Green Infrastructure Implementation in Urban Parks for Stormwater Management

Prior to the monitoring period, the study rain garden was constructed in Shoelace Park, adjacent to the intersection of East 228th Street and Bronx Boulevard [Fig. 1(a)] in Bronx, NY. Stormwater originating on 228th Street is intercepted by a stormwater inlet installed as part of the rain garden project in the streetscape just east of an existing catch basin [Combined Sewer Inlet A, as shown in Fig. 1(a)] located on the southeast corner of the intersection of 228th Street with Bronx Boulevard. A 26.8-m-long, 30.5-cm-diameter ductile iron inflow pipe directs water from the inlet first into a shallow manhole and then into a surcharge pit located on the opposite side of Bronx Boulevard in the rain garden. The inflow is distributed across the flat rain garden surface at a depth of 0.5 m vis-à-vis a 10.2-cm-diameter perforated polyvinyl chloride (PVC) distribution pipe. An overflow (domed) riser positioned inside the rain garden conveys water ponded more than 8.9 cm over the surface, with the top of the riser still below grade, to Combined Sewer Inlet B located 3 m to the west [as shown in Fig. 1(a)]. It does so by means of a 30.5-cm-diameter PVC pipe with 26 perforations in its cap.

The oblong rain garden is $37{\mathrm{m}}^2$ in area and was constructed by removing the in-situ soils and replacing them with 45.7 cm of engineered soil on top of 30.5 cm of crushed stone. The facility was designed to receive runoff from an approximately $600{\mathrm{m}}^2$ tributary area, including the streets and sidewalk of the south side of 228th Street between Bronx Boulevard and Carpenter Avenue [Catchment C1 as presented in Table 1 and depicted in Fig. 1(b)], as well as from a sloped lawn inside the park (Catchment C2). This configuration would have resulted in a hydraulic loading ratio (HLR) of $16:1$ (i.e., total tributary area divided by rain garden area). However, postconstruction field observations during an 88.9-mm storm event on April 20, 2015, revealed that the actual tributary area of the rain garden is much larger, or approximately $\mathrm{4,850}{\mathrm{m}}^2$. The tributary area includes an extended portion of the sloped lawn (Catchment C3), a section of Bronx Boulevard extending from 226th to 229th Street (Catchment C4), and a segment of the pedestrian trail inside the park (Catchment C5). These additional tributary areas are shown in Table 1 and Fig. 1(b). The aforementioned site visit also revealed several roof drains contributing to the C1 runoff entering the stormwater inlet. However, these tributary contributions were not estimated because it was not possible to gain roof access to those areas. The actual, as-built, hydraulic loading ratio was thus assumed to be at least $131:1$.

The monitoring system included a Thel-Mar weir (Brevard, North Carolina), two Global Water pressure transducers (College Station, Texas), and a rain gauge. The Thel-Mar weir was fitted into the upstream end of the 30.5-cm pipe leading to the rain garden from the shallow manhole. Pressure transducer A (PT-A) was placed within the shallow manhole and provided measurements at 5-min time steps (reporting time average over the previous 5 min) of the water depth upgradient of the weir. Depths exceeding 34.1 cm (weir invert depth $D_I$) over the manhole bottom exceeded the weir invert, allowing flow to the rain garden. Due to the imperviousness of the tributary area, storms with rapidly varying intensities can translate to large water level variations during the time step. These variations are smoothed by the time-averaging measurement technique. Weir rating curves made available by Thel-Mar, LLC, were used to convert pressure transducer readings into volumetric flow rates using polynomial curve fitting when PT-A readings exceeded $D_I$ (Fig. S1 in Supplemental Data). The effects of inflow turbulence and flow through a potential manhole crack on the PT-A measurements are assessed in the supplemental materials. Some lower-intensity storms within the monitoring period resulted in minimal runoff from 228th Street and small increases in PT-A measurements. In these cases, the water level often remained below $D_I$, generating no flow over the weir.

