Quantitative Analysis of the Water Budget of a Rain Garden in Pennsylvania

The reduction of pervious surfaces and plant cover because of development results in increased runoff and decreased infiltration. Stormwater control measures (SCMs) that promote infiltration have become a common approach in stormwater management because they are able to reduce runoff volumes and remove many pollutants through adsorption (Komlos and Traver 2012; Kabelkova et al. 2015; Mermillod-Blondin et al. 2019; Guo 2020; Kordana and Słyś 2020; Bastia et al. 2021; Hayes et al. 2021; Martin et al. 2021; Xu et al. 2022; Lebon et al. 2023). These SCMs also have the potential to return water to the groundwater table through recharge (Rowney et al. 2008). Some examples of infiltration SCMs include infiltration trenches, pervious pavements, infiltration basins, rain gardens, and swales. Infiltration practices mitigate excessive stormwater flows, improve water quality, and ameliorate groundwater depletion problems that coincide with land development. Although each of the aforementioned infiltration practices contributes to the sustainable management of stormwater, rain gardens are especially beneficial because they can reduce volume not only through infiltration, but also through evapotranspiration (ET).

A rain garden is a surface depression that collects runoff from the surrounding area, and is planted with vegetation that is appropriate to the climate and resilient in wet and dry conditions. A rain garden has its own hydrologic budget that varies depending on the specifics of the design. Design components include the volume of the bowl, soil characteristics, and inflow and overflow mechanisms (PaDEP 2006; National Research Council 2009; Malaviya et al. 2019; Sharma and Malaviya 2021). In relation to infiltration, the characteristics of the underlying soil are the most important aspect in facilitating volume removal (Carpenter and Hallam 2010; Lee et al. 2016a, b). When choosing a site for a SCM that relies on infiltration, the soil should be inspected to ensure that infiltration is a viable option. Even if a porous engineered mix is used within the garden, the interface with the in situ soil will influence the pond’s recession rate (Lee et al. 2013). If the capacity of the rain garden is exceeded, and overflow will pass through to the existing storm sewer systems. The location of the groundwater table is another critical component that can affect rain garden performance. If the groundwater table is too close to the surface, it could intersect the rain garden and hamper infiltration. Ideally, the rain garden should have adequate depth to allow for a vadose zone in which pollutants have a chance to be filtered out and plants have the opportunity for root uptake.

When water enters an unlined rain garden, it can infiltrate into the soil, evaporate into the air, or exit through an outflow pipe if one is present (Bethke et al. 2022). The water that infiltrates the soil media can either replenish the groundwater table (recharge) or transpire through vegetation. The surface components of inflow and outflow can be determined using basic hydrologic principles. Inflow can be determined by calculating runoff from the connected watershed using mathematical models such as SWMM (Rossman and Huber 2015) or HEC-HMS (Heasom et al. 2006), the Natural Resources Conservation Service (NRCS) curve number method (Cronshey 1986a), the kinematic wave method (Miller 1984), or through monitoring of an inflow channel. Singular points of entry from inflow channels are rare, so engineers often estimate inflows. Outflows can be monitored using weirs or orifices in conjunction with the outflow infrastructure. The amount of water that recharges the groundwater or evapotranspires is more difficult to quantify. Recent research has sought to determine the amount of groundwater mounding beneath infiltration SCMs (Machusick et al. 2011; Nemirovsky et al. 2015; Tu and Traver 2019) and the percentage of water controlled through evapotranspiration (Denich and Bradford 2010; Hess et al. 2017, 2019; Delvecchio et al. 2020; Hess et al. 2021, 2023).

The heavily instrumented and well documented rain garden described in this paper has been the subject of many studies (Heasom et al. 2006; Emerson and Traver 2008; Jenkins et al. 2010; Heasom and Traver 2012; Hickman et al. 2012; Komlos et al. 2012; Komlos and Traver 2012; Lee et al. 2012, 2013, 2016a; Flynn and Traver 2013; Lord et al. 2013; Nemirovsky et al. 2015; Zukowski et al. 2016; Barr et al. 2017; Wadzuk et al. 2017; Ebrahimian et al. 2020; Amur et al. 2022; Hess et al. 2023; Smith et al. 2023). Previous research conducted at this site on groundwater mounding found that the mounding was localized and dissipated quickly (Machusick et al. 2011; Nemirovsky et al. 2015). These researchers reported that more than 18 mm (0.71 in.) of rain was needed in order to see a response in the groundwater table. According to an analysis of rainfall events in the Philadelphia region conducted by the Philadelphia Water Department (PWD) (2009), approximately 70% of all rain events in this region are 18 mm or less. Hess et al. (2017) conducted a study at this location using weighing lysimeters with the same soil as that found in the rain garden that is the subject of the present study and reported that nearly half of the water budget could be attributed to ET.

