Environmental Justice and Stream Restoration: A Multiscale Spatial Analysis for North Carolina

Abstract A multiscale spatial analysis was conducted for the state of North Carolina, which provides centralized state funding for stream restoration through the North Carolina Land and Water Fund (NCLWF). A total of 86 NCLWF restoration projects from 2013 to 2020 were geolocated and analyzed in relation to environmental justice metrics, both for a state-wide analysis using counties and for a within-county analysis using census block groups. State-wide, there was a strong negative association between a county’s proportion of people of color and its likelihood of having a restoration project (p = 0.001) and in the per-capita number of restoration projects (p = 0.000) and amount of project funding (p = 0.004). Within counties, there was weaker evidence for disparities. Finally, at the municipal scale, qualitative analysis is provided through a case study and discussion of recent stream restoration projects in Greensboro, North Carolina’s third largest city.


Introduction
For much of its history, stream restoration has been primarily focused on physical and ecological objectives, such as reducing flooding, improving water quality, and enhancing aquatic habitat (Bernhardt et al. 2005).There has, however, been a lack of empirical evidence to suggest that stream restoration projects have been successful in their stated objectives of improving ecological function, in large part because the driving forces of degradation are watershed-scale processes that are not affected by common restoration activities like channel reconfiguration (Vietz, Walsh, and Fletcher 2016).This is not to say that there are no benefits to in-channel stream restoration projects.Such projects can improve the esthetics, recreational opportunities, and accessibility of streams.In an inversion of the traditional biophysically based stream restoration paradigm, then, the primary benefits of in-channel restoration are arguably social rather than ecological.Accordingly, in recent years, the societal benefits of stream restoration have been increasingly emphasized by both researchers and practitioners (Chen, Wang, and Mo 2022).This shifting balance places a greater emphasis on societal benefits as major objectives of restoration projects, whereas traditionally these benefits have seen as secondary to ecological objectives.
In this paper, the theoretical connection between environmental justice and stream restoration is reviewed.Then an empirical case study is presented, focusing on the distributional and procedural justice elements of stream restoration in North Carolina.

