Introduction

The link between surface water quality and land-cover has been firmly established through studies demonstrating that impervious surfaces contribute to declining stream health and water quality (Schueler 1994; Arnold and Gibbons 1996; Booth and Jackson 1997). Impervious surfaces amplify the transport of polluted stormwater runoff into urban streams, often directly through a network of channels and pipes (Arnold and Gibbons 1996; Paul and Meyer 2001; Walsh 2004). Generally, surface water quality degrades when impervious surfaces comprise 10 % of a watershed, and becomes significantly degraded as that level approaches 20–30 % (Arnold and Gibbons 1996; Booth and Jackson 1997). Impervious cover also replaces and fragments vegetation that could otherwise maintain or improve stream health (Bolund and Hunhammar 1999; King et al. 2005; Nowak and Greenfield 2012). Thus, management suggestions often emphasize limiting imperviousness and utilizing alternative development patterns that preserve contiguous areas of vegetation, especially in and adjacent to riparian zones (Arnold and Gibbons 1996).

Although the general relationship between water quality and impervious cover is well known, water quality varies greatly, in unexplained ways, among watersheds with similar levels of impervious cover. In a recent meta-analysis of empirical studies examining this relationship, Schueler et al. (2009) found variation in stream and water quality indicators to be so pervasive that it prompted a specific revision to the impervious cover model (Schueler 1994) that highlights this variation (Schueler 1994; Booth et al. 2002; Schiff and Benoit 2007). While many dynamic elements likely play a role in this process (Schueler et al. 2009), the arrangement of land-cover types and stormwater infrastructure is particularly interesting indicators that have recently been used to explore variation in stream quality (Walsh et al. 2004; Zampella et al. 2007; Alberti et al. 2007; Shandas and Alberti 2009).

Current research suggest that protective vegetated buffer zones are less effective in urban and suburban watersheds due to the cumulative effect of watershed-scale, non-point source pollutants (Pratt and Chang 2012), and stormwater infrastructure that bypasses riparian buffers (Walsh et al. 2004). These mechanisms essentially extend the riparian boundary throughout a given watershed, which emphasizes the need for more advanced land-cover pattern analyses. For example, an index of upland vegetation fragmentation and the percentage of vegetation in the riparian corridor explained a significant proportion of variation in stream biological conditions among 51 nested watersheds in coastal Washington (Shandas and Alberti 2009). Similarly, a set of variables describing the extent and distribution of urban development best explained the water quality variation among six watersheds in Durham, NC (Carle et al. 2005). Patterns of imperviousness and forest cover have proven to be important water quality predictors (Alberti et al. 2007), as have quantitative measures of imperiousness with direct connections to stormwater infrastructure (Walsh et al. 2004). Most importantly, these studies demonstrate the underappreciated effects that land-cover patterns can have on water quality.

Despite these advances, the methodologies used to link landscape patterns to water quality variation are often disparate and generally focus on some combination of landscape metrics describing imperviousness, infrastructure, or aggregated measures of vegetation or forest cover patterns. These approaches, although conceptually innovative, overlook potentially significant landscape elements, such as grass and buildings that contribute to the heterogeneous nature of urban systems (Cadenasso et al. 2007). We suspect that residual unexplained water quality variation among prior studies is likely an artifact of unrecognized variation in land-cover patterns among watersheds within the same range of imperviousness. For instance, upland and riparian forest fragmentation certainly influences stream conditions and explains a proportion of water quality variation among watersheds (Alberti et al. 2007). Additional variation might be explained by exploring the land-cover dynamics that contribute to this fragmentation, such as the composition and configuration of grass, buildings, and pavement. If land-cover patterns do indeed vary among urbanized watersheds with similar levels of impervious cover, future studies would greatly benefit from understanding this variation prior to testing water quality.

