Introduction

The United States (U.S.) is the world’s second-largest livestock producer and the largest beef producer1. Nearly all U.S. livestock, including 70% of cattle, and 98% of hogs, are reared at Animal Feeding Operations (AFOs)2. AFOs pack animals into confined areas where they are quickly brought to slaughter weight using energy-dense diets. Because of high stocking densities at AFOs, vegetation is scarce, and animal hooves often kick up dust. The sheer number of animals also produce immense volumes of manure and urine. These activities generate airborne particulate matter, both directly as dust and indirectly through precursors volatilized at manure lagoons.

Airborne pollution around open or semi-confined AFOs poses a health risk to neighboring communities3,4,5. Particulate matter below 2.5 microns diameter (PM2.5) is of particular concern as it lingers in the air for extended periods and is easily inhaled. Numerous studies link PM2.5 exposure with serious health conditions, including asthma, bronchitis, cardiovascular disease, and leukemia6,7. Research conducted by the American Medical Association has estimated that for every 10 micrograms per cubic meter (μg/m3) increase in PM2.5, there is a corresponding 4% increase in overall mortality, with a 6% increase in cardiopulmonary mortality and an 8% increase in lung cancer mortality8. These health impacts have been observed at communities up to 19 kilometers from AFOs9,10.

Research indicates that communities near AFOs are often lower income, less educated, and composed of higher proportions of racial minorities compared to more distant neighborhoods5,6,9,10,11. While the concentration of cattle AFOs and hog farms and associated air pollution in disadvantaged communities underscores a clear environmental injustice, critical knowledge gaps persist. Studies have primarily focused on individual U.S. states or counties, excluding some of the largest beef, dairy and hog-producing states12,13. Although nationwide assessments have highlighted the correlations between agriculture, PM2.5 levels, and heightened mortality rates in communities, they do not differentiate between livestock and other agricultural activities10,14,15. Conducting national assessments utilizing precise cattle AFO and hog farms locations and sizes would support more robust statistical analysis linking these facilities to nearby particulate levels and would clarify if environmental injustices are systemic or localized across the U.S.

To date, access to comprehensive data on cattle AFOs and hog farms location remains siloed at the individual state level, is often kept confidential by state agencies, or is incomplete. Accessible nationwide datasets only include coordinates of production facilities or livestock counts for large areas such as counties. Researchers have also downscaled agricultural censuses to estimate livestock density per area, but the underlying data are from 2010 and do not capture the current geography of livestock production in the U.S.15. These data limitations have hampered the deployment of targeted federal policies, as the livestock industry has long maintained that evidence linking AFOs to air pollution remains inconclusive nationally. A key challenge in making this link has been the lack of high-resolution data on both locations and sizes of cattle AFOs and hog farms, since omitting size in statistical models can downplay the influence of these operations on local environments. Our study provides a comprehensive, polygon-based dataset, including both size and location of feeding operations, that is necessary to refute this erroneous assertion and to develop sound evidence-based policies to reduce air pollution impacts associated with livestock production.

Building on this effort, we conduct a nationwide study specifically assessing PM2.5 around AFOs as opposed to PM2.5 from agriculture or livestock production more broadly10,16,17,18. We focus on pollution near two categories of AFOs: (1) cattle AFOs, defined here as feedlots producing beef and dairy farms producing milk (and beef as a byproduct), and (2) hog farms producing pigs. We employ remote sensing to develop the largest spatial dataset to date on AFOs, containing the precise locations and sizes of 15,726 cattle AFOs and hog farms across the U.S. This is a systematic, national-scale assessment of PM2.5 pollution near AFOs, spanning thousands of census tracts across the contiguous United States and capturing a wide diversity of geographic and demographic contexts. The inclusion of size is a key advance over comparable data products, as it facilitates statistical analysis linking cattle AFOs and hog farms to air pollution than previously possible. We then use geo-statistical models and statistical matching to establish connections between these facilities and PM2.5 pollution. Finally, we draw socio-demographic profiles of communities affected by AFO pollution. This analysis serves to further three policy objectives: (1) to support Clean Air Act (CAA) regulations for livestock operations, (2) to identify vulnerable populations bearing undue burdens associated with this pollution, and (3) to identify priority counties for pollution reduction measures.

