reverts back to solidly summer peaking.

Total Load (Residential + Non-Residential)
Figure 5 now considers the seasonality of peak total load in the ERCOT system.Just as in the residential load data shown in Figure 4, we see that the switch to HP adoption scenarios increases the frequency of winter peaking events.Likewise, we see that our definition of peak load matters a great deal.Assuming that peak load is equal to the highest "Single Hour" results in more winter peaking events than if we define peak load as the "Top 0.5" or "Top 1%".But, overall, it is clear that the effect of residential HP adoption on the seasonality of total ERCOT load is much less pronounced than if we just consider the effects on the residential load sector.Depending on how we define peak load, ERCOT may go from experiencing 0-3 years that are winter peaking to 2-14 years that are winter peaking.
Winter peaking events under HP adoption futures are generally shorter in duration compared to summer peaking events (i.e., heat waves last longer than cold snaps) (Figure 4).This explains why defining peak load as the top "Single Hour" most commonly yields a winter peaking grid compared to the "Top 0.5%" or "Top 1%".From a planning perspective, the difference in time dynamics between heat waves and cold snaps may influence utilities' choices about building new firm capacity, implementing demand response, planning for sufficient duration storage, or a combination of these measures to ensure reliability during winter peaking versus summer peaking years.Particularly, as more dispatchable fossil fuel generators are retired in favor of cleaner sources, proper resource adequacy planning must be put in place to ensure that the capacity replacing these resources is able to provide power during both winter and summer demand surges.
Diving deeper into the cause of these peaking patterns, Figure 6 analyzes the trend in the historical (2016) hourly time series of temperature and load, and compares them to the simulated results for a selected future year that we observed to alternate between summer and winter peaking, depending on the efficiency of the HPs adopted.The selected future year is 2058, and we examined it under an RCP 4.5 "Hotter" scenario.In the second row of Figure 6, the legends also indicate the season of peaking for each scenario.
A clear pattern is that the introduction of HPs reduces summer load but increases the winter peak load.The size of these effects, and their influence on the timing of peak load, is mediated by HP efficiency.For example, we see that both the historical ( 2016) load data and the future year 2058 under climate change are summer peaking under the base case.In both weather years, adoption of Standard and High Efficiency HPs causes the system to become winter peaking.
The winter peak occurs in December in the 2016 weather year, while in the 2058 case it occurs in January.Note that the severe cold temperatures that cause the system to peak in winter under Standard and High Efficiency HP adoption occur for only a few hours of the year.The Ultra-High Efficiency HP causes the system to remain summer peaking under the 2058 simulation year, though the difference between the highest hourly winter load and highest summer load is small (55 MW).This is because the higher degree of efficiency of the Ultra-High HP caused a considerable decrease in peak load in both the summer and the winter.This suggests that potential policy measures that do (or do not) seek to promote the adoption of higher efficiency HPs could have consequential effects on ERCOT's capacity needs.

Certain HP Efficiencies Could Introduce Intermittent Severe Winter Peaking and Make Future Planning a Challenge
Figure 7 shows the year-to-year variability in peak hourly total load, color-coded depending on the season in which the peak occurred.From the base case to the Ultra-High Efficiency HP case, there is a reduction in summer peak loads.Conversely, winter peak load frequency generally increases with HP adoption, with that increase moderated by higher HP efficiency.Winter peaking years are rare, but when they occur, peak loads can be much higher than in summer peaking years.
Total annual electricity demand also decreases when HPs are used (especially the High and Ultra-High HPs) (see Figure S2 in Supporting Information S1).This result is significant because it means using more efficient HPs could carry some emissions benefits, which would help in decarbonization efforts.It would eliminate the direct combustion of fossil fuels in homes for providing space heating, while also reducing the total energy consumption, meaning that less electricity generation would be needed (White et al., 2021a).However, when the focus is on reliability impacts, then the peak load is of more importance, because it informs system planners on how much infrastructure and firm capacity to build/make available.
The reduction in summer peak loads is due to the fact that the homes that switch to HP use would replace their current use of air conditioners (ACs) for cooling with the more energy efficient HPs, thus reducing energy use (BDC, 2023;White et al., 2021a).Additionally, higher efficiency HPs perform better at dehumidifying than the standard AC, which reduces summer energy consumption (U.S. DOE, n.d.).
Comparing across climate scenarios (columns), going from RCP 4.5 Cooler (left) to RCP 8.5 Hotter (right), summer peak loads increase.Additionally, the magnitude of peak load during the winter peaking years in the HP adoption cases is slightly reduced across columns with an increase in the climate warming intensity because the winter temperatures for the same year were not as cold.Under each of the HP technology scenarios, the highest peak load values across all 80 future years were experienced in the winter.In addition, the magnitude of these winter peaks is generally much higher than the highest summer peaks across all years.This means that as resource planners and policymakers propose to transition to cleaner future grids (with a higher renewable portfolio) and while governments push for HP adoption, planners should consider the need for increased winter firm capacity (NERC, 2023) to withstand infrequent, short duration, but very high magnitude winter peaking events that could arise due to widespread HP adoption.
Comparing all 80-year simulations of peak load under future pathways, the Standard Efficiency HP under an RCP 4.5 Cooler climate scenario had the highest hourly peak load value (91,658 MW), which occurred in the winter.It is also worth noting that, as shown in Figure S3 of the Supporting Information S1, the load factors (defined as the ratio of hourly average load to peak load for a year) in certain years are considerably lower in the HP scenarios compared to the base case, due to the sporadic yet severe winter peaking behavior of the system.Correspondingly, the number of years with winter peaks that are higher than the highest summer/fall peak as a percentage of all years (Table 2) were only as high as 9% (RCP 4.5 Cooler under the Standard and High Efficiency HP) and as low as 3% (RCP 8.5 Hotter under the High and Ultra-High Efficiency HP).The implication here is that system planners might need to build new costly firm capacity to protect against relatively rare, but much larger winter peak demand events, and then this new capacity could be underutilized.