In this study, our goal is to analyze and compare land-cover patterns among forested urban watersheds falling within a 20–30 % (degraded) range of impervious cover, and determine their structural contrasts. We hypothesized that the interactions among land-cover types within the study watersheds are complex and that patterns vary significantly despite relatively consistent levels of impervious surface. Additionally, we hypothesized that grass and buildings, which are often aggregated into more general measures of impervious surface or vegetation, contribute meaningfully to the predicted overall variation among land-cover patterns. The strength of our analysis is that we included all land-cover types represented within the urban matrix (pavement, buildings, grass, and trees), as the relationships among these complex land-covers have yet to be implemented in the design of water quality studies. Considering there is not yet sufficient evidence describing the role of land-cover patterns in shaping catchment-scale water quality, our final goal was to draw from theoretical and empirical examples in the literature to examine the implications of our results for urban watershed management.

Materials and Methods

Study Area

Centrally located within the North Carolina Piedmont, the Triangle region encompasses portions of Wake, Durham, and Orange Counties. The three primary Triangle cities are Durham (Durham County), Chapel Hill/Carrboro (Orange County), and Raleigh (the state capital in Wake County). The Triangle is one of the fastest growing urban areas in the country (Fisher 2012) largely due to the influx of high-technology workers to the Research Triangle Park industrial center, and the presence of three top-tier research universities: NC State (Raleigh), Duke (Durham), and the University of North Carolina (Chapel Hill) (Rohe 2011).

Situated in the US Forest Services’ Southern Mixed Forest Province, the region is characterized by woody vegetation, subtropical rainfall levels (between 1016 and 1524 mm/year), and numerous streams, pocosins, swamps, ponds, and lakes (Bailey 1995). Triangle municipalities are dotted with moderate to dense tree cover and an abundance of perennial and ephemeral streams. A recent study has shown that Raleigh boasts more than 50 % canopy cover, making it one of the most tree-covered cities in the country (Bigsby et al. 2013). These factors, coupled with the region’s high rate of growth, make it an ideal location to study complex urban land-cover patterns and their potential effects on stream water quality.

Watershed Selection

To select watersheds of similar size and imperviousness across the Triangle, Somers et al. (in prep) delineated all first-order streams for which stormwater infrastructure data were available within the municipal boundaries of Chapel Hill, Carrboro, Durham, and Raleigh. These streams were defined as first-order using available stream data, not including piped streams, and represent the best approximations of headwater, first-order streams in an urban environment. Of these delineated watersheds, 32 fell within the significantly degraded (20–30 %) range of imperviousness as measured by high-resolution land-cover data, and were selected for analysis; three in Chapel Hill/Carrboro, nine in Durham, 19 in Raleigh, and one that is split equally between Durham and Chapel Hill (Fig. 1; Table 1).

Fig. 1
figure 1

Map of study area, watersheds, and major hydrography shows locations of small-scale first-order watersheds within Raleigh, Durham, and Chapel Hill/Carrboro. There are three in Carrboro/Chapel Hill, 9 in Durham, and 19 in Raleigh. One is split between Chapel Hill and Durham. Major hydrography shows the significant water bodies and rivers in the region. The analysis watersheds amount to sewer-sheds, where the stormwater infrastructure reroutes natural flow patterns

Full size image
Table 1 Watershed Summary Statistics
Full size table

Land-Cover Classification

We used United States Department of Agriculture (USDA) 2009 color infrared National Agricultural Imagery Program (NAIP) aerial photo mosaics to create high-resolution (1 m) land-cover classifications for the Triangle cities of Raleigh, Durham, and Chapel Hill. These classifications were created using Erdas IMAGINE’s unsupervised classification protocol, setting 1200 individual spectral classes. A custom post-processing model that relies on ancillary vector pavement and building footprint data was applied to generate six land-cover classes, which are most useful for biophysical analyses in cities (Ridd 1995; Cadenasso et al. 2007): (1) coarse vegetation (trees); (2) water; (3) pavement impervious (roads, parking lots, driveways); (4) buildings; (5) fine vegetation (grass, non-woody vegetation); and (6) bare earth (Fig. 2). Accuracies were assessed using ultra-high-resolution (0.5 ft2) imagery with 50 random points assigned to each class in each municipality, checked against the classified land-cover (Congalton 1991). Overall accuracies were 89.33 % in Chapel Hill (KHAT = 87.2 %), 90.67 % in 117 Durham (KHAT = 88.80 %), and 89.33 % in Raleigh (KHAT = 87.2 %).

