High-resolution climate model datasets for energy infrastructure planning in a renewable-dependent future Open Access

In an effort to mitigate climate change, energy systems worldwide are making a rapid push to decarbonize. Decarbonization efforts focus on combining electrification, or the transition from fossil-fuel-based to electric energy sources, with the deployment of renewable generators. At the same time, the interdependence of the climate and energy systems is growing as the climate impacts both the supply of and demand for energy. On the demand side, rising temperatures and changing weather patterns will affect electricity use habits, while on the supply side, renewable generation sources will be vulnerable to a changing earth system. For example, extended wind droughts could limit wind energy production, or increased cloud cover could limit solar energy production. Future climate projections are therefore becoming increasingly important for energy infrastructure planning.

A recent report by the Energy Systems Integration Group (ESIG) stresses the need for weather and climate datasets that meet the requirements of the energy infrastructure planning community (ESIG, 2023). The datasets should include high spatial resolution to capture local effects, as well as high temporal resolution to capture transient events and diurnal variability. Additionally, the relevant variables should be coincident and physically consistent to ensure realistic downstream effects in power system models. Finally, and most importantly, the datasets should be validated to ensure accurate representation of weather and climate processes. Together, these requirements help to ensure that interactions between the climate and energy systems are represented accurately in infrastructure planning models.

The ESIG report argues that in the near-term future, the combination of electrification and weather-dependent capacity outweighs the threat of climate change on the energy system. It therefore focuses on recommendations for historical and present-day datasets based on numerical weather prediction (NWP) models. NWP models are physics-based models of atmospheric dynamics that are commonly used for short-term (hours to days) and medium-term (weeks) forecasting, hindcasting of past events, and long-term (multi-year) historical reanalysis. The latter combines available observations with NWP modeling to produce a gridded weather dataset with as much accuracy as possible. Reanalysis datasets are especially useful for energy system analysis and planning due to their physical consistency, extended time coverage, and basis in reality.

The ESIG report also acknowledges the growing importance of climate change impacts on the energy system. In fact, recent increases in climate model resolution are blurring the line between climate modeling and traditional NWP applications. Climate models are similar to NWP models in that they are physics-based models of atmospheric and other earth system dynamics. However, they are used to make projections decades into the future based on possible changes in greenhouse gas emissions and other forcing agents. Increases in climate model resolution have been driven by increases in computing power, as well as new exascale computing architectures and software development frameworks that enable the use of graphics processing units (GPUs) for scientific applications.

The larger energy community is coalescing around the need to incorporate climate model projections into infrastructure planning. For example, the Electric Power Research Institute (EPRI) Climate READi initiative aims to establish a common framework for accessing climate data and applying it to energy infrastructure applications (EPRI, 2025). Regional initiatives such as Cal-Adapt, funded by the California Energy Commission (CEC), provide decision makers access to climate information and guidance on the appropriate use of climate model data (Cal-Adapt, 2025).

Increases in climate model resolution present an opportunity to use future climate projections more directly for energy infrastructure planning. In this article, we review recent and ongoing advancements in high-resolution climate modeling, with a specific focus on energy infrastructure applications. We then look ahead to discuss how high-resolution climate data can be used to plan for a renewable-dependent future, and envision a unified climate-energy model framework that captures the two-way feedbacks between these interdependent systems.

Climate effects on the energy system will vary regionally (see, e.g., Gernaat , 2021), requiring future projections with high spatial resolution to predict local impacts. However, global climate model (GCM) simulations are typically run with a horizontal resolution of approximately 100 km, especially for large ensembles such as the Coupled Model Intercomparison Project Phase 6 (CMIP6; Eyring , 2016). Although these simulations capture general trends in temperature and precipitation, regional or local effects are unresolved, leading to reduced accuracy for renewable-energy-relevant quantities. As noted by Pryor (2020), simulations with resolutions of 50 km or less provide limited utility in projecting future wind resources because many of the drivers of wind variability, such as terrain features and storm systems, are unresolved at this scale. Similarly, although coarse climate simulations capture general trends in solar radiation, they do not resolve fine-scale cloud variability, which remains a large source of uncertainty in future solar energy projections (e.g., Wild , 2015), and in climate projections more broadly (e.g., Zelinka , 2020).