We find that cattle AFOs are primarily concentrated in select counties within California, Wisconsin, Nebraska, Idaho, Iowa, and Texas, while hog farms are mostly clustered in Iowa, North Carolina, Minnesota, Oklahoma, and Missouri. Environmental justice concerns linked to these facilities are both concrete and widespread. Our analysis confirms that these facilities are predominantly situated in vulnerable communities, serving as focal points for heightened PM2.5 levels. Spatial clustering of cattle AFOs and hog farms underscores the potential for targeted interventions in these jurisdictions, such as enhanced air quality monitoring, adoption of cleaner technologies, and alternative farming methods. Whereas previous work has shown that racial and income-based disparities in PM2.5 exposure persist and have even worsened nationwide, our study narrows the scope to livestock-specific sources, providing new evidence that AFO-related air pollution is a significant and systemic contributor to environmental injustice14. Extending regulations like the CAA to cover livestock production could mitigate these disparate health impacts, particularly in counties with high concentrations of these facilities would further alleviate the burden on local communities.

Results

Our findings indicate that the geographical distribution of cattle AFOs differs from that of hog farms, as they are not typically located in the same regions. Our detailed dataset enables robust national analysis of whether AFO-related pollution is localized or broadly distributed. It also affirms known livestock production patterns and reveals clear links between AFO presence and disparities in pollution exposure, highlighting environmental justice concerns. Using geographically optimized matching, we find that AFOs raise local PM2.5 levels by 28% near cattle operations and 11% near hog farms, even after accounting for urban and industrial confounders.

The Geography of Cattle AFOs in the United States

We identified and digitized 8763 cattle AFOs spanning 212,687 hectares (ha) across the contiguous U.S. (Fig. 1). Most are concentrated in the Midwest, with 3614 facilities (maximum: Wisconsin with 869, minimum: Missouri with 20), and the West, with 3434 facilities (maximum: California with 1385, minimum: Nevada with 32) (Fig. 1). These regions host nine of the top 10 states with the most cattle AFOs. New York in the Northeast, which has a large dairy industry, is an exception. The top 10 states house 6129 cattle AFOs (70% of total) over 149,130 hectares (70% of total area), akin to the combined land area of Rhode Island and approaching that of Delaware. California leads in both number of AFOs and total area (34,592 ha), though differences in average size of these operations impact this ranking. For instance, although Texas has fewer facilities than Wisconsin (567 vs. 869), they cover a larger area (29,384 ha vs. 6689 ha) (Supplementary Table 1a).

Fig. 1: Spatial distribution of Animal Feeding Operations (AFOs) for cattle and hogs in the contiguous United States.
figure 1

Left top: Aggregated size of cattle AFOs per county, Right top: Number of cattle AFOs per county. Left bottom: Aggregated size of hog farms per county, Right bottom: Number of hog farms per county.

Full size image

At the county level, a mere 21 counties across the contiguous U.S. have over 50 cattle AFOs, collectively hosting 26% of facilities (2256) and 23% of total cattle-AFO area (48,537 ha) (Supplementary Table 2a). Six counties in California’s Central Valley alone accommodate half of these facilities (1033), covering 24,893 hectares. Notably, five of the top ten counties (Tulare, Stanislaus, Merced, Kings in California, and Weld in Colorado) have more area dedicated to cattle AFOs than some of the top states. For instance, Tulare County, CA, would rank sixth in the cattle AFO area and eighth in number of operations if treated as a state (Table 1).

