Abstract

Green infrastructure (GI), such as green roofs, rain gardens, and porous pavement, is a stormwater management strategy designed to capture rain where it falls and allow it to soak into the ground rather than running off into a stream channel, thus reducing flooding and improving water quality. While there has been a lot of research into the performance of individual GI projects, much less is known about its performance at the catchment scale. This study uses a US EPA SWMM model to examine the effectiveness of GI in improving water quality and reducing flooding at the catchment scale. Results show that in the study catchment, a large centralized wetland was the most effective at reducing and slowing peak discharge. Infiltration based decentralized GI best reduced flood volumes. In addition to changes in effective impervious area, flood volumes were also reduced due to differences in drainage network structure and modifications to the pervious portions of the catchment. Reductions in flood volumes resulted in lower pollutant loads, except for pollutants that are particularly efficiently removed by wetlands. Routing runoff through a large, centralized wetland removed more nitrate load than letting rain infiltrate where it falls.

1 Introduction

Flooding and poor water quality in urban areas are typically caused by impervious surfaces such as roads, buildings and parking lots (Leopold 1968; Schueler 2003). These hard surfaces prevent rain from soaking into the ground. As rain runs off over the land surface, it takes with it pollutants that eventually enter the surface water system. Green stormwater infrastructure (GI), also referred to as low impact development (LID) (Fletcher et al. 2015), is a strategy that can be used to prevent flooding and improve water quality by capturing rain where it falls. GI strategies such as green roofs, rain gardens, and porous pavement can also provide ecosystem services and human health benefits (Coutts and Hahn 2015). Stormwater wetlands, defined by the US EPA as landscape or watershed level GI (US EPA 2021), are often designed to detain runoff during storm events and release it slowly, but can also promote infiltration and evapotranspiration.

Many studies have examined the performance of individual GI projects (Hathaway et al. 2008; Sychterz et al. 2014; Shuster et al. 2017; and Chowdhury and Abaya 2018 are among the many examples). While a single GI project can effectively reduce flooding and improve water quality, work that summarizes and synthesizes these studies suggests that the effectiveness of any given site level GI project often depends on the type of GI examined, site characteristics, and the pollutant of concern (Center for Watershed Protection 2007; Koch et al. 2014; Leisenring et al. 2014). Furthermore, the scaling of the performance of GI practices from individual or lot scales to catchment scales is not well understood (Ahiablame et al. 2012; Golden and Hoghooghi 2018).

Catchment scale GI performance can be evaluated by collecting flooding and water quality data in the field. This requires monitoring a watershed before and after implementation of GI (Shuster et al. 2013) or through multi-watershed studies (Pennino et al. 2016). At the neighborhood scale, some paired-catchment studies have shown that GI within a residential development can reduce runoff and improve water quality (Dietz 2007; Dietz and Clausen 2008; Hood et al. 2007), yet similar studies have shown that GI cannot mitigate the impacts of residential development and the associated impervious surfaces (Yang and Li 2013). Multi-watershed studies have found that pollutant loads were lower in watersheds with more GI (Pennino et al. 2016), and Loperfido et al. (2014) found that a watershed with distributed GI reduced runoff volumes and peaks better than a paired watershed with centralized GI, but more work is needed to assess whether these results are consistent across geographic regions (Jefferson et al. 2017; Golden and Hoghooghi 2018). Jefferson et al. 2017 also note confounding factors can complicate the interpretation of catchment scale field studies.

An alternative approach to empirical and field studies for assessing catchment scale GI performance is to use a rainfall–runoff model (see for example Di Vittorio and Ahiablame 2015; Carvallo Aceves et al. 2017). As with site level studies, the results of these modeling studies also vary with GI type, site and pollutant as well as land use type and location (Jefferson et al. 2017). Scale dependence on the performance of GI at the watershed scale (Javier et al. 2007; Meierdiercks et al. 2010) may explain some of the inconsistencies (Smith et al. 2015).