The Environmental Justice of Stream Restoration
As stream restoration increasingly focuses on people as project beneficiaries, questions of power and justice become increasingly critical.Since the 1980s, the environmental justice movement has revealed the unequal distribution of environmental hazards and benefits based on sociodemographic variables like race, ethnicity, and class (Mohai, Pellow, and Timmons Roberts 2009).Environmental justice includes multiple forms of justice, including distributive justice (equal apportionment of social benefits and burdens) and procedural justice (equal rights to self-determination in societal decisionmaking).Over time, "first-generation" environmental justice, which focused on spatial proximities to hazardous waste and industrial sites among different groups, has evolved into more complex understandings of how structures of power are reproduced across space in multiple ways (Walker 2009).For example, Lane et al. (2022) explored how the historic racially discriminatory mortgage appraisal practice of redlining contributed to modern-day intraurban disparities in air pollution in US cities. Attending to the structural forces that result in unjust outcomes is critical to understanding the observed patterns of disparities.
In stream restoration, there are several aspects of environmental (in)justice that must be considered.The first is that, in line with first-generation environmental justice research, there is strong evidence that water pollution is disproportionately concentrated in poor communities of color.For example, in an analysis of the correlation between stream health indicators and demographic data in Michigan's Saginaw River watershed, Sanchez et al. (2014) found that census tracts with high nonwhite populations were more likely to be associated with poor stream health metrics, or alternatively with no streams at all.Another aspect of environmental justice relevant to stream restoration is flood risk.Particularly in non-coastal areas of the US South, low-lying flood-prone areas have largely been populated by poor people of color who were systematically excluded from safer areas with greater land value because of residential segregation (Ueland and Warf 2006).Finally, although first-generation environmental justice focused largely on distribution of environmental risks, the distribution of environmental benefits is equally important.Parks and green space are generally less accessible to minority and lowincome communities than to higher-income White communities (Riley and Gardiner 2020).Such disparities in opportunities for outdoor recreation and meaningful connections to nature are associated with disparities in physical and mental health (Jennings, Johnson Gaither, and Gragg 2012).By making streams more accessible, visible, and enjoyable, stream restoration projects have the potential to increase green space, with its attendant benefits.
From a distributional justice perspective, a just outcome would be to disproportionately focus restoration efforts on streams in marginalized communities to remedy these existing disparities.From a procedural justice standpoint, however, marginalized people also have less economic and political power to advocate for restoration projects in their communities.The technical and outcome-driven nature of stream restoration often excludes questions of justice (Hillman 2004).While early stream restoration projects were often initiated by citizens in a bottom-up fashion (Moran 2007), today restoration is more centralized by state agencies, meaning that existing inequities in public spending, the built environment, and land ownership can perpetuate injustices in site selection for restoration projects (Baptiste and Moran 2018).For example, Bradbury (2023) studied two urban streams in the Miami metropolitan area to determine why one in a wealthy and majority-White neighborhood was selected for a restoration project while another in a low-income neighborhood with a higher proportion of Black and Hispanic residents was considered but ultimately not selected.One key factor in the differing outcomes was the greater availability of undeveloped public land in the wealthier neighborhood, which greatly simplified the project with major cost savings.Even when restoration projects are implemented in marginalized communities, their qualitative character can be different from those in other communities.D ıaz-Pascacio et al. ( 2022) examined two urban streams in Austin, both of which were the recipients of flood-control projects.A stream in the more affluent and whiter part of the city got an esthetically pleasing vegetated Riverwalk, while in a part of the city that is dominated by African American, Hispanic, and poor White residents, a stream was instead outfitted with a trapezoidal concrete channel with no pedestrian access.Such disparate outcomes are not inevitable.Using a case study of an urban river in Milwaukee, Holifield and Schuelke (2015) argued for the need to consider variations in the desired trajectories of social-environmental systems, how those desires are formalized and circulated, and how to select among conflicting desires.Moran, Perreault, and Smardon (2019) provided a case study of a successful stream restoration project in Syracuse, New York, in which participatory co-production methods were used for sustained community engagement in the planning process.
Despite the extensive theoretical literature on stream restoration and environmental justice, there have been comparatively few empirical studies of the distribution of stream restoration projects and their relation to demographic variables.In a study of stream restoration projects in Pennsylvania, Moran (2010) found that they were disproportionately located in rural areas, thus neglecting the benefits urban communities could gain from projects.Stanford, Zavaleta, and Millard-Ball (2018) examined the relationship between stream restoration projects on the Central Coast of California and biophysical and social-political characteristics at the catchment scale.They found that projects were concentrated in catchments with impaired water quality, high population density, high pro-environmental voting, and high levels of education, income, and non-Hispanic White population.These studies provide evidence for disparities in the distribution of stream restoration projects within US states or regions based on race and income.Because these studies were conducted at a single scale, they do not address whether, for example, there are disparities in project siting within smaller spatial units, such as within urban areas.To address the question of whether disparities exist at multiple scales, a multi-scalar spatial analysis is needed.
This research aims to answer the following questions: 1) What is the distribution of stream restoration projects in the state of North Carolina and how does it relate to sociodemographic environmental justice metrics at the scale of both counties and census block groups?2) What elements of the stream restoration process result in the observed distribution of projects?

Study Area
The study area for this research is the US state of North Carolina, which holds special significance as the widely acknowledged birthplace of the environmental justice movement following community protests against a hazardous waste landfill in the poor and mostly Black Warren County in the 1980s.The state is divided into three physiographic provinces that broadly coincide with cultural regions as well (Figure 1).The western part of the state, which is mostly rural and has a larger proportion of White residents than the state average, is dominated by the Blue Ridge Mountains.The central Piedmont region, a low-relief plateau, is the most densely populated part of North Carolina, containing all three of the state's largest metropolitan areas (Charlotte, the Research Triangle of Raleigh/Durham/Chapel Hill, and the Piedmont Triad of Greensboro/Winston-Salem/High Point), with racially diverse populations and the state's highest average incomes.Eastern North Carolina consists of the low-lying Coastal Plain, a rural region with high proportions of Black residents and the highest poverty rates in the state.