Fig. 2
figure 2

High-resolution land-cover snapshot shows fine detail of 1 m land-cover classification. With this, we are able to accurately measure urban land-cover patterns within our study watersheds. Coarse vegetation is represented in dark green, pavement impervious surfaces are gray, building impervious surfaces are red, open water is blue, fine vegetation is light green, and bare earth is tan/brown

Full size image

Storm Water Infrastructure

We collected vector data delineating stormwater pipes, inlets, and outfalls directly from city stormwater management and engineering departments within the study area. This is the most inclusive stormwater network coverage for Triangle municipalities to date, though completeness was not guarantee due to ongoing projects and maintenance.

Metric Calculations

We extracted four individual land-cover types (trees, pavement, buildings, and grass) from the high-resolution (1 m2) land-cover classification into the 32 selected urban watershed boundaries using the spatial analyst extraction toolset in ArcGIS 10.1. The two remaining land-cover types, water and bare earth, were not sufficiently represented within these watersheds, and did not warrant consideration in this analysis. Stormwater inlets, outlets, and pipe networks were isolated to watershed boundaries to quantify density measurements.

Land-cover configuration pattern metrics were derived using FRAGSTATS 4.0 (McGarigal et al. 2012). FRAGSTATS is a commonly used, fully automated spatial statistics tool originally developed for the USDA Forest Service to quantify spatial patterns for ecological assessments (McGarigal and Marks 1995).

We calculated five class-level metrics—patch density, edge density, mean patch area, mean contiguity, and land-cover percentage—for each of the four land-cover class types using the high-resolution (1 m2) classification dataset (Table 2). Stormwater inlets, outlets, and pipe densities were also calculated, creating a total of 23 analysis metrics for each of the 32 study watersheds (Table 2). Patch and edge densities, mean patch area, and mean contiguity are relatively simplified metrics that describe the fragmentation, heterogeneity, and connectivity of urban watersheds. While the percentage of land-cover types tells us little about landscape patterns, it does aid us in our interpretation of landscape variation by serving as a measure of land-cover composition.

Table 2 Analysis metric list/measurements
Full size table

Statistical Analysis

To test for the presence of structural variations in land-cover among urban watersheds, all 23 metrics (5 pattern metrics for each of 4 land-cover types and 3 stormwater infrastructure density variables) were subjected to three multivariate statistical analyses: (1) multivariate 152 scatter plots/correlations (JMP 10.0; SAS); (2) a hierarchical cluster analysis with heat map (JMP 10.0); and (3) a Principal Components Analysis (SAS 2012). These three tests constitute a comprehensive, though interpretive, method for uncovering structural variations in land-cover configuration, allowing us to make inferences about potential sources of water quality variation among urban watersheds.

Results

Watershed Clustering

Four distinct catchment clusters emerged from this analysis, which suggests that land-cover patterns do vary among potentially degraded watersheds (Fig. 3). We examined the land-cover patterns in each cluster and describe the four categories below. We include short descriptive names for each category.

Fig. 3
figure 3

Watershed clustering shows four distinct watershed groupings (JMP). Land-cover from select watersheds from cluster groups 1–4 (Group 1 left, group 4 right) are shown at the bottom of the dendrogram. Analysis metrics are shown in rows, numerical watershed identifiers are shown on the bottom as columns. High metric values are depicted as red, low metrics values are blue, and middle range values are green to orange. Clusters of values can be easily viewed in the heat map, and the general landscape structure of each cluster group can be interpreted based on these metric values. This confirms the existence of land-cover pattern variation among watersheds within a 20–30 % gradient of impervious surface cover

Full size image

Cluster 1: Semi-City Living (9 Watersheds)

These watersheds comprise well-connected forest patches and relatively moderate levels of pavement. Buildings are large and fairly close together, and grass patches are sparse and fragmented by other cover types. These watersheds boast moderately dense developments, many of which are likely commercial (large buildings), mixed with clustered residential developments that are obscured by tree canopy as indicated by low grass metric values (Fig. 4).