Table 1 A) Top 10 states with highest number of cattle AFOs; B) Top 10 states with largest Average Size (ha) of cattle AFOs; C) Top 10 counties with highest number of cattle AFOs; D) Top 10 counties with largest Average Size (ha) of cattle AFOs
Full size table

Cattle AFO size across the United States

Across the U.S. the average cattle AFO covers 24.3 ± 38.7 hectares, with sizes ranging from large (656 ha in Yuma, CO) to small (0.24 ha in Texas). The 50 largest cattle AFOs average 319 ha and are concentrated in 12 states (Arizona, California, Colorado, Florida, Idaho, Kansas, Nebraska, New Mexico, Oklahoma, Oregon, Texas, and Washington) (Supplementary Table 3a). Southern states boast the largest operations (37.5 ± 55 ha), attributed to lower population densities, arid climates, and ample flat land, which reduce land and labor costs19,20. Moreover, a history of large-scale livestock production and established infrastructure support large-scale operations. Conversely, the Northeast hosts the smallest operations (8.34 ± 8.3 hectares), reflecting limited land availability and higher population densities20. Pairwise correlations confirm this trend, with Southern operations being significantly larger than those in the Northeast, Midwest, and West (P < 0.0001) (Supplementary Table 4b).

Within these regions, there is notable variation by state and county. The top five states in average cattle AFO size are Florida (69 ha), Arizona (62 ha), Oklahoma (54 ha), Kansas (53 ha), and Texas (52 ha). Interestingly, in four of the top 10 states, Wisconsin (8 ha), Iowa (12 ha), Washington (13 ha), and New York (10 ha), the average size is below the U.S. average due to stringent environmental regulations that limit AFO size and strictly control pollution21. These regulations encompass waste generation rates and requirements for waste treatment and disposal.

Several counties, even outside the top 10 states, host very large cattle AFOs, such as Duval, FL, Morrow, OR, and Lane, KS. Counties with some of the largest operations are clustered in California’s Central Valley, northern Texas, and southwest Kansas. The Central Valley, known for its dairy industry, along with northern Texas and southwest Kansas, offers suitable climates and available land conducive to large operations. Additionally, the presence of processing and transportation infrastructure, coupled with relatively lax regulatory environments22,23,24, contributes to the concentration of large cattle producers in these areas.

The geography of hog farms in the United States

We identified 6963 hog farms, spanning 53,847 hectares of land across the contiguous U.S. (Fig. 2). Hog farms exhibit a pronounced concentration in the Midwest, with 4325 sites (Max: Iowa with 2119, Min: North Dakota with 6), and in the South with 2381 sites (Max: North Carolina with 2328, Min: Florida with 3). These areas encapsulate the top 10 U.S. states with the highest density of hog farms (Table 2). The collective total of hog farms in the top 10 states amounts to 6019 (86% of the total), covering 41,272 hectares (77% of the total area), approximately equivalent to the land area of the District of Columbia. Iowa holds the lead in both the count of hog farms and the total area covered (10,197 ha) (Supplementary Table 1b).

Table 2 A) Top 10 states with highest number of hog farms; B) Top 10 states with largest Average Size (ha) of hog farms; C) Top 10 counties with highest number of hog farms; D) Top 10 counties with largest Average Size (ha) of hog farms
Full size table
Fig. 2: Comparison of 17-year average PM2.5 concentration between 1418 similar census tracts with and without presence of cattle AFOs and 1682 similar census tracts with and without presence of hog farms.
figure 2

The error bars show the range of most data points, with dots representing unusually high or low values.

Full size image

Only 28 counties across the contiguous U.S. boast over 50 hog production sites. These counties host 41% of all identified sites (2879) and 28% of the total hog farms area (15,277 ha) (Supplementary Table 2b). Over half of these sites (1486) are concentrated in seven counties spread across Iowa, North Carolina, Oklahoma, and Minnesota, covering 7291 hectares of land (Table 2).

