Table  summarizes the key data sources used in this study. These sources include information on cost-viable DW energy generation potential, economic indicators, and demographic characteristics, all of which are essential for our analysis. This study focuses on data for the contiguous United States, which excludes Alaska and Hawaii as data for those states are not yet available in the DWEFS. While containing data for nearly 150 million parcels, a small fraction of missing parcel data results in select counties that are not represented in this analysis, primarily located in New Mexico, Kentucky, Indiana, South Dakota, Utah, Nevada, and Colorado. We have used an updated version of the DWEFS data from , which has not yet been published but was developed with a similar methodology. Key improvements from the previously published data include updating technology costs, electricity rates, and state compensation mechanisms to the latest available data (2024) and a more comprehensive evaluation of federal incentives. Importantly, these data assume a federal investment tax credit equivalent to 30 % of project capital expenditures and location-dependent credit bonuses (for energy communities, low-income communities, and tribal lands) which were present in 2024. In light of the federal policy changes in 2025, the availability of these tax credits is expected to be reduced in future years, in turn reducing modeled cost-viable AEP. These potential changes in tax credit policy are beyond the scope of this study but may be considered in future or more localized analyses. The results presented here can be interpreted as an average or above-average estimate of market opportunity in a favorable policy environment.

This analysis does not directly incorporate weather-related variables (e.g., heating/cooling degree days) or housing characteristics (e.g., age, thermal efficiency), both of which can influence EB independent of income. Because EB is calculated from realized electricity expenditure, some of this variation is implicitly embedded in the data, but the models do not isolate it, so its influence on covariates such as poverty and agricultural employment cannot be ruled out. Future work incorporating these variables directly could refine the predictors identified here.

Economic and demographic data used in this analysis reflect the most recent available estimates rather than year-specific projections, as projected demographic and economic data at comparable county-level resolution were not available for this study. Only DW generation potential and electricity demand vary by scenario year. This represents a data availability constraint rather than an assumption of no future change, and the framework presented here could be readily applied with updated demographic projections as they become available.

Data on energy generation potential, EB, and modeled energy demand offsets are used by their corresponding scenario year (2025 or 2035), while other socioeconomic and demographic data correspond to the latest available datasets to reflect current demographics and economic characteristics across the contiguous United States.

To enable spatial consistency with representative datasets, we aggregate parcel-level data to the county level. The data include sector-specific electricity metrics for residential (household), commercial, and industrial consumers, which are also aggregated to county level by sector and normalized where appropriate.

We define the AEP-to-demand ratio (AEP_demand) as 1AEPdemand=AEPtotal electricity demand, where total electricity demand is the sum of residential, commercial, and industrial electricity demand at the county level. For sector-specific analyses, the denominator is replaced by the corresponding sector's demand alone; for example, residential AEP_demand divides cost-viable residential AEP by residential electricity demand only. All demand values are drawn from the NLR SLOPE dataset, which provides county-level electricity consumption disaggregated by sector . This ratio standardizes DW generation relative to total electricity demand across residential, commercial, and industrial sectors, accounting for variation in population density and consumption levels. Sector-specific analyses also use disaggregated demand components as needed. Higher values of AEP_demand indicate that potential DW generation could supply a relatively larger share of local electricity demand, suggesting greater opportunity for demand-offsetting generation. Lower values indicate that even substantial wind resources would cover only a small fraction of demand, limiting system-level impact of DW deployment.

Our primary affordability metric is energy burden (EB), which conceptually reflects residential-level affordability. In our analysis, we compute EB at the county level as the ratio of total residential electricity expenditure to the product of median household income and the number of residential units in the county: 2EB=total residential electricity expendituremedian household income×number of residential units. Higher EB values indicate greater energy cost burdens and lower affordability, meaning households must devote a larger fraction of income to electricity. Lower EB values indicate comparatively affordable electricity, where energy costs consume a smaller share of household income. This county-level aggregate formulation differs from conventional household-level EB metrics typically derived from survey data or census-tract-scale estimates . The county-level aggregate approach is adopted here to maintain spatial consistency with the DWEFS parcel-level data aggregated to the county level and to enable the spatial prioritization analysis that is the primary objective of this study.