Stream Restoration Projects
In 1996, the North Carolina General Assembly created the Clean Water Management Trust Fund, now known as the North Carolina Land and Water Fund (NCLWF).Through the fund, eligible applicants (state agencies, local governments, and nonprofit organizations) can apply for grants for projects to conserve North Carolina's land and water resources.Project categories include land acquisition, stormwater, and stream restoration.In its twenty-five years of existence, NCLWF projects have protected or restored over 4800 kilometers of streams and rivers (NCLWF 2022).In an analysis of stream restoration in the Southeast US, Sudduth, Meyer, and Bernhardt (2007) found that North Carolina had by far the most projects of the states in the region and attributed the large number of projects to North Carolina's funding model.NCLWF provides a unique opportunity to analyze stream restoration projects in North Carolina because 1) many restoration projects in the state receive funding from NCLWF, making it a highly centralized source of funding; 2) NCLWF has been operating for many years, allowing for a comprehensive assessment of past restoration projects; and 3) a database of funded projects is publicly available.Although compensatory mitigation, a separate funding mechanism, is also responsible for many stream (and wetland) restoration projects in North Carolina (Sudduth, Meyer, andBernhardt 2007, Lave 2018), this study focuses on NCLWF projects because they are more often publicly accessible sites that potentially provide public benefits.
In this study, the public database of NCLWF-funded projects, covering the years 2013 to 2020, was used as the data source for stream restoration projects in North Carolina (NCLWF 2022).The database provides the name, award amount, and county for each funded project.The projects are not geolocated in the database, so it was necessary to manually assign a geographic location to each project in ArcGIS at the level of National Hydrography Dataset Plus v2 (NHDPlus v2; EPA 2021) reach.In most cases, the project name contained the name of the stream which, combined with the county name, narrowed down the project location to one or several NHDPlus reaches.In cases in which the project stream consisted of multiple NHDPlus reaches, the correct reach was identified by searching for news stories or other public information about the restoration project, which would then lead to additional location information such as project maps, street names, confluences, or parks.In cases in which the specific NHDPlus reach of the project location could not be identified, a representative reach from near the middle of the stream's course was selected.In some cases, either the project stream could not be identified from the available information, or the project was state-wide, in which case those projects were omitted from the analysis dataset.Also, some multi-stage projects were funded in multiple years at the same location.For those projects, only the single project location was used, but the funding amounts from all project stages were summed to a single project cost.From the original 136 stream restoration projects in the database, after eliminating projects that had unidentifiable locations, were state-wide, or were repeats of multi-year projects, 86 projects (63%) were successfully geolocated to NHDPlus reaches in ArcGIS.

Environmental Justice Metrics
To characterize sociodemographic variables in North Carolina, the Environmental Protection Agency's EJScreen dataset was used (EPA 2022).Six demographic indicators from EJScreen were included: 1) people of color (the percent of individuals in a block group who list their racial status as a race other than White alone and/or list their ethnicity as Hispanic or Latino); 2) low-income (the percent of a block group's population in households where the household income is less than or equal to twice the federal poverty level); 3) linguistic isolation (the percent of people in a block group living in households in which all members age 14 and over speak a non-English language and also speak English less than "very well"); 4) less than high school education (percent of people age 25 or older in a block group whose education is short of a high school diploma); 5) under age 5 (the percent of people in a block group under the age of 5); and 6) over age 64 (the percent of people in a block group over the age of 64).
Instead of using raw percentages for the above indicators, EJScreen uses percentiles that compare a particular block group to the rest of the state, EPA region, or nation.In this study, state percentiles were used.That is, EJScreen assigns each block group in North Carolina a percentile score to indicate how its percentage people of color, for example, compares to other block groups in North Carolina.This percentile-based approach allows for the demographics of a particular location to be considered in the state-wide context of North Carolina, which is a more meaningful indicator of the patterns of environmental justice or injustice than raw percentages.
The analysis was conducted at two spatial scales.The first, the state-wide analysis, compares different counties in the state, with the objective of determining whether NCLWF stream restoration projects are more likely to be located in counties with particular environmental justice metrics.Because EJScreen data are available at the census block level only, ArcGIS was used to aggregate EJScreen data to the county level.Select by Location was used to select all the block groups that are within the county boundaries, then the demographic indicators for the selected block groups were averaged to create a county-level value for each variable.The second scale of analysis, the withincounty analysis, compared block groups within a county to the county averages, with the objective of determining whether stream restoration projects were more likely to be located in areas within a county with specific environmental justice metrics.