Hog farms size across the United States

U.S. hog farms average 12.3 ± 27.1 hectares, or roughly half the average cattle AFO. As with cattle AFOs, hog farms exhibit significant variation, with sizes ranging from 621 hectares for a single hog farm in Woodward, OK to 1ha for a hog farm in Tennessee. The 30 largest hog farms are concentrated in 14 states (Oklahoma, Utah, Texas, Nebraska, Minnesota, Kansas, Missouri, South Dakota, Maryland, Florida, Alabama, Kentucky, and Georgia), averaging 182 hectares (Supplementary Table 3b). Hog farms in the West are larger on average, but unlike cattle AFOs, there’s less variation across regions (Supplementary Table 4b). This consistency across the South and Midwest can be attributed to the nature of hog production facilities and the land requirements for hog farming. However, compared to the Northeast, hog farms are larger overall. This difference may stem from the same pressures faced by beef and dairy producers.

Within these regions, there is notable variation by state and county. The top 5 states in terms of average farm size are Utah (102 ha), Montana (41 ha), Florida (42 ha), Arizona (31 ha), and New Mexico (28 ha). Counties where some of the largest hog farms are clustered in eastern North Carolina, Iowa, and eastern Nebraska. These areas have led U.S. hog production for decades and have mature infrastructure, strong industry presence, and political clout25. Below, we illustrate that areas such as these with high concentrations of cattle and hog AFOs are hotspots of environmental inequities stemming from PM2.5 emissions.

Cattle AFOs and hog farms correlated with higher PM2.5 concentrations

We employed a geographically optimized matching approach to examine correlations between presence of cattle AFOs, hog farms and PM2.5 levels across the conterminous U.S. This approach filtered out comparisons that were not meaningful, for example comparing PM2.5 levels over a small or isolated farm against an industrial site or a heavily urbanized area. Thus, we devised an approach to isolate areas where the largest farms are clustered. Using a univariate Getis-Ord General G model (P-value threshold 0.005), we identified 1418 census tracts that form geographic clusters with large total areas of cattle AFOs and 1682 census tracts that form clusters with large total areas of hog farm. The inclusion of total AFO area, as opposed to AFO presence, is key because larger AFOs house more animals and are at risk of generating more air pollution than smaller operations. Within each cluster, a statistical matching model allowed us to identify census tracts without the presence of AFOs but with similar characteristics such as climate, land use composition, other forms of industrial activity, transport infrastructure, and socio-demographics (Supplementary Table 5).

Our geographically optimized matching isolates AFO-specific impacts from urban and industrial confounders, providing a systematic and statistically robust evidence of livestock-induced air pollution on the national scale. Our analysis reveals that census tracts with cattle AFOs exhibit an average PM2.5 concentration of 7.7 μg/m3, 1.7 μg/m3 higher than identical tracts without these facilities (P < 0.0001) (Fig. 2). Similarly, census tracts with hog farms show an average PM2.5 concentration of 9.2 μg/m3, 0.9 μg/m3 higher than identical tracts without farms (P < 0.0001) (Fig. 2). These findings confirm that AFOs are associated with elevated local PM2.5 levels. Given the absence of a safe PM2.5 exposure limits, even slight increases in particulate pollution from AFOs pose health risks to nearby populations26.

Who lives near cattle AFOs and hog farms?

Our high-resolution, coast-to-coast mapping of cattle AFOs and hog farms facilitates a nationwide assessment of communities living near these facilities. Figure 3 shows the results of multinomial regression models of aggregate size as a function of demographic characteristics (Fig. 3). We ran this model in the previously identified large cattle AFOs and hog farms cluster and used a 10-mile (16.25 km) threshold for what constitutes nearby, a distance consistent with similar studies27.

Fig. 3: Results of the multinomial regression model.
figure 3

The census tracts that are within a 10 miles radius from a Cattle AFOs N = 1844 (Adjusted R-squared = 0.27), and b Hog farms N = 2391 (Adjusted R-squared = 0.24). Notes: The aggregated size of the AFOs per census tract was the response variable in the models. POC refers to People Of Color “POC” including Black or African American alone, American Indian and Alaska Native alone, Asian alone, Native Hawaiian and Other Pacific Islander alone.

