As roofs can represent up to 32% of horizontal surfaces in a typical urban setting, they are ideal candidates for green infrastructure. But despite their potential and growing implementation globally, green roofs are poorly researched, particularly regarding management practices.
Methods
This study examines biodiversity found on green roofs in 13 urban locations across three states in the Southeastern United States: Asheville, NC; Atlanta, GA; Chattanooga, TN; Knoxville, TN; and Nashville, TN. Samples were collected over the course of 6 months from March to August 2024.
Results
We find evidence that plant diversity is positively influenced by: soil temperature, roof size, replanting and age. It is negatively influenced by: substrate moisture, roof elevation, irrigation and chemical treatments. Plant frequency (cover) is positively influenced by: substrate temperature, moisture, and elevation. Invertebrate diversity is also positively influenced by soil temperature and age, and negatively influenced by irrigation. Invertebrate abundance is positively influenced by the presence of management and negatively influenced by: replanting, weeding, and substrate depth.
Discussion
As green roofs in the Southeastern United States are sparsely researched, this study helps examine the complexity of green roof ecosystems and provides preliminary insight about the best green roof features and management practices to promote biodiversity.
ecologyecosystem servicesgreen infrastructuremanagementnature-based solutionssustainabilityurban environmentsurbanizationThe author(s) declared that financial support was not received for this work and/or its publication.section-at-acceptanceUrban EcologyIntroduction
With roofs representing up to 32% of the horizontal surface of urban or built areas (Oberndorfer et al., 2007), roofs are prime locations for green infrastructure. Green roofs are specially designed roofs planted with several different vegetation types on top of a substrate, or growth medium (Shafique et al., 2018). Green roofs offer a wide range of benefits compared to a conventional roof, including environmental, social, and economic benefits (Nguyen et al., 2021). In terms of ecosystem services, green roofs provide habitats for pollinators, thermal regulation of cities, and can provide a sense of social cohesion within a community (Langemeyer et al., 2020). In this context, green roofs can be understood as a form of nature-based solution, functioning as designed urban ecosystems that support biodiversity and ecosystem services; while not explicitly focusing on unmanaged vegetation, they also offer insights relevant to refined wilding and urban rewilding within highly managed urban landscapes.
A review of the green roof literature found that many challenges in green roof implementation can be traced to research gaps involving costs, insufficient knowledge about construction mechanics, maintenance, and a lack of coordination between different disciplines (Shafique et al., 2018). Indeed, green roofs are a relatively understudied topic of the urban ecosystem (Langemeyer et al., 2020), with much of the research being done from small test plots in controlled environments and not on large established green roofs (Snodgrass and Mcintyre, 2010).
To study green roof benefits in a rigorous and quantifiable way, bioindicators, or proxies of green roof ecosystem services can be used (Roeland et al., 2019). For example, a review of 77 studies showed that many papers utilized taxonomic diversity, species richness and diversity as indicators for various kinds of ecosystem services (Schwarz et al., 2017). Invertebrates are often used as a proxy for ecosystem services because they provide a key ecological role in terrestrial ecosystems such as economic and non-economic benefits in crop production (Noriega et al., 2018). Ecosystem services for invertebrates include: pollination and pest control, and many financial, food and health benefits to humans (MacIvor and Ksiazek, 2015; Noriega et al., 2018). Invertebrates can also provide ecosystem services such as soil turnover, decomposition and nutrient cycling and they play an important role in local food webs (Gill et al., 2016).
Recent studies document hundreds of species on green roofs indicating how green infrastructure can contribute to regional biodiversity by providing resources to a variety of species (MacIvor and Ksiazek, 2015). Invertebrates such as mollusks and annelids also have key ecological roles as they fill many niches and are globally abundant and diverse, and their acute responsiveness to weather helps us monitor and evaluate the rate of climate change (Prather et al., 2012). Terrestrial mollusks and annelids provide such ecological services as seed dispersal, pollination, decomposition, nutrient cycling, among many others (Prather et al., 2012). Quantifying diversity and abundance of these groups can thus provide key insights into the many benefits of green roofs.
In addition to invertebrates, plants can provide information as proxies for ecosystem services (Robinson and Lundholm, 2012; Buchanan et al., 2020; Walston et al., 2021). For example, plant diversity has been shown to provide improvements in: regulating urban climates (Lehmann et al., 2014), food production, raw materials, medicines, and aesthetics (Robinson and Lundholm, 2012) and above ground biomass increases which, in turn, provides more ecosystem services (Quijas et al., 2010). Plant diversity can also provide erosion control, resistance to plant invasions, regulation of plant pathogens, pollinator supply (Quijas et al., 2010; Walston et al., 2021), and nutrient cycling (Buchanan et al., 2020). Importantly, ecological succession often alters biodiversity and biomass through time on green roofs, especially where roof ecosystems have little to no management (e.g., Madre et al., 2013; Catalano et al., 2016; Ksiazek-Mikenas et al., 2018; Zhang et al., 2021). This can have important implications for changing ecosystem service provisioning over time.
Thus, by analyzing green roof invertebrate and plant communities through time and among different green roofs, conclusions can be drawn about ecological succession and how green roofs can provide benefits relative to a natural ecosystem. We can also compare community changes between green roofs to study the effects of different roof features and management practices.
Specifically, this research aimed to answer the questions outlined below.
How does invertebrate and plant biodiversity vary between management practices (e.g., high or low maintenance) among 13 green roofs in five major US cities?
How does green roof age and physical variables (e.g., soil temperature and moisture) affect biodiversity in these locations?
MethodsStudy area & sample size
In January 2024, 35 roofs were selected in five Southeastern US cities: Asheville, NC, Atlanta, GA, Chattanooga, TN, Knoxville, TN, and Nashville, TN. These cities were chosen due to their drivable proximity of 2–3 hours from the University of Tennessee, Knoxville. Also, these are the largest nearby cities and thus have the highest concentration of green roofs (McKinney et al., 2019). Access for research was secured through cold calls to each building, resulting in 13 of 35 contacted roofs accepting the research proposal to participate in this study.
The 13 chosen locations are illustrated in Figure 1, with red points indicating their locations across the five cities. The specific green roof location details including acronyms and coordinates are outlined in Supplementary Table S1.
Locations of 13 studied green roofs, denoted by a red dot in each of five cities: Asheville, Atlanta, Chattanooga, Knoxville, and Nashville. Locations adapted from McKinney et al., (2019) and found through internet searches (e.g., www.greenroofs.com). Roof locations are denoted by red points. Figure also found in Swanson & McKinney 2025.
Composite map graphic showing detailed street maps for Knoxville, Nashville, Atlanta, Asheville, and Chattanooga, each marked with red dots at specific locations. An inset map highlights the southeastern United States, with a red rectangle outlining the broader region depicted in the individual city panels.Sample determination
Each of the 13 roofs vary in size, ranging from 191,000 ft2 to 100 ft2. To ensure proportional sampling across the different roofs, a systematic approach was adopted. Pitfall traps were set at intervals of one every 200 ft2. For every 400 ft2, 20 total teabags were buried, one plant survey was taken, and one soil sample was collected. Given the variability in roof sizes, samples were collected up to a maximum area of 4,800 ft2 per roof. This was done to ensure a feasible workload. Specifically, sampling was done with a minimum of two traps, two plant surveys, two soil collections and twelve teabags per roof. There was a maximum of twenty-four traps, twelve plant surveys, twelve soil collections, and one hundred-twenty teabags for the large roofs. For roofs exceeding 4,800 ft2, only 4,800 ft2 total was sampled to ensure manageable data collection. Roofs exceeding 4,800 ft2 were randomly sampled from across the entirety of the roof.
Roof features data collection
Each green roof was surveyed regarding structural attributes. These independent attributes are: soil temperature, soil moisture, age (years), elevation (stories), substrate depth (inches), irrigation, replanting, weeding, substrate addition, chemical treatments, and management classification. Each management attribute is defined below:
Irrigation was described as whether the roof used any sort of irrigation system and how often the roof was irrigated.
Replanting asked whether the roof had ever been replanted at any point in time.
Weeding patterns were described by asking if routine weeding was conducted throughout the year.
Substrate additions were defined as whether the roof had ever had extra substrate added or substrate replaced since being built.
Chemical treatment asked if the roof uses any chemical treatments such as herbicides and pesticides.
Each of these independent attributes are collected through field surveys and questionnaires. Substrate temperature and moisture were collected each month during roof visits by utilizing a handheld probe. Using this probe, measurements were collected at the beginning of the visit, ensuring they were measured around the same time of day for each collection. The age, elevation, substrate depth, and various management strategies were collected using a questionnaire sent to each roof manager. These questions are found in Supplementary Table S2, and the responses are found in Table 1. The questionnaire was designed to classify each roof into one of two management classifications: high or low. High management is defined as regular irrigation of more than once a month and weeded every 3–12 weeks and low management when the roof has less than those frequencies or no regular irrigation or weeding throughout the year. This detailed categorization was the foundation for analyzing the impact of different management practices on the ecosystem services provided by the green roofs.
Summary of location information including: age, size, elevation, substrate depth, and the different management practices.