Spatial Analysis
For the state-wide analysis, the number of stream restoration projects was summed by county.First, non-parametric Mann-Whitney U-tests were used to test for significant differences in each of the environmental justice metrics between counties that had projects compared to those that had no projects over the study period.Logistic regression was used to assess the significance of the environmental justice metrics in predicting whether counties had projects.Then, additional analysis was completed for the counties with projects only.The number of projects and the total amount of project funding were normalized both by county population (US Census Bureau 2022) and by county total stream length (derived from NHDPlus; EPA 2021).The rationale for this normalization is that more populous counties or counties with more streams may be likely to have more restoration projects, all else being equal, so normalization allows for identification of areas in which there are more or fewer projects than expected.The significance of the bivariate correlations between county-level environmental justice metrics and normalized number of projects and total project funding were assessed using the non-parametric Spearman's rho coefficient.Multiple curves were fit to the relationship between project frequency/funding and the environmental justice metrics, and a power function was found to provide the best fit.Accordingly, multiple regression was run on the log-transformed environmental justice metrics to predict project frequency and funding.
For the within-county analysis, the block groups in which projects were located were extracted in ArcGIS, using Select by Location to identify the block group that contains or intersects the project reach.The environmental justice metrics for project-containing block groups were compared to the average values for the county in which the block group is located.Because of the small number of projects per county (maximum 8), it was not possible to test for significant differences in the environmental justice metrics between block groups with projects and block groups with no projects.Instead, the comparison was quantified descriptively using z-scores, which characterize the deviation of a given value (in this case, the value of the environmental justice metric for the project-containing block group) from the mean of the distribution (in this case, the mean environmental justice metric value for the county in which the block group is located).

Stream Restoration Projects
Of North Carolina's 100 counties, 40 had funded NCLWF stream restoration projects during the 2013-2020 study period (Figure 2).Funding for all included projects totaled approximately $34.8 million.Surry County, on the border of the Piedmont and Blue Ridge physiographic provinces, had both the largest number of projects at 8 and the largest amount of total funding at $6.2 million.When normalized by population, rural Ashe County, located in mountainous Western North Carolina, had the highest frequency of projects at 22.1 per 100,000 people.Surry County, however, still had the highest amount of project funding per capita, at $86.32 per person.When normalized by stream length, the highly urbanized Mecklenburg County, home to Charlotte, had both the highest frequency of projects at 8.3 per 1000 stream kilometers and the most project funding at $4081.33 per 1000 stream kilometers.

State-Wide Analysis
Mann-Whitney U-tests revealed that the percentile score for people of color was significantly (p ¼ 0.001) higher in counties with no NCLWF stream restoration projects (53.8) compared to counties with projects (37.1) (Figure 3 and Table 1).The percentile score for less than high school education, meanwhile, was significantly (p ¼ 0.011) higher in counties with projects (67.1) compared to counties with no projects (60.0).Counties with projects also had significantly (p ¼ 0.019) higher populations than counties without projects (mean of 162,448 compared to 66,503).There were no significant differences (at the 0.05 level) between counties with and without projects in the low-income, linguistically isolated, under age 5, over age 64, or stream length variables.
The logistic regression model correctly classified 78% of counties as having or not having projects.In the regression equation, the only variables that emerged as significant in a stepwise selection procedure were people of color (p ¼ 0.000) and population (p ¼ 0.006) (Table 2).This indicates that the effect of the education variable disappears when controlling for population and that having a high proportion of people of color is strongly negatively associated with the likelihood of a county having stream restoration projects.
For counties with stream restoration projects, there was a significant negative correlation between people of color and both the number of projects per 100,000 people (q ¼ À0.588, p ¼ 0.000) and the funding dollars per capita (q ¼ À0.450, p ¼ 0.004).The low-income metric was positively associated with project frequency per capita (q ¼ 0.319, p ¼ 0.045), but not significantly associated with project funding per capita at the 0.05 level (q ¼ 0.289, p ¼ 0.070).The linguistically isolated metric was negatively correlated with both per capita project frequency (q ¼ À0.409, p ¼ 0.009) and per capita project funding (q ¼ À0.315, p ¼ 0.047).Less than high school education was positively associated with per capita project frequency (q ¼ 0.414, p ¼ 0.008) but not significantly associated with per capita project funding (q ¼ 0.221, p ¼ 0.171).Under age 5 was negatively correlated with both per capita project frequency (q ¼ À0.643, p ¼ 0.000) and per capita project funding (q ¼ À0.568, p ¼ 0.000).Over age 64 was positively correlated with both per capita project frequency (q ¼ 0.624, p ¼ 0.000) and per capita project funding (q ¼ 0.527, p ¼ 0.000).Normalizing by stream length did not produce any significant correlations.
In the multiple regression model of per-capita project frequency against the log-transformed environmental justice metrics (R 2 ¼ 0.428, p ¼ 0.003), the most significant predictor variable was people of color (b ¼ À0.474, p ¼ 0.027) (Table 3).When per-capita project funding is instead used as the response variable in the model (R 2 ¼ 0.370, p ¼ 0.013), the most significant predictor variable is under age 5 (b ¼ À0.643, p ¼ 0.011).