Full size image

In line with earlier studies, we find that the total area of cattle AFOs and hog farms in a census tract tends to rise near socially and economically disadvantaged communities5,10,12. For instance, residents without high-school diplomas are more likely to reside nearby cattle AFOs (P < 0.0001). The same holds true for Latino residents (P < 0.0001). Somewhat unexpectedly, our model showed that the aggregated size of cattle AFOs is associated with lower poverty rates (P < 0.0001) in nearby communities and an explanation would be that these facilities could be a source of income and employment. However, official statistics might overlook undocumented or seasonal livestock laborers who are more likely to live below the poverty line28. Lastly, we find that health-insurance coverage rates tend to drop significantly in areas near cattle AFOs and hog farms (P < 0.0001). In the U.S., due to lack of nationalized health care, uninsured residents are less likely to seek medical care29. Given the link between these facilities and PM2.5, this co-location of cattle AFOs, hog farms and uninsured residents represents a potential health risk and significant environmental injustice (Supplementary Table 6).

We estimate that 632,216 residents without health insurance live in close proximity (<10 miles) to cattle AFOs and 763,588 residents without health insurance live close to hog farms. The results from the Local Indicators for Spatial Association model add geographic specificity of this coexistence and are illustrated in Fig. 4. These high-risk areas of cattle AFOs are clustered in the Central Valley of California, the Texas panhandle, and lower Kansas. Additional smaller high-risk clusters are in Oklahoma, eastern Colorado, eastern Missouri, and southern Idaho. Two large clusters where high AFOs concentrations overlap with high insurance coverage exist; one in the plains where Iowa, South Dakota, Minnesota, and Nebraska meet, and another in Upper Wisconsin. The high-risk areas of hog farms are clustered in northern Texas, southern Oklahoma and North Carolina. There are also small clusters of uninsured populations in Utah, Iowa and Missouri, where targeted policies are needed to reduce emissions and implement public health interventions.

Fig. 4: Bivariate Local Indicators of Spatial Autocorrelation of aggregated size of cattle AFOs and hog farms and percentage of uninsured population.
figure 4

The High-High cluster (in red) highlights census tracts where targeted policies are needed to reduce emissions and implement public health interventions.

Full size image

Discussion

We developed a comprehensive, national dataset of cattle AFOs and hog farms to evaluate their contribution to ambient PM2.5 concentrations across the U.S. Using this high-resolution, facility-specific dataset, we apply a geographically optimized matching method to isolate AFO-specific pollution impacts. Spatial clusters of large AFOs were identified using the Getis-Ord General G statistic (P < 0.005), yielding 1418 census tracts for cattle and 1682 for hog operations based on total AFO area, a proxy for herd size and pollution potential. Within these clusters, we matched AFO-exposed tracts to unexposed counterparts with similar climate, land use, industrial activity, infrastructure, and socio-demographic profiles. Census tracts with cattle operations exhibit 28% higher PM2.5 levels and those with hog farms 11% higher than matched controls (P < 0.0001). Our results provide a statistically robust, national-scale evidence of livestock-induced air pollution and show that socially vulnerable and minority populations—often with limited health insurance—are disproportionately affected.

Unlike prior small-scale studies, our facility-level national analysis provides systematic, policy-relevant evidence and pinpoints regions where targeted interventions may yield the highest public health gains. Our approach also enables assessment of whether environmental injustices linked to AFOs are systemic or location-specific. The dataset we developed can serve as a foundational resource for future research into other environmental impacts of industrial livestock production, including water contamination, eutrophication, and methane emissions30,31,32. It also supports scenario modeling to evaluate the potential benefits and trade-offs of transitioning to more sustainable systems, such as regenerative agriculture and agroecology, to advance a resilient and equitable food future.