Fig. 4
figure 4

Cluster Group 1 watersheds are comprised of large, well-connected forest patches, and relatively moderate levels of pavement surfaces. Buildings are large and fairly close together, and grass patches are sparse and fragmented by other cover types. These watersheds boast moderately dense developments, which are likely commercial due to building size, mixed with clustered residential developments that are obscured by tree canopy as indicated by low grass metric values

Full size image

Cluster 2: Suburban Lawns (11 Watersheds)

These watersheds are developed with dense residential parcels that have ample grass cover (primarily lawns). Buildings are small, numerous, and close together. Tree patches are large and contiguous, though there are fewer trees than in the semi-city living watersheds. Pavement is spread out across the landscape and not obscured by tree canopy due to the dense residential nature of development (Fig. 5).

Fig. 5
figure 5

Cluster group 2 watersheds are developed with dense residential parcels that have ample grass cover, or lawns. Buildings are small, numerous, and close together. Tree patches are large and contiguous, despite lower tree cover levels. Pavement is spread out across the landscape, and not obscured by tree canopy due to the dense residential nature of development in these watersheds

Full size image

Cluster 3: Shaded Urban Homesteads (10 Watersheds)

These watersheds are fairly fragmented by the complex interaction of land-cover types. Tree canopy obscures roads in some areas, while roads cut tree patches where they are not obscured by canopy. Buildings are spread out and partially obscured by canopy, which gives the impression that residential zones in these watersheds consist of larger parcels and less treeless development. This implies that the majority of development in these watersheds is dispersed residential, though there are likely some commercial areas with parking lots as indicated by moderate pavement levels (Fig. 6).

Fig. 6
figure 6

Cluster group 3 watersheds are fairly fragmented by the complex interaction of land-cover types. Tree canopy obscures roads at certain points, while roads cut tree patches where they are not obscured by canopy. Buildings are cordoned off and partially obscured by canopy, which gives the impression that residential zones in these watersheds consist of larger parcels, and less clear-cut development. The majority of development in these watersheds is dispersed residential, though there are likely small areas of commercially zoned development as indicated by moderate percent pavement levels

Full size image

Cluster 4: Well-Drained (2 Watersheds)

These two watersheds are somewhat similar in structure to the Shaded Urban Homesteads. Significantly higher storm water infrastructure metric values are what separate these watersheds from the other groups. While inlet and outlet densities for nearly all other watersheds are very low, they are much higher in these two watersheds. Both are more urbanized, with large parking lots and buildings mixed into clusters of residential neighborhoods. This density of urbanization requires significantly more storm water management than the other groups (Fig. 7).

Fig. 7
figure 7

Cluster group 4 watersheds are very similar in structure to watersheds in cluster group 3. The difference is that these two watersheds have very high stormwater infrastructure metric values, where all other watersheds have low values for inlet and outlet densities. Both of these watersheds are located in more highly developed urban areas, where the majority of development is dense and non-residential

Full size image

Observed Cluster Variation

The cluster groupings indicate that variation occurs in landscape structure among the analysis watersheds. However, metric variation also occurs within cluster groups. Interestingly, the density and shape of grass patches across all clusters are highly inconsistent. Within groups that have the highest grass patch and edge density variability, the shape and density of tree patches are also variable. Watersheds with smaller, more fragmented tree patches had larger, more complexly shaped grass patches (r = 0.91). This interaction suggests that grass affects forests in a similar manner as buildings and roads and could ultimately serve as a measure of urbanization alongside imperviousness. The principal components analysis further explains inter-cluster land-cover variation.

