Regular issues
Special Issues

Urban Naturalist
    URNA Home
    Aim and Scope
    Board of Editors
    Editorial Workflow
    Publication Charges

Other Eagle Hill Journals
    Northeastern Naturalist
    Southeastern Naturalist
    Caribbean Naturalist
    Neotropical Naturalist
    Prairie Naturalist
    Eastern Paleontologist
    Journal of the North Atlantic

Eagle Hill Institute Home

Systematic Variation in Roof Spontaneous Vegetation: Residential “Low Rise” Versus Commercial “High Rise” Buildings
Michael L. McKinney and Nicholas D. Sisco

Urban Naturalist, Special Issue No. 1 (2018): 73–88

Full-text pdf.


Site by Bennett Web & Design Co.
Urban Naturalist 73 M.L. McKinney and N.D. Sisco 2018 Special Issue No. 1 URBAN NATURALIST 2018 Special Issue No. 1:73–88 Systematic Variation in Roof Spontaneous Vegetation: Residential “Low Rise” Versus Commercial “High Rise” Buildings Michael L. McKinney1,* and Nicholas D. Sisco2 Abstract - The ability of roof habitats to support vegetation depends on many local and landscape characteristics that have rarely been studied. We statistically analyzed GIS data and found that ~3.5% of Knox County, TN, is covered by roof surface. We also showed that tall “high-rise” buildings (≥6.1 m [20 ft] high) tend to have more roof area and less surrounding green space than shorter “low-rise” buildings (less than 6.1 m [20 ft] high ). These patterns of building geometry almost certainly strongly affect the composition of spontaneous plant communities that colonize, survive, and reproduce on the roof habitats. A vegetation survey of roofs in Knox County, TN, documented 26 plant species, most of which were common urban plants, likely sourced from nearby vegetation and representing many life habits (trees, shrubs, grasses) and dispersal modes. Most species (61.5%) were not native to the US and 42% were classified as “weeds”. Roofs on low-rise buildings have different spontaneous vegetation communities than high-rise buildings, and vegetation tends to be more abundant on low-rise buildings. Introduction Green roofs have great potential for increasing urban biodiversity and many ecosystem services (Oberndorfer et al. 2007). For example, documented ecosystem services provided by green roofs include stormwater capture (Mentens et al. 2006, Stovin 2010) and reduction of urban temperatures (Bowler et al. 2010, Santamouris 2014). Green roofs may help reduce pollution and climate change (Li and Babcock 2014) and have many other potential benefits and uses relating to urban infrastructure in various ecoregions (Dvorak and Volder 2010). However, more research is needed to evaluate the ability of green roofs to meet other goals such as habitat provisioning (Williams et al. 2014). There are several types of buildings and rooftops in urban areas; thus, one important research topic is to assess which specific roof characteristics affect biodiversity, ecosystem services, and other green-roof functions. As described by MacIvor and Ksiazek (2015), these characteristics include local abiotic factors (e.g., roof area, percent vegetated, age, maintenance, irrigation, substrate composition and depth) and landscape factors (e.g., roof height, distance to source propagules, and percent green space). Local abiotic and landscape factors then influence which plants can colonize and persist 1Earth and Planetary Sciences, University of Tennessee, Knoxville, TN 37916, USA. 2School of Planning, College of Design, Architecture, Art, and Planning, University of Cincinnati, Cincinnati, OH 45221, USA. *Corresponding author - Manuscript Editor: Sonja Knapp Green Roofs and Urban Biodiversity Urban Naturalist M.L. McKinney and N.D. Sisco 2018 Special Issue No. 1 74 on the roof habitat, and these plants strongly influence the animal community that subsequently develops (Faeth et al. 2011). Put another way, green roofs act as filters whereby the habitat properties of each roof, along with the properties of the surrounding landscape, determine which groups of organisms can colonize (or be cultivated), persist, and reproduce (Madre et al. 2014). Spontaneous vegetation—plants that are dispersed to and grow on green roofs—are excellent indicators of the types of plants that can potentially colonize and sometimes, persist and reproduce on specific habitats (Archibold and Wagner 2007, Jim and Chen 2011, Madre et al. 2014). These roof colonizers are quite diverse (Dunnett 2015) and often provide many ecosystem services and ecological functions (Del Tredici 2010, Robinson and Lundholm 2012). Any complete estimate of green-roof habitat in an urban area must therefore include as many of these local and landscape variables as possible. It is well known that urban infrastructure is the template that shapes urban ecological processes at many scales (Forman 2014), and we know that many factors influence that infrastructure. For example, building age and construction type can affect both the appropriateness of retrofitting buildings with green roofs and the type of green roof that can be safely installed (Castleton et al. 2010). Here, we examine how spatial building geometry affects the biodiversity and ecology of unmanaged green roofs. Our basic hypothesis is that spatial associations relating to building geometry are