Many studies have pointed to the need to better understand the cumulative impact of GI on the catchment scale (Fahy and Chang 2019; Golden and Hoghooghi 2018; Loperfido et al. 2014, for example). While it is generally understood that as the percentage of the watershed treated by GI increases, so do the hydrological and water quality benefits (Jefferson et al. 2017) as the percentage of effective imperviousness decreases (Palla and Gnecco 2015). However the cumulative impact of the spatial configuration of GI, particularly centralized versus distributed GI, is less straightforward. Catchment scale modeling studies suggest that centralized GI can reduce flood peaks and volumes when storage capacity or watershed capacitance is not exceeded (Lim and Welty 2017), but distributed GI may be more effective when runoff volume exceeds the storage capacity of centralized GI (Antolini and Tate 2021; Fahy and Chang 2019). More work is needed to better understand the cumulative impacts of GI, particularly on water quality, at the catchment scale. This study addresses these research gaps through the following objectives: (1) to use a stormwater model to examine the impacts of GI (green roofs, wetlands and soil management improvements) on water quality and flooding in the study catchment; and (2) to identify performance differences between centralized versus decentralized GI in the study catchment.

2 Methods

2.1 Study area

The study area for this project was the Siena College campus catchment (Figure 1), a subwatershed of the larger Kromma Kill Watershed. The 0.86 km² study area is 23% impervious (covered by roads, buildings, and parking lots) and contains five stormwater wetlands (P-1, P-2, P-3, P-5, and P-6 in Figure 1) and one detention pond (P-4). The total 8.7 km drainage network consists of 95% pipes and 5% surface channels. The predominant soil type is loamy sand. Of the 0.20 km² of impervious surfaces in the study area, 11.5% are buildings. The Kromma Kill stream channel originates on campus and flows 4.8 km downstream to the Hudson River. The Kromma Kill is an impaired waterbody and prone to flooding and other water quality problems. Though polychlorinated biphenyls (PCBs) and heavy metals have been detected in the sediments downstream of the study area, near a Brownfield site (Al Tech Specialty Steel, Site Code 401003), according to water quality sampling during 2014 and 2015, runoff from campus is typical of institutional dominated land uses. For example, during 2014 baseflow sampling, nitrate and total nitrogen concentrations were higher than other sampling sites throughout the area, but still well below the New York State Department of Environmental Conservation’s standards for aquatic health.

Figure 1 Map of the study area.

2.2 Model development

A US EPA SWMM v5.1 stormwater model was built from field observations and geospatial datasets for the study area (Table 1, Meierdiercks et al. 2017). See James et al. (2010) for a full description of the mathematical equations used by SWMM. Field observations of rainfall were used as model forcing, and streamflow and water quality measurements were used to calibrate the model. The model was then modified to represent several design scenarios. In each scenario, one or more GI strategies were added to or removed from the catchment and the impact on water quantity and quality was examined.

The 0.86 km² study area includes the main catchment (0.52 km²) that drains to the outlet indicated in Figure 1. Seven additional subareas (ranging in size from 0.008 km² to 0.1 km²) were included in model simulations. Each subarea drains to an outlet which then flows to the main Kromma Kill stream channel. Subareas are further divided into subcatchments ranging in size from 61.5 m² to 97 700 m² (average 8390 m²) and in percent impervious from 0 to 88.5% (average 27.7%). Both the impervious and pervious areas of each subcatchment runoff directly into the storm drain network (subarea routing to the outlet).

The model includes a large centralized stormwater wetland, 4 additional smaller wetlands, and 1 stormwater detention pond (Table 2). The stormwater wetlands and detention pond were modeled in SWMM as storage units with removal efficiencies (Table 3) determined from Leisenring et al. (2014). The volume of the wetlands were determined from 61 cm (2 ft) contour lidar-derived topography data. Outflow was modeled using a depth based rating curve provided in stormwater pond design documentation.

Table 1 SWMM model parameters.