At the county level, we extend this to EB relative to GDP (EB_GDP), which provides a broader perspective on county-level economic performance. Rather than focusing solely on residential metrics, this approach considers the county as an economic unit, viewing its GDP and energy expenditures as components of overall economic output: 3EBGDP=total county energy expenditureGDP. Higher values of EB_GDP indicate that a larger share of county-level economic output is absorbed by energy expenditures, reflecting greater macroeconomic energy intensity or stress. Lower values indicate that energy costs represent a smaller fraction of economic activity, suggesting greater economic resilience to energy prices.

We retain the EB nomenclature rather than alternative framings, such as energy cost intensity, to maintain consistency with the residential EB metric. Both metrics share the same ratio structure of energy expenditure relative to an income or output measure, and the parallel naming reflects this parallel construction within the study's analytical framework.

While EB measures the proportion of income spent on energy – offering a direct sense of affordability – it can produce distorted or infinite values for households with very low or near-zero income . NER addresses this by remaining mathematically defined and interpretable across the full income range while also providing smoother statistical properties and a complementary interpretive angle. NER reflects the economic return or residual income relative to energy expenditure: 4NER=household income-energy expenditureenergy expenditure. Because NER is an algebraic transformation of EB, they are mathematically linked as 5NER=1EB-1. Higher NER values indicate greater residual income relative to energy spending and thus lower effective EB. Lower NER values indicate limited residual income after paying for electricity, reflecting higher affordability stress.

To parallel our EB_GDP measure and further extend the analysis to the macroeconomic level, we define a GDP-based version of NER: NER_GDP. This metric evaluates the return of economic output over energy spending at the county level: 6NERGDP=GDP-total county energy expendituretotal county energy expenditure. Higher values of NER_GDP indicate that economic output substantially exceeds energy expenditures, suggesting stronger economic returns per unit of energy cost. Lower values indicate that energy expenditures consume a relatively large share of economic output, signaling reduced economic efficiency.

While EB_GDP and NER_GDP are mathematically equivalent transformations of one another, they offer complementary interpretive framings: EB_GDP emphasizes the share of economic output absorbed by energy costs (a burden-oriented framing), while NER_GDP emphasizes economic return relative to energy spending (a resilience-oriented framing). Retaining both is consistent with the residential-level EB/NER pairing established above.

Together, these metrics – EB, EB_GDP, NER, and NER_GDP – provide a multidimensional view of energy affordability and serve as key response variables in our spatial correlation and modeling analyses.

While national policies typically define energy poverty as EB > 6 % or NER < 16 % , these thresholds include all energy sources (e.g., natural gas), whereas we calculate EB for electricity only. This is because DW electricity generation is only comparable to electricity costs, not other costs like natural gas consumption. Therefore, we define extreme values using state-specific distributions – identifying the top 10th percentile for EB and bottom 10th percentile for NER – to better capture localized energy stress in an electricity-specific context.

This analysis examines correlations among NER, EB, and AEP_Demand for residential, commercial, industrial, and combined-sector demand at the county level. Notably, the spatial distribution of AEP_Demand might not necessarily align with areas of high EB. A state or county may exhibit high energy demand, but if the associated EB is low, the relationship between the two remains weak. Strong correlations emerge only when both conditions – high DW potential and high EB – are present, a pattern not uniformly observed across the country.

To assess the relationships between key variables, both parametric and nonparametric correlation methods were employed. This dual approach accounts for the non-normal distribution of EB and AEP_Demand in their raw forms. Box–Cox transformations were applied to normalize the data for Pearson correlation analysis, while Kendall's tau was used as a robust, nonparametric alternative to capture rank-based associations under non-normal conditions.

Kendall's tau evaluates ordinal association by comparing all possible observation pairs . A pair (xi,yi),(xj,yj) is concordant if the ranks of both variables move in the same direction and discordant if they move oppositely. The coefficient is calculated as 7τ=C-D(C+D+T1)(C+D+T2), where C and D represent the number of concordant and discordant pairs, respectively, and T1 and T2 account for ties in each variable. Positive values of τ indicate a direct association; negative values indicate an inverse relationship.

After computing both parametric and nonparametric correlations, we construct rankings at the state level and then combine them into a weighted ranking to identify states where DW deployment opportunity could be most strongly aligned with EB. The purpose of this ranking is not to maximize wind resource alone but to prioritize locations where DW potential spatially coincides with affordability challenges, consistent with the primary objective of this study.