non-randomly distributed in urban space. For example, taller buildings also tend to have greater roof area because of scaling effects (e.g., taller buildings need larger foundations, and therefore, greater roof size) and functional reasons (e.g., taller buildings are often intended to have more volume so that roof area is increased as well as height). Furthermore, we suggest that spatial associations among these traits arise because, for example, large commercial buildings tend to be clustered into localized commercial parts of the urban area, whereas smaller residential buildings are clustered into localized residential areas. In addition, we hypothesize that these structural and spatial relationships will translate into nonrandom patterns on rooftop biological communities, in this case, spontaneous vegetation. For example, it has been shown that both roof height (MacIvor 2016) and green-roof substrate depth (Gabrych et al. 2016) influence insect and plant communities, respectively. As potential habitats, larger buildings tend to have more roof area with more structural support for deeper, moister, growing substrates with more moderate temperatures than smaller buildings, which have less roof area and thinner growing-substrates. In terms of dispersal potential, larger buildings will have different landscape characteristics than smaller ones, and higher roofs may be surrounded by less green space, which makes plant and animal dispersal to roofs on larger buildings more difficult and thus less likely. Methods We measured several aspects of building geometry (such as building height, roof area, and surrounding green space) and related these to spontaneously growing plant communities that were found on the roofs of surveyed buildings. Our study Urban Naturalist 75 M.L. McKinney and N.D. Sisco 2018 Special Issue No. 1 focused on Knox County, TN, USA, which contains the rapidly growing city of Knoxville. Settled in 1792, Knoxville has a population of 184,281 and is the 3rdlargest city in the state (US Census Bureau 2015). Due to its moderate climate and proximity to a major national park, the rapid growth of Knoxville is typical of many cities in the Southeastern US that are characterized by extensive urban sprawl (Cho et al. 2010). Knox County has no cultivated or maintained green roofs, so our survey focused on spontaneous vegetation that was found growing on roofs (usually abandoned structures) that were not designed specifically as vegetative habitat. This site characteristic is a critical point, expanded on below, because our findings focus on which plants can disperse to and germinate on these roof habitats; they cannot address how many of these plants can survive and reproduce over a long period, compared to constructed green roofs that are specifically designed and maintained as plant habitats (e.g., as described by Oberndorfer et al. 200 7). Geospatial data We acquired the geospatial data used to conduct this research from the Knoxville Geographic Information Systems (KGIS) group from the year 2010. The shapefiles included: building footprints, roads (edge of pavement), structures, water, and county/city boundaries. In 2010, there were ~230,000 buildings in Knox County. We conducted all geospatial analyses in ESRI ArcGIS software. We projected the data in Tennessee State Plane (feet) and then converted to metric values. The building- footprint shape-file displays every building in Knox County as rooftop shapes traced from aerial imagery. In order to process such a large dataset, we used the grid-index features tool to divide Knox County into 1-km2 cells, which produced a total of 1232 cells (Fig. 1). Building footprints were added to the map and were spliced by the grid lines to accurately measure roof area and percent roof-cover per cell. After using the intersect tool to splice the polygons, we recalculated the geometry to obtain area (m2) per roof. We used the dissolve tool to create a single multipart shapefile per cell, which aggregated the total roof-area per cell. We divided roof area by the area of the cell (1 km2) to calculate the percent roof-cover per cell and for all of Knox County. The individual area of each roof was also calculated and exported i nto an excel file. In addition to roof area, we used the KGIS database to analyze roof height. Height for each building is an attribute calculated by KGIS prior to our analysis. This database also provided information on nearest-neighbor distance, which was calculated as the shortest distance between each building and the closest adjacent building. Finally, we calculated green space per cell by creating a shape file of impervious and non-vegetated objects by merging the transportation shape-file (which included roads, bridges, and driveways) with the building footprints shape-file, the miscellaneous structures shape-file (for example, storage tanks and other non-vegetated objects), and the hydrological shape file (creeks, rivers, lakes). We dissolved and clipped per sample cell the non-vegetated shape file in order to complete the geoprocessing in an efficient and timely manner. We then figured green-space area using basic geometry calculations. Urban Naturalist M.L. McKinney and N.D. Sisco 2018 Special Issue No. 1 76 We employed distributional