Within-County Analysis
Z-scores were used to determine how much a NCLWF stream restoration project-containing block group differs from non-project-containing block groups in the same county and in which direction.Because the environmental justice metrics are not normally distributed, these z-scores cannot be interpreted in probabilistic terms as with normally distributed data.Instead, the z-scores are interpreted descriptively, with a score of >1 or <-1 indicating that the project-containing block group is somewhat different from the rest of the county with regard to the selected environmental justice metric, and a score of >2 or <-2 indicating that the project-containing block group is very different.Using these standards, of the 86 projects, 21 (24%) had z-scores exceeding the threshold for education.Of these, 13 were located in block groups in which the proportion of people with less than a high school education was less than the county average and 8 were in block groups in which there were more people with less than a high school education.For both people of color and low-income, there were 20 projects (23%) exceeding the z-score threshold, and in both cases 15 were in block groups with a lower proportion of people of color/low-income people than the county average and 5 with a higher proportion.For over age 64, 17 projects (20%) met the threshold, with 10 having higher than expected proportions of people over age 64 and 7 having lower than expected.For under age 5, 16 projects (19%) met the threshold, with equal numbers (8 each) of higher and lower than expected proportions of people under age 5.For linguistically isolated, only 4 projects (5%) met the threshold, with 3 having higher than expected proportions of linguistically isolated people and 1 having lower.

Distributive Justice
This study provides evidence of racial disparities in the siting of NCLWF stream restoration projects across North Carolina at a state-wide scale.The evidence includes 1) counties with no NCLWF restoration projects over the study period had significantly higher proportions of nonwhite populations than counties with projects; 2) in counties with projects, the proportion of people of color was strongly negatively correlated with both per capita project frequency and per capita project funding in bivariate and multivariate analyses; and 3) other variables that were correlated with project frequency in a bivariate analysis (i.e.income, language, education, age) did not emerge as significant predictor variables in the multivariate analysis, indicating that these variables were not significant independent predictors of project frequency.Population, however, was independently predictive of whether a county had restoration projects or not, with more populous counties being more likely to have projects.The most populous counties in the state also have relatively high proportions of people of color (e.g., 55% for Mecklenburg, 43% for Wake, 53% for Guilford) (US Census Bureau 2022).These urban counties were likely to have restoration projects, but the per capita frequency and funding of projects was relatively low because of the large total populations.The counties with the highest scores on the proportion of people of color metric, meanwhile, are on the Coastal Plain of Eastern North Carolina.For example, Robeson County is a majority-minority county, with 68 percent of its population identifying as either Native American (37% of the county population, mostly from the state-recognized Lumbee Tribe), Black or African American (22%), or Hispanic or Latino (10%).There were no NCLWF restoration projects in Robeson County during the study period.Other examples of counties with no projects and high proportions of people of color include Hertford (57% Black or African American), Edgecombe (56%), and Halifax (51%).These are all mostly rural counties on the Coastal Plain.In contrast, many of the counties with the lowest scores on the proportion of people of color metric are rural counties in the Blue Ridge.Examples include Madison (91% White), Haywood (90%), Ashe (90%), Yancey (90%), Mitchell (91%), and Watauga (83%), all of which had one or more NCLWF stream restoration projects during the study period.Because of the small populations in these counties, their per capita project frequencies are high.Overall, then, there is a pattern of high per capita project frequency in the mostly White counties of the Blue Ridge, high total frequency but low per capita frequency in the racially diverse urban Piedmont counties, and low absolute and per capita frequency in the Coastal Plain counties with high proportions of Black, Native American, and Hispanic populations.
The study provided weaker evidence for any disparities in the siting of stream restoration projects within North Carolina counties.Most restoration projects (>75%) were located in block groups that were demographically representative of their counties.The interpretation is that, once the first-order controls on the state-wide distribution of restoration projects described above are accounted for, there appear to be only minimal further disparities within smaller spatial units like metropolitan areas.It is worth noting, however, that for the projects located in block groups that are somewhat or very different from their counties, the majority (approximately two-thirds) of those block groups have lower proportions of people of color, low-income people, and people with less than a high school education than the counties in which they are located.This suggests that there may be at least a minor effect in which restoration projects are more likely to be sited in whiter, more affluent, and more educated areas within a county.