The goal of this research was to use measured values for precipitation, overflow, and rise in groundwater level to estimate the amount of ET for an unlined rain garden located on Villanova University’s campus in Pennsylvania.

This study was conducted on an instrumented rain garden on the campus of Villanova University in southeastern Pennsylvania, which is located approximately 18 km west of Philadelphia (Fig. 1). The rain garden was retrofitted into a traffic island in 2001 and designed to capture and infiltrate the stormwater runoff from storms up to 25.4 mm. The rain garden has a total drainage area of 5,300 ${\mathrm{m}}^2$ (0.53 ha), 44% of which is impervious. The rain garden was constructed by excavating an existing traffic island to a depth of 1.8 m (6 ft). The rain garden soil media was made by mixing the excavated soil with concrete sand at a 1:1 ratio and filled to a depth of 1.2 m (4 ft) in the excavated area, leaving 0.6 m (2 ft) of depression depth. The native soil was classified as sandy silt (ML) according to the Unified Soil Classification System (USCS), and after mixing it was classified as a silty sand (SM). The soil profile below the rain garden includes the soil mantel [sandy silt (ML) and silty sand (SM)] which is underlain by mica schist bedrock (Fig. 2) (Nemirovsky et al. 2015). Vegetation native to coastal New Jersey was chosen for the bowl based on the ability of this type of vegetation to survive in dry and wet conditions and on their tolerance for salt (Nemirovsky et al. 2015). A slug test determined that the conductivity of the native soil in the aquifer in the direction of flow is $6.3\times {10}^{-5}\mathrm{cm}/\mathrm{s}$ ($2.5\times {10}^{-5}\mathrm{in.}/\mathrm{s}$). The gradient was calculated to be 0.0125. The groundwater moves from right to left on cross-section B-B′ (Figs. 1 and 2). Infiltration capacity has been analyzed frequently by observing the soil moisture and ponding during and following storms and performing on-site infiltration tests in the rain garden. The geometric mean hydraulic conductivity ($K_{\mathrm{sat}}$) value of the rain garden soils, as determined by single-ring infiltrometer testing, is approximately $2.9\times {10}^{-4}\mathrm{cm}/\mathrm{s}$ ($1.14\times {10}^{-4}\mathrm{in.}/\mathrm{s}$) (Press 2019). This value is similar to the average infiltration rate of $1.9\times {10}^{-4}\mathrm{cm}/\mathrm{s}$ ($7.5\times {10}^{-5}\mathrm{in.}/\mathrm{s}$) estimated from the recession rate data (Zukowski et al. 2016). For most storm events, the basin drained within 48 h (Emerson and Traver 2008; Lee et al. 2013; Zukowski et al. 2016).

This site was instrumented with sensors to monitor rainfall, ponding levels, soil moisture, groundwater levels, and overflow volumes (Table 1). Although the soil moisture sensors were not used directly in the calculations for this study, they were indirectly used to validate and confirm changes in the groundwater level. Data collected between May 2014 and December 2016 were used for this analysis. Eight groundwater monitoring wells (MWs) were installed around the site to monitor the response of the water table (Fig. 1). MW 1 and MW 5 were installed upgradient of the rain garden. MW 2, MW 4, MW 6, and MW 7 were installed adjacent to the rain garden, and MW 3 and MW 8 were installed downgradient. MW 4 was installed to a depth of 10.13 m, and MW 1, MW 2, and MW 3 were installed to depths of 11.20, 9.23, and 9.19 m, respectively. MW 5, MW 6, MW 7, and MW 8 were installed to a depth of 10.97 m.

Quantifying the hydrologic budget within the rain garden was accomplished by calculating each term in the mass balance. For all rain gardens, the change in storage is equal to the inflow minus the outflow, such thatwhere $S$ = storage; $I$ = inflow; and $O$ = outflow.

The inflow is composed of direct rainfall on the rain garden and runoff that is directed into the rain garden from nearby surfaces. The outflow is composed of groundwater recharge ($\mathrm{\Delta }{S}_g$), evapotranspiration (ET), and overflow ($O_{\mathrm{over}}$)

Groundwater recharge is the amount of water that moves through the pores of the soil in the unsaturated zone and recharges the groundwater table. After a storm event is over, and after the bowl has infiltrated all ponded water, and the soil reaches field capacity, the change in storage is assumed to be zero, thus

The NRCS curve number method was used to estimate the inflow volume (Cronshey 1986b; Farnsworth 2017; Amur et al. 2022). The impervious and pervious areas were assigned a curve number of 98 and 80, respectively, based on the land use. A V-notch weir with a 60° angle and a maximum head of 38.1 cm (15 in.) was used in conjunction with a pressure transducer to calculate the overflow volumes when they occurred.