statistics (central tendencies, skewness, histograms) and bivariate and multivariate methods to analyze the data on roof area, roof height, nearest neighbor, and green space for each cell to identify any relationships among them. All analyses were run in SAS software, version 9.0 (SAS 2016). For multivariate analysis, we utilized principle component and factor analysis to examine any covariations between building height, area, nearest neighbor, and surrounding green space. Our goal was to quantify any non-random patterns in building geometry. Where needed, we normalized the per-cell data using log transformation. We log10-transformed all variables except percent green space for the correlation analysis of the 1232 cells to meet the conditions of normality. We compared patterns of roof area, height, nearest neighbor, and adjacent green space in high-rise buildings versus low-rise buildings for the multivariate cell-analysis and the vegetation survey. We chose the 6.1-m (20-ft) cut-off to differentiate low- and high-rise buildings because residential homes are typically 2 stories or less, which usually translates into a height of ≤6.1 m (20 ft) (http://www. In contrast, buildings taller than 6.1 m (20 ft) tend to be larger commercial buildings (e.g., apartments, offices, retail) not found in typical mass suburban-housing subdivisions. We derived a subset from the KGIS data (not based on per-cell data) of roof area, height, and green space to perform a finer-scale analysis. We employed the subset features tool (ArcGIS) to select 100 randomly sampled buildings and create a 3.05-m (10-ft) buffer around each building. The impervious surface shapefile was then Figure 1. Example of a 1-km2 cell showing University of Tennessee campus (including football stadium.) Dark areas are building footprints (roof areas). Gray shading indicates impervious and other non-vegetated areas. White areas represent vegetation (green space). This cell (Cell 902) has 19.0% roof cover. Urban Naturalist 77 M.L. McKinney and N.D. Sisco 2018 Special Issue No. 1 intersected and clipped to the 3.05-m (10-ft) buffer. We used the buffer to measure the percentage of green space around each of the chosen buildings (Fig. 2). These data were analyzed using the bivariate and multivariate methods described above to examine finer-scale relationships among roof area, height, and green space. Spontaneous vegetation survey To directly examine the effects of roof area, height, and surrounding green space on roof vegetation, we surveyed spontaneous vegetation growing on roof habitats. Based on our analysis of the KGIS data, we selected 12 high-rise and 12 low-rise cells for subsequent survey. We surveyed all 24 cells between 3 April 2015 and 12 July 2015 to locate as many plants as possible growing on building roofs in those cells. Our surveys included habitats on any roof structures that occurred at or above roof level, such as rain gutters and chimneys. There are no designed or cultivated green roofs in Knox County; thus, these plants were generally growing in locations where soil had somehow accumulated naturally, a process which occurs through time (Getter et al. 2007). We used a laser distance-meter with a 200-m (656-ft) range (Leica Disto E7500i meter, Leica Geosystems, Norcross, GA) to measure the building height of buildings with vegetation. We followed 2 basic survey methods. For buildings with roof tops visible from the ground, such as most residential houses, we could often confirm plant presence with walking or driving surveys. However, for high-rise buildings, we had to rely Figure 2. 3.05-m(10-ft) buffer area denoted by gray line. We subtracted the area of impervious surface (striped) from the total buffer area to derive green-space area within the buffer zone and converted this value to a percentage. In this case: buffer area (382.80 m2) - impervious area (248.71m2) = green-space area (134.09 m2). Green-space percentage = green space area/buffer area = 35.03%. Urban Naturalist M.L. McKinney and N.D. Sisco 2018 Special Issue No. 1 78 on observations from elevated areas, which often included gaining access to rooftops of commercial buildings to examine them and to view surrounding roof tops. When we observed plants growing on rooftops, we typically gained access to the roof to identify or collect them directly. In some cases, however, especially for very tall buildings or derelict buildings with unsafe access, we used telephoto lenses to obtain photographs that we used for identification. In an effort to standardize for sampling effort, we devoted roughly 18 person-hours to each of the 24 cells. Our roof survey encountered 2 basic types of roof habitats that supported spontaneous plants. All of the houses surveyed had angled roofs constructed of asphalt roof shingles. In these cases, plants were growing in crevices between the shingles, most often around or on a chimney, and (commonly) in gutters. We sometimes observed vines growing on the shingle exterior. The second type of roof habitat commonly found was a flat-top roof typical on commercial buildings. Surfaces of these roofs were paved with concrete, asphalt, or pebble-covered tar paper. We documented plants on these flat roofs in many areas, usually where some sediment accumulation