Without Careful Planning, Widespread Adoption of Some HP Types Could Increase Reliability Risks
From among all 1,280 years (combinations of future climate and technology efficiency), we used the DC OPF model of the ERCOT system to identify the largest summer loss of load event and the largest winter loss of load event.These events are shown in Figure 8.The summer day is 3rd August 2091 under RCP 8.5 Hotter and a base case scenario.The winter day is 23rd December 2069 under RCP 4.5 Cooler, Standard Efficiency HP scenario.Plot (a) and (c) in the figure represent the nodal location and magnitude of load-shedding outages (i.e., rolling blackouts) on the ERCOT grid associated with each day, and (b) and (d) show the most extreme temperatures experienced on those days, which are concentrated in the landlocked Northeastern part of the state.Load shedding occurs at individual nodes when there is no feasible way for the DC OPF model to meet electricity demand.In these hours, the model activates a last resort, "slack" or load shedding variable that is prohibitively expensive.This allows us to quantify the magnitude of electricity load shedding that may be required to maintain nodal energy balances and system stability.
We track instances of load shedding across the full range of scenarios and simulation years on an hourly basis, giving us a detailed estimate of the net impacts of climate change and heating electrification on system reliability.
Figure 9 shows the total annual unserved load in the summer (red) and winter (blue) seasons under all scenarios (varying the HP adoption scenario and climate RCP) and future years.
The impacts of full space heating electrification (modeled here as adoption of HPs to replace the remaining 40% non-electrified heating currently in ERCOT) on system reliability are complex.In cooler climate futures (e.g., RCP 4.5 scenarios) where summer heat waves are comparatively less intense and winter cold snaps are more severe, if lower efficiency (i.e., Standard) HPs are adopted we observe large increases in total unserved load.This is due primarily to the increased severity of winter peaking events driven by electrification.
Increasing HP efficiency (e.g., moving to Ultra) reduces the size of both winter and summer loss of load events, regardless of climate scenario.However, the higher cost of this more efficient equipment (Bonk, 2024;Nerkar & Ngo, 2023) could inhibit some consumers from adopting them.The benefits of HPs, in terms of minimizing overall loss of load, are largest toward the end of the century and under hotter future climate scenarios (like RCP 8.5).
System planners have always needed to balance the dual goals of minimizing costs for consumers and meeting reliability standards.Given the (somewhat independent) uncertainties in future climate pathways and technology adoption, Figure 9 points to unprecedented challenges that system planners will face in anticipating peak load behavior and maintaining reliability.