The analysis of recharge at the rain garden was based on the water table fluctuation method (Meinzer 1923; Healy and Cook 2002). This method uses groundwater level data to estimate the volume of water that enters the groundwater table through recharge in unconfined aquifers. This method assumes that all rises in groundwater levels in an unconfined aquifer are a result of water that passed through the unsaturated zone vertically, and that no other parts of the water cycle are contributing during the period of recharge. Because of the latter assumption, the water table fluctuation method is more accurate over shorter periods of time than other methods such as unsaturated zone water balance, age dating, or recession-curve-displacement methods (Rutledge 1998; Delin et al. 2007).

Because the groundwater level fluctuates naturally before a recharge period, it would not be accurate to define the change in the water table as the change from the start of the period to the maximum level (Fig. 3). Instead, the recession rate before the groundwater rise (antecedent recession) must be extrapolated to determine what the groundwater level would have been at the time of the peak groundwater level. The difference between this extrapolated value and the peak observation should be used as the change in groundwater level, $\mathrm{\Delta }h(t_j)$ (Healy and Cook 2002). A number of methods can be used to determine the antecedent recession rate, including manual graphical extrapolation, developing a master recession curve to be used over a larger period, or computer programs (Delin et al. 2007).

The water table fluctuation method can be used to model a small area to large regions, if enough spatial data are available. The model can be accurate up to a few square meters for one well because an observation well is representative to this level. The method is more accurate with shallower water tables that experience sharper rises in water levels, but has been applied to deeper aquifers as well. Temporally, the model can be used with data ranging from event-based to seasonal (Delin et al. 2007). Recharge is estimated using the following equation:where $R(t_j)$ = recharge between times $t_0$ and $t_j$ (m); $S_y$ = specific yield of soil in aquifer; and $\mathrm{\Delta }h(t_j)$ = water level rise due to recharge (m).

To calculate recharge volumes for a single event, the following process was used. Time-series data for each event were collected and prescreened to ensure that the declining trend of the groundwater levels in each well had enough data points to calculate the slope. If more points were needed, the event range was expanded by as much as 1 day in advance of the event. After the data were acquired, they were processed using the pivot table function in Microsoft Excel to summarize the data on an hourly basis. The maximum groundwater levels were used during each hour for the analysis. The slope of the premounding recession limb was calculated by finding the maximum groundwater level at the start of the event and the minimum groundwater level before the mounding was observed. The change in the water table elevation was divided by the time between the two points to produce the slope. This slope was used to extrapolate the recession limb from the time of the minimum groundwater level before the mound was observed to the time of the peak observation of the mound. During this period, the groundwater recharge was calculated for each hour using Eq. (4). Based on soil type, a specific yield ($S_y$) of 0.25 was assumed in this work (Johnson 1967; Woessner and Poeter 2020). This value of specific yield was validated by slug testing. The calculated recharge then was multiplied by the surface area of the rain garden to represent a recharge volume across the entire site. These hourly recharge volumes were graphed alongside the hourly inflow and overflow volumes in order to examine the timing of each in comparison with each other. To determine the total recharge throughout the event, the same calculation was performed, using the change in elevation between the minimum and maximum observations and the timing between the two. For each of the wells available for the event, the recharge totals were averaged to determine a representative recharge approximation for the site.

Groundwater fluctuations may not always be a result of recharge from the soil directly above it. Water also can enter laterally or be injected and withdrawn. In this case, injection and withdrawal can be ignored because the site is in a residential and recreational area on a university campus where there are no wells. For this study, lateral movement of water within the aquifer was negligible over the period considered for the calculations due to the low hydraulic conductivity of the aquifer and the low gradient of the groundwater table. Instrumentation error also can contribute to rises and falls in the recorded water level. Aqua TROLL 200 sensors (In-Situ, Fort Collins, Colorado) were used in this study, with an accuracy of $\pm 1.1\mathrm{cm}$ (0.39 in.), which is small compared with the rises observed in this study. The data at the site were averaged over 15-min intervals, and this study used the maximum reading over the hour. This removed any low spikes that may have been recorded in that period. Any abnormally high elevations were removed manually.