Principal Component Analysis

To identify the metrics potentially responsible for land-cover variation within watersheds, we applied a principal components analysis. Each principal component constitutes a unique landscape measurement. The first four principal components accounted for 82.20 % of total variance in the land-cover metric dataset (Table 3) and allowed us to identify those measurements that are responsible for a majority of this variation. These PCA results allowed us to distinguish which aspects of land-cover might be important factors in future urban water quality assessments.

Table 3 PCA eigenvalues for first four principal components show the increasing cumulative percentages of variance explained by each principal component (SAS)
Full size table

Using eigenvector scores (Table 4) to aggregate metrics into principal component measurements, we determined that the first principal component (PC1) measures the patch connectivity of all land-cover types, and is responsible for 39.55 % of the variance in land-cover among watersheds (Table 5). Levels of connectivity are highly influenced by the extent of development, as indicated by the inclusion of mean patch area metrics for pavement and building surfaces. In watersheds with high values for these metrics, development is more concentrated and denser in some areas, and vegetation patches are larger and more connected. Metrics describing area, number, and shape of building and grass patches show that the second principal component (PC2) measures the type of development in watersheds (single-family residential vs. multi-family or commercial). More buildings mean more grass, and the size and density of building patches influence the size and shape of grass cover. The number and size of tree patches constitute the third principal component (PC3). However, at least three metrics must be assigned to a component to interpret a measurement (SAS 2012). Stormwater infrastructure metrics were assigned to principal component four (PC4). Pipe densities are highly variable, although inlet and outlet variation is very low. Inlet and outlet variability was inflated due to extremely high numbers in two watersheds. Despite their lack of variability, stormwater metrics could still contribute to water quality variability.

Table 4 PCA eigenvector scores for first four principal components show high and low values for metrics in each principal component
Full size table
Table 5 Land-cover configuration PCA for 20–30 % impervious watersheds (n = 32)
Full size table

Discussion

Pavement, buildings, and grass are all features of urban systems that contribute to urban runoff pollution and work together to displace tree cover in forested watersheds. Previous studies have recognized impervious and forest patterns as important predictors of water quality variation (Carle et al. 2005; Alberti et al. 2007; Shandas and Alberti 2009), but despite the large number of studies dedicated to the biogeochemistry of lawns (Zhu et al. 2004; Pickett et al. 2008; Cheng et al. 2014) grass and buildings have been largely ignored as water quality predictors in urban landscapes. Our results demonstrate that grass and buildings are important sources of landscape variation in watersheds falling within a 20–30 % range of impervious cover, which suggests that their exclusion from future analyses exploring linkages between land-cover patterns and water quality could be a critical oversight. In this section, we further explore the implications of our findings for watershed management, and argue for a continued push toward more robust analyses that account for the heterogeneous land-cover interactions inherent within urban landscapes.

Lawns are a Measurement of Urbanization

In many studies that evaluate the effects of development on stream ecosystems, “urbanization” usually refers to some quantified measurement of impervious surface (Carle et al. 2005) or a combination of development and human population density (Jones and Christopher 1987). However, our cluster groups suggest that ‘urban’—at least in ecological terms—should ultimately refer to a more complex interaction among land-covers. Our analysis shows that although the typical indicator of urbanization (pavement) remains relatively static, comparisons among watersheds reveal stark and varied changes to the vegetative landscape that appear to be influenced by the size, shape, and number of buildings. In particular, our results suggest that linkages between building and grass patterns are especially strong.

This complex relationship becomes more apparent when comparing the percentages of each land-cover type to the percentage of total impervious area across watersheds (Fig. 8). There are three important trends to notice. (1) Even though total vegetation declines as the amount of imperviousness increases, the disaggregated measurements of forest and grass cover follow diverging patterns. (2) Forest cover declines at a rate similar to total vegetation cover; however, grass cover increases with pavement and building cover. (3) Grass and forest cover are both highly variable, while pavement and buildings are far less variable. These three results show that lawns in urban systems represent a significant proportion of the land-cover variation among these watersheds. For these reasons, we argue that grass is not only a factor of urbanization, but it may also be influencing forest fragmentation and loss more dramatically than pavement alone.