Procedural Justice
At a micro-level, the application and rating system for the NCLWF can be understood as a control on the narrow issue of which projects are funded through this specific mechanism.According to the stream restoration applicant's funding manual, applications are rated based on resource significance (15% of the total score), effectiveness/measurable outcomes (45%), other public benefits (10%), readiness (10%), and value (20%) (Bevington 2021).The "primary resource benefits" category awards points for criteria such as classification as Outstanding Resource Waters, wild trout habitat, or supporting species listed as federally threatened or endangered.This heavy emphasis on the biophysical condition of the stream potentially undermines projects that have more of a societal focus.The emphasis on trout habitat gives an advantage to the cold-water streams in Western North Carolina that can support trout, potentially diverting resources from the Coastal Plain.In addition, because restoration projects in North Carolina typically focus on single-thread meandering channels, this discourages projects on the multi-threaded channels and wetlands typical of Coastal Plain streams.The "effectiveness/measurable outcomes" criteria are largely based on technical computations of habitat or ecological uplift, but also includes a "readiness" criterion, which includes a rating of landowner interest.Areas of the state with higher proportions of public land, like western North Carolina, have more potential sites available for NCLWF restoration projects because of the greater ease of working with state or local land-management agencies rather than private landowners.This contrasts with compensatory mitigation projects, which are often located on private land through conservation easements established with landowners (Lave 2018).The "other public benefits" category awards points for recreational uses and public access, providing public or scientific education, development of riparian greenways, and being located near a public drinking water supply.These are the only explicitly societal criteria in the entire application, and they are only collectively worth 10% of the total score.Moreover, there is nothing in these criteria that specifically addresses the diversity of communities served by the project.Finally, the "value" criterion is based on the percentage and type of matching funds available for the project.Obtaining matching funds may be a significant barrier to implementing restoration projects through the NCLWF, especially for poor and rural parts of the state where local governments face major fiscal problems.
Overall, then, the rating system for NCLWF poses several challenges to the procedural environmental justice of stream restoration in North Carolina.Some of the challengesthe biophysical condition of streams, the unequal distribution of public land in the state, and the fiscal limitations of local governmentsare exogenous to the NCLWF.Others, however, might potentially be addressed within the NCLWF application and rating system.For example, increasing the weighting of societal benefits and including explicit criteria encouraging projects that serve racially diverse populations might help ameliorate some of the disparities observed in the state-wide distribution of NCLWF projects.
Broader elements of the procedural justice of NCLWF stream restoration projects align with those noted in the literature (e.g., Hillman 2004;Moran 2007).These include the technical expertise needed to develop a successful application and the need for state agencies, local governments, or nonprofit organizations to apply.State agencies, which initiate most stream restoration projects (Sudduth, Meyer, and Bernhardt 2007), may replicate patterns of environmental injustice by actively shifting environmental risk to marginalized communities.Local governments may lack the resources to develop stream restoration applications, provide matching funds, or acquire land (Moran 2007).Environmentally oriented nonprofit organizations, meanwhile, are likely to be concentrated in areas of the state where the local population and income levels can support them.Despite recent progress on racial diversity of staff and board members of environmental organizations, in a national analysis, Taylor (2015) found that minorities remained underrepresented at 11.7% of all staff and 6.1% of all board members of mainstream environmental organizations.These demographic disparities may translate into disparities in the types of communities served by the stream restoration projects these organizations choose to prioritize.