After the groundwater recharge was determined, the only unknown parameter was ET, which was estimated by rearranging Eq. (3) to yield

The data were analyzed by separating individual rain events and determining their impact on groundwater mounding. A rainfall event was defined by cumulative precipitation of greater than 2.54 mm (0.1 in.) preceded by at least 6 h of dry time (Machusick et al. 2011). To examine further the mechanics of the water flow at the site, the events were divided into three categories based on how the water was moving through the garden. The separating rainfall volume was informed by previous studies conducted at the site (Machusick et al. 2011; Nemirovsky et al. 2015)

Although the classification of small and medium events is clear from a volume standpoint, medium and large events had an overlap in their ranges. This was because if the mounding event includes only one rainfall event and it is greater than 24.5 mm (1 in.), overflow will occur, but if the mounding event includes several rainfall events of less than 24.5 mm (1 in.), and the cumulative total rainfall is a larger value, overflow will not necessarily occur.

Although storm events typically are classified in terms of rainfall volume, in this application the rainfall volume does not necessarily correlate to the mounding event type. Because the events can be much longer and may contain two rainfall events within one mounding event, the rainfall volume may not behave the same way between events.

Water levels in the wells directly surrounding the rain garden (MW 2, MW 6, and MW 7) were used in these calculations. It was determined previously (Nemirovsky et al. 2015) that the water levels in the wells upstream and downstream had minimal influence from the rain garden, so they were not used in the study. The events chosen for analysis were based on the availability of data and sufficient separation between events. Priority was placed on events for which data were available at all three wells. The events were screened to determine if there were multiple mounding peaks throughout the event. Events with a single mounding peak were chosen for this study. The recharge calculations also were compared with rainfall totals, inflow calculations, and overflow calculations at the site, so events were chosen based on availability of these data as well. Throughout the study period, May 2014 to December 2016, there were periods in which sensors produced bad data, were removed for the winter, or were removed for maintenance. After filtering the events with these restrictions, 26 events were chosen for analysis (Fig. 4); 22 of the events occurred during the growing season (April 15–November 15) and 4 (Fig. 4, shaded regions) occurred during the winter months when the vegetation was dormant and ET was insignificant.

Specific events are highlighted herein to describe the observed data, the calculated recharge from the observed data, and the evapotranspiration estimated from the recharge results. Although the recharge and evapotranspiration occur mainly within the rain garden perimeter, not all rain that falls on the drainage area enters the rain garden as inflow. Some is lost as initial abstractions, which likely contribute to additional evapotranspiration over the watershed separate from that facilitated by the rain garden. Because about half of the watershed area is pervious, it is expected that it will not produce much runoff, especially for small events. Thus, the values reported underestimate the volume of ET over the entire drainage area.

Rain gardens play a vital role in sustainable stormwater management. These innovative green spaces utilize natural processes, such as infiltration and evapotranspiration, to effectively control and mitigate runoff from rain events. An instrumented rain garden on Villanova University’s campus provided a unique opportunity to use the hydrologic budget to estimate the volume of evapotranspiration occurring at this unlined rain garden in Pennsylvania. In agreement with prior studies completed at this site (Machusick et al. 2011; Nemirovsky et al. 2015), most of the water (83%, on average) that entered the rain garden for storms with rainfall of less than 18 mm (0.71 in.) was evapotranspired and did not reach the groundwater table. In this region, 82% of storms are 18 mm (0.71 in.) or less. For storms with rainfall of more than 18 mm (0.71 in.), water that infiltrated either recharged the groundwater table or was removed by ET. On average, for storms with rainfall exceeding 18 mm (0.71 in.) that were not intense or large enough to cause overflow, 62% of the inflow was evapotranspired and 38% of water recharged the groundwater table. Although only one large storm was observed during the study period, the percentage of water evapotranspired (73%) was similar to the percentages found for the small and medium storms. As found with previous studies, any groundwater mounding that occurred dissipated within 2 to 3 days (Machusick et al. 2011; Nemirovsky et al. 2015). The percentage of retained inflow determined using the water budget method presented in this paper was similar to the results reported by Hess et al. (2017), who used weighing lysimeters to model this rain garden, thus validating the results. These findings indicate that because ET accounts for a significant portion of the hydrologic budget, especially for medium and smaller rainfalls, the volume going to deep infiltration and mounding is considerably less than the current perception. Because the infiltration volume is part of any mass loading estimate, this reduction also would be reflected in water quality mass loading analysis. Moreover, it can be inferred that groundwater mounding beneath rain gardens in similar climates is unlikely to be a concern. In addition, this also indicates that pollutant mass transport from rain gardens is smaller than currently is expected.

All data generated in the study are available from the corresponding author by request.

This study was supported by the Pennsylvania Department of Environmental Protection (PADEP) 319 Nonpoint Source Implementation Grant and Coastal Zone Management Grant, The William Penn Foundation, and the partners of the Villanova Urban Stormwater Partnership (VUSP). The opinions and findings expressed here represent those of the authors and not the funding agencies.