Limitations
Our methods draw from the findings of White et al. (2021a), but go further to investigate several system-wide impacts, providing granular but likely transferrable insights.We explore the effects of climate change under the different pathways.We also assess how different HP adoption scenarios affect the reliability of the bulk electric grid, measured in terms of the frequency and magnitude of load shedding.
This experiment has some limitations relating to both data and scope.Since our model utilizes data from White et al. (2021a), the assumptions used therein also apply here.For example, the grid impacts of electrification of water heating, and electrification of household cooking are not considered.Additionally, the study only electrifies residential space heating and does not factor in the electrification of commercial heating, or changes in demand patterns from other sectors such as commercial and industrial.Just like White et al. (2021a)  electricity demand such as electrifying transportation (e.g., plug-in electric vehicles) (Mai et al., 2018).Researchers interested in these aspects are encouraged to build on our model and incorporate these factors.
The ML models were trained using a limited range of data (hourly variables for a single historical meteorological year, 2016).The results of cumulative loss of load are also sensitive to the GO ERCOT model topology chosen (Figure 3).It was noticed that when a low topology network (e.g., 75 nodes) was used, the loss of load intensity results were higher compared to a larger network (like 150 nodes).However, the effects of HP adoption and climate change were consistent regardless of topology chosen.
Also, the GO ERCOT model we used does not account for capacity expansion and/or future changes in generation resource mix, transmission, or demand response.The only variable changing from one simulation year to another in the GO ERCOT model is the load (the aim of this was to isolate and attribute the changes in peaking behavior to dual influences of HP adoption and climate change).Adding firm (dependable and dispatchable) capacity would reduce the modeled impacts of HP adoption and climate change on system reliability.The impact of adding variable renewable energy (wind and solar capacity) would improve reliability to the extent that it reduces net load (total load minus solar and wind) in peak periods (there is thus some uncertainty about wind and solar's contribution to overall system reliability).Wind and solar exhibit different expected seasonal patterns and corresponding uncertainties, so it's important to note that shifts in seasonal peaking behavior (e.g., from summer to winter) may be particularly important to predict in systems decarbonizing via wind and solar.
It is important to note that the climate change data used (Jones et al., 2022) are bounded by the assumptions inherent in a Thermodynamic Global Warming modeling approach.Namely, they assume warmer winters on average, but otherwise do not account for potential dynamical shifts in atmospheric circulation that could alter the frequency of cold snaps in the U.S.This means that our results may underestimate the impacts of winter peaking events in the ERCOT system.Finally, while this experiment explores scenarios of fully electrified residential space heating (replacing existing non-electric heating with HPs), a remaining question is how peak load dynamics would change under 100% HP adoption (i.e., replacing existing electric resistance heating with HPs, too).Broad HP adoption is recognized as a potential tool for increasing energy efficiency and reducing overall carbon emissions (Wilson et al., 2024).There is a growing effort at the policy level (federal, state, and local) to incentivize, encourage, or mandate HP installation in new buildings, and replacement of other heating equipment and air conditioning units (especially when they are older, due for upgrade/replacement) with electric HPs (BDC, 2023;ENERGY STAR, n.d.;EN-ERGY STAR, 2024;Jeff St, 2023;U.S. DOE, 2024).Given that cooling demand is the largest contributor to summer grid stress (BDC, 2023), broad HP adoption could be key to reducing summer peak loads in a warmer world and improving resource adequacy.Recent work (Wilson et al., 2024) demonstrated that 100% residential adoption of HPs could result in a reduction in average greenhouse gas emissions and energy bill reductions in a majority of households.
We do not consider 100% HP adoption in this paper due to the unavailability of modeled data (such as those provided by White et al. (2021a)), on which we rely to train our ML load prediction.However, we can estimate how our results may change under 100% HP adoption.Compared to the partial HP adoption scenarios explored in our study, we would expect 100% HP adoption to reduce winter and summer electricity demands (including the system peak), given that higher efficiency HPs would be replacing electric resistance heating and lower SEER AC units in ERCOT.
However, without explicitly modeling 100% HP adoption, it is difficult to quantify how the seasonality and magnitude of peak loads in such a scenario would compare to the current base case.This is an area for future investigation.

Conclusions
This study provides new insights on the potential outcomes of fully electrifying residential space heating on demand and reliability using a wide heterogeneous range of pathways and data at high resolutions.
The effects of adoption of HPs and climate change on peaking behavior and reliability of future grids is uncertain.
In addition to the rate of adoption, these effects will depend on climate warming intensity, HP efficiency (measured in terms of HSPF and SEER), the potential for electrification outside the residential sector, and how system planners define "peak" load.
In the ERCOT system, we observed that electrifying the remaining 40% of non-electrified residential space heating via HPs would likely reduce loss of load probability in hotter future years, because HPs reduce cooling demand.However, during relatively infrequent years that experience severe winter temperatures, lower efficiency HPs (if widely adopted) increase hourly peak load by up to 13%, thereby augmenting the risk of very large supply shortfalls.These severe winter peaking events would be more muted if higher efficient HPs were widely adopted.
The challenge is that higher efficiency HPs are currently expensive.
HP adoption could cause a switch from a generally summer peaking system in the base case to a mix of summer and winter peaking years when we define peak load as "Single Hour" peak load across all sectors.The winter peaking years are less frequent, but they can exhibit higher peak demand values relative to the base case, especially if lower efficiency HPs are adopted.If low efficiency HPs are adopted, system planners may be forced to increase winter firm capacity and plan for appropriate mitigation/adaptation strategies, such as demand response, energy efficiency, weatherization of generation infrastructure, long-duration storage, transmission expansion, and emergency measures to reduce coincident industrial and commercial load.At the same time, because HPs increase efficiency and replace fossil fuel usage, they likely entail significant emissions benefits (here again, higher HP efficiency is more beneficial).
As system planners consider ways to increase decarbonization of space heating and cooling, particularly through electrification using HPs, understanding the system-wide grid impacts of these load shifts is important for reliability assessments, setting reserve margin targets, and integrated resource planning.Concurrently, climate change will also be a driver of future changes in load patterns.Assessing the net effect of these phenomena (higher HP penetration alongside the synchronous impact of climate change) on load is important for future grid planning and firm capacity assessments.The use of highly resolved data provided valuable insights into the gradual and stepwise changes in these trends over the different conditions, pathways, and over longer periods.These methods Earth's Future 10.1029/2024EF004443 can also be transferred across other U.S. interconnections, to be able to understand the long-term demand and reliability dynamics.