Pressure transducer B (PT-B) was placed within a 10.2-cm, perforated PVC pipe inside the rain garden to measure the ponding depth at 5-min intervals. PT-A and PT-B readings were calibrated with manual measurements (Table S1 in Supplemental Data) as well as validated with manual in-situ measurements throughout the monitoring period (Tables S2 and S3 in Supplemental Data). Standard hydraulic equations were used to compute rain garden effluent flow rates to Combined Sewer Inlet B for all ponding depths exceeding 8.9 cm above the flat rain garden surface. A Global Water GL500 data logger was mounted on the western curb of Bronx Boulevard to log the pressure transducer data. A tipping bucket rain gauge was positioned nearby at the intersection of 224th Street and Bronx Boulevard.

The rain garden was monitored between October 2014 and July 2015, with gaps in monitoring due to equipment failure. Due to a prolonged period of freezing temperatures, data from PT-A and PT-B were not deemed reliable in February 2015. Additionally, the onsite rain gauge did not function properly between April 17 and June 25, and storms occurring during this time period were not included in the analysis. During the time that the rain gauge was functioning, the average rainfall was $3.15\mathrm{mm}/\mathrm{day}$, consistent with the long-term NYC average of $3.20\mathrm{mm}/\mathrm{day}$, suggesting relatively normal climatological conditions over the monitoring period (US Climate Data 2018). See the supplemental materials for further discussion of monitoring considerations.

The continuous rainfall record was discretized into individual events based on the inflow into the rain garden through the Thel-Mar weir. The beginning of each storm was defined as the onset of recorded precipitation, and the end of each storm was defined as the time corresponding to when the poststorm water level in the manhole reached the Thel-Mar weir invert elevation $D_I$. The storm depth was computed as the total precipitation recorded during this period. Discretization of storms using a 4-h interevent dry period yielded the same results. The same storm profile was used in the analysis of the PT-B readings, given that a short time-lag was observed between the two measurement locations. Rain garden outflow to Combined Sewer Inlet B included all flow through the riser that was recorded between the onset of recorded precipitation and the first time the ponded water level subsided to the elevation corresponding to the invert of the riser pipe.

Inflows to the rain garden included direct precipitation onto the rain garden surface ($I_P$), offsite street runoff generated on tributary area C1 ($I_{228}$), and onsite parkland runoff generated on C2–C5 tributary areas ($I_{\mathrm{Sh}}$). Outflows from the rain garden included infiltration, evapotranspiration, and flow to Combined Sewer Inlet B ($O_{\mathrm{RG}}$). These components of the rain garden water balance are shown in Fig. 2.

Considering only the storm periods, the volume of stormwater retained in the rain garden ($V_R$) can be computed by subtracting the rain garden outflow to Combined Sewer Inlet B from the sum of all inflows (Eq. 1)where $V_R$ = rain garden volume retained (${\mathrm{m}}^3$); $I_{228}=228$th Street Inflow (C1) (${\mathrm{m}}^3$); $I_{\mathrm{Sh}}$ = additional Shoelace Park inflow (C2 to C5) (${\mathrm{m}}^3$); $I_P$ = direct rainfall on the rain garden (${\mathrm{m}}^3$); $O_{\mathrm{RG}}$ = rain garden outflow (${\mathrm{m}}^3$).

Both infiltration and evapotranspiration during the storm periods contributed to the rain garden’s retention capability and are thus included in $V_R$. The ability of the rain garden to retain stormwater varied from storm to storm due to the antecedent conditions that determined the available storage capacity. The methodology used to compute each of the inflows and outflows is described subsequently.

Inflow from C1 ($I_{228}$) was estimated using the weir manufacturer’s equation and the PT-A measurements. Two different flow situations were considered in the calculations, as in Fig. 3. In Case A, inflow occurred through the triangular or rectangular section of the Thel-Mar weir only. In such cases, inflow could be directly estimated using the manufacturer’s weir experimental data. Case B occurred when the manhole water level rose above the upper edge of the Thel-Mar weir. This was calculated from the addition of orifice flow through the Thel-Mar weir opening and open channel flow through the section above the Thel-Mar weir. This situation only occurred during a short period in two separate storms and thus did not significantly influence results.