As shown in Eq. (), the composite ranking assigns 50 % of the total weight to correlation-based metrics. This emphasis reflects the central research question of this analysis: identifying where DW potential and EB are spatially aligned, rather than where either factor is independently high. Correlation strength is therefore treated as the primary diagnostic indicator of potential relevance for affordability-oriented DW deployment.

A weight of 50 % was selected to ensure that correlation serves as the dominant but not decisive criterion in the composite ranking. Assigning less than half the total weight would allow scale-based metrics to outweigh alignment between DW potential and EB, undermining the central objective of the study. Conversely, assigning substantially more than half the weight would risk over-prioritizing statistical alignment at the expense of deployment relevance and practical impact. The 50 % threshold therefore represents a balance point at which correlation is guaranteed to influence rankings while preserving sensitivity to demand magnitude and generation scale.

To ensure that correlation-based rankings remain meaningful in terms of deployment scale and practical impact, we incorporate two additional counterbalancing components. First, AEP_Demand for both the 2025 and the 2035 scenarios is included to represent relative DW opportunity normalized by electricity demand. Second, absolute electricity demand is included to account for scale effects and avoid disproportionately prioritizing low demand where standardized ratios may appear favorable despite limited potential impact.

Each component is ranked independently and combined using this weighted formulation: 8rankcomposite=0.5⋅(rank25+rank35)+0.25⋅(rankAEP/D,25+rankAEP/D,35)+0.25⋅(rankDem,25+rankDem,35), where rank25 and rank35 are the correlation-based ranks of DW potential and EB alignment for 2025 and 2035, rankAEP/D,25 and rankAEP/D,35 are the ranks of DW opportunity normalized by electricity demand (AEP_Demand), and rankDem,25 and rankDem,35 are the ranks of absolute electricity demand.

The selected weights are intentionally goal driven and reflect a balance between (i) emphasizing spatial alignment between EB and DW potential and (ii) maintaining sensitivity to deployment scale and demand magnitude. These weights are not derived from statistical optimization, and alternative weighting schemes are possible; however, the chosen structure directly supports the study's focus on affordability-relevant siting rather than maximum energy generation. To assess the sensitivity of the rankings to the chosen weighting scheme, we evaluated several alternative formulations by varying the components and weights assigned to correlation, AEP_Demand, and absolute demand.

Based on this framework, we classify states into two groups and a set of special cases, each designed to highlight a distinct type of deployment-relevant context.

Group 1 consists of states with the highest composite rankings in Eq. (), reflecting a favorable combination of spatial alignment between EB and residential DW potential together with substantial demand scale. These states represent locations where affordability challenges and DW opportunity coincide geographically, making them priority candidates for affordability-oriented DW deployment. While correlation strength is a primary component of the ranking, inclusion in Group 1 does not require uniformly strong correlations across all states but rather a balance between alignment and deployment relevance.

Special cases include states that do not clearly fall into either group due to scale effects, extreme EB values, or weak spatial alignment between EB and DW potential. These states illustrate important boundary conditions of the framework, where high need or high resource availability exists, but the spatial coincidence required for prioritization under the composite ranking is limited. Identifying these cases helps avoid overgeneralization and highlights the importance of sub-state or targeted deployment strategies. This grouping framework is intended as an interpretive tool rather than a prescriptive prioritization scheme, providing a transparent and reproducible way to compare states across multiple, policy-relevant dimensions.

While AEP_Demand and EB-related metrics are computed for residential, commercial, and industrial sectors (Sect. ), the composite ranking and grouping framework is based specifically on residential-sector metrics. This choice is guided by the empirical finding that residential AEP_Demand consistently shows the most robust correlation with residential EB across all sectors analyzed (Sect. ), making it the most informative metric for identifying states where DW potential spatially aligns with energy affordability challenges. Commercial and industrial sector results are presented as descriptive context but do not directly influence group classification.

To identify key predictors of residential EB at the county level, we implement a mixed-effect linear regression model using nationwide data (Table ). In this framework, county-level covariates – including poverty rate, percentage of agricultural employment, and percentage of the population identifying as non-white racial and ethnic groups – are treated as fixed effects, meaning that in this national model (granularity is studied later), their estimated relationships with EB are assumed to be consistent across the United States.