Location
Age (years)
Size (ft2)
Elevation (stories)
Substrate depth (in)
Irrigation
Replant
Weed
Chemical treatment
Substrate addition
Mgmt. score
Mgmt. classification
MCC
14
191,000
9
2.5
Yes
Yes
Yes
Yes
Yes
5
High
FW
15
5000
3
5
Yes
Yes
Yes
Yes
No
4
High
RMH
9
2200
1
4.25
Yes
Yes
Yes
No
No
3
High
HCH
13
4200
2
6
Yes
Yes
Yes
No
No
3
High
GS
1
200
1
4
Yes
Yes
No
No
Yes
3
High
ZEC
3
4000
5
10
Yes
No
Yes
No
No
2
Low
ABG
16
200
1
24
No
Yes
Yes
No
No
2
Low
SF
15
1200
2
6
No
No
Yes
No
Yes
2
Low
SB
15
2500
2
6
No
No
Yes
No
No
1
Low
BLR
19
9000
1
6
No
No
No
Yes
No
1
Low
CCB
6
1200
4
6
Yes
No
No
No
No
1
Low
LSH
16
200
1
4
No
No
No
No
No
0
Low
GR
2
100
1
12
No
No
No
No
No
0
Low
Each yes for management is scored as a 1 and no as a 0. This calculation provides a score, where values greater than 2.5 indicate ‘high’ management and values below 2.5 indicate ‘low’ management.
Vascular plant collection
Random sampling of vegetation was completed during the growing season at each location every 50 days from March until August utilizing 0.7m x 0.75m plastic quadrat frames on the roof, with the same sampled location being surveyed each collection (Bagella et al., 2020). Quadrat size and shape can vary with vegetation; with herbaceous vegetation, quadrats are recommended to be 1-2m2 (Higgins et al., 1996). For this study the quadrat size was smaller due to the size restraints of many green roofs. These quadrats were placed at random locations across the roof, marking the location to ensure the same spot was sampled each time. Vegetation frequency was estimated visually through 15cm x 15cm grids in each quadrat (Lönnqvist et al., 2021). Occurrence of each plant species was used to estimate percent frequency. For example, where if there were 12 squares and one species occurs in 6 of them, the frequency would be 6/12 x 100, or 50% (Higgins et al., 1996). Thus, plant frequency is a rough estimate of how widespread the occurrence of each species is.
Plant identification followed keys or online databases such as wildflowersearch.org. Problematic identifications were made by botany faculty at the University of Tennessee. All the plants identified within the quadrat were recorded to genus or species and number of occurrences of each species within the grid (Zhao et al., 2009; Wang et al., 2015; Lönnqvist et al., 2021). Diversity was then calculated using the Shannon Index, Equation 1, to evaluate the diversity, using richness, of the sampled areas (Wang et al., 2015). This diversity index was calculated on each roof for each collected quadrat.
Equation 1The Shannon Index (Wang et al., 2015).
He'=−∑i−1SPilnPi
Where pi is the number of individuals of the ith species and S is the total number of species (Lönnqvist et al., 2021). The Shannon index, H’, ranges from 0-3.5 with lower number indicating less diversity. This index was calculated between green roofs to compare diversity.
Invertebrate collection
To study invertebrate diversity and abundance, pitfall traps were used. These were chosen as they are very commonly used. Pitfall traps were used in almost 90% of field studies on ground beetles from 2008 to 2010 (Knapp and Ruzicka, 2012; Knapp et al., 2016). They were used in over 200 studies in 2014 alone, remaining one of the most popular sampling techniques (Knapp and Ruzicka, 2012; Knapp et al., 2016). Typically, on each of the roof’s traps were set randomly, keeping each trap a minimum of about 4 meters apart (MacIvor and Lundholm, 2010; Fabián et al., 2021). Trap design is variable between studies and has been shown to affect catch size, where traps with no lid obtained higher catches than lidded traps and traps made with a smoother material showed higher catch efficiencies as well (Knapp et al., 2016). Samples have also been shown to be affected by trap size (Knapp et al., 2016), trap color (Buchholz et al., 2010), and preservative fluid (Knapp and Ruzicka, 2012). When choosing the preservation fluid, it is important to consider the efficiency as well as the environmental impact and risk (Knapp et al., 2016). Compared to other options propylene glycol is low in toxicity and has a slower evaporation rate maintaining its capture efficiency and preservation (Martin and Murphy, 2000). Knapp and Ruzicka, 2012 and Knapp et al., 2016 showed propylene glycol to have a higher trapping efficiency than the compared preservatives.
For this study, pitfall traps were made from 9 oz smooth plastic beverage cups, with one cup buried, lip flush with the ground, and a second cup placed inside it, covered about 5cm above with a white plastic lid sized 15cm x 15cm (MacIvor and Lundholm, 2010; Jacobs et al., 2023). The lidded design was used due to the longer collection intervals to keep the traps free of debris and rainfall. The cup is then filled with about 2 oz of propylene glycol (Jacobs et al., 2023). Pitfall traps are typically collected every ten days to three weeks for five months (MacIvor and Lundholm, 2010; Jacobs et al., 2023). Due to scheduling restraints and distance from the University of Tennessee, Knoxville campus, traps were collected every 25–30 days. The propylene glycol and captured invertebrates were drained through a coffee filter in efforts to reuse the propylene glycol. After the invertebrates from each pitfall trap were collected in the coffee filter, they were washed with clean water, air dried, sorted, and stored in 99.9% ethanol to be later be counted and identified (MacIvor and Lundholm, 2010; Fabián et al., 2021; Jacobs et al., 2023). The invertebrates were identified using dichotomous keys and online databases such as bugguide.net to the family level and genus where applicable (MacIvor and Lundholm, 2010; Fabián et al., 2021). Questionable identifications were made by entomology faculty at the University of Tennessee.
Invertebrate’s richness and abundance metrics were used to determine diversity in the sampled area by using the Shannon index outlined in Equation 1 (MacIvor and Lundholm, 2010; Wang et al., 2015). Invertebrate abundance was calculated by the total invertebrates found in each individual trap during any given sample collection.
Statistical analyses
Statistical analysis of these data was carried out with the statistical software package IBM SPSS Statistics Version 29.0.2.0 (20) (Phoomirat et al., 2020). A Principal Component Analysis (PCA) was implemented to identify covarying components and create groupings of the independent variables that explain much of the variance in the data set. These independent variables are: soil temperature, soil moisture, age (years), elevation (stories), substrate depth (inches), irrigation, replanting, weeding, substrate addition, chemical treatments, and management classification. Within the analyses each data set was checked for normality. For non-normal distributions, a logarithmic transformation was used to produce normality. If the transformation did not work, nonparametric tests were used utilizing the Spearman correlation. The normal values were compared via the Pearson correlation.
Linear mixed models were then produced for each ecosystem service proxy variable against each of the independent variable groupings. Linear mixed models were utilized over a univariate ANOVA or other test as it considers the location and sample dates of each data point. Using a univariate ANOVA does not account for location differences and uses an inflated sample size, leading to misleading results. The linear mixed models were completed using a manual stepwise regression to determine which of the independent variables showed the highest significance. Statistical significance was evaluated using two alpha levels (α = 0.05 and α = 0.10) to reflect differing levels of stringency in hypothesis testing. The conventional α = 0.05 threshold was used to identify strong evidence of effects, while α = 0.10 was applied to highlight marginal trends that may warrant further investigation. To complete the stepwise regression, each time the model was run, the least significant variable was removed until the final variables showed the most statistical significance.
Since the independent variables in the groupings found during the PCA were compared to the proxy’s using covariance, this provides a better view of what is influencing each proxy while holding all other variables constant. This was done to find the statistical significance and effect of the environmental, management, and site characteristics on the dependent ecosystem service proxy variables (e.g., invertebrate and plant diversity). Finally, to check the assumptions of statistically significant influences of each component a linear mixed model was carried out using the component scores found from the PCA.
Results
Table 2 shows the PCA analysis for each of the independent variables. Extraction numbers represent the proportion of each variable’s variance that can be explained by the principal components. PCA values range from 0 to 1, where values closer to 1 represent higher communality meaning the variable is well represented by the components and a significant portion of its variance is captured. Based on the values in Table 2, variables with high communalities and are thus well represented by the extracted components include: soil moisture, soil temperature, age, size, elevation, irrigation, replanting, chemical treatment, and substrate addition. Management classification variance is very highly captured with an extraction of 0.966. Variables that show moderate communalities are substrate depth and weeding. In sum, the PCA shows all variables to be well represented by the selected components with communalities above 0.70, indicating that the principal components capture most of the variance in each variable.
Extraction results from the completed PCA.
Independent variable
Extraction
Substrate Moisture (%)
0.831
Substrate Temperature (°C)
0.731
Age (yrs)
0.885
Size (ft2)
0.890
Elevation (stories)
0.899
Substrate Depth (in)
0.774
Irrigation (Y/N)
0.896
Replanting (Y/N)
0.878
Weeding (Y/N)
0.756
Chemical Treatments (Y/N)
0.807
Substrate Addition (Y/N)
0.884
Management Classification (high/low)
0.966
Values closer to 0 represent higher communality.