The Soil Conservation Service (SCS) curve number method (Viessman and Lewis 2003) was used to estimate additional observed inflow into the rain garden from tributary areas C2–C5 ($I_{\mathrm{Sh}}$). The method is summarized subsequentlyvalid forwhere $P$ = precipitation (m); $I_a$ = initial abstraction (m); CN = curve number; $S$ = watershed storage (m); $A$ = tributary area (${\mathrm{m}}^2$); $Q$ = direct runoff (${\mathrm{m}}^3$).

Table 2 tabulates the area, type of surface, and curve number corresponding to all contributing surfaces. 228th Street (C1), Bronx Boulevard (C4), and the Shoelace Park walkway (C5) all have a curve number of 98 (Viessman and Lewis 2003), reflecting their predominantly asphaltic surfaces. The sloped lawn and adjacent slope (C2 and C3), which are covered by more than 75% grass and are best represented by Soil Group B, have an estimated curve number of 61 (Viessman and Lewis 2003). Runoff values were also estimated for C1 to validate the Thel-Mar weir inflows. These assumptions are further discussed subsequently.

The direct precipitation on the rain garden area ($I_P$) was calculated for each storm as the product of the storm depth and rain garden area.

Domed riser flow ($O_{\mathrm{RG}}$) was calculated as sum of the flow rates through the individual holes of the perforated cap, as shown in Fig. 4. There were two separate rows of 1.27-cm-diameter holes on the cap. The lower and upper rows included 15 and 11 holes, respectively. PT-B measured the head as the distance from the observation well sump to the water table in the rain garden. The orifice flow equation was used to calculate the flow rate through these perforations as shown in Eq. 4. The flow through the holes is divided into a piecewise function based on whether flow is through the first row or both first and second rows of perforations according to $H$where $O_{\mathrm{RG}}$ = rain garden outflow (${\mathrm{m}}^3/\mathrm{s}$); $C$ = coefficient of discharge [0.61 for sharp-edged perforations (Street et al. 1996)]; $A_1$ = lower row perforation area ${\mathrm{m}}^2=1.9\times {10}^{-3}{\mathrm{m}}^2$; $A_2$ = upper row perforation area ${\mathrm{m}}^2=1.4\times {10}^{-3}{\mathrm{m}}^2$; $g=9.81\mathrm{m}/{\mathrm{s}}^2$; $H$ = head (m); and $\mathrm{\Delta }t$ = PT-B time step = 300 s.

Eqs. 5 and 6 were developed to roughly estimate the amount of rain garden infiltration over the duration of each storm and do not influence computation of $V_R$where $V_{\mathrm{IN}}$ = infiltration (${\mathrm{m}}^3$); $V_{\mathrm{Stored}}$ = water volume stored over duration of storm (${\mathrm{m}}^3$); ${\mathrm{\theta }}_{\mathrm{FC}}$ = soil field capacity; ${\theta }_{\mathrm{WP}}$ = soil wilting point; $V_{\mathrm{RG}}$ = rain garden volume (${\mathrm{m}}^3$).

A constant maximum water volume stored of $1.7{\mathrm{m}}^3$ during a storm ($V_{\mathrm{Stored}}$) is computed using Eq. 6 along with the assumption of field capacity (${\mathrm{\theta }}_{\mathrm{FC}}$) and wilting point (${\mathrm{\theta }}_{\mathrm{WP}}$) equal to 0.3 and 0.25, respectively, based on engineering soil type and rain garden volume ($V_{\mathrm{RG}}$) of $34{\mathrm{m}}^3$. Inflow in excess of this volume is assumed to infiltrate over the duration of the storm (Eq. 5). Because this calculation assumes the rain garden soil dries to the wilting point by the beginning of each storm, it is considered an upper-limit estimate of the static water storage capacity. This consequently yields a lower-limit estimate of infiltration during and after the storm. In the absence of nested piezometers or soil moisture sensors, this is considered the best possible estimate of infiltration volume per event.