Figure 1 .
Figure1.ERCOT historical and future climate change peak hourly annual temperatures weighted by SSP3 population projections(Burleyson et al., 2023).Top: ERCOT maximum hourly annual population-weighted temperature under historical and future climate change scenarios RCP 8.5 Hotter and RCP 4.5 Cooler.Bottom: ERCOT minimum hourly annual population-weighted temperature under historical and future climate change scenarios RCP 8.5 Hotter and RCP 4.5 Cooler.Representative Concentration Pathways (RCPs) 4.5 and 8.5 are projections of plausible future climate conditions under different levels of anthropogenic activity.Hotter and Cooler are distinctions of High-and Low-sensitivity models respectively under the same RCP(Jones et al., 2023).Shared Socioeconomic Pathways (SSPs) track different socioeconomic scenarios.Historical U.S. population projections are based on the U.S. census; future populations are consistent with the SSPs(Jiang et al., 2020;  Zoraghein & O'Neill, 2020).Also see Table1.

Figure 2 .
Figure 2. Modeling framework showing data inputs, heating electrification scenarios, models, and methods used.The flowchart also delineates the contribution of White et al. (2021a) (gray) and our contribution (blue) to the framework.The four data sets of residential electricity demand from White et al. (2021a), together with the 2016 weather data were used to fit 4 ML models under four scenarios of space heating electrification: Base, Standard Efficiency, High Efficiency, and Ultra-High Efficiency HP adoption.A fifth ML model was trained based on non-residential demand data and 2016 weather data.The 5 ML models were then used to predict electricity demand using future climate data.The output from those models was then passed through a direct current optimal power flow model for the ERCOT grid to simulate reliability impacts.

Figure 4 .
Figure 4. Season for the peak hourly annual RESIDENTIAL load under different heating electrification scenarios.Top: Considering the "single hour" with the highest load, Middle: Considering the "top 0.5%" peak load hours, Bottom: Considering the "top 1%" peak load hours.Base means the Base Case scenario.Standard, High, and Ultra are the three HP adoption scenarios.

Figure 5 .
Figure 5. Season for the peak hourly annual TOTAL load under different heating electrification scenarios.Top: Considering the "single hour" with the highest load, Middle: Considering the "top 0.5%" peak load hours, Bottom: Considering the "top 1%" peak load hours.Base means the Base Case scenario.Standard, High, and Ultra are the three HP adoption scenarios.

Figure 6 .
Figure 6.Temperature (top row) and hourly total load (bottom row) under the four heating electrification scenarios and 2 weather years: The left column is the 2016 historical year and the right column is the 2058 future climate year under RCP 4.5 Hotter.The legends on the bottom row panels indicate the electrification scenario and the season of peaking under each scenario.

Figure 7 .
Figure 7. Peak total load marked by season, under the different scenarios of climate change, and heating electrification.

Figure 8 .
Figure 8.(a) Location and magnitude of outages using climate change data on the single summer day that experienced the most outages (RCP 8.5 Hotter, Base Case, 3 August 2091).(b) Distribution of temperature at the hottest hour of the day in (a).(c) Location and magnitude of outages on the single winter day that experienced the most outages (RCP 4.5 Cooler, Standard Eff.HP, 23 December 2069).(d) Distribution of temperature at the coldest hour of the day in (c).

Figure 9 .
Figure 9. Cumulative loss of load grouped by season, under the different scenarios of climate change, and heating electrification.

Table 1
The Range of Simulation Data Used in the Experiment It includes four climate scenarios (RCP) weighted using 1 socioeconomic scenario (SSP)'s population projections, 80 simulation years, and four heating electrification scenarios under each case.
Figure 3. Reduced network topology of selected model consisting of 150 nodes and a network of transmission lines.SSEMBATYA ET AL.

Table 2
Number of Winter Peaking Years That Exceeded the Highest Summer/Fall Peaking Year as a Percentage of All Years SSEMBATYA ET AL.
, the model does not account for future HP technology improvements.It does not account for future changes in building stock energy efficiencies, or population shifts.The model does not consider other potential future drivers of residential