Green infrastructure’s capacity to mitigate urban environmental problems, like heat island effects and excessive stormwater runoff, is partially governed by its plant community. Traditionally, green infrastructure design has focused on engineered aspects, such as substrate and drainage, rather than on the properties of its living components. Since the functioning of these plant assemblages is controlled by ecophysiological processes which differ by species, the identity and relative abundance of the species used will influence green infrastructure performance. We used trait-based modeling to derive principles for the effective composition of green infrastructure plant assemblages, parameterizing our model using the vegetation and ecophysiological traits of the species within New York City rain gardens. Focusing on two plant traits that influence rain garden performance, leaf surface temperature and stomatal conductance, we simulated the cumulative temperature and transpiration for plant communities of differing species composition and diversity. The outcomes of the model demonstrate that plant species composition, species identity, selection effects, and interspecific complementarity increase green infrastructure performance much the way biodiversity affects ecosystem functioning in natural systems. More diverse assemblages resulted in more consistent transpiration and surface temperatures, with the former showing a positive, saturating curve as diversity increased. While the dominant factors governing individual species' leaf temperature were abiotic, transpiration was more influential at the community level, suggesting that plants within diverse communities may be cooler in aggregate than any individual species on its own. This implies green infrastructure should employ a variety of vegetation; particularly plants with different statures and physical attributes, such as low-growing ground covers, erect herbaceous perennials, and shrubs.

Ecophysiological measurements on New York City rain gardens

Ecophysiological measurements were collected on the vegetation from six New York City Department of Environmental Protection (NYC DEP) rain gardens in Rego Park, Queens, New York, NY. The plant ecophysiological traits of surface temperature and stomatal conductance were measured on a total of six species, using four individuals of each species. The six species consisted of: two shrubs (Aronia melanocarpa and Spiraea nipponica), two non-leguminous forbs (Eupatorium dubium and Nepeta faassenii), a C4 grass (Panicum virgatum) and a C3 grass (Calamagrostis x acutiflora). The first five of these species occurred in three rain gardens, while Calamagrostis was present as a monoculture in three others, as specified by NYC DEP planting design. These species encompassed all intentionally planted non-tree vegetation within their respective rain garden; volunteer species occupied less than 2% of total cover. 

Data were collected between 10am and 2pm over ten days from July to September 2016. For each species within each rain garden for each sampling day, two thermal images and a stomatal conductance measurement were taken on the second fully expanded leaf of four individual plants. The same plants, but different leaves, were used during the ten study days. The thermal images were taken prior to measuring stomatal conductance. For the first image, the leaf was undisturbed. In the second image, a plastic tray with a piece of white filter paper half wet (wet reference) and half dry (dry reference, Whatman 1443-125) was held behind and at the same angle as the leaf. A weather station (HOBO U30 Data Logger, Onset Computer Corporation) measuring wind speed, photosynthetically active radiation (PAR), barometric pressure, relative humidity, and temperature was mounted to a hand truck and positioned immediately adjacent to the rain garden where measurements were being taken to collect local climate information. The weather station recorded environmental variables every second and logged the mean of these data each minute and was wheeled to each rain garden over the course of the day. At the beginning of each study day, the thermal camera (FLIR T650sc, FLIR Systems, Inc.) was calibrated with temperature and humidity readings from the portable weather station. Emissivity was assumed to be 95% (Jones, 2004). Time was synchronized between the thermal camera, weather station, and porometer (SC-1 Leaf Porometer, Decagon Devices),

Thermograms were converted to comma-separated value (CSV) files (FLIR ExaminIR Pro software) and read into ImageJ (Schneider et al. 2012) where the leaf and, if applicable, dry and wet references were manually outlined to extract temperature information. Mean temperature of these areas was used for later analyses; all references to leaf and reference temperatures refer to these thermographically-determined means. Since the leaf included both areas in shade and those in sun, a mean of the minimum temperatures was used as a “shade” temperature for later analyses. A paired t-test on a 17-leaf subset of the data was used to compare the temperature of leaves with references behind them to those without. Time of stomatal conductance readings was rounded to the nearest minute and matched with the weather station data from the same time to calculate transpiration (E) according to Lambers, Chapin, and Pons (2008) as: (VPleaf - VPair)/barometric pressure*gw. gw is stomatal conductance, VPleaf is the saturation vapor pressure of the leaf, using the temperature from the porometer, and VPair is the vapor pressure of the air (saturation vapor pressure of the air, using temperature from the weather station, multiplied by relative humidity). Transpiration data were associated to corresponding thermal images taken on the same day for each individual leaf. 

Simulations

All simulations and statistical analyses were performed in R version 3.2.3 (2015-12-10). Simulated assemblages, parameterized from the NYC rain gardens, ranged from one to six species. These assemblages consisted of randomly drawn species from the pool of six, without replacement. Each species was assumed to occupy an equal portion of the rain garden; e.g., if an assemblage consisted of three species, each covered a third of the rain garden, if four species, each filled a quarter, etc. Total rain garden size was considered to be 9.3 m², the most common area for the NYC rain gardens. As the species differ in physical structure, rather than using vegetation cover, the volume of leaves in each portion was calculated as follows: height was binned into five levels and, at each of these levels, both total coverage (woody and leaf) and percent coverage transpiring/temperature regulating (leaves only) were determined based on personal observation. Whole rain garden transpiration and surface temperature were assumed to be the mean calculated from stomatal conductance and thermography measurements, respectively, both scaled up using the area of the rain garden occupied by a species and percent leaf coverage at each level. Whole rain garden coverage can exceed 100% since it was calculated as the sum of coverage at each of the five levels. Total transpiration was additive, whereas surface temperature was calculated as a weighted mean based on proportion of leaf cover at each level.

To allow for structural complementarity between species, if a species did not fully cover its rain garden portion at the ground level, the other species in the community were sequentially allowed to “fill in” its portion until either the ground level was fully occupied or there were no new species left. The pool and order of species were determined by the initial random draw. For example, a spreading shrub with a woody base might have 100% leaf coverage at its top, but only 10% stem coverage at ground level. Here, creeping ground vegetation could enter into the remaining 90% unoccupied by the stem, if this species occurred subsequently within the species draw. If coverage of this creeping vegetation was 100% at ground level, no further species could fill in; if not, another species from the community was allowed to occupy the remaining portion, until no new species were available. In the model for transpiration, structural complementarity added species-specific transpiration in the amount of the additional percent cover provided by the new species; in the model for temperature, a weighted average based on additional percent cover was used. To simulate competition for light, inter-specific shading was simulated by using a shade temperature, calculated from the mean shade temperature (by species) from the thermography measurements and the relative heights of the species within the communities. If any of the species filling in the unoccupied ground level was shorter than the original plant (e.g., the creeping ground vegetation in the above example), a shade temperature was used. If the filling-in species was taller, the model assigned a shade temperature to the original plant in proportion to the percentage occupied by the new species. For example, if a flowering herbaceous species occupied 90% of the ground and a taller spreading shrub filled in the remaining 10%, the entire rain garden temperature would be a weighted average of 10% of the herbaceous shade temperature, 80% if its regular temperature, and 10% of the shrub’s regular temperature. Since the order in which species fill in each other’s portions matters, each community of two through six species was iterated 100,000 times.

Weather station data and ecophysiological measurements for plant species within New York City rain gardens are provided as a .csv. These data were used to simulate communities of one to six species. These simulations are provided as two separate .rds files: one for transpiration and one for temperature.