Fig. 8
figure 8

Percent land-cover types vs. percent total impervious surface: tree cover drives the relationship between total vegetation and impervious surface. High grass values and low tree values still equate to high vegetation values, even though water quality control potential is minimized as forest patches shrink. Grass contributes to variation in tree cover, similarly to buildings and roads, although the effects of grass seem to compound tree loss in urbanized watershed

Full size image

Implications for Watershed Management

Framing these findings within the context of urban watershed management means that the interactions among vegetation patterns have several significant implications for explaining water quality variation. First, forests absorb and process runoff and associated pollutants with far greater efficiency than grass (Brabec et al. 2002). While it might seem logical that any vegetation is beneficial to the health of a given watershed, approaches that aggregate trees and grass into a single vegetative class (Ridd 1995; Li et al. 2008) are assuming a false equivalency in their contributions to runoff control. Second, grass in urban and suburban areas tends to be treated with fertilizers and other chemicals, which echo agricultural management strategies that have been linked to decreased diversity of macro-invertebrate assemblages (Potter et al. 2004). Thus, it is no surprise that lawns intensify pollution concentrations in surface water runoff in suburban residential zones (Bannerman et al. 1993; Göbel et al. 2007). Given these factors, coupled with the dynamic relationship between grass and tree cover, it is safe to predict that urban watersheds with more complex grass patterns (e.g., “Suburban Lawn Mowers” cluster) will have poorer water quality than those with higher tree cover (e.g., “Shaded Urban Homesteads” Clusters).

While vegetation patterns within urban watersheds are no doubt critical to water quality outcomes, it is impossible to deny the role of impervious cover. However, it is also essential to consider the relative influence of impervious cover type on water quality in the same way we would consider the relative influence of vegetation type. For example, grass and herbaceous cover are the known sources of nutrient pollution in urban systems (Bannerman et al. 1993; Göbel et al. 2007), while forests are generally sinks (Nowak and Greenfield 2012). In similar fashion, pavement carries the highest runoff pollutant loads, but buildings are additional and variable sources of urban runoff pollution (Bannerman et al. 1993; Davis et al. 2001; Göbel et al. 2007). For example, differences in building and roofing material translate to changes in pollutant types and loadings, including heavy metals and other toxic chemicals (Pitt et al. 1995; Clark et al. 2008), while pollutants and loadings from pavement are closely tied to surrounding land uses and traffic volume (Pitt et al. 1995; Davis et al. 2001).The heterogeneous patterns among all land-cover types identified by our results, coupled with their unique polluting or mitigating potential, highlight the need for scientific and management approaches that appreciate the complexity of these relationships. This becomes increasingly important when accounting for the role of stormwater infrastructure.

A recent study has shown that the cumulative effects of non-point source pollution within a catchment cannot be negated by protective buffer zones (Pratt and Chang 2012). This could be a direct impact of stormwater pipe connectivity, which is often a better predictor of some water quality parameters than the percentages of total and directly connected impervious area (Hatt et al. 2004). The piping of stormwater extends the riparian boundary throughout an urban catchment, which facilitates a full exchange of protective buffers for the land-cover patterns surrounding pipe inlets. Although many studies and management approaches quantify directly connected impervious areas (Walsh et al. 2004; Hatt et al. 2004; EPA 2005), they are not inclusive of the land-covers adjacent to them that could be additional sources or sinks of pollutants. Although our analysis showed a surprising lack of variation in stormwater infrastructure metrics among the study watersheds, the land-cover types and patterns feeding these systems do indeed vary. Furthermore, there are likely additional modifications, like the size, material, design, age, and maintenance records of stormwater pipes themselves that could influence water quality. For these reasons, we contend that approaches in which all land-cover types and stormwater characteristics are considered will have a greater potential to provide insights into observed variations in water quality than those relying on aggregated measures of impervious or vegetative cover.