States are modeled as random effects to account for unobserved, state-specific factors that systematically influence EB but are not explicitly included in the model, such as differences in energy policy, utility regulation, climate, or electricity pricing structures. Modeling states as random effects allows the intercept of the regression to vary by state, capturing baseline differences in EB while still estimating a single set of national-level fixed-effect relationships.

The significance of the random state effect is supported by the intraclass correlation (ICC), which quantifies the proportion of total variance attributable to state-level differences, as reported in Sect. .

Following the identification of state-level variation, we conduct separate fixed-effect linear regressions (with and without interactions) for representative states from each of the groups defined in the ranking analysis (see Sect. ). States were selected based on the composite ranking and grouping framework described in Sect. , with one or more representative states chosen from Group 1, Group 2, and the special cases. These states are not a random sample but rather illustrative of distinct deployment-relevant contexts identified through the spatial correlation and ranking analysis. These models provide more granular insights into how key predictors of EB differ across regional contexts and support the interpretation of group-level patterns observed in the correlation and ranking results. To ensure the validity of these models, we evaluate key diagnostics, including adjusted R2, residual distributions, variance homogeneity (homoskedasticity), and multicollinearity. Multicollinearity was assessed in the models without interactions using the variance inflation factor (VIF) to verify that the main predictors were not highly correlated before including interaction terms.

This modeling framework identifies key demographic and economic factors statistically associated with EB. It does not imply that interventions such as DW deployment will directly alter the covariates involved. The mixed-effect and state-level regression models are therefore used as explanatory and diagnostic tools to contextualize the spatial correlation results and composite ranking framework. These models are not used to determine the weights, thresholds, or group classifications directly, nor are they intended for prediction or causal inference. Instead, they serve three purposes: (i) to confirm that EB exhibits statistically significant variation across states, motivating a state-specific analysis framework; (ii) to identify socioeconomic factors consistently associated with elevated EB, providing interpretive context about high-EB regions that emerge in the correlation analysis; and (iii) to assess whether states grouped by correlation patterns exhibit distinct EB drivers. In this way, the regression results support interpretation and validation of the grouping framework rather than defining it.

We analyze spatial patterns in EB and NER to understand how energy and cost burdens vary across the United States and how they may relate to DW opportunity. EB and NER vary significantly across United States counties and states. Southeastern states exhibit above-median residential EB (Alabama has the highest EB), while many western states fall below the national average in both scenarios. However, county-level analysis reveals substantial intrastate variation, suggesting that state-level aggregates may obscure local disparities. Figure  presents residential EB distributions and spatial patterns for the 2025 and 2035 scenarios.

The EB distribution (Fig. e–f) is slightly right-skewed, with a median and mean around 1.7 % for both scenarios (2025 and 2035) and a 90th percentile near 2.5 % for 2025 and 2.4 % for 2035. The geographic distribution is not even, which reinforces the earlier observation of intrastate heterogeneity.

Regarding the reported percentages, it is important to note a methodological distinction discussed in Sect. . Because DW is only comparable with electricity costs, and not other household energy uses such as natural gas or heating fuels, EB is calculated here using electricity expenditures only rather than total household energy costs. As a result, the reported EB values may appear lower than commonly cited EB estimates that account for all residential energy sources.

Moreover, as shown in Fig. c–d, the southeastern states of Alabama, South Carolina, and Georgia consistently exhibit the highest percentages of counties with EB above the 90th percentile nationwide. These patterns are further highlighted in Fig. , which presents the top 10 states with the largest share of counties exceeding this high-EB threshold.

To further understand how regional economic capacity determines energy affordability and its implications for DW opportunity, we explore EB_GDP and compare it to residential EB. The EB_GDP metric shows less heterogeneity within states (Fig. a–b) compared to the residential EB (Fig. ), with state values generally clustering closer to the national median. However, noticeable regional patterns at a county level (Fig. c–d) still emerge, often crossing state boundaries. This result underscores the importance of conducting sub-state analyses. Taken together, these trends suggest that understanding energy affordability requires not only county-level granularity but also consideration of broader economic and geographic factors.

To enable consistent comparisons across places with different populations and demands, we apply the AEP-to-demand normalization described in Sect. , yielding the AEP_Demand metric. This ratio, like raw AEP, is highly right-skewed, with most parcels exhibiting low values relative to demand and a minority of outliers with high relative potential.