Greensboro Case Study
Examining specific restoration projects provides some qualitative interpretation to support the findings of the quantitative analysis.Here, a case study is presented for Greensboro, North Carolina's third largest city.Greensboro, located in the Piedmont Triad of central North Carolina, has a population of nearly 300,000 in its city limits and over 700,000 in its metropolitan area (US Census Bureau 2022).The city is in the headwaters of the Cape Fear River basin and, like other Piedmont cities, has an extensive network of small urban streams.Greensboro's two major streams are North and South Buffalo Creek.
Two stream restoration projects in the North Buffalo Creek watershed were recently funded by NCLWF (Figure S1).The first, funded in 2019 in the amount of $400,000, is on the mainstem of North Buffalo Creek in the Revolution Mill development.Revolution Mill is the site of a former cotton mill that was in operation from the late 1890s until 1982.In 2013, the mill complex was renovated into a multi-use development that includes luxury apartments, offices, restaurant and retail space, and art galleries.North Buffalo Creek runs right through the site, including under the building at one point.The in-progress restoration project will treat 610 linear meters of stream with bank stabilization, native plantings, and constructed riffles and cross-vanes (PCC 2022).The census block group where the project is located has relatively lower proportions of people of color (EJScreen score of 44) compared to the average for Guilford County (70).
The second recent NCLWF project in the North Buffalo Creek watershed is on College Branch, a small tributary that flows through downtown Greensboro.This project was funded in 2020, also for $400,000.College Branch flows alongside an abandoned rail corridor that will become the final leg of Greensboro's Downtown Greenway.When completed, the restoration project will enhance 622 linear meters of the stream with bank stabilization, improved bed substrate, and in-stream habitat structures (PCC 2022).The block group where this project is located also has much lower proportions of people of color than the county average (EJScreen score of 37, compared to 70 for Guilford County).
South Buffalo Creek runs through mostly commercial and industrial areas south of the city center.It was also the site of a NCLWF stream restoration project, funded in four phases from 2013 to 2018 for a total project cost of $2.5 million.The primary goal of this project was to reconnect the stream with its floodplain to spread out and slow down the flow while filtering sediment and pollutants (City of Greensboro 2022).Unlike the other two Greensboro projects, this one is in a block group that has a higher proportion of people of color (EJScreen score of 87) than the Guilford County average (70).Also unlike the other two projects, which are closely surrounded by residential and commercial areas, the South Buffalo Creek project took place on a mostly undeveloped floodplain surrounded by industrial areas and Interstate 40.
These three projects illustrate some of the elements of the environmental justice of stream restoration in North Carolina.The North Buffalo and College Branch projects are both in parts of the city that are whiter than the county average.They are also in affluent areas (a luxury housing development and downtown) and are geared toward facilitating the city's economic development (via the Revolution Mill complex and the Downtown Greenway).Both projects are largely focused on channel reconfiguration and improving the esthetic experience of people who live or recreate on the streams.In contrast, the South Buffalo Creek project, which is in a part of the city with a higher proportion of people of color than the county average, has a different focus, in that it is mostly focused on reducing problems of flooding, erosion, and water quality at a watershed scale.The amount of funding for the South Buffalo Creek project was much higher than for the other two Greensboro projects, but that is because it was a much largerscale project focused on watershed impacts rather than on the local experience of people living near or visiting the stream.Although the project does include some in-channel habitat structures and allows for public access because it is in a city park, the on-site esthetic and recreational benefits of the project are not prioritized in the same way that they are for the two projects in the North Buffalo Creek watershed.In particular, because the South Buffalo Creek project is on a mostly undeveloped floodplain with no residential areas immediately adjacent to the stream, the overall experience of neighborhood residents is not affected by the project to the same extent.Even when a quantitative analysis suggests parity in the siting of restoration projects, there can still be disparities in the qualitative nature of those projects, such as in who the primary beneficiaries are.
Moreover, to understand environmental justice of stream restoration more comprehensively, it is not enough to look at where restoration projects are.It is also necessary to look at where they are not.A prominent example is the Cottage Grove neighborhood of East Greensboro.For much of the twentieth century, Cottage Grove was a thriving center of Black-owned businesses and single-family homes, with many of its residents employed by the nearby mill.By the 1980s, with the closure of the mill and the advent of low-income public housing, the neighborhood began suffering from crime and disinvestment (Barnes 2020).Today, the census block group for Cottage Grove is in EJScreen's 93 rd percentile for people of color (compared to 70 for Guilford County), the 97 th percentile for low-income (55 for Guilford County), and the 97 th percentile for linguistically isolated (71 for Guilford County).
Several small unnamed tributaries of South Buffalo Creek run through the Cottage Grove neighborhood.These streams are all highly degraded, with significant incision, visible signs of pollution including thick mats of algae and abundant litter, and unpleasant odors.In a survey of Cottage Grove residents, only 25% of respondents strongly felt that there were safe places for children to play outside in the neighborhood, and over 81% said they would be much or somewhat more likely to spend more time outdoors if there were more walking trails and parks (Sills 2021).The two parks that are found in the neighborhood, Apache Park and Bingham Park, are both little-used and lacking in amenities.Apache Park is a narrow green space surrounding a highly degraded and flood-prone stream, adjacent to a public housing complex.Bingham Park is a larger park, surrounded by houses and apartments, where an extremely incised and polluted stream flows through.If these South Buffalo Creek tributaries were to be restored, it would have major benefits for the quality of life of neighborhood residents.In the case of the Bingham Park tributary, restoration is currently impossible because the park is built on top of a pre-regulatory landfill (i.e. a landfill that was established prior to environmental regulations governing safe operation) for incinerated municipal waste that was in operation from 1922 to 1953 (Barnes 2020).A study commissioned by the North Carolina Department of Environmental Quality (NCDEQ) found that the stream sediment contains excessive levels of heavy metals (NCDEQ 2019).Signs have been posted in the park warning visitors not to disturb the soil or come in contact with the stream water.This existing contamination limits the potential for stream restoration in the neighborhood and further restricts access to outdoor spaces for residents until the site is remediated.Meanwhile, areas of Cottage Grove outside the landfill-contaminated zone, like Apache Park, remain under-resourced and unrestored because stream restoration efforts have been focused on more affluent parts of the city like College Branch and Revolution Mill.
The Bingham Park case reveals additional dimensions of the environmental justice of stream restoration in North Carolina.The Bingham Park site is too degraded for stream restoration through typical mechanisms.Such streams are more likely to be located in poor communities of color like Cottage Grove because structural racism led to the initial siting of polluting facilities like incinerators and landfills and to the decades of segregation and disinvestment that followed.Even sites in Cottage Grove that are potentially eligible, like Apache Park, have not been the recipient of NCLWF projects despite being in serious need of restoration, thus severely limiting the opportunities for neighborhood residents to enjoy the benefits of restored streams found in other parts of Greensboro.An approach that specifically sites stream restoration projects in marginalized communities like Cottage Grove would not only be more environmentally just, it would have the potential to move toward a paradigm of reparative justice (Almassi 2017).