Table 3 shows the total variance explained, including the proportion of variance each principal component captures in the data set. The initial eigenvalues show the amount of variance in the data each component explains indicating the percentage of the total variance for each individual component. Only the first 5 components have significance, with eigenvalues over 1. These 5 components explain 84.983% of the total variance, indicating a strong representation. A Varimax rotation was applied to the PCA to help separate the variables by redistributing the variance more evenly across the components. Rotation sums of squared loadings show the total variance explained by each component after rotation, leading to a more balanced solution. Based on these findings, component 1 explains about 23% of the variance, component 2 explains about 18% of the variance, component 3 explains 17% of the variance, component 4 explains 14% of the variance and component 5 explains 11% of the variance.
The results of the PCA, outlining the explained variance from each component.
Initial eigenvalues
Extraction of squared loadings
Rotation sums of squared loadings
Component number
Total
% of variance
Cumulative %
Total
% of variance
Cumulative %
Total
% of variance
Cumulative %
1
3.790
31.587
31.587
3.790
31.587
31.587
2.766
23.053
23.053
2
2.186
18.219
49.806
2.186
18.219
49.806
2.175
18.122
41.175
3
1.812
15.099
64.905
1.812
15.099
64.905
2.112
17.598
58.774
4
1.382
11.519
76.423
1.382
11.519
76.423
1.763
14.694
73.468
5
1.027
8.560
84.983
1.027
8.560
84.983
1.382
11.515
84.983
6
0.658
5.483
90.466
7
0.476
3.967
94.433
8
0.295
2.460
96.894
9
0.229
1.905
98.798
10
0.131
1.092
99.891
11
0.009
0.076
99.967
12
0.004
0.033
100
Note that components 1–5 are the only ones that have a total eigenvalue over 1, indicating more significance.
The varimax rotation yielded the component matrix in Table 4. High loadings, with values over 0.50, indicate that variable is strongly associated with a component. High loadings on component 1 are: irrigation, replanting and management classification. We thus summarize this group of loadings as the management practice component. Component 2 shows a high loading by soil moisture and temperature which we summarize as the substrate condition component. Component 3 has high loadings on: roof size, elevation, and substrate additions, summarized as the roof features component. Component 4 has high loadings on age and chemical treatment, labelled simply as the age and chemical management component. Finally, component 5 shows high loading on weeding and substrate depth, labelled as the substrate and vegetation management component.
Rotated component matrix results from the PCA.
Rotated component matrix
Variable
Component 1
Component 2
Component 3
Component 4
Component 5
Substrate Moisture (%)
0.141
0.889
−0.099
−0.078
0.68
Substrate Temperature (°C)
0.197
−0.792
−0.156
0.068
0.189
Age (yrs)
−0.177
−0.283
−0.058
0.851
0.214
Size (ft2)
0.200
0.251
0.795
0.390
0.063
Elevation (stories)
0.182
0.555
0.698
0.233
0.127
Substrate Depth (in)
−0.319
−0.154
−0.232
−0.221
0.739
Irrigation (Y/N)
0.825
0.396
0.143
−0.164
−0.108
Replanting (Y/N)
0.855
−0.205
0.117
0.080
0.291
Weeding (Y/N)
0.35
0.061
0.149
0.220
0.747
Chemical Treatments (Y/N)
0.200
0.097
0.251
0.813
−0.185
Substrate Addition (Y/N)
0.138
−0.259
0.866
−0.141
−0.167
Management Classification (high/low)
0.953
−0.055
0.179
0.102
−0.11
Highlighted values indicate the most significant variables (>0.50). These variables are then grouped together for the corresponding component by their communalities.
In summary, based on the findings of the varimax rotated PCA it is convenient and helpful to use these 5 components as groupings for linear mixed models using the variables in each component as covariant groups. This analysis thus reduced data complexity by identifying five key components, each representing a unique aspect of the independent variables. This will, in turn, indicate the influence of these groups of green roof management and physio-chemical variables on the proxy variables (e.g., invertebrate and plant diversity).
Temporal trends in roof substrate moisture and temperature over the course of the study are outlined in Figure 2. There is a generally inverse pattern: as mean roof temperature increased throughout the season; moisture tended to decline. Both maximum substrate temperature and minimum substrate moisture approximately occurred in late June/early July.
Collected soil (substrate) moisture and temperature from March until August 2024. The blue points and line indicate the collected moisture levels (%) and the red points and line indicate the collected temperature levels in Celsius. The quadratic trend line represents the mean with a 95% confidence interval.
Scatter plot showing blue dots for percent soil moisture and red dots for soil temperature in degrees Celsius collected between February and August 2024, with trendlines indicating decreasing soil moisture and increasing then decreasing soil temperature over time.Vascular plant community
For each collection cycle, 87 vegetation surveys were conducted across the thirteen roofs. At the end of the four collection cycles 348 total surveys had been conducted. Of the vascular plants in the Tracheophyta phylum, a total of 102 different species were identified. The most common families across the roofs found were: Crassulaceae, Poaceae, Fabaceae, and Asteraceae. All found species can be found in the supplemental materials. Figure 3 illustrates the plant frequency (i.e., number of grids occupied) and diversity found at each roof throughout the study. The trend in frequency and diversity are similar in that they increase until June and then start to decline in July, which is in line with the seasonality from March until August.
Plant diversity and frequency on each roof from March until August 2024. The blue points and line indicate the vascular plant frequency found at each roof and the red points and line indicate the diversity of the vascular plant community found atop each roof. The quadratic lines represent the mean with a 95% confidence interval.
Scatter plot with trendlines showing vascular plant log frequency (blue points and line) and diversity (red points and line) at multiple dates from February to August 2024, with quadratic R squared values near zero.
Table 5 outlines the significance of each of the highly correlated variables found in the PCA. Regarding management practices, irrigation is highly significant (0.05 and 0.10) as a covariate effect on plant diversity. Interestingly, the effect of irrigation on diversity is negative. Although it is only near significant (0.124), there is some indication that replanting, as a covariate, has a positive effect on plant diversity. Management intensity appears to have no effect on plant diversity. For plant frequency, none of the management practice variables are even near-significance. For substrate condition effects on plant diversity, soil moisture and temperature alone are each statistically significant to the 0.05 level. Soil moisture shows a negative effect on plant diversity and temperature shows a positive effect on plant diversity. For plant frequency effects, we find that soil temperature and moisture both have a significant positive effect on frequency when considered together but not alone.
Results of the linear mixed model comparing the component variables determined by the PCA.
Plant diversity component 1: Management practices
Irrigation (Y/N)
0.070 (−0.413)
0.048 (−0.358)
0.167 (−0.209)
Replant (Y/N)
0.539 (0.176)
0.124 (0.270)
0.615 (0.080)
Management (high/low)
0.676 (0.150)
0.992 (−0.002)
Plant frequency component 1: Management practices
Irrigation (Y/N)
0.253 (16.3)
0.243 (15.4)
0.447 (6.7)
Management (high/low)
0.449 (−17.4)
0.363 (−12.2)
0.954 (−0.526)
Replant (Y/N)
0.764 (5.4)
0.979 (−0.240)
Plant diversity component 2: Substrate conditions
Soil Moisture (%)
0.196 (−0.010)
0.016 (−0.014)
Soil Temperature (°C)
0.514 (0.005)
0.032 (0.012)
Plant frequency component 2: Substrate conditions
Soil Temperature (°C)
0.036 (0.617)
0.254 (0.241)
Soil Moisture (%)
0.063 (0.617)
0.589 (0.119)
Plant diversity component 3: Roof features
Elevation (stories)
0.062 (−0.117)
0.042 (−0.124)
0.294 (−0.036)
Size (ft2)
0.195 (3E−6)
0.079 (4E−6)
0.975 (4E−8)
Substrate Additions (Y/N)
0.463 (0.146)
0.374 (0.158)
Plant frequency component 3: Roof features
Substrate Additions (Y/N)
0.304 (−12.3)
0.253 (−11.9)
0.649 (−4.8)
Elevation (stories)
0.310 (3.6)
0.065 (3.8)
0.112 (3.0)
Size (ft2)
0.938 (1E−5)
0.281 (9E−5)
Plant diversity component 4: Age and chemical management
Chemical Treatments (Y/N)
0.005 (-0.515)
0.032 (-0.364)
Age (yrs)
0.059 (0.023)
0.620 (0.007)
Plant frequency component 4: Age and chemical management
Age (yrs)
0.658 (0.390)
0.538 (0.463)
Chemical Treatments (Y/N)
0.856 (2.2)
0.656 (4.7)
Plant diversity component 5: Substrate and vegetation management
Weeding (Y/N)
0.612 (0.086)
0.561 (0.095)
Substrate Depth (in)
0.650 (0.007)
0.583 (0.008)
Plant frequency component 5: Substrate and vegetation management
Weeding (Y/N)
0.132 (13.9)
0.125 (13.4)
Substrate Depth (in)
0.743 (−0.259)
0.922 (−0.081)
Dark gray shading represents significance to the 0.05 level and light gray indicates significance to the 0.10 level. Parentheses indicate positive or negative effect on the dependent variable.