To facilitate statistical analysis, we apply the Box–Cox transformation (Fig. ) solely for descriptive visualization and Pearson correlation diagnostics. These Pearson correlations (using Box–Cox-transformed variables) and Kendall's tau correlations (using untransformed variables) produced similar results on the rankings that we analyze later, indicating that the results are robust to transformation choice. However, all state rankings and group classifications are based on untransformed metrics and nonparametric (Kendall's tau) correlations, ensuring that the transformation does not influence the study's prioritization results.

Several states, including Kansas, Illinois, Minnesota, and Nebraska, have both an absolute and a proportionally higher number of counties exceeding the national median of AEP_Demand (Fig. ). These states also consistently rank within the top 10 in total AEP_Demand, although in a different order. These are areas of the United States where economically feasible DW opportunity meets or exceeds electrical demand, suggesting a strong potential for its deployment. Economically feasible DW potential in these states can result from a combination of numerous factors captured in the DWEFS data, including strong wind resources, favorable policies like net metering or local ordinances impacting siting, and elevated utility rates that make DW costs competitive, amongst other factors .

In contrast, Texas and Georgia – while also ranking in the top 10 for total AEP_Demand by number of counties – achieve this largely due to their high overall county count. When considering the proportion of counties exceeding the national median, both states fall to midrange positions. This suggests that although total AEP is high, the distribution of DW potential across counties is uneven. As a result, these states may still offer strong opportunities for DW deployment but require more localized, county-level assessment to identify the most suitable areas.

We disaggregate AEP_Demand further by sector (residential, industrial, and commercial) to capture sector-specific dynamics. Across scenarios the residential sector consistently emerges as the most influential in explaining AEP_Demand variation in each scenario. Significant differences between scenarios are observed in all pairwise comparisons, confirmed using both parametric and nonparametric tests. These results highlight the dominant role of residential AEP_Demand in shaping spatial and temporal patterns in DW deployment potential.

Correlations were examined across NER, EB (both residential and GDP-based), and AEP_Demand (residential and county total). Among these, the major finding is that residential AEP_Demand shows the strongest correlation with residential EB. Figure a–b show the correlation between EB and residential AEP_Demand.

To further investigate the nature of these correlations, we also analyzed the nonstandardized AEP values and the demand by sector. In some cases, states with large and internally diverse populations, such as Texas, exhibit high absolute demand and significant EB in certain areas yet show lower overall correlation. This internal variability, driven by differences in population density and economic capacity across regions within the state, dilutes the strength of the observed associations. These differences in correlation behavior motivated the classification of states into two distinct groups based on shared statistical and spatial patterns; Alabama and Texas were identified as special cases.

Spatial analysis of baseline scenarios for 2025 and 2035 revealed two distinct state groupings and special cases based on the correlation patterns between EB and AEP_Demand. Sensitivity analysis of the composite ranking using alternative weighting formulations showed that while the exact composition of groups varied across schemes, the overall framework consistently distinguished states with high spatial alignment between EB and DW potential from those with high DW potential but weaker EB alignment, supporting the interpretive value of the grouping approach even if specific state assignments may shift under alternative weights.

Mixed-effect models reveal that poverty rate and the presence of the agriculture industry at the county level are both significantly and positively associated with higher EB, with substantial variability across states. The state-level random effect accounts for approximately 40 % of the total variance in EB (intraclass correlation = 0.40), confirming substantial interstate variability beyond what is explained by the fixed-effect covariates alone, which motivates the subsequent state-level analyses presented below and indicates that regional economic and policy differences play a meaningful role in explaining this variation. Diagnostics of the mixed-effect model show no multicollinearity but reveal slight deviations from normality and significant heteroskedasticity in the residuals. These issues arise because the model treats covariates as fixed at the national level, pooling variation across all states. As described in the following paragraph, these limitations are addressed by conducting separate state-level regressions, which improve residual diagnostics and overall model performance.

To further investigate variation at the state level, we conducted separate linear regressions for key states representing the two identified groups and the special cases under the two scenarios (2025 and 2035). These models demonstrate stronger statistical performance – exhibiting higher coefficients of determination, reduced heteroskedasticity (constant variance of residuals), no multicollinearity among main-effect predictors (VIF≤5), and more normally distributed residuals – enhancing the reliability of their results. Overall, the state-level regressions broadly align with national trends, consistently identifying poverty and agricultural employment as key covariates of the EB model. Summaries of these models, including coefficients and correlation metrics, are provided in Tables , , and . County-level correlation maps are presented in Figs.  and .