With recent advancements in computing, global climate simulations are more frequently being run with higher spatial resolution. For example, the High-Resolution Model Intercomparison Project (HighResMIP, a subset of CMIP6; Haarsma , 2016) includes global simulations with 25–50 km resolution. On the leading edge of high-resolution GCMs, the Energy Exascale Earth System Model (E3SM, the U.S. Department of Energy's climate model; see Leung , 2020) has a configuration for global cloud-resolving simulations. Known as the Simple Cloud Resolving E3SM Atmosphere Model (SCREAM), this configuration leverages GPU acceleration and has been tested at 3 km resolution (Caldwell , 2021; Donahue , 2024).

Despite computational advancements, the expense of high-resolution global climate simulations has led to the development of regional climate model (RCM) configurations that focus increased resolution in particular areas of interest (e.g., Giorgi and Gutowski, 2015). RCMs generally use a coarse GCM to provide initial and boundary conditions to an NWP model such as the commonly used Weather Research and Forecasting model (WRF; Skamarock , 2019) through a process known as dynamical downscaling. RCMs have been run with a range of resolutions, from 50 km down to 3 km. Pryor (2020) provides a wind resource-specific summary of RCM results (see Table III therein). Additional examples of energy-focused RCM studies include the U.S.-wide hydropower assessment of Kao (2022), which uses 4 km resolution, and several studies associated with the Cal-Adapt (2025) initiative, which cover the western U.S. with 9 km resolution (Rahimi , 2024) and California with 3 km resolution (Rahimi, 2022).

An alternative approach to RCM simulations that has been adopted by the developers of E3SM is to run a global model with a regionally refined mesh (RRM), rather than a separate dynamical downscaling model, to focus resolution in a region of interest. E3SM version 2 (Golaz , 2022) and the upcoming E3SM version 3 include a North America RRM configuration with grid refinement down to 25 km (Tang , 2023). The E3SM RRM has also been refined down to 3 km over California for both hindcasts (Bogenschutz , 2024) and future projections (Zhang , 2024), as shown in Fig. 1. The RRM approach eliminates the need for model-to-model coupling, but will require “scale-aware” (i.e., resolution-dependent) atmospheric physics parameterizations as grid refinement increases. The development of scale-aware schemes is an active area of research in the climate community; however, the E3SM RRM configuration is not yet fully scale-aware (see discussion in Zhang , 2024).

Statistical or machine-learning methods may also be used to downscale coarse climate data to a region of interest, but are not a focus of this article. These methods are generally more computationally efficient than dynamical methods, allowing for larger ensembles, but come with their own sets of assumptions and limitations. One example of statistical downscaling that has been used for energy applications is the Localized Constructed Analogs (LOCA; Pierce , 2014, 2023) method, which has datasets available at 6 km over North America and 3 km over California (see Pierce, 2025). Additionally, the super-resolution for renewable energy resource data with climate change impacts (Sup3rCC; Buster , 2024) method uses generative machine learning informed by the Wind Integration National Dataset (WIND) Toolkit (Draxl , 2015a, 2015b) and the National Solar Radiation Database (NSRDB; Sengupta , 2018) to downscale coarse GCM data to 4 km over the U.S. Some of the aforementioned studies (e.g., Kao , 2022) and initiatives (e.g., Cal-Adapt, 2025) use a combination of dynamical and statistical downscaling to leverage the benefits of both approaches. Interested readers are also referred to the companion papers of Pryor (2023) and Coburn and Pryor (2023), which provide a comparison between dynamical and statistical downscaling methods for wind energy in North America.

Many previous energy infrastructure studies have relied on reanalysis datasets for a variety of applications such as extreme weather impacts (e.g., Panteli , 2016; van der Most , 2022; and Yang , 2023), renewable resource drought assessment (e.g., Rinaldi , 2021; Brown , 2021), resource adequacy assessment (e.g., Sundar , 2023), and capacity expansion planning (e.g., Chen , 2018; Radu , 2021; Henry , 2021; Zhou , 2021; and Sundar , 2023). These studies use one of several global reanalysis products, including the Modern-Era Retrospective Analysis for Research and Applications (MERRA; Rienecker , 2011) or its successor MERRA-2 (Gelaro , 2017), both of which have roughly 50 km resolution, or the European Centre for Medium-Range Weather Forecasts Reanalysis version 5 (ERA5; Hersbach , 2020), which has 31 km resolution.

While useful for understanding the intersection of the climate and energy systems, such studies cannot account explicitly for changes in the future climate. Now, however, the resolution of global and regional climate models is approaching that of the reanalysis models most commonly used by the energy community. The highest-resolution climate models are even on par with operational NWP models such as the 3 km High-Resolution Rapid Refresh model (HRRR; Benjamin , 2016; Dowell , 2022). This convergence of model resolution is opening new doors for studies of climate impacts on the energy system.