Regarding roof feature effects on plant diversity, elevation is significant at the 0.10 level when all of component 3 variables (roof size, elevation, and substrate additions) are utilized as covariates. When the least significant variable, substrate additions, is removed elevation and size gain stronger statistical significance with elevation becoming significant at the 0.05 level and size becoming significant at the 0.10 level. Elevation has a negative effect on plant diversity whereas roof size as a covariate has a positive effect. For plant frequency, the only evidence of an effect is for elevation which shows a significant positive effect when substrate additions are included and a near-significant positive effect (0.112) when analyzed alone. Age and chemical management both show a significant effect on plant diversity. Chemical treatments have a strong negative impact alone and even when age is a covariate. Age has a significant positive effect as a covariate with chemical treatment. For plant frequency, neither age nor chemical treatments show any significant or near-significant effect.
Neither weeding nor substrate depth have a significant effect on plant diversity or frequency, with or without covariance. However, there is a near-significant positive effect (0.125) of weeding alone on plant frequency.
Invertebrate community
There was a total of six sample collection points, where the pitfall traps collected data on the abundance and diversity of Arthropods, Mollusks, and Annelids on each roof. Of the 160 total pitfall traps set out in March, an average of only 135 traps were collected at each sample interval. This occurred as some traps could not be sampled for various reasons including human and animal disturbance of traps, bad weather, and vegetation overgrowth. By the end of the six collections, a total of 807 traps were retrieved, capturing a total of more than 136,500 individuals from three respective phyla and 193 different genera. The number of individuals captured, by phyla, were: 13,296 Mollusca, 123,194 Arthropoda, and 22 Annelida. Within these phyla are multiple ground dwelling orders, the most notable of which are: 5,273 Coleoptera, 2,778 Araneae, 38,666 Entomobryomorpha, 5,107 Polydesmida, 4,887 Orthoptera, and 180 larvae and caterpillars. Notable flying orders within these phyla were 24,665 Hymenoptera and 4,727 Diptera. All discovered genera can be found in the supplemental material. Figure 4 illustrates the trends of the invertebrate community throughout the duration of this study. Interestingly, invertebrate abundance on green roofs began to decline in May but diversity increased until June before beginning to decline.
Invertebrate diversity and abundance from March until August 2024. The blue points and line indicate invertebrate diversity and the red points and line indicate invertebrate abundance. The quadratic line represents the mean of all the points with a 95% confidence interval.
Scatter plot with polynomial trend lines depicts invertebrate diversity (blue) and log-transformed abundance (red) by sampling date from late March to August 2024, indicating slight declines over time for both variables.
Table 6 outlines the significance of each highly correlated variable in the PCA. Of the management practice variables (Table 6), invertebrate diversity is significantly negatively influenced by irrigation when replanting is a covariate at the 0.05 level. Indeed, irrigation has a near-significantly negative effect at the 0.10 level on diversity by itself (-0.112). All other management practices (management intensity and replanting) show no statistical significance on invertebrate diversity. For invertebrate, both replanting and management are significant to the 0.10 level when included as covariates of each other, i.e., removing irrigation as an influence. Interestingly, replanting shows a negative influence and management shows a positive influence on invertebrate abundance.
Results of the linear mixed model comparing the component variables outlined by the PCA.
Invertebrate diversity component 4: Age and chemical management
Age (yrs)
0.050 (0.028)
0.128 (0.019)
Chemical Treatments (Y/N)
0.170 (−0.259)
0.654 (−0.081)
Invertebrate abundance component 4: Age and chemical management
Age (yrs)
0.216 (0.019)
0.197 (0.017)
Chemical Treatments (Y/N)
0.772 (−0.060)
0.743 (0.061)
Invertebrate diversity component 5: Substrate and vegetation management
Substrate Depth (in)
0.375 (0.013)
0.360 (0.013)
Weeding (Y/N)
0.857 (−0.29)
0.957 (−0.008)
Invertebrate abundance component 5: Substrate and vegetation management
Substrate Depth (in)
0.003 (−.037)
0.009 (−0.34)
Weeding (Y/N)
0.055 (−0.060)
0.217 (0.199)
Dark gray squares represent significance to the 0.05 level and light gray squares indicate significance to the 0.10 level. The number in parentheses indicates positive or negative effect on the dependent variable.
Regarding the two substrate condition variables, temperature and moisture, temperature alone is statistically positively significant (0.05 level) for invertebrate diversity (Table 6). However, moisture alone does have a near-significant negative effect (-0.123) on invertebrate diversity. For invertebrate abundance, neither substrate moisture nor temperature is found to be statistically significant or even near significant. Of the roof feature variables, when all three variables are compared to diversity, elevation is found to be the most significant with a p-value of 0.138 and becomes more (near- significant) with a p-value of 0.110 when substrate additions are removed. In all cases, the effect of roof elevation on invertebrate diversity is negative, indicating a general trend of less diversity with increasing roof height. Interestingly, none of the three roof features (elevation, size, and substrate additions) examined here show any significant or even near-significant effect on invertebrate abundance. Regarding age and chemical treatments, age has a significant and positive effect on diversity as a covariate at the 0.05 level. Chemical treatment values imply a negative effect on diversity, but their statistical significance is too low to be confidently interpreted. Age and chemical treatments have no evident impact on invertebrate abundance.
For substrate depth and weeding, there is no evidence for a significant or even near-significant effect on invertebrate diversity. Interestingly however, there is apparent evidence for a strong (0.05 level) negative effect of substrate depth on abundance, both alone and as a covariate. Weeding also has a significant (0.10) negative effect on invertebrate abundance when analyzed as a covariate.
Overall component score analysis
A PCA was completed for an overall component score analysis to compare each of the independent variables as a whole while utilizing the other scores as covariates (Table 7). This was done for both the plant and invertebrate communities. This gives us insight onto how the components as a whole effect the plant and invertebrate communities, whereas the above analyses dive into how each individual variable affects these communities.
Results of the linear mixed model comparing the component variables outlined by the PCA.
Variables
Component 1
Component 2
Component 3
Component 4
Component 5
Plant Diversity
0.720 (-0.026)
0.060 (-0.110)
0.507 (0.041)
0.358 (-0.067)
0.294 (0.077)
Plant Frequency
0.539 (3.0)
0.815 (0.620)
0.917 (0.496)
0.420 (4.1)
0.220 (6.2)
Invertebrate Diversity
0.453 (-0.057)
0.015 (-0.126)
0.781 (0.018)
0.648 (0.030)
0.314 (0.066)
Invertebrate Abundance
0.767 (0.024)
0.315 (0.049)
0.532 (0.050)
0.299 (0.050)
0.578 (-0.045)
Dark gray squares represent significance to the 0.05 level and light gray squares indicate significance to the 0.10 level. The number in parentheses indicates positive or negative effect on the dependent variable.
When all other variables are held the same, we found that only component 2, substrate conditions, shows significance to the 0.10 level for plant diversity. Indicating the substrate conditions to have an overall negative effect on plant diversity. These findings indicate that substrate conditions are moderately significant and account for a substantial portion of the variance in plant diversity. Although the other components do not show significance, roof features and substrate and vegetation management show a positive effect on diversity and frequency. Whereas high management and age and chemical management shows a negative effect on plant diversity but a positive effect on frequency.
When all other variables are held constant, the only significant component is component 2, substrate conditions, which shows a negative effect on invertebrate diversity. This indicates that substrate conditions are highly significant and explain much of the variance found in invertebrate diversity. Although the other components do not show significance it is found management practices have a negative influence, and roof features, age and chemical management, and substrate and vegetation management all have a positive influence on invertebrate diversity. For invertebrate abundance it shows that: management practices, substrate conditions, roof features, and age and chemical management have a positive influence and substrate and vegetation management have a negative influence as a whole component, rather than the individual features as shown in the above models.
Summary of results
Plant diversity: soil temperature (+), roof size (+), replanting (+), age (-), irrigation (-), soil moisture (-), chemical treatment (-), elevation (-)
Plant frequency: soil temperature (+), soil moisture (+), elevation (+),
Invertebrate diversity: soil temperature (+), age (+), irrigation (-)
DiscussionTemporal patterns in temperature, moisture, plants, and invertebrates
Over the course of this study (March-August), most of the key variables measured peaked in mid- late June: maximum substrate temperature, minimum substrate moisture, plant diversity and frequency and invertebrate diversity. Only invertebrate abundance, which peaked in May, deviated from this pattern. Thus, it appears that temperature has a strong positive influence on the diversity of plants and invertebrates. Perhaps more surprisingly, moisture has a negative influence. Possible reasons for these relationships are discussed below.
Green roof features: effects on plant and invertebrate communities
The variables comprising green roof features, and their relationships to plant and invertebrate community variables are all found in components 2, 3, and 4 which we discuss next.
Substrate temperature and moisture effects on plant diversity and frequency
Regarding the substrate conditions effects of component 2 (soil temperature and soil moisture), we find that soil moisture alone has a very significant negative effect on plant diversity. This negative effect of moisture is not only counterintuitive, but it contradicts a sizable literature on green roofs that documents the importance of soil moisture and supplemental irrigation to promote plant diversity (Saraeian et al., 2022; Vandegrift et al., 2019). We expand on this apparent contradiction below in our discussion of management practices because we also found irrigation to have a negative effect on plant diversity.