A key contribution of this study is demonstrating that DW deployment relevance depends not only on absolute wind resource, but also on how DW opportunity spatially aligns with EB within states in many cases. The grouping framework intentionally prioritizes spatial correlation between EB and AEP_Demand, rather than maximum DW potential alone. Also, an important result from this study is the heterogeneity of results both between and within states. We identified two distinct groupings of states based on the EB–AEP_Demand relationship and two special cases:

Group 1 (high need/demand, favorably correlated potential) (e.g., North Carolina, Georgia, Iowa, Louisiana, and California) ranks highest in the weighted score (Eq. ) due to the most pronounced positive alignment between residential EB and residential AEP_Demand at the county level. Inclusion in this group does not imply uniformly high or utility-scale wind resources. Instead, it indicates that, within these states, counties with higher EB systematically coincide with counties that have relatively greater DW opportunity, while lower-EB counties coincide with lower opportunity. This distinction is particularly important for states such as Georgia and Louisiana. Although these states have lower absolute wind resources compared to traditional wind states, the observed correlations indicate that the limited DW potential that does exist is geographically well matched to areas of high EB. In such contexts, even modest BTM or small community-scale systems could potentially provide energy affordability benefits relative to their generation size.

States such as Iowa, North Carolina, and California appear in Group 1 for complementary reasons. Iowa and North Carolina rank highly in both absolute and standardized DW metrics, reinforcing their suitability across multiple deployment scales. California, despite lower standardized AEP_Demand due to exceptionally high electricity demand, remains strategically important because of its large population, economic scale, and localized pockets where EB and DW opportunity intersects.

Louisiana, while ranked only 28th in residential AEP_Demand and 32nd in total AEP, exhibits one of the strongest EB–AEP_Demand correlations nationally. In this context, “potential” reflects relative, localized suitability rather than statewide wind abundance, suggesting that targeted, BTM, or community-scale deployment in specific counties may be impactful even in low-wind-resource states.

These findings illustrate the diverse nature of EB and demand dynamics across the United States. Interpreting these relationships effectively requires a geographically nuanced approach, often down to the county level or beyond, since DW siting depends on localized wind resource and economics. Local economic structure, energy infrastructure, and policy environment all shape how EB and energy demand interact. The influence of agricultural economies, in particular, often crosses state lines, as explored in the mixed-effect modeling.

Consistent with prior research , our models show that poverty and agricultural employment are reliable associated factors of higher EB, although some global studies report mixed results . Agricultural employment likely reflects broader rural economic characteristics (such as lower and more variable incomes, older housing stock, or limited access to utility programs) rather than a direct causal effect of agricultural activity on energy costs and should therefore be interpreted as a proxy for rural economic disadvantage. The interaction between unemployment and agriculture in states like Wisconsin also highlights complex dynamics between industry structure and energy vulnerability. These findings suggest that DW planning should account for both demographic and economic context, not just technical feasibility.

It is important to clarify that while variables like unemployment and poverty frequently emerge as key covariates in explaining EB, this does not imply that DW deployment will directly reduce those underlying socioeconomic conditions. However, by potentially lowering energy costs in areas where these conditions are prevalent, DW may help ease energy-related hardship through activities like income generation and workforce development and indirectly contribute to improved quality of life.

This study demonstrates that a significant opportunity exists for targeted DW deployment to potentially alleviate economic hardships in high-EB states and counties with viable DW potential. However, implementation must consider disparities in conditions and environmental factors. Even in states with high overall AEP, deployment may be directed to counties where the benefits of energy cost reduction are most needed. Similarly, DW potential does not by default indicate energy cost savings, so care must be taken to ensure there will be reasonable savings before DW project design or implementation. In some areas, strong wind resources exist alongside relatively low EB; these contexts may require different strategies or benefit rationales, such as other forms of economic development, independence, or resilience objectives rather than energy affordability. Households or communities with elevated EB may also find investments in DW projects unaffordable, especially without suitable financing or funding options, even if there are identified long-term economic benefits.