had occurred in the corners or depressions on the roof. Both of these roof habitats have characteristics that are very different from those of designed green-roof habitats, which has important implications as discussed below. We consulted the website of the Knox County tax assessor (http://tn-knoxassessor. to determine the age of 64 surveyed buildings (year constructed) with spontaneous vegetation. We assessed the effects of building age on spontaneous vegetation. To specifically detect if high-rise roofs are proportionately younger or older than low-rise roofs, we performed a chi-square test using our high- and low-rise definitions; we further defined a building as old if building age was ≥40 y and as young if building age was less than 40 y. We compared plant species occurrence on high-rise buildings versus low-rise buildings. To reduce the statistical “noise”, we excluded from analysis species with fewer than 2 total occurrences. We used the weediness, life-habit, and nativity designations of all species found as listed by the USDA ( We obtained seed-dispersal information from the Dispersal and Diaspore Database (Hintze et al. 2013). Weeds are defined here as those plants causing significant economic or ecological impacts. We define native species as those occurring in North America before European colonization. Using bivariate and multivariate factor analysis, we examined the effects of building height, building area, and green space on the composition of plant associations found on the buildings we surveyed. Results Roof cover in each 1-km2 cell studied in Knox County, TN, ranged from much less than 1% to over 22% (Fig. 3). The sum of the roof cover for all cells is 47.72 km2, indicating that ~3.5% of Knox County was covered by roof area (47.72 km2 roof cover/1362 km2 total area). The densest roof cover is in historic downtown Knoxville in the county center and trends SW along the interstate-highway corridor (Fig. 3). Cells in the urban core had the highest roof cover, with a maximum Urban Naturalist 79 M.L. McKinney and N.D. Sisco 2018 Special Issue No. 1 of 22.7% and an average for the urban core of roughly 16% roof cover. Cells in the suburban areas typically averaged around 11% roof cover but varied from 4% to 17%, depending on the specific housing subdivisions analyzed. Cells sampled in rural parts of the county, far from the urban core, varied from 0% to 7% roof cover, with an average of ~2%. The frequency distribution of roof area in highly urban (downtown) cells shows a strongly right-skewed distribution (median = 177.93 [SD = 98.44], skewness = 4.62 [SE = 0.120], kurtosis = 33.04 [SE = 0.240]). This result indicates that, for highly urban areas, available roof area consists of many small roofs and a few very large roof tops. In suburban cells, the frequency distribution of roof area is less right-skewed and reflects decreased presence of large buildings (median = 253.87 [SD = 88.63], skewness = 2.50 [SE = 0.364], kurtosis = 14.98 [SE = 0.364]). Correlation analysis showed that percent green space was significantly negatively correlated with both median roof height and median roof area for the cells Figure 3. Knox County, TN: Percent roof cover in each 1-km2 cell ranges from much less than 1% to over 22%. Cumulative roof cover for the County is about 3.5% of total county area. Table 1. Correlation matrix of median variables for 1232 cells. All variables except percent green space are log10 transformed. Roof height = distance from the ground to the roof, roof area = area of roof top, percent greenspace = percentage area per cell covered with vegetation, and nearest neighbor = distance to nearest building. * indicates values that are different from 0 at a significance level alpha = 0.05 Variables Roof height Roof area Percent Greenspace Nearest neighbor Roof height 1* Roof area 0.756* 1* Percent greenspace -0.411* -0.472* 1* Nearest neighbor -0.012 0.048 0.131 1* Urban Naturalist M.L. McKinney and N.D. Sisco 2018 Special Issue No. 1 80 (Table 1). Also, median roof area and median roof height were strongly positively correlated (Table 1). Nearest-neighbor analysis showed no significant relationships (Table 1) among any of the variables studied here. Analysis of the 100 randomly selected buildings also indicated a strong positive relationship between roof area and roof height (Fig. 4A). There is some evidence that the relationship is curvilinear, as demonstrated by the better fit of the power function. Also, both roof area and roof height showed a strong negative and curvilinear relationship to percent green buffer (Fig. 4). Figure 4. (A) Positive relationship between roof height and roof area for 100 randomly selected buildings (y = 0.936x1.66, R² = 0.636, P < 0.01). Note negative relationship between percent green-buffer percentage (x 10 for visual display on graph) and roof height for 100 randomly selected buildings (y = 0.495x2 - 30.338x + 895.42, R² = 0.279, P < 0.01). (B) Negative relationship between percent green-buffer and roof area for 100 randomly selected buildings (y = 155.13x-0.234, R² = 0.422, P < 0.01). Urban Naturalist 81 M.L. McKinney and N.D. Sisco 2018 Special Issue No. 1 For the 1232 cells, a principal component analysis (PCA) summarized these patterns with roof