There is a strong significantly positive relationship between soil temperature alone and plant diversity. This is also a finding that contradicts some of the current literature on green roofs which indicates that shading on extensive green roofs promotes species richness in plants by reducing substrate temperature (Van der Kolk et al., 2020). Indeed, as discussed further below, one of the main reasons why deeper roof substrates promote plant diversity is that they provide a buffer against extreme, high temperatures (Vandegrift et al., 2019). Considering that our temperature data relies on the substrate temperature at one single point in time, taken several weeks apart, this may be a spurious result of having only a few data points. This inference is bolstered by the fact that plant diversity for our data peaked in June and then declines (Figure 4) whereas peak summer temperature from weather monitoring stations for these locations occurred in July and August (National Oceanic and Atmospheric Administration, 2025). A fuller understanding of this relationship will require future study based on continuous temperature measurements.
For plant frequency, we found that neither soil moisture nor temperature alone had an effect. However, when considered together both had an interactive positive influence on frequency. That moisture positively influences plant growth (frequency) aligns well with several previous studies showing the benefits of irrigation (discussed below). However, the positive relationship of temperature with plant growth does not align well with previous work and as noted above, requires further study with more continuous temperature data.
Substrate temperature and moisture effects on invertebrate diversity and abundance
For invertebrate diversity, we find that temperature has a significant positive effect both alone and as an interaction with moisture. As with plant diversity, this does not agree with (the few) previous studies and one review paper, which show that arthropod diversity is negatively correlated with soil temperature (Salman and Blaustein, 2018; Wang et al., 2022). Such a pattern would make sense given that invertebrate diversity tends to be correlated with the diversity of plants, which invertebrates utilize as habitat (Braaker et al., 2017; Joimel et al., 2018; Salman and Blaustein, 2018; Tonietto et al., 2011). Thus, we suspect that our finding to the contrary may indicate a spurious outcome based on our few, widely spaced temperature data points noted above. We find no relationship between invertebrate diversity and soil moisture, which agrees with the one previous study we are aware of (Salman and Blaustein, 2018).
For invertebrate abundance, we find no significant relationships with soil temperature or moisture. For moisture, this agrees with a previous study showing no relationship with abundance (Salman and Blaustein, 2018). For temperature effects on invertebrate abundance, of the two studies reviewed by Wang et al. (2022), one also found no correlation and one found a negative correlation between soil temperature and abundance.
Roof elevation, size and substrate addition effects on plant diversity and frequency
Regarding the roof features effects of component 3 (roof elevation, size, and substrate additions), we find that roof elevation alone does not have a significant effect on plant diversity. This aligns with the review by Wang et al. (2022) who found that none of the 4 studies on elevation and plant diversity they examined showed a relationship between roof height (elevation) and some metric of plant diversity. However, we did find that roof height has a negative effect on plant diversity when the interactive (covariate) effects of roof size are considered. We suspect this is because the more elevated roofs in our study, which were significantly larger in size (area), were more intensely managed sedum-based monocultures with relatively low diversity. Thus, roof area did have a positive effect on plant diversity when these covariate effects of height were accounted for. Substrate additions (a management variable) had no apparent effect on plant diversity.
For plant frequency, we found that no roof feature alone had an effect. This aligns with previous studies, found in the review paper Wang et al. (2022), where neither roof size nor roof height influenced plant abundance or cover. The only detectable effect on plant frequency was a moderate positive effect from roof elevation when the interactive impacts of substrate additions were considered. Again, we suspect this is because the more elevated roofs in our study were more intensely managed sedum-based monocultures where monocultural plant cover was a main goal of management.
Roof elevation, size and substrate addition effects on invertebrate diversity and abundance
Regarding invertebrates, we find a striking lack of significant effects of any of these variables, either alone or as covariates, on invertebrate diversity or abundance. These findings are in general agreement with previous studies on arthropod diversity and elevation and/or area where 8 of 10 studies reviewed found no relationship between arthropod diversity and roof height and 6 of 8 found no relationship between diversity and roof area (Wang et al., 2022).
For invertebrate abundance, we again find a lack of significant effects for any of the roof feature variables, either as covariates or alone. For roof elevation, these findings generally agree with previous studies of arthropod abundance where 4 of 5 studies reviewed found no relationship between arthropod abundance and roof height (Wang et al., 2022). However, one study in the review did find a negative correlation between abundance and height, as did a more recent study of pollinators (bees, wasps, and flies) (Underwood et al., 2025). For roof area, previous studies reveal a more complex pattern, with 2 studies reporting a positive correlation between arthropod abundance and area, 1 study showing a negative correlation and 1 study showing no relationship (Wang et al., 2022).
Age and chemical effects on plant diversity and frequency
Regarding the effects of component 4 (age and chemical management), we find that roof age does not have a significant effect on plant diversity by itself. This seems counterintuitive given that natural ecological succession through time typically increases diversity as organisms colonize or recolonize an area. However, a review of comparative green roof studies on age and plant diversity found that only 2 out of 9 of these reported such a positive effect, with 7 showing no effect of age on plant diversity (Wang et al., 2022). Droz et al. (2021) suggested that this lack of an age effect on plant diversity may result from the intense and often differing human management of green roof ecosystems over time which frequently alters natural successional processes. There is much variation in the amount of irrigation, weeding, replanting and other management strategies which can selectively reduce or increase plant diversity through time. This limitation could be addressed in future studies of roof age and diversity through longitudinal monitoring of vegetation, together with explicit consideration of planting configuration, establishment density, and post-installation management practices (Droz et al., 2021).
Roof age, however, does have a positive effect on plant diversity when considered as a covariate with chemical treatment. This is interesting given that chemical treatments have a significant negative effect on diversity, both alone and as a covariate with age. This finding seems to imply that roof age only increases diversity when the negative diversity impacts of chemical treatments are considered. This provides statistical support for the idea that different management regimes are affecting the role of age so that age becomes significant when some of the management effects (e.g., chemical treatments) are accounted for.
In contrast to diversity, we did not find any relationship between age and plant frequency. This finding is consistent with most previous studies that examined plant cover (as we can find none that looked at plant frequency and age). The review by Wang et al. (2022) found that 3 of 4 studies on the relationship between plant cover and age reported no relation between the two. Similarly, a study of 10 extensive green roofs in North Carolina (not included in the Wang et al. (2022) review), found no relationship between roof age and plant cover (Sherk et al., 2020).
Not surprisingly, we found that chemical treatments alone had a strong negative effect on plant diversity. Most of these treatments were likely selective herbicides targeted at weedy species and so we would expect a reduction of diversity. That the treatments did not significantly affect plant frequency could imply an “ecological release” effect whereby unaffected species replace the affected species as vegetative cover.
Age and chemical effects on invertebrate diversity and abundance
Our finding that roof age alone has no significant effect on invertebrate diversity is also consistent with most previous studies. The review by Wang et al. (2022) found that all 10 studies on diversity and age examined found no significant relationship between roof age and invertebrate (arthropod) diversity. Somewhat similarly, a study of 19 extensive green roofs in Germany (not included in the Wang et al. (2022) review) found that roof age had a very small effect on overall invertebrate diversity, but this effect disappeared when individual taxic (ants, bees, beetles, spiders) diversity was statistically analyzed (Ksiazek-Mikenas et al., 2018). As with plants, we infer that this lack of an age effect is produced by the widely different management regimes of the roofs compared, pointing to the need for more longitudinal studies (Droz et al., 2021).
Age does have a positive effect on invertebrate diversity when considered as a covariate with chemical treatment. Again, this seems to imply that roof age increases diversity when the negative diversity impacts of chemical treatments are considered. Importantly, these effects of age and chemical treatments on invertebrate diversity do not appear to be a matter of invertebrates simply tracking the effects of plant diversity: only 3 of 14 studies that examine this find that arthropod diversity on green roofs over time is statistically affected by plant diversity (Wang et al., 2022).
In contrast to diversity, we did not find any relationship between age and invertebrate abundance. This finding is consistent with most previous studies. Wang et al. (2022) found that 4 of 5 studies on age and invertebrate abundance reported no relation between the two, with the remaining study reported a negative effect of age. Ksiazek-Mikenas et al. (2018) found a small positive effect of age on abundance for 19 German extensive roofs but this effect not clear at the level of individual taxa (ants, bees, beetles, spiders). That we found no effect of chemical treatments on invertebrate diversity or abundance may occur because most treatments were herbicides and thus targeted for plants.
The general lack of roof feature effects on plant and invertebrate diversity and frequency/abundance found here and in many other studies reviewed by Wang et al. (2022) is likely attributable to the intense and often widely differing human management of green roof ecosystems which can overwhelm the differences in diversity and abundance caused by roof height, area and substrate additions (Droz et al., 2021). For example, while we might expect more diversity of plant and invertebrate species at lower elevations due to easier dispersal (MacIvor, 2016) or on larger roofs from the species-area effect (Ksiazek-Mikenas et al., 2018), these patterns can be greatly distorted if lower or larger roofs undergo management practices such as weeding and herbicides aimed at reducing diversity (Droz et al., 2021).