Two different performance indicators were defined to assess the rain garden performance. The performance efficiency (PE) is intended as a measure of the ability of the 228th Street stormwater inlet to capture C1 runoff. It is defined as the ratio between the PT-A measured inflows from 228th Street and the CN estimated runoff produced over C1 ($V_{228\_\mathrm{CN}}$) as shown in Eq. 7. PT-A measured inflows include the combined calculated volume of 228th Street inflow and water volume accumulated in the stormwater inlet catch basin ($V_{\mathrm{Bas}}$). $V_{\mathrm{Bas}}$ is computed in Eq. 8where $V_{\mathrm{Bas}}$ = volume accumulated in stormwater inlet basin below $D_I$ (${\mathrm{m}}^3$); $V_{228\text{\_}\mathrm{CN}}$ = curve number estimated runoff generated on C1 (${\mathrm{m}}^3$); $A_{\mathrm{Bas}}$ = stormwater inlet basin area (1.49 ${\mathrm{m}}^2$); $H_{\mathrm{Peak}}$ = peak water depth below weir invert during storm (m); and $H_{\mathrm{Init}}$ = initial water depth before storm (m).

$V_{\mathrm{Bas}}$ is, in other words, the runoff volume captured from C1 by the 228th Street inlet that accumulates below the $D_I$ invert that does not ultimately flow through the Thel-Mar weir. $V_{228\text{\_}\mathrm{CN}}$ is estimated using Eq. 2. The peak water depth below the weir invert ($H_{\mathrm{Peak}}$) and initial water depth ($H_{\mathrm{Init}}$) are measured using PT-A. For storms that produce flow over the Thel-Mar weir, $H_{\mathrm{Peak}}$ equals $D_I$. Although $V_{\mathrm{Bas}}$ does not contribute to $V_R$, it significantly affected the PE values.

When the PE is 100%, the site inlet has successfully captured all C1 runoff. When the PE is below 100%, it is likely that there was bypass at the stormwater inlet leading to the rain garden. In the case where the PE is greater than 100%, the effective C1 catchment area was likely larger than assumed, for example, due to bypass of other combined sewer inlets located further upslope beyond Carpenter Avenue or lot-level runoff discharged to this segment of 228th Street.

The second indicator of the rain garden stormwater capture performance is the percent retained (PR) considering all inflows and outflows to the rain garden during each storm

First, relationships between hyetographs and inflow and outflow hydrographs were established. The relationship between performance indexes and precipitation event characteristics was evaluated using Spearman correlation coefficients with all stronger correlations ($|\rho \mathrm{|}>0.4$) showing significance ($p<0.05$). Finally, the rain garden performance was extrapolated for a simple estimate of stormwater capture possible from similar facilities installed in all NYC parklands.

The comparison between the hydrograph and hyetograph for storms on November 17, 2014 [Figs. 5(a and b)]; March 31, 2015 [Figs. 5(c and d)]; and June 27 and 28, 2015 [Figs. 5(e and f)], depicts the typical hydraulic response of the system to precipitation. There is a delay between the storm beginning and non-zero $I_{228}$ and $O_{\mathrm{RG}}$ incremental flow rates, which is a function of the storm intensity and preceding head within the stormwater inlet basin (measured with PT-A). In each case, the $O_{\mathrm{RG}}$ flow rate exceeds the $I_{228}$ incremental flow rate, showing a strong contribution of runoff from Shoelace Park tributaries (C2 to C5). However, 35% of storms resulted in nonzero $I_{228}$ but zero $O_{\mathrm{RG}}$ where the retention capacity of the rain garden is evident.

PE results and relationships are shown in Tables 4–7 and in Fig. 6. The overall average PE was 182%, and it tended to decrease with increasing precipitation and storm duration, as suggested by significant negative correlations (Table 6 and Fig. 6). There appeared to be a seasonal dependence on PE [Fig. 6(c) and Table 7], with higher values during spring and summer (377% and 425%) than fall and winter (44% and 50%). PE exceedance of 100% means that the site receives more runoff from C1 than estimated, which is consistent with observations. During the site visit on April 20, 2015, additional tributary areas, such as roof drains, not accounted for in the original catchment area were observed contributing to the stormwater inlet, thereby justifying typical PE near 200%. However, the results should be interpreted with caution because some significant deviations from 100% were observed (Table 4) with no clear explanation after evaluating four scenarios (discussed in supplemental materials).