area and roof height loading very highly (0.614, 0.597, respectively) on Axis 1, with green space (- 0.515) showing an opposite pattern (Fig. 5). Overall, Axis 1 explains over 50% (51.8%) of the variation in the data. Roof nearest neighbor shows no consistent relationship to those variables and loads very highly on Axis 2 (0.973), explaining 26.0% of the total variation. A plot of factor scores of all 1232 cells onto Axis 1 and Axis 2 shows that high-rise cells form a separate population from low-rise cells (Fig. 5). The latter are characterized by lower roofarea and height and greater green space (because they score lower on Axis 1, which is negatively correlated with green space). The spontaneous vegetation survey found 26 plant species with an occurrence >1. These 26 species represent a wide variety of life habits; 38.5% were native, 61.5% were non-native, 42.3% were weeds, and 57.7% were non-weeds. Of these, 34.6% were animal dispersed and 65.4% were mechanically dispersed (Table 2). A plot of the factor loadings for the plants co-occurring in the 24 cells indicates which plants tend to occur together: species to the right are increasingly associated with Figure 5. Plot of factor scores of all 1232 cells onto Axis 1 and Axis 2. For this PCA, roofarea and roof-height loading vary highly (0.614, 0.597, respectively) on Axis 1, with green space (- 0.515) showing an opposite pattern. Centroid envelope encompasses 95% of score values for respective category. The major axis (1) contains over half of the variation and shows the high covariation of median roof area with median roof height and the negative relationship of those 2 variables with greenspace. Urban Naturalist M.L. McKinney and N.D. Sisco 2018 Special Issue No. 1 82 Table 2. Plants identified in this study, with various attributes shown. Weed, growth habit, and nativity status from USDA classifications ( Seed-dispersal information was obtained from the Dispersal and Diaspore Database (Hintze et al. 2013). Native = native to US. High- or low-rise trend derived from Figure 6 as explained in text. * = listed as common spontaneously growing urban plants of the northeastern US (Del Tredici 2010). Species Common name Abundance Life habit Nativity Weed Dispersal High-rise trend Ailanthus altissima (Mill.) Swingle* Tree of Heaven 22 Tree Non Y Mechanical Ambrosia artemisiifolia* Besser Ragweed 17 Herb Native Y Mechanical Buddleja davidii Franch. Butterfly Bush 6 Shrub Non Y Mechanical Ligustrum sinense Lour.* Chinese Privet 19 Shrub Non Y Animal Paulownia tomentosa (Thunb.) Steud.* Princess Tree 19 Tree Non Y Mechanical Pellaea atropurpurea (L.) Link Cliff Fern 4 Fern Native N Mechanical Phytolacca americana* L. Pokeweed 14 Herb Native N Animal Solidago canadensis* L. Canada Goldenrod 14 Herb Native N Mechanical Low-rise trend Acer saccharinum* L. Silver Maple 4 Tree Native N Mechanical Cercis canadenisis L. Redbud 7 Tree Native N Mechanical Daucus carota* L. Wild Carrot 10 Herb Non Y Mechanical Parthenocissus quinquefolia* (L.) Planch. Virginia Creeper 7 Vine Native N Animal Plantago major* L. Common Plantain 19 Grass Non N Animal Pyrus calleryana Decne. Bradford Pear 8 Tree Non N Animal Rubus allegheniensis* Porter Blackberry 11 Shrub Native N Animal Setaria faberi* R.A.W. Herrm. Giant Foxtail 11 Grass Non Y Mechanical Taraxacum officinale* F.H. Wigg. Dandelion 22 Herb Non N Mechanical No clear trend Asplenium platyneuron D.C. Eaton Ebony Spleenwort 4 Fern Native N Mechanical Campsis radicans (L.) Bureau Trumpet Vine 18 Vine Native N Mechanical Cirsium vulgare* (Savi) Ten. Bull Thistle 4 Herb Non Y Mechanical Hedera helix Lowe English Ivy 18 Vine Non Y Animal Lonicera japonica* Thunb. Japanese Honeysuckle 13 Vine Non Y Animal Morus alba* Sudw. White Mulberry 13 Tree Non N Animal Paspalum dilatatum Poir. Dallisgrass 13 Grass Non N Mechanical Urtica dioica* L. Stinging Nettle 2 Herb Non N Mechanical Verbascum thapsus* L. Great Mullein 3 Herb Non Y Mechanical Urban Naturalist 83 M.L. McKinney and N.D. Sisco 2018 Special Issue No. 1 high-rise buildings and species to the left are increasingly associated with low-rise buildings (Fig. 6). A plot of the factor scores of those plants indicates that plant species assemblages on high-rise buildings seem to differ (have little overlap) relative to the assemblages on low-rise buildings (Fig. 6). Table 2 shows the 26 species on the basis of whether they occur mostly on lowrise (loadings <- 0.3 on Fig. 6) or high rise buildings (loadings >0.3 on Fig. 6). Ten species (38.5%) are native to the US, 11 species (42.3%) are classified as weeds, and 17 species (65.4%) are considered common urban plants in North America (Del Tredici 2010). Our analyses identified no association between height and life history, nativity, weed status, or main dispersal mode. The sample size was relatively small for each category (9–11, out of 26 total); thus, we performed a Fisher exact probability test to relate each of the 4 attributes to high- or low-rise status. None of the attributes were non-randomly distributed at the 0.05 level of confidence. Analyses of the vegetation-survey data identified a strong effect of building height on plant abundance and plant species richness (Fig. 7). Plant abundance is the total number of occurrences (observations) for all plants on all roofs at each height interval. For example, interval 10 = total occurrences at 0–3.05 m (0–10 ft), interval 20 = total occurrences at 3.06–6.1 m (>10 to 20 ft.) and so on. Similarly, Richness equals the total number of species found for all plants on all roofs at each interval. There is a roughly exponential decline in both abundance and richness with increasing building height (Fig. 7). Figure 6. Plot of factor loadings for plant species on F1 and F2, based on their co-occurrences on the 24 cells that were surveyed. Species to the right are increasingly associated with high-rise buildings. Species to the left are increasingly associated with low-rise buildings. Urban Naturalist M.L. McKinney and N.D. Sisco 2018 Special Issue No. 1 84 Regarding the effects of building age on spontaneous vegetation, regression of building age (y) onto building height indicated no relationship (n = 64, R2 = 0.004, F = 0.001, P = 0.976). Multiple regression of plant species richness onto age (y) and building height (m) indicated that both had an effect on species richness: Richness = - 0.0146Height + 0.061Age + 2.75 (n = 64, adj R2 = 0.683, F = 68.81, P < 0.0001). Richness increased with age but declined with height. Multiple regression of plant species abundance onto age (y) and building height (m) indicated that both had an effect on abundance: Abundance = - 0.0232Height + 0.073Age + 4.95 (n = 64, adj R2 = 0.702, F = 75.07, P < 0.0001). Abundance increased with age but declined with height. The chi-square distribution for the 64 buildings of known age was as follows: old/low rise = 11, old/high-rise = 12, young/low-rise = 18, young/high-rise = 23. The results are: chi-square = 0.092 (critical value = 3.841), P = 0.762 with alpha = 0.05. The computed P-value is greater than the significance level alpha = 0.05; thus, one cannot reject the null hypothesis H0 that the rows and columns are independent, i.e., there is no apparent pattern in age distribution between high-rise and low-rise buildings. Discussion The KGIS data indicated that, for all of the total area of Knox County, there is a great deal of potential green-roof habitat, with a total roof area that exceeds 3.5% of the Knox County (Fig. 3). This finding is significant because Knoxville is Figure 7 Plant species abundance and species richness decline w ith building height. Urban Naturalist 85 M.L. McKinney and N.D. Sisco 2018 Special Issue No. 1 typical of many cities in the US, especially in the southeastern US, in experiencing substantial urban sprawl (Cho et al. 2010). This form of growth is notorious for its negative environmental impacts via extensive devegetation and habitat loss (Martinuzzi et al. 2015). That such a large area (3.5%) of the Knox County is covered by roof area implies there is a very significant amount of potentially available habitat that could mitigate the effects of sprawl on biological systems. In addition, we showed that this potential green-roof habitat differs systematically on the basis of height, area, and surrounding green space. Specifically, tall high-rise buildings tend to have more roof area and less green space compared to shorter low-rise buildings (Figs. 4 and 5). Also, the distribution of green-roof habitat is variable; highly urbanized downtown areas have many smaller buildings interspersed with relatively few very large roofs. Roof area in more suburban areas tends to be somewhat more evenly distributed (less right skewed ). These building and roof patterns have significant implications for the plant communities that can colonize, survive, and reproduce on these habitats. Our tentative findings from the plant surveys on roof tops indicate that different communities are found on the taller high-rise buildings than on the low-rise roof tops (Fig. 6). We did not, however, find any clear trend in life history, nativity, dispersal mode, or weediness among these 2 communities (Table 2). Our findings also indicate that taller buildings have lower plant abundance and lower plant species richness (Fig. 7). Increasing roof height will, therefore, shape green-roof plant communities by reducing the number of species that can disperse to high-rise sites. Madre et al. (2014), for example, found that roof height was a significant factor in shaping the taxonomic and functional compositions of the colonizing plant communities of 115 green roofs in northern France. Our finding that increasing roof height reduced spontaneous plant diversity and abundance was also shown for bees and wasps (MacIvor 2016) and for arboreal flora in Hong Kong (Jim and Chen 2011). As previously documented by Liu et al. (2016) for other urban areas, we found that percent green space is negatively related to building height. For bees and wasps, reduced surrounding ground-level green space is related to decreasing species richness and abundance on vegetated roofs (MacIvor and Ksiazek 2015, Tonietto et al. 2011). Our results also indicate that older buildings tend to have more roof vegetation, a pattern also noted by Jim and Chen (2011) in Hong Kong. Importantly, our study does not directly address the ability of the spontaneous vegetation found here to survive and reproduce on roofs over the long term. Many of the plants