Green roof management: effects on plant and invertebrate communities
We now discuss the influence of components 1 and 5, which cover nearly all the management practices (except for chemical treatments and substrate addition noted above). Component 1 includes: management presence (yes/no), irrigation (yes/no), and replanting (yes/no). Component 5 includes weeding and substrate depth. (While the latter is not directly a management practice, it was statistically grouped into this component by the analysis and can be affected by management).
Management effects on plant diversity and frequency
Regarding component 1, we find that management presence has no significant effect on plant diversity or plant frequency. This lack of significant impact by management presence is likely because our study included green roofs with a diverse array of management strategies that affected plant diversity in different ways. For example, some management strategies included weeding and chemical treatments, which would reduce plant diversity, whereas others included reseeding and replanting, which could increase diversity. Thus, replanting has a near-significant (0.0124) positive effect on plant diversity when measured as a covariate with irrigation.
In addition to these divergent effects of different management strategies on plant diversity and frequency, another factor is that no management at all can have divergent effects. Several studies show that green roof management intensity often declines over time, and, in many cases, roofs become entirely neglected and unmanaged (reviewed by Schrieke et al., 2021). In some cases, this loss of management can lead to a loss of plant diversity, e.g., death of drought-intolerant species when irrigation is no longer carried out (Vandegrift et al., 2019). However, in many cases such neglected roofs gain species as they become colonized by a diverse community of stress-tolerant species that are more diverse than their original planted communities (Schrieke et al., 2021).
Interestingly, irrigation has a significant negative effect on plant diversity as a covariate with management occurrence and replanting, and at the 0.167 level alone. This aligns with the above finding of a negative relationship between moisture and diversity. Such a negative influence of irrigation and moisture is surprising given previous studies showing the importance of irrigation and moisture, especially in promoting plant diversity in the initial phase of green roof establishment (Saraeian et al., 2022; Vandegrift et al., 2019). This apparent discrepancy with our findings may occur because our study includes older roofs with a long history of establishment that often have little or no irrigation for several years especially in the later stages. Under these conditions, green roof communities often undergo successional replacement of planted and ruderal species with stress-tolerant colonizers which can tolerate hotter and drier non-irrigated conditions (Madre et al., 2013; Catalano et al., 2016; Ksiazek-Mikenas et al., 2018). As these stress-tolerant communities on non-irrigated green roofs are often highly diverse, especially compared to traditional sedum-based planted communities (Heim et al., 2024; Schrieke et al., 2021; Zhang et al., 2021), this could produce the negative correlation between irrigation/moisture and diversity found here. Irrigation and replanting have no significant effect on plant frequency.
Regarding component 5 (weeding and substrate depth), we find that neither weeding nor substrate depth has any significant effect on plant diversity or plant frequency. The lack of a consistent effect of mechanical weeding may result from differing weeding regimes that are not included in our data, which simply asks if weeding is done. For example, weeding that selectively targets invasive species can increase plant diversity by reducing competition and allowing for other species to become established (Iddris et al., 2023). However, in the case of a sedum monoculture, managers may selectively remove any spontaneous vegetation that is not part of the sedum community and thereby reduce overall plant diversity.
In the case of substrate depth, the lack of an effect is found in quite a few previous green roof studies. In their review, Wang et al. (2022) report that 7 of 17 studies that focus on substrate depth found no relationship between plant diversity and substrate depth while 10 of 17 of those studies found a positive relationship. That same review also found that 6 of 10 studies found a positive relationship between substrate depth and plant abundance/cover, with the other 4 studies finding no relationship (3) or a negative relationship (1) (Wang et al., 2022).
Where a positive relationship has been found between depth and plant diversity and abundance variables (cover, frequency), it has been attributed to any of several benefits of a deeper substrate: increased water holding capacity (Van Mechelen et al., 2015), as a buffer for plants at extreme temperatures (Vandegrift et al., 2019), better support of root stratification (Jauni et al., 2020) and access to more nutrients in deeper soils (Van der Kolk et al., 2020). In addition, deeper substrates can allow for different combinations of substrate depths and soil properties that provide a greater variety of ecological niches (Madre et al., 2013).
Given these suggested benefits of deeper substrates and that over half of reviewed studies focusing on this (Wang et al., 2022) do find that depth increases plant diversity and abundance variables, the question arises why a significant number of studies, including this study, do not find a significant positive influence of depth. We suggest this occurs because of the diverse methods historically used in green roof studies, as well as the different types of roofs and management strategies employed by these different roofs. In particular, plant diversity and abundance patterns can be highly influenced by intensive management practices in ways that can obscure or overwhelm the effects of substrate depth alone on plant diversity and abundance (Droz et al., 2021).
Management effects on invertebrate diversity and abundance
For component 1 (management presence, irrigation, and replanting), we find, as with plants, that management presence has no significant effect on diversity. Similarly, replanting has no significant effect. However, irrigation is significant as a covariate and a near-significant variable (0.112) alone with a negative impact on diversity, as is the case with irrigation impacts on plant diversity. As invertebrate diversity often tracks plant diversity (Ksiazek-Mikenas et al., 2018; Salman and Blaustein, 2018), this finding is not surprising.
For invertebrate abundance, the only significant relationship in component 1 (management presence, irrigation, and replanting) is an interactive one between replanting which decreases abundance and management which increases abundance. Further study is needed to see if this is a robust finding as there is no prior study reporting effects of these variables for invertebrate abundance on green roofs. However, if it is valid, one explanation could be that replanting reduces abundance by removing source populations of invertebrates and requiring recolonization. Explaining the positive effect of management is more difficult since, as noted above, there are several types of activities that are merged under this general term. However, perhaps surprisingly, many specific management activities discussed above (irrigation, chemical treatments, substrate additions, replanting) do not show a significant effect on invertebrate abundance. We suggest this is at least partly a result of the diversity of ages and features of the roofs being combined in this study. But one management activity, weeding, does seem to have an impact on abundance, as discussed below.
Regarding component 5 (weeding and substrate depth), we find that neither weeding nor substrate depth has any significant effect on invertebrate diversity. This result for substrate depth generally aligns with previous studies on depth and invertebrate diversity: 3 out of 4 of these studies found no relationship between substrate depth and arthropod diversity, with the remaining study showing a positive relationship (Wang et al., 2022). In contrast to diversity, we find that invertebrate abundance is significantly negatively impacted by both weeding and substrate depth. Neither of these is easy to explain. We know of no previous studies relating invertebrate abundance to mechanical weeding but perhaps this is caused by loss of invertebrate habitat from removing weeds. The decrease in abundance with substrate depth, both alone as an interactive covariate with weeding, is even more puzzling as 2 of 3 previous studies on this topic report that abundance increases with depth, with the third study showing no relationship (Wang et al., 2022).
Conclusions
Our study attempted to identify a suite of independent variables that influence plant and invertebrate diversity and abundance on green roofs in the Southeastern United States. Our multivariate analysis grouped these independent roof variables into three general groups, age, physical characteristics, and management, which are largely independent of each other. Perhaps our most basic conclusion is that comparative studies, which are the most common green roof studies attempting to identify the role of these factors on biodiversity, encounter the fundamental problem of conflating these three groups of variables. By comparing roofs of different ages, different physical features and different (and often changing) management regimes, it becomes difficult to identify (tease out) clear causal relationships. However, despite these confounding effects, we were able to identify some statistically significant patterns that often support previous studies.
Roof age
Our findings that roof age does not increase plant or invertebrate diversity or abundance agrees with the large majority of most previous studies. That this counterintuitive result may be related to the confounding effect of comparing different management regimes on different roofs is suggested by our finding that age does have a positive effect on both plant and invertebrate diversity when considered as a covariate with chemical treatment. This aligns with the idea that different management regimes are confounding the role of age so that age only becomes significant when some of the management effects (e.g., chemical treatments) are accounted for.
Roof physical features
Most previous studies indicate that roof elevation and area have no significant discernable effect on plant and invertebrate diversity or abundance. While we find that elevation or area alone have no clear effect, we did find that roof elevation seems to reduce plant diversity when the effects of roof area are accounted for. We suspect this is because the more elevated roofs in our study, which were significantly larger in area and are more intensely managed sedum-based monocultures with relatively low diversity. Thus, it again appears that when the confounding effects of comparing roofs with different management regimes are removed, we may have a fuller understanding of individual independent variables. This would explain why so many previous comparative studies have counterintuitive findings that lower elevation (easier dispersal) and larger surface area (more habitat) do not promote biodiversity (MacIvor, 2016; Ksiazek-Mikenas et al, 2018).
Although many previous studies also find (as we did) that substrate (soil) depth does not visibly increase plant or invertebrate diversity, many other studies do find that depth has a positive influence on diversity. Given the documented benefits of a deeper substrate on: water holding capacity, temperature buffering, root support and nutrient access, it may be that such negative findings reflect the confounding effects of many other variables (e.g., differing roof types, age, plant communities, and especially management strategies) used in comparative studies of several roofs.
Substrate (soil) temperature is obviously an extremely critical variable for green roof biodiversity. Our finding of a strong significantly positive relationship between soil temperature and plant and invertebrate diversity contradicts some of the current literature on green roofs which indicates that shading on green roofs promotes plant diversity (and thus insect diversity) by reducing substrate temperature. We suspect our results reflect the discontinuous nature of the temperature data which rely on the substrate temperature at one single point in time taken several weeks apart.