Before unleashing high-resolution climate models on energy infrastructure applications, accurate simulation of the quantities of interest, such as wind and solar resources, must be demonstrated. Climate models typically rely on extensive validation against observed data from historical and present-day sources. This validation ensures that the models are representing the dynamics of the climate system accurately enough to trust future projections. However, validation studies usually focus on temperature and precipitation, which are primary indicators of climate change.

Recent work by Golaz (2024) and Lee (2025) leverages publicly available generation data from the U.S. Energy Information Administration (EIA) to evaluate climate model representation of wind and solar resources. Based on public reporting by utility-scale generators, the EIA provides site-specific, monthly production data (via Form 923) and regional, hourly production data (via Form 930). Golaz (2024) used the former to evaluate the E3SM version 2 North America RRM for both seasonal and multi-year average wind and solar production. Lee (2025) extended this work by comparing E3SM-predicted wind and solar capacity factors to those of industry-standard benchmark datasets including the WIND Toolkit (Draxl , 2015a, 2015b), PLUSWIND (Millstein , 2023), and NSRDB (Sengupta , 2018).

These preliminary studies establish a baseline for E3SM-based predictions of wind and solar energy production, and provide an evaluation framework that can be used for other high-resolution climate models. Such studies will highlight regional and seasonal biases that require further examination. For example, on the U.S. west coast, the E3SM North America RRM tends to underpredict wind energy capacity factors in the summer and overpredict them in the winter (Lee , 2025). Continued work is required to pinpoint the physical causes of these and other biases, which will inform future climate model development. Furthermore, the downstream effects of climate model biases must be quantified for different energy infrastructure applications.

Additional energy-focused validation studies for a variety of climate models and configurations will help to increase confidence in future projections. Climate datasets can then be applied to infrastructure studies just as reanalysis or NWP datasets were applied in the past. As an initial example of this approach, Lee (2024) used data from HighResMIP to analyze changes in wind and solar capacity factors in the western U.S. between 1979 and 2099. This study quantified regional and seasonal differences in projected capacity factors, as well as potential increases in wind and solar resource droughts, over the coming century.

Taking another a step in complexity, high-resolution climate projections can be connected to capacity expansion models for infrastructure planning in a more renewable-dependent future. Capacity expansion models are especially well-suited to consider climate effects because they are used to optimize infrastructure investments multiple decades into the future. The main links between climate and capacity expansion models are projections of energy demand, based primarily on surface temperature, and renewable supply, based primarily on hub-height wind speeds and surface solar radiation. Establishing these links requires careful communication between climate and energy infrastructure experts to ensure proper climate data selection and transfer between models. Moreover, capacity expansion models require additional constraints related to costs, siting, transmission routing, and renewable portfolio standards, among others.

The recent work of Zuluaga (2024) shows what is possible by combining state-of-the-science techniques in both climate modeling and capacity expansion planning (see illustration in Fig. 2). This study demonstrates a nodal generation and transmission planning model for California that is informed by climate projections for the year 2045, as simulated in the 3 km E3SM California RRM. The capacity expansion model is stochastic in that it optimizes over multiple representative days from the climate dataset, and therefore inherently quantifies uncertainty related to climate inputs. By leveraging parallel computing, Zuluaga (2024) included up to 360 representative days in the optimization, reaching converged solutions within several hours.

These advancements enable a wide range of new research possibilities at the climate-energy nexus. In terms of climate effects on energy infrastructure planning, optimal infrastructure investment strategies could be explored for different emissions scenarios, timelines, or siting and routing constraints. In terms of the model framework itself, the sensitivity of the capacity expansion solution could be explored for different climate data inputs, including the choices of future scenario, climate model, spatial resolution, and representative days. For example, Zheng (2023) examined the impact of climate model resolution on a future resource adequacy assessment for the Texas interconnection. Ultimately, such sensitivity tests should reveal the most important climate-related factors for capacity expansion, thus informing future research directions and planning decisions.

In a broader sense, any application of climate model projections to energy infrastructure planning should, to the extent possible, include uncertainty quantification. Although increases in model resolution are expected to reduce uncertainty related to small-scale physical processes, many sources remain. These are primarily related to future scenario, internal climate variability, and model-to-model differences. By adopting the ensemble-based approach of the climate community, these uncertainties can be quantified, propagated through energy infrastructure models, and ultimately communicated to stakeholders.