Public health researchers can extend our work by establishing connections between AFOs, disease burdens, and health outcomes. While deterministic models suggest these linkages, empirical evidence is lacking33. The public health ramifications of livestock production are extensive and diverse, encompassing concerns from antimicrobial resistance to health risks associated with air pollution6,34,35,36. Addressing these urgent challenges requires evidence-based policies, including stringent regulations on air pollutant emissions from AFOs and targeted public health measures to mitigate health disparities and protect affected populations37,38. Such interventions not only safeguard public health but also promote environmental justice and social equity in rural areas39,40. Additionally, our dataset can be used to study the impacts of AFOs across multiple sustainability dimensions, including economic (real estate values, employment), ecosystem services, and ecosystem functioning.

Our data and findings offer valuable insights to address the health impacts linked with livestock production. We pinpoint areas where cattle AFOs or hog farms, elevated pollution levels, and vulnerable populations converge. While the presence of an AFO can offer employment opportunities and decrease poverty rates, our study reveals that uninsured, less-educated, and minority residents in these areas face heightened exposure to PM2.5 and associated health risks. Unfortunately, they often lack access to medical resources or knowledge to take preventive measures. This increases the likelihood of emergency room visits, hospitalizations, and poor health outcomes, incurring substantial financial costs for the healthcare system41,42.

Public health authorities and advocates should prioritize targeted interventions in these high-risk communities we identified. These interventions might involve increased visits from free mobile health clinics and public health information campaigns designed for vulnerable populations (e.g. in Spanish for migrant workers). Potential dietary and pharmacological methods, such as antioxidants, could address the toxicological mechanisms associated with PM2.5 pollution for the people in these areas43. On-site strategies at AFOs such as lagoon covers, wind breaks, and buffers could mitigate the direct impact of PM2.5 on proximate communities44. Some technical approaches to reduce PM2.5 include air scrubbing, manure solid-liquid separation, manure acidification and digestion, optimizing diets to reduce enteric fermentation, diversification of livestock animals and crops and utilizing fertilizers more efficiently45,46.

The dataset also holds potential for training automated systems to identify AFOs using remote sensing and machine learning, enabling near real-time updates. However, this approach may compromise accuracy, particularly in distinguishing between hog and cattle AFOs. Existing efforts using ML techniques could only identify less than 3000 locations of AFOs in the United States47. An additional advancement would involve incorporating poultry farms into our dataset. Including poultry would facilitate comparisons with official state-level statistics, which often exclude AFO location, offering a comprehensive overview of the geographic distribution, environmental impacts, and distributional justice within the entire livestock industry in the U.S.

Our findings have implications for strengthening policies aimed at mitigating air pollution from existing AFOs. The livestock industry has stymied attempts to include AFOs under the CAA48 by arguing that links between these facilities and air pollution are inconclusive49. However, our study clearly demonstrates the existence of this link. Residents, NGOs, and local governments should advocate for national regulations on PM2.5 emissions from AFOs based on our findings. Additionally, livestock producers can take steps to reduce PM2.5 emissions, such as by using sprinklers to keep manure and soil wet, particularly in highly polluted areas. Even if implemented in a piecemeal fashion, these efforts would eventually achieve meaningful reductions in air pollution. Our dataset offers a valuable tool for establishing National Pollutant Discharge Elimination System compliance priorities. The EPA has highlighted the lack of reliable size-related data on AFOs, which has hampered efforts to prioritize different jurisdictions50.

The geographic distribution of cattle AFOs and hog farms warrants heightened scrutiny. Our dataset provides clear identification of their locations, enabling communities and NGOs to search for evidence of regulatory capture. This may involve examining spatial clustering of AFOs in regions where livestock companies have made substantial political donations or lobbied for lenient environmental regulations5. Governments could also consider imposing moratoriums on AFO expansion in areas of existing concentration or where residents are most vulnerable. Such measures would avoid exacerbating existing environmental injustices51.