Substrate (soil) moisture is also an extremely important variable for green roof habitats. Our finding that soil moisture has a very significant negative effect on plant diversity is not only counterintuitive, but it contradicts a sizable literature on green roofs that documents the importance of soil moisture. We expand on this apparent contradiction below in our discussion of management practices because we also found irrigation to have a negative effect on plant diversity.
Management
Our finding that management presence has no significant effect on plant diversity and frequency and invertebrate diversity seems counterintuitive considering the strong impacts that management activities such as weeding, fertilizing and irrigation can obviously have on these plants and invertebrates. However, this result is almost certainly attributable to the fact that our study included green roofs with many distinct management strategies that affected diversity in different ways. Thus, some management activities (e.g., weeding and herbicides) act to reduce diversity whereas other activities (e.g., fertilizing) may often act to increase diversity. In addition, green roof management often declines over time and this loss of management can have divergent effects. Depending on variables such as local climatic conditions (e.g., rainfall and temperature) and substrate depth, some roofs may gain species and other roofs lose species over time as management is reduced.
For these reasons (and no doubt others), it is therefore more informative to focus on our findings of the effects of specific types of management activities rather than the mere presence of management. Thus, we find that chemical treatments alone had a strong negative effect on plant diversity, as would be expected from the use of selective herbicides. Similarly, our finding that replanting often increases plant diversity aligns with expectations.
Less intuitively, we also found that chemical treatments did not significantly affect plant frequency implying an “ecological release” effect whereby unaffected species replace the affected species as vegetative cover. Also less intuitive are our findings that weeding and substrate depth have no significant effect on plant and invertebrate diversity and plant frequency. Again, we attribute some of this to the confounding effects of other management practices occurring on the different roofs being compared at different times. This conclusion is reinforced by the fact that several other previous comparative studies also found no relationships between these management variables and plant or invertebrate diversity.
Finally, our most anomalous finding is that irrigation has a negative impact on plant and invertebrate diversity which contradicts much of the green roof literature. As our study includes several roofs with a long history of declining management (including irrigation) this may reflect their successional replacement with high-diversity communities of stress-tolerant colonizers which can tolerate hotter and drier non-irrigated environments.
In conclusion, our study adds to the growing literature examining the effects of time (age), physical features (area, elevation, soil depth), and management practices on plant and invertebrate diversity. However, the vast majority of these studies, including our own, are based on a comparative approach which analyzes different roofs that typically vary in their age, physical features and management regimes. This blending of several confounding variables makes it more difficult to tease out specific causal relationships. Thus, as noted several times in our discussion and in the review by Wang et al. (2022), previous studies often report different results when comparing the effects of these variables on plant and invertebrate diversity. Variation among other published results likely reflects differences in climate, management, and experimental design across studies, though evaluating these factors in detail is beyond the scope of this work. We therefore strongly agree with Droz et al. (2021) who noted that future green roof studies attempting to identify these causal relationships will need to utilize long-term longitudinal studies of the same roofs over the course of many years. This will provide data on the specific kinds of management practices being implemented and describing the specific outcomes of those practices. Overall, these findings highlight the role of green roofs as nature-based solutions in urban environments demonstrating how vegetation and invertebrate communities contribute to ecosystem services and resilience within managed cities.
Data availability statement
The original contributions presented in the study are included in the article/Supplementary Material. Further inquiries can be directed to the corresponding author.
Ethics statement
The manuscript presents research on animals that do not require ethical approval for their study.
My green roof field and lab team: Maddy Anyan, Orchid Gingham, Kyra Harrington, Zachary Kolluri, Jessica Lopez, Carolyn Rezler, Laila Stempkowski, Gregory Swanson. University of Tennessee, Knoxville staff members: Daniel Hembree, Ernest Bernard, Jennifer DeBruyn, Jennifer Franklin, Michael McKinney, and Mike Ross. External collaborators: Atlanta Botanical Gardens, Blueridge Parkway Visitor Center, City of Knoxville, Freeman Webb Company, Hamilton County Health Department, Lucy S. Herring Elementary School, Music City Center, Ronald McDonald House Charities, Shelby Bottoms Nature Center, Southface Energy Institute, and The University of Tennessee. This research was made possible by the collaboration of these buildings but also the staff and security that allowed me the access to the green roofs, including Angela Collins and Michael Del Valle, Bob Freeman and Robby Hutcherson, Chris Gallop and Jennifer Twachtman, Chris Ulrey, Drew Bryant, Joel Rummage and others, John Michael Cassidy, Jordan Diamond, Sonia Calvin, Stephen Ward, and finally Terry McConnell and others.
Conflict of interest
The author(s) declared that this work was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
Generative AI statement
The author(s) declared that generative AI was not used in the creation of this manuscript.
Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.
Publisher’s note
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.
Supplementary material
The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fevo.2026.1748660/full#supplementary-material
Location acronyms and coordinates.
Survey questionnaire distributed to each of the study roofs.
ReferencesBagellaS.CariaM. C.SeddaiuG.LeitesL.RoggeroP. P. (2020).
Patchy landscapes support more plant diversity and ecosystem services than wood grasslands in Mediterranean silvopastoral agroforestry systems. Agric. Syst.185, 102945. doi: 10.1016/j.agsy.2020.102945BraakerS.ObristM. K.GhazoulJ.MorettiM. (2017).
Habitat connectivity and local conditions shape taxo- nomic and functional diversity of arthropods on green roofs. J. Anim. Ecol.86, 521–531. doi: 10.1111/1365-2656.12648, PMID: 28164299BuchananS. W.BaskervilleM.OelbermannM.GordonA.ThevathasanN. V.IsaacM. E. (2020).
Plant diversity and agroecosystem function in riparian agroforests: providing ecosystem services and land-use transition. Sustainability12, 568–568. doi: 10.3390/su12020568BuchholzS.JessA.-M.HertensteinF.SchirmelJ. (2010).
Effect of the colour of pitfall traps on their capture efficiency of carabid beetles (Coleoptera: Carabidae), spiders (Araneae) and other arthropods. Eur. J. Entomology107, 277–280. doi: 10.14411/eje.2010.036CatalanoC.MarcenòC.LaudicinaV. A.GuarinoR. (2016).
Thirty years unmanaged green roofs: Ecological research and design implications. Landscape Urban Plann.149, 11–19. doi: 10.1016/j.landurbplan.2016.01.003DrozA. G.CoffmanR. R.BlackwoodC. B. (2021).
Plant diversity on green roofs in the wild: testing practitioner and ecological predictions in three midwestern (USA) cities. Urban Urban Green60, 127079. doi: 10.1016/j.ufug.2021.127079FabiánD.GonzálezE.Sánchez DomínguezM. V.SalvoA.FenoglioM. S. (2021).
Towards the design of biodiverse green roofs in Argentina: Assessing key elements for different functional groups of arthropods. Urban Forestry Urban Greening61, 127107. doi: 10.1016/j.ufug.2021.127107GillR. J.BaldockK. C. R.BrownM. J. F.CresswellJ. E.DicksL. V.FountainM. T.. (2016).
Protecting an ecosystem service. Ecosystem Services: From Biodiversity to Society Part 254, 135–206. doi: 10.1016/bs.aecr.2015.10.007HeimA.BiermannB.HicksT.BuffamI.LundholmJ. (2024).
More than sedum: Colonizing weedy species can provide equivalent green roof ecosystem services. Nature-Based Solutions5, 100101. doi: 10.1016/j.nbsj.2023.100101HigginsK. F.JenkinsK. J.ClambleyG. K.UreskD. W.NaugleD. E. (1996).
Vegetation sampling and measurement. Res. Manage. Techniques Wildlife Habitats5, 567–591.
IddrisN. A. A.FormaglioG.PaulC.von GroßV.ChenG.Angulo-RubianoA.. (2023).
Mechanical weeding enhances ecosystem multifunctionality and profit in industrial oil palm. Nat. Sustainability6, 683–695. doi: 10.1038/s41893-023-01076-xJacobsJ.BeenaertsN.ArtoisT. (2023).
Green roofs and pollinators, useful green spots for some wild bee species (Hymenoptera: Anthophila), but not so much for hoverflies (Diptera: Syrphidae). Sci. Rep., 13. doi: 10.1038/s41598-023-28698-7, PMID: 36702922JauniM.KuoppamakiK.HagnerM.PrassM.SuonioM.FranssonA. M.. (2020).
Alkaline habitat for vegetated roofs? Ecosystem dynamics in a vegetated roof with crushed concrete-based substrate. Ecol. Eng.157, 105970. doi: 10.1016/j.ecoleng.2020.105970JoimelS.GrardB.AuclercA.HeddeM.Le DoaréN.SalmonS.. (2018).
Are Collembola “flying” onto green roofs? Ecol. Eng.111, 117–124. doi: 10.1016/j.ecoleng.2017.12.002KnappM.BaranovskáE.JakubecP. (2016).