observed (especially trees and shrubs) were in the early stages of growth and we do not know if they could survive for many years and reproduce on roof habitats where soil was sometimes much thinner than on the ground. Indeed, Miller et al. (2014) found that most rooftop tree seedlings did not survive for more than 1 year. Furthermore, the survival and reproductive potential of these species will likely be different for plants on designed green-roof habitats. As described by Oberndorfer et al. (2007), there are both extensive and intensive designs for such green roofs. The extensive roof, with an average substrate of 2–20 cm and requiring little irrigation or maintenance (Oberndorfer et al. 2007), is most similar to the Urban Naturalist M.L. McKinney and N.D. Sisco 2018 Special Issue No. 1 86 kinds of neglected roof habitats analyzed here. However, there are still many differences between designed extensive roofs and the residential and commercial roof tops studied here. Most notable perhaps is that (where they are not regularly “weeded”) extensive roofs would likely generally provide better habitat for spontaneous vegetation because of the relatively less hostile conditions (e.g., better growing media) of such roofs. Most of the plants in this study were growing in small areas with sparse soil accumulation and no intentional design for such habitat conditions as shade, soil thickness, or soil quality (including nutrients and moisture). Several studies have shown that substrate depth plays a major role in shaping roof-plant community composition (Brown and Lundholm 2015, Gabrych et al. 2016, Ksiazek et al. 2017), including colonizing plant communities (Madre at al. 2014). Although our findings cannot provide information on which plants can survive and reproduce on designed green roofs, we suggest that they provide information on a “first filter” for green-roof communities, that of dispersal and germination. That is, they shed light on which local plant species may be most capable of dispersing to these rooftop habitats. For plants that are able to colonize high-rise habitats via natural dispersal or cultivation, our results show they will likely find more roof area for growth and reproduction than in residential and other low-rise roof habitats. Roof area has been shown to affect roof-community composition in spontaneous vegetation (Madre et al. 2014). In addition to more area, roofs on larger buildings have greater structural support with the potential to hold much deeper soils. We cannot yet verify specific attributes (life habit, nativity, weed status, and dispersal mode) that can explain why certain plant species tend to colonize and grow on the roofs studied here. However, there are some general patterns about these roof colonizers that we hope will stimulate further research. One pattern is that roughly two-thirds (65.4%) of the spontaneous vegetation was composed of plants that are common in many urban habitats. These plants are familiar species in many hardscape and disturbed urban areas, such as roadsides, sidewalk crevices, and vacant lots, and they are well adapted to the urban landscape with their high tolerance for dry, hot, and sunny habitats (Archibold and Wagner 2007, Del Tredici 2010). As argued by Madre et al. (2014), such urban species have value in their ease of maintenance combined with their performance of ecosystem services and ecological functions (see also Robinson and Lundholm 2012). Interestingly, although 86% of their urban roof species were native, we found a much lower percentage (38.5%) of spontaneous roof species that were native. Another pattern in our study was the presence of several species that are commonly cultivated in landscaping (e.g., Buddleja davidii [Butterfly Bush], Campsis radicans [Common Trumpet Creeper], Pyrus calleryana [Bradford Pear]) pointing to this activity as a potentially important avenue for green-roof dispersal. In a study on a Canadian University campus, Archibold and Wagner (2007) also noted the importance of ornamental species dispersing to rooftops. In conclusion, our results support the predictions that building configurations affecting roof habitats (e.g., roof area, height, and surrounding green space) occur Urban Naturalist 87 M.L. McKinney and N.D. Sisco 2018 Special Issue No. 1 in non-random spatial associations that produce different communities of spontaneous vegetation. Our findings also confirm a larger view that recommends that vegetation on roofs should not be considered in isolation, but should be considered as part of a much larger “green and gray infrastructure” with a large-scale dynamic that has many implications for designing sustainable cities (Forman 2014). Clearly, however, many roof-habitat variables, in addition to those studied here, affect patterns of dispersal and persistence in roof vegetation. Specific recommendations for further work include determining survival and reproduction in rooftop spontaneous vegetation, and making detailed estimates of the ecosystem services provided by these species. Acknowledgments Assistance with plant identification was provided by the University of Tennessee (UT) Botany Department, especially Aaron Floden, the UT Forestry Department, and the UT Arboretum. Literature Cited Archibold, O.W., and L. Wagner. 