In addition to exploring new questions with our dataset, future research should also improve the data and try to causally link AFOs and pollution. One advance would be to automate AFO identification with ML. Manually identifying AFOs is labor-intensive, even if it is more accurate than current ML models50. Another challenge is visually distinguishing between hog farms and cattle AFOs in satellite imagery. To address this, we implemented a systematic approach to manually validate our findings against county-level statistics on counts of AFOs. As ML models improve in accuracy and computational constraints to nationwide analysis are lifted, the AFO identification and segmentation process should be automated. A related limitation is the impracticality of conducting statistical matching for the entire U.S. To overcome this, we employed a geographically optimized matching approach using a subset of counties with high concentrations of AFOs. Another limitation of matching is that it does not causally link AFOs to degraded air quality. To establish causal links, different statistical models such as Regression Discontinuity Design, Difference-in-Differences, and Randomized Controlled Trials are required. If methodological challenges can be overcome (e.g., need for longitudinal data), researchers can use these methods in nationwide studies exploring causal relationships between industrial livestock farming, environmental change, and community health.

Livestock farms play a vital role in the economies of many rural communities and are expected to continue doing so for the foreseeable future. Our study highlights how these facilities increase PM2.5 levels and contribute to enduring distributional justice issues in affected communities. These challenges are widespread and impact millions of Americans nationwide. This study serves as a starting point for future research on environmental justice issues related to livestock production and identifies priority counties where stakeholders should concentrate their efforts to promote sustainability and equity in livestock production.

Materials and methods

Generating the dataset

To develop a large dataset of cattle AFOs and hog farms in the U.S., we merged three incomplete existing datasets with an extended survey using high-resolution satellite imagery and systematic visual inspection. Initially, we utilized multiple EPA sources at state and regional levels, supplementing these with inquiries to all six EPA regions for any available non-public AFOs datasets. The consolidated dataset, termed EPA Dataset, comprises 7058 cattle AFOs and 2378 hog farms, each georeferenced. We validated each entry using high-resolution Google satellite imagery (nominal years 2017–2019) via the HCMGIS plugin in QGIS. However, limitations include variations in reporting standards across EPA regions and underreporting by several states, affecting the dataset’s representativeness. Additionally, we identified and removed duplicate entries from our sample.

We augmented the EPA dataset with information from the USA Business database sourced from Data Axle52. Using the NAICS code, we extracted 4138 cattle facilities and 317 hog farms. These entries were cross-checked against Google satellite imagery. However, the Data Axle dataset displayed systematic underrepresentation and missing data, particularly for smaller AFOs, and contained numerous false positives, such as company headquarters and slaughterhouses, which we filtered out. Subsequently, we integrated data from CounterGlow53, a crowd-sourced dataset, identifying 10,299 cattle AFOs and 7800 hog farms. After removing duplicates and false positives, we combined all three datasets into a single layer, resulting in the locations of 6018 cattle AFOs and 4163 hog farms.

To enhance the dataset’s representativeness, we extended it to include farms systematically omitted in all three databases. Initially, we identified meat-producing regions based on USDA cattle and hog counts per county, comparing them with previously identified AFOs. Regions underreported by the databases were pinpointed, and a nationwide 10 × 10 km grid was established. Each grid cell within the identified regions underwent visual inspection by zooming in through each grid via Google satellite imagery to identify additional AFOs, aiming to better represent all meat-producing regions. A secondary objective was to digitize each AFO to ascertain its size, serving as a proxy for capacity and environmental impact. We classified a land patch as an AFO when there was: (1) clear evidence of cattle or hog and (2) clear evidence of manure piles, or paddocks, Fig. 5. We reviewed validated images of cattle AFOs, and hog farms in different U.S. states to be able to accurately differentiate between AFO types. Through this process, we visually identified 2746 cattle AFOs and 2800 hog farms, ultimately digitizing a total of 15,726 AFOs.

Fig. 5: Examples of identified cattle and hog AFOs through visual inspection of high-resolution satellite imagery from Google Earth.
figure 5

Red outlines indicate the digitized boundaries of each AFO. (Satellite imagery courtesy of Google and Maxar Technologies, 2022).