Effects of bait presence and type of preservative fluid on ground and carrion beetle samples collected by pitfall trapping. Environ. Entomology45, 1022–1028. doi: 10.1093/ee/nvw047, PMID: 27260789KnappM.RuzickaJ. (2012).
The effect of pitfall trap construction and preservative on catch size, species richness and species composition of ground beetles (Coleoptera: Carabidae). Eur. J. Entomology109, 419–426. doi: 10.14411/eje.2012.054Ksiazek-MikenasK.HerrmannJ.MenkeS. B.KöhlerM. (2018).
If you build it, will they come? Plant and arthropod diversity on urban green roofs over time. Urban Nat.1, 52–72.
LangemeyerJ.WedgwoodD.McPhearsonT.BaróF.MadsenA. L.BartonD. N. (2020).
Creating urban green infrastructure where it is needed – A spatial ecosystem service-based decision analysis of green roofs in Barcelona. Sci. Total Environ.707, 135487. doi: 10.1016/j.scitotenv.2019.135487, PMID: 31759703LehmannI.MatheyJ.RößlerS.BräuerA.GoldbergV. (2014).
Urban vegetation structure types as a methodological approach for identifying ecosystem services – Application to the analysis of micro-climatic effects. Ecol. Indic.42, 58–72. doi: 10.1016/j.ecolind.2014.02.036LönnqvistJ.BleckenG.-T.ViklanderM. (2021).
Vegetation cover and plant diversity on cold climate green roofs. J. Urban Ecol.7. doi: 10.1093/jue/juaa035MacIvorJ. S. (2016).
Building height matters: nesting activity of bees and wasps on vegetated roofs. Isr. J. Ecol. Evol.62, 88–96. doi: 10.1080/15659801.2015.1052635MacIvorJ. S.KsiazekK. (2015).
Invertebrates on green roofs. Ecol. Stud.333–355. doi: 10.1007/978-3-319-14983-7_14MacIvorJ. S.LundholmJ. (2010).
Insect species composition and diversity on intensive green roofs and adjacent level-ground habitats. Urban Ecosyst.14, 225–241. doi: 10.1007/s11252-010-0149-0MadreF.VergnesA.MachonN.ClergeauP. (2013).
A comparison of 3 types of green roof as habitats for arthropods. Ecol. Eng.57, 109–117. doi: 10.1016/j.ecoleng.2013.04.029MartinA. E.MurphyF. H. (2000). “
Glycols, propylene glycols,” in Kirk-othmer encyclopedia of chemical technology. Hoboken, NJ:
John Wiley & Sons, Inc. doi: 10.1002/0471238961.1618151613011820.a01McKinneyM. L.GladstoneN. S.LentzJ. G.JacksonF. A. (2019).
Land snail dispersal, abundance, and diversity on green roofs. PloS One14, e0221135. doi: 10.1371/journal.pone.0221135, PMID: 31725718National Oceanic and Atmospheric Administration. Available online at: https://www.noaa.gov/ (Accessed November 15, 2025).
NguyenC. N.MuttilN.TariqM. A. U. R.NgA. W. M. (2021).
Quantifying the benefits and ecosystem services provided by green roofs—A review. Water14, 68. doi: 10.3390/w14010068NoriegaJ. A.HortalJ.AzcárateF. M.BergM. P.BonadaN.BrionesM. J. I.. (2018).
Research trends in ecosystem services provided by insects. Basic Appl. Ecol.26, 8–23. doi: 10.1016/j.baae.2017.09.006OberndorferE.LundholmJ.BassB.CoffmanR. R.DoshiH.DunnettN.. (2007).
Green roofs as urban ecosystems: ecological structures, functions, and services. BioScience57, 823–833. doi: 10.1641/b571005PhoomiratR.Nipada Ruankaew DisyatatN. R.ParkT. Y.LeeD. K.DumrongrojwatthanaP. (2020).
Rapid assessment checklist for green roof ecosystem services in Bangkok, Thailand. Ecol. Processes9. doi: 10.1186/s13717-020-00222-zPratherC. M.PeliniS. L.LawsA.RivestE.WoltzM.BlochC. P.. (2012).
Invertebrates, ecosystem services and climate change. Biol. Rev.88, 327–348. doi: 10.1111/brv.12002, PMID: 23217156QuijasS.SchmidB.BalvaneraP. (2010).
Plant diversity enhances provision of ecosystem services: A new synthesis. Basic Appl. Ecol.11, 582–593. doi: 10.1016/j.baae.2010.06.009RobinsonS. L.LundholmJ. T. (2012).
Ecosystem services provided by urban spontaneous vegetation. Urban Ecosyst.15, 545–557. doi: 10.1007/s11252-012-0225-8RoelandS.MorettiM.AmorimJ. H.BranquinhoC.FaresS.MorelliF.. (2019).
Towards an integrative approach to evaluate the environmental ecosystem services provided by urban forest. J. Forestry Res.30, 1981–1996. doi: 10.1007/s11676-019-00916-xSalmanI. N. A.BlausteinL. (2018).
Vegetation cover drives arthropod communities in mediterranean/sub- tropical green roof habitats. Sustainability10, 4209. doi: 10.3390/su10114209SaraeianZ.FarrellC.WilliamsN. S. (2022).
Green roofs sown with an annual plant mix attain high cover and functional diversity regardless of irrigation frequency. Urban Forestry Urban Greening73. doi: 10.1016/j.ufug.2022.127594SchriekeD.LönnqvistJ.BleckenG.-T.WilliamsN. S.FarrellC. (2021).
Socio-ecological dimensions of spontaneous plants on green roofs. Front. Sustain. Cities3. doi: 10.3389/frsc.2021.777128SchwarzN.MorettiM.BugalhoM. N.DaviesZ. G.HaaseD.HackJ.. (2017).
Understanding biodiversity-ecosystem service relationships in urban areas: A comprehensive literature review. Ecosystem Serv.27, 161–171. doi: 10.1016/j.ecoser.2017.08.014ShafiqueM.KimR.RafiqM. (2018).
Green roof benefits, opportunities and challenges – A review. Renewable Sustain. Energy Rev.90, 757–773. doi: 10.1016/j.rser.2018.04.006SherkJ. T.FuW.NealJ. C. (2020).
Site conditions, maintenance costs, and plant performance of ten extensive green roofs in the research triangle area of central North Carolina. HortTechnology.761–769. doi: 10.21273/HORTTECH04565-20SnodgrassE. C.McintyreL. (2010). The green roof manual: a professional guide to design, installation, and maintenance (Portland, Oregon and London:
Timber Press).
ToniettoR.FantJ.AscherJ.EllisK.LarkinD. (2011).
A comparison of bee communities of Chicago green roofs, parks and prairies. Landsc Urban Plan103, 102–108. doi: 10.1016/j.landurbplan.2011.07.004UnderwoodS. M.SookhanN.HungK. J.MacIvorJ. S. (2025).
Pollinator visits increase with bloom amount but decline with building height on extensive green roofs. Insect Conservation and Diversity18(3), 438–445. doi: 10.1111/icad.12806VandegriftD. A.RoweD. B.CreggB. M.LiangD. (2019).
Effect of substrate depth on plant community development on a Michigan green roof. Ecol. Eng.138, 264–273. doi: 10.1016/j.ecoleng.2019.07.032Van der KolkH.-J.van den BergP.KorthalsG.BezemerT. M. (2020).
Shading enhances plant species richness and diversity on an extensive green roof. Urban Ecosyst.23. doi: 10.1007/s11252-020-00980-wVan MechelenC.DutoitT.HermyM. (2015).
Vegetation development on different extensive green roof types in a Mediterranean and temperate maritime climate. Ecol. Eng.82, 571–582. doi: 10.1016/j.ecoleng.2015.05.011WalstonL. J.LiY.HartmannH. M.MacknickJ.HansonA.NootenboomC.. (2021).
Modeling the ecosystem services of native vegetation management practices at solar energy facilities in the Midwestern United States. Ecosystem Serv.47, 101227. doi: 10.1016/j.ecoser.2020.101227WangH.-F.QureshiS.KnappS.FriedmanC. R.HubacekK. (2015).
A basic assessment of residential plant diversity and its ecosystem services and disservices in Beijing, China. Appl. Geogr.64, 121–131. doi: 10.1016/j.apgeog.2015.08.006WangL.WangH.WangY.CheY.GeZ.MaoL. (2022).
The relationship between green roofs and urban biodiversity: a systematic review. Biodiversity Conserv.31, 1771–1796. doi: 10.1007/s10531-022-02436-3ZhangH.FanX.RenL.JiangY.WuJ.ZhaoH. (2021).
Crassulacean plant succession over eight years on an unirrigated green roof in Beijing, Urban. For. Urban. Green.63, 127189. doi: 10.1016/j.ufug.2021.127189ZhaoJ.OuyangZ.ZhengH.ZhouW.WangX.XuW.. (2009).
Plant species composition in green spaces within the built-up areas of Beijing, China. Plant Ecol.209, 189–204. doi: 10.1007/s11258-009-9675-3
Edited by: Melissa Vogt, Independent researcher, Sydney, NSW, Australia
Reviewed by: Leigh Whittinghill, Connecticut Agricultural Experiment Station, United States
Chelsea Butcher, Northwood University, United States