PR results are shown in Tables 5, 6, and 8 and Fig. 7. The rain garden successfully retained an average of 78% of inflows for all storms and 100% of inflows for 58% of storms. Increased precipitation depths, outflow volumes, and storm durations tended to decrease PR (Fig. 7 and Table 8). Additionally, the same behavior as Davis et al. (2012) observed is noted where outflows are linearly related to inflows above an inflow threshold (not shown). PR values were generally higher in the spring and summer, which may be partially explained by shorter storm durations with no change in intensity (Table 6). Finally, five storms were removed from the analysis (reducing the number of storms from 31 to 26) due to obstruction of Combined Sewer Inlet B and consequent invalidation of rain garden outflow computations. Because these events resulted in erroneously negative PR, removing these events increased average PR (discussed in supplemental materials).

The rain garden retained nearly all inflows (average PR of 96%) in storm events less than approximately 10 mm [Fig. 7(a)]. Above this precipitation depth, the rain garden began losing water through the domed riser, likely from inflows exceeding the rate of rain garden infiltration with saturation of superficial layers. Storm events exceeding 10 mm are relatively frequent in NYC, making up 35% of events during the monitoring period and historically about 20% of events (Catalano de Sousa et al. 2016). Given that additional runoff within Shoelace Park (C3 to C5) increased the design tributary area by a factor of eight (contributing 70% of the inflow on average during the monitoring period), it is expected that this rain garden can fully retain runoff from more significant storms if runoff contributions are only from street-level contributions (C1). Therefore, despite managing inflows beyond its design capacity with a high hydraulic loading ratio under typical NYC climatology, the rain garden manages 78% of inflows, performing similarly to other rain gardens (Davis 2008; Davis et al. 2012; Dietz and Clausen 2005; Hunt et al. 2006; Kurtz 2008; Schlea et al. 2014; Tang et al. 2016).

Combined sewers serve 60% of NYC’s 72% impervious covered surfaces: an area approximately $340{\mathrm{km}}^2$ in size (NYC DEP 2018a, b). The NYC Department of Environmental Protection (DEP) goal is to capture the first 25 mm of runoff from 10% of this using GI (Bloomberg and Strickland 2012). Additionally, an estimated $92{\mathrm{km}}^2$ of parkland space is within the combined sewer area of NYC (NYC Parks 2018). The upper limit of full rain garden runoff retention in this study is a storm depth of approximately 10 mm, translating to runoff volume retention of about $8.6{\mathrm{m}}^3$ (discussed in supplemental materials). Linear scaling is assumed in which the $8.6{\mathrm{m}}^3$ retained runoff volume on the study tributary area is equivalent to a scenario in which $8.6{\mathrm{m}}^3$ of only street-level runoff is fully captured and retained by the rain garden in a 25-mm storm. Thus, the rain garden can fully retain inflows from a 25-mm storm on approximately nine times its area (hydraulic loading ratio of $9:1$) from only street-level runoff. Assuming all hypothetically implemented GI in NYC urban parks performs similarly to the Shoelace Park rain garden under similar conditions, only about 4% of NYC urban park space allocated for GI would be necessary to meet the NYC DEP runoff reduction goal, at least at the municipal level. Furthermore, 5%, 10%, and 15% of NYC urban park space allocated for GI would treat runoff from 12%, 25%, and 37% of impervious coverage served by combined sewers, respectively, significantly exceeding the NYC DEP goal. Retrofitting less than 15% of parkland space with GI, therefore, would result in large stormwater retention with minimal to no influence on the current use of the park. This analysis, however, assumes that the rain garden is only retaining inflow from street-level runoff. Given the unexpected inflows from within Shoelace Park, this study emphasizes the importance of accurate tributary area delineation during site planning and the need to make provisions to exclude pervious area runoff from entering parkland GI. Additional spatial analyses are needed to determine whether potential parks are positioned ideally for stormwater capture given the variable conveyance capacities of the city’s many combined sewersheds, and the desired pollutant load reductions for each of its receiving water bodies.