Conclusion
This study found evidence of distributive and procedural injustice in stream restoration in North Carolina.Counties with large proportions of people of color, especially in the rural Coastal Plain, were less likely to be the sites of NCLWF stream restoration projects, and race was the most significant predictor of both number of projects and amount of funding state-wide.These distributional inequities can be partially attributed to elements of the stream restoration process both endogenous to NCLWF and exogenous to it.The Greensboro case study provides some qualitative analysis of the nature of stream restoration projects that are and are not being completed in North Carolina.
Although the findings of this study are specific to North Carolina, some of the themes may apply to other geographic contexts as well.There is therefore a need for more empirical studies examining the distribution of stream restoration projects at multiple spatial scales and for research on how the procedural elements of restoration influence where restoration projects are done and where they are not.

Figure 1 .
Figure 1.Physiographic provinces and major rivers of North Carolina.

Figure 2 .
Figure 2. NCLWF-funded stream restoration projects over the 2013-2020 study period with EJScreen percentiles of (a) people of color; (b) low-income; (c) linguistic isolation; (d) less than high school education; (e) under age 5; and (f) over age 64.

Figure 3 .
Figure 3. Differences between North Carolina counties with and without NCLWF-funded stream restoration projects over the 2013-2020 study period in (a) people of color; (b) less than high school education; and (c) population.

Table 1 .
Results of Mann-Whitney U-test for differences between North Carolina counties with and without NCLWF-funded stream restoration projects over the 2013-2020 study period.

Table 2 .
Results of logistic regression model predicting the likelihood of North Carolina counties having NCLWF-funded restoration projects over the 2013-2020 study period based on environmental justice metrics, population, and stream length.

Table 3 .
Results of multiple regression model of the relationship between NCLWF stream restoration project frequency/funding in North Carolina counties over the 2013-2020 study period and environmental justice predictor variables.