Full size image

Matching analysis

We employed statistical matching, a modeling approach, to assess the impact of AFO presence on PM2.5 concentration. This method involves matching census tracts with cattle AFOs or hog farms to similar tracts without these facilities, facilitating a meaningful and valid comparison. By reducing bias introduced by covariates, matching enables a comparison of PM2.5 concentration among these census tracts54. We obtained PM2.5 concentration data spanning 2000–2016, at a 1 km spatial resolution for the contiguous U.S., from an Ensemble-based Model55. We calculated the 17-year average and spatially joined the results to the census tracts. Additionally, we created a binary indicator of cattle AFOs or hog farms presence or absence for each census tract.

To focus on regions with concentrated large AFOs and control for geography, we computed the total size of cattle AFOs or hog farms per census tract. Next, we extracted polygon centroids from the count points in the polygon function in QGIS, with AFO size as the weight field. The outcome was the aggregated size of all cattle AFOs or hog farms within each census tract. This served as the response variable in our models. To pinpoint clusters with large areas of cattle AFOs or hog farms across the U.S., we utilized the univariate Getis-Ord Local G model56, considering P values less than 0.005 as significant. This analysis identified 1,418 census tracts for cattle AFOs and 1682 census tracts for hog farms (Supplementary Fig. 1).

We ran optimal matching models within these regions for cattle and hogs separately using the matchit package in R. A cohort of demographics, climate, geomorphology, land cover, built environment, and proximity to water served as control variables (Supplementary Table 5). AFO presence or absence was treated as the outcome variable for matching. This process entailed comparing 623 census tracts with at least one cattle AFO to 623 identical census tracts without any cattle AFO, and 584 census tract with at least one hog farm to 584 identical census tracts without any hog farms (refer to Supplementary Fig. 2 and Supplementary Table 5 in the Supplementary Materials for diagnostics). To ensure matching reliability, tests on covariate distribution were conducted to detect significant differences in other influential factors. Post-matching, most covariates displayed a standardized bias of less than 10%. Standardized deviation in control variables after matching fell within an acceptable range, affirming the dependability of covariate matching and passing the balance test. Visual representation of bias improvement is provided in Fig. SI1. Post-matching, we introduced the 17-year average PM2.5 and calculated concentration differences between matched census tracts with cattle AFOs or hog farms and those without, using logistic regression.

Statistical analysis

Focusing on the largest AFOs clusters (separate models for cattle and hog) identified from the Getis-Ord Local G model56, we aimed to characterize the socio-demographic profiles of communities residing within a 16 km radius from the nearest AFOs. We employed multinomial regression models to explore the relationship between socio-demographics and the aggregated size of AFOs. Census tract-level data from the 2015–2019 U.S. census57 and the 2020 U.S. Social Vulnerability Index served as predictors58. The variables included: Percentage of Latino, Percentage of People of Color (Black and African American, American Indian and Alaska Native, Asian, Native Hawaiian and Other Pacific Islander), Percentage of Persons (age 25+) with no high school diploma, Percentage of persons below 150% poverty, Percentage of Uninsured in the total civilian noninstitutionalized population, and Percentage of civilian (age 16+) unemployed. To select our sample, we computed the Euclidean distance of each census tract to the five nearest cattle AFOs or hog farms and calculated the average. We included census tracts within a 16 km radius from the nearest AFO, resulting in a sample of 1844 census tracts for cattle AFOs and 2391 census tracts for hog farms. For hog farms, we used the size of the farm as a weight in our analysis because hog farms are typically much smaller than cattle AFOs, and the size of the farm has a significant impact on the surrounding communities.

To enhance the geographical specificity of the regression findings, we augmented the analysis with a bivariate spatial clustering model. This model facilitated the identification of clusters where high concentrations of AFOs and high rates of individuals lacking health insurance coexist across the U.S. We employed the bivariate Local Indicators of Spatial Association (LISA) with a P value threshold of 0.005 and adhered to the queen’s rule of adjacency, considering one census tract59.