Beyond the current state-of-the-science, future studies should take a more holistic approach to climate and energy modeling. The energy system is becoming more intertwined with the Earth system due to electrification and continued renewables deployment, and climate change will only magnify this connection. A unified model framework that captures the dynamics of both systems, as well as the feedbacks between them, will facilitate increased understanding of their interdependence. The advent of high-resolution climate simulations enables such a framework, which is not feasible with coarse climate information (see illustration in Fig. 3).

A unified climate-energy model would allow for investigations into how the energy system will respond to different climate change scenarios, or how planned renewable deployments will perform under future climate conditions. It would also support ensemble-based uncertainty quantification. Specific applications of the unified model framework would depend on the scales and dynamics of interest, and would require representation of energy supply, demand, and transmission at comparable fidelity to that of the climate model. For example, larger-scale (e.g., national-level) studies could use regionally aggregated energy system models, while smaller-scale (e.g., state- or local-level) studies could use higher-resolution energy system models. Synthetic-but-realistic representations of the electrical grid (e.g., Taylor , 2023; TAMU, 2024) could be used to facilitate open research, due to the protected nature of energy infrastructure information. However, the unified model framework could also leverage real grid models when appropriate.

A pressing challenge for infrastructure planning that could be addressed with the unified model is site selection for renewable generators and energy storage technologies. New wind and solar energy deployments must simultaneously consider the available resource along with connections to load centers, including transmission capacity on the existing grid and routing constraints for new transmission lines. These considerations can help reduce curtailment under projected supply and demand scenarios. Moreover, as the energy system becomes more dependent on renewables, large-scale energy storage deployments will be necessary to mitigate wind and solar resource intermittency, which can have impacts on diurnal to seasonal time scales. To support the economic viability of storage under different future scenarios, the location, capacity, and potential utilization of new deployments should be informed by coupled projections of energy supply, demand, and transmission. Ultimately, optimal renewable and storage integration will depend on an improved understanding of feedbacks between the climate and energy systems.

We envision two forcing paradigms for the unified climate-energy model. The first could build upon the shared societal pathway (SSP; see, e.g., Riahi , 2017) framework used by the climate science community to relate greenhouse gas emissions to plausible socioeconomic conditions. A unified model could explicitly represent changes to the energy system that would be required to achieve a given level of emissions. This approach would enable studies that link renewable portfolio standards to different future scenarios, while also considering energy system performance and reliability. Baulenas and Bojovic (2023) present the results of a recent workshop examining such an approach to scenario development.

The first paradigm, by nature, would focus on general climate trends and their impacts on the energy system. However, a more event-based approach would also be beneficial because energy system planning is often driven by reliability during infrequent yet impactful events. Thus, we envision a second forcing paradigm focused on so-called storyline simulations (Shepherd , 2018), which demonstrate how realistic events from the past may unfold under future climate conditions. In the unified model, events such as heat waves, storms, wind droughts, or others that have outsized impact on the energy system could be simulated in both present-day and future climate conditions. Prospective future energy infrastructure investments could then be “stress-tested” in plausible event-based scenarios.

In conclusion, high-resolution climate models will be essential tools for energy system planning in a renewable-dependent future. In the near term, high-resolution climate datasets can be used to analyze future changes in energy-relevant quantities, such as temperature and renewable generator capacity factors. These values can then be applied to resource adequacy studies, or used as inputs to capacity expansion models. Looking ahead, we envision a more unified climate-energy modeling approach that captures the two-way feedbacks between these increasingly interdependent systems. However, any application of high-resolution climate modeling to energy infrastructure planning will require careful model validation and communication of uncertainties to ensure appropriate use of the data.

This work was performed under the auspices of the U.S. Department of Energy by Lawrence Livermore National Laboratory under Contract No. DE-AC52-07NA27344. The authors gratefully acknowledge the support of Laboratory Directed Research and Development (LDRD) funding through Project No. 23-ERD-010. The authors also thank two anonymous reviewers for their thoughtful comments on the manuscript.

The authors have no conflicts to disclose.

Robert S. Arthur: Conceptualization (equal); Funding acquisition (lead); Visualization (lead); Writing – original draft (lead). Jean-Christophe Golaz: Conceptualization (equal); Writing – review & editing (equal). Hsiang-He Lee: Conceptualization (equal); Writing – review & editing (equal). Jessica Wert: Conceptualization (equal); Writing – review & editing (equal). Matthew Signorotti: Conceptualization (equal); Writing – review & editing (equal). Jean-Paul Watson: Conceptualization (equal); Project administration (supporting); Writing – review & editing (equal).

Data sharing is not applicable to this article as no new data were created or analyzed in this study.