2007. Volunteer vascular plant establishment on roofs at the University of Saskatchewan. Landscape and Urban Planning 79 :20–28. Bowler, D.E., L. Buyung-Ali, and T.M. Knight. 2010. Urban greening to cool towns and cities: A systematic review of the empirical evidence. Landscape and Urban Planning 97:147–155. Brown, C., and J.T. Lundholm. 2015. Microclimate and substrate depth influence greenroof plant community dynamics. Landscape and Urban Planning 143 :134–142. Castleton, H.F., V. Stovin, and S.B. Beck. 2010. Green roofs: Building ener gy savings and the potential for retrofit. Energy and Buildings 42:1582–1591. Cho, S.H., D.M. Lambert, R.K. Roberts, and S.G. Kim. 2010. Demand for open space and urban sprawl: The case of Knox County, Tennessee. Pp. 171–193, In A. Paez, J. LeGallo, R.N. Buliung, and S. Dallerba (Eds.). Progress in Spatial Analysis: Methods and Applications. Springer-Verlag, Berlin, Germany. 492 pp. Del Tredici, P. 2010. Wild Urban Plants of the Northeast. Cornell University Press, Ithaca, NY, USA. 392 pp. Dunnett, N. 2015. Ruderal green roofs. Ecological Studies 223:2 33–255. Dvorak, B., and A. Volder. 2010. Green-roof vegetation for North American ecoregions: A literature review. Landscape and Urban Planning 96:197–213. Faeth, S.H., C. Bang, and S. Saari. 2011. Urban biodiversity: Patterns and mechanisms. Annals of the New York Academy of Science 1223:69–81. Forman, R.T.T. 2014. Urban Ecology. Cambridge University Press, Cambridge, UK. 476 pp. Gabrych, M., D.J. Kotze, and S. Lehvävirta. 2016. Substrate depth and roof age strongly affect plant abundances on sedum–moss and meadow green-roofs in Helsinki, Finland. Ecological Engineering 86:95–104. Getter, K. L., D.B. Rowe, and J.A. Andresen. 2007. Quantifying the effect of slope on extensive green-roof stormwater retention. Ecological Engineering 31:225–231. Hintze, C., F. Heydel, C., Hoppe, S. Cunze, A. König, and O. Tackenberg. 2013. D³: The dispersal and diaspore database—baseline data and statistics on seed dispersal. Perspectives on Plant Ecology, Evolution, and Systematics 15:180–192. Available online at Accessed on 11 November 2016. Urban Naturalist M.L. McKinney and N.D. Sisco 2018 Special Issue No. 1 88 Jim, C.Y., and W.Y. Chen. 2011. Bioreceptivity of buildings for spontaneous arboreal flora in compact city environment. Urban Forestry and Urban Greening 10:19–28. Ksiazek, K., J. Herrmann, Sean B. Menke, and Manfred Köhler. 2017. If you build it, will they come? plant and arthropod diversity on urban green roofs over time. Urban Naturalist Special Issue 1:52–72. Li, Y.L., and R. W. Babcock. 2014. Green roofs against pollution and climate change: A review. Agronomy and Sustainable Development 34:695–705. Liu, Y.Q., Q.Y. Meng, J.H. Zhang, L.L. Zhang, T. Jancso, and R. Vatseva. 2016. An effective building-neighborhood green-index model for measuring urban green space. International Journal of Digital Earth 9:387–409. MacIvor, J.S. 2016. Building height matters: Nesting activity of bees and wasps on vegetated roofs. Israel Journal of Ecology and Evolution 62:88–96. MacIvor, J.S., and K. Ksiazek. 2015. Invertebrates on green roofs. Pp. 333–355, In R. Sutton (Ed.) Green Roof Ecosystems. Springer International Publishing, Cham, Switzerland. DOI:10.1007/978-3-319-14983-7_14. Madre, F., A. Vergnes, N. Machon, and P. Clergeau. 2014. Green roofs as habitats for wild plant species in urban landscapes: First insights from a large-scale sampling. Landscape and Urban Planning 122:100–107. Martinuzzi, S.J., J.C. Withey, A.M. Pidgeon, A.J. McKerrow, and V.C. Radeloff. 2015. Future land-use scenarios and the loss of wildlife habitats in the southeastern United States. Ecological Applications 25:160–171. Mentens, J., D. Raes, and M. Hermy. 2006. Green roofs as a tool for solving the rainwater runoff problem in the urbanized 21st century? Landscape and Urban Planning 77:217–226. Miller, L.E., A.E. Heim, and J. Lundolm. 2014. Green-roof vegetation type affects germination and initial survival of colonizing woody species. Urban Forestry and Urban Greening 13:892–899. Oberndorfer, E., J. Lundholm, B. Bass, R.R. Coffman, H. Doshi, N. Dunnett, S. Gaffin, M. Köhler, K.K.Y. Liu, and B. Rowe. 2007. Green roofs as urban ecosystems: Ecological structures, functions, and services. Bioscience 57:823–833. Robinson, S.L., and J.T. Lundholm. 2012. Ecosystem services provided by urban spontaneous vegetation. Urban Ecosystems 15:545–557. SAS. 2016. Statistical Analysis System, 9.0. Cary, NC. Santamouris, M. 2014. Cooling the cities: A review of reflective and green-roof mitigation technologies to fight heat island and improve comfort in urban environments. Solar Energy 103:682–703. Stovin, V. 2010. The potential of green roofs to manage urban stormwater. Water and Environment Journal 24:192–199. Tonietto, R., J. Fant, J. Ascher, K. Ellis, and D. Larkin. 2011. A comparison of bee communities of Chicago green roofs, parks, and prairies. Landscape and Urban Planning 103:102–108. US Census Bureau. 2015. Urban population by state. Available online at http://www.census. gov/. Accessed 12 March 2016. Williams, N.S.G., J.T. Lundholm, and J.S. MacIvor. 2014. Do green roofs provide habitat for biodiversity? Journal of Applied Ecology 51:1643–1649.