Social-Ecological Drivers of Land-Cover Change

Ultimately, variation in urban land-cover is linked to human decision making and we would be remiss to not acknowledge a few of the socio-ecological drivers that make urban systems so dynamic. For example, recent research in Raleigh, NC, showed that urban morphology is a driver of tree cover patterns, but management of residential parcels is based on a variety of neighborhood level social characteristics such as lifestyles, preferences, and historical legacies (Bigsby et al. 2013). Factors such as lifestyle play an important role in shaping the vegetative patterns of neighborhoods (Grove et al. 2006; Boone et al. 2009).

Demographics, including socio-economic status, are also deterministic elements of land-cover (Grove et al. 2006). Low socio-economic status neighborhoods typically feature an over-abundance of pavement and very few trees (Pickett et al. 2008). In these areas, grass cover provides dual benefits as a pervious surface to slow runoff, and a source neighborhood pride (Pickett et al. 2008). This suggests that the potential disservices of grass cover within watersheds only extends so far along the impervious gradient—some pervious cover is better than no pervious cover. Regardless of the reasoning, understanding how and why people interact with and alter their landscapes is critical for developing effective management solutions.

Limitations

There are limitations to this approach. For example, the catchment level land-cover patterns observed in the piedmont of North Carolina may or may not be consistent with patterns observed elsewhere. Additionally, land-cover likely affects water quality differently across climate and ecosystem zones. Infrastructure, development, and vegetation patterns in Phoenix, AZ, where rainfall events are sparse and large, will influence water quality differently than in Seattle, WA, where it rains continuously throughout the year. Also, there are factors beyond land-cover that could explain some variation in water quality, including household age, legacy of uses, or morphological characteristics. These limitations are easily overcome through replication of our methods to determine pattern variability, which could lead to a common understanding of how best to measure land-cover patterns and ecosystem processes across cities. This logic is similar to the reasoning behind the recent LCZ classification system (Bechtel et al. 2015). While land-cover patterns alone cannot fully predict variation in water quality, they can serve as indicators for potentially important development metrics that will boost the predictive power of regression models.

Conclusions

Urban land-cover patterns are dynamic and vary significantly among developed watersheds. This variation has implications for watershed management because the polluting or mitigating characteristics of alternative land-cover patterns influence water quality. Indeed, a meta-analysis by Schueler et al. (2009) contends that impervious cover will fall short as a predictor in watersheds with lower impervious rates (<40 %) due to the wide range of variability in observed water quality. While there are certainly additional contributing factors to urban water quality variation, land-cover patterns are easily quantified using high-resolution remote sensing data, and basic land-cover types are also unique contributing sources of pollution. The recent advancements in urban high-resolution land-cover classifications that disaggregate land-cover types into their most basic components, and the movement away from reliance on impervious cover as the principal predictor of urban water quality seem to portend a paradigm shift in the way we approach the management of urban watersheds. In anticipation of continued inquiries into the best management practices for urban watersheds, we offer three recommendations for future studies:

  1. (1)

    Not all vegetation is created equally—especially in urban systems. It is important to consider that tree cover has far greater pollutant mitigating and runoff controlling potential than lawns (grass cover). At this point, it should be a standard practice to assess vegetation types separately, at least in the context of urban catchment management.

  2. (2)

    Pavement, buildings, and grass are all features of urban systems that contribute to urban runoff pollution and work together to displace tree cover in forested watersheds. While impervious cover is an obvious predictor of water quality, disaggregating impervious cover into its basic components (buildings or pavement/concrete) could improve the predictive potential of regression models. Additionally, trends point to linkages between building and grass patterns, which are often overlooked components of the urban landscape that should warrant more consideration moving forward.