Parameter	Type	Value	Source
Rain gauges	Input/forcing	2.24 cm, 4.04 cm and 7.04 cm	60 min 1 y, 10 y and 100 y design storms from: http://precip.eas.cornell.edu/
Area	Model	Variable	Measured using GIS
Width	Model	Variable	Measured using GIS
% Slope	Model	Variable	Computed using 3 m lidar-derived digital elevation model from Albany County, NY
% Imperv	Model	Variable	Computed using 30 m NLCD 2011, verified and corrected as needed using aerial photos
N-Imperv	Model	0.02	Calibration parameter
N-Perv	Model	0.3	Calibration parameter
%Zero-Imperv		25	James et al. (2010)
Subarea Routing	Model	OUTLET	Observed in the field
Percent Routed	Model	100	Observed in the field
Node location and geometry	Model	Variable	Siena College stormwater pipe data
Link location and geometry	Model	Variable	Siena College stormwater pipe data and Albany County lidar derived 2 ft contour data
Storage unit geometry	Model	Variable	Albany County lidar derived 2 ft contour data
Storage unit outlet geometry	Model	Variable	Siena College stormwater pond design data
Infiltration	Model	Green–Ampt	NA
Suction Head	Model	6.096 cm (2.4 in.)	Value for loamy sand (James et al. 2010)
Conductivity	Calibration	4.7625 mm/h	Calibration parameter
Initial Deficit	Calibration	0.146	Calibration parameter
TN EMC (mg/L)	Calibration	2.1	James et al. (2010)
TDS EMC (mg/L)	Calibration	178.1	Measured in the field
NO3- EMC (mg/L)	Calibration	1.46	Measured in the field
TP EMC (mg/L)	Calibration	0.263	James et al. (2010)

Table 2 Stormwater pond dimensions.

Name	Surface area (m2)	Max depth (m)	Max outflow (m3/s)	Depth of max outflow (m)
P-1 Wetland	2530	2.13	2.36	1.1
P-2 Wetland	2500	0.61	0.184	0.61
P-3 Central wetland	18 200	7.32	1.61	1.58
P-4 Detention pond	662	1.23	0.55	1.23
P-5 Wetland	232	0.305	No outlet	NA
P-6 Wetland	502	0.305	No outlet	NA
Decentralized wetlands (9)	6080	2.54	0.18	0.521

Table 3 Pollutant removal efficiencies (Leisenring et al. 2014).

	TN	TDS	NO3−	TP
Wetland	−0.053	−0.75	0.73	0.46
Stormwater Pond	−0.21	0.052	0.18	0.22

Calibration of the model is described in Meierdiercks et al. (2017). Five storms close in magnitude to the 60 min 1 y design storm were used to calibrate the model. Saturated hydraulic conductivity, Manning’s roughness of the pervious and impervious portions of the study area and initial deficit (Table 1) were manually adjusted to match the timing and volume of the observed discharge hydrographs that were measured at the drainage outlet indicated in Figure 1. Calibration results are shown in Table 4. For water quality, total dissolved solids (TDS), nitrate (NO₃⁻), total nitrogen (TN) and total phosphorus (TP) were added to the system, initial buildup was set to zero, and the event mean concentration (EMC) function was used to compute pollutant washoff during rain events. EMC is the average pollutant concentration during a storm event and is computed as the pollutant load divided by the volume of runoff (James et al. 2010). EMC was chosen as the water quality washoff method rather than an exponential function or rating curve due to the lack of high frequency calibration data. The model was calibrated by adjusting EMC to match concentrations measured in the field during a storm event close in magnitude to the design storm or to match values typical of urban runoff (Table 1).

Table 4 SWMM model calibration performance; numbers shown as percentages are the percentage bias between modeled and observed for hydrograph volume and peak discharge, and difference in time to peak over storm duration; NRMSE between modeled and observed values is also shown.

Storm event date	Storm total precipitation (cm)	Hydrograph volume percent bias	Peak discharge percent bias	Difference in time to peak
2013-06-07	1.85	13.7%	17.4%	5.7%
2013-06-13	4.18	−32.4%	−33.7%	−3.1%
2013-06-27	1.98	−17.9%	−14.5%	−3.6%
2013-06-28	2.18	23.3%	−4.2%	−0.3%
2013-07-09	3.94	58.9%	30.5%	−4.8%
NRMSE	2.83	0.44	0.27	0.23

The calibrated model was then modified to represent a number of alternative design scenarios (Figure 2, Table 5) to test the relative impact on water quality and quantity of centralized versus decentralized and detention based versus retention based GI at the catchment scale. For the decentralized wetlands scenarios, the large centralized wetland was replaced with 9 smaller wetlands, each with 11.11% of the volume of the wetland. The outlet rating curve was modified so each decentralized wetland provided 11.11% of the outflow of the centralized wetland. Two decentralized wetlands scenarios were tested. The first keeps the distribution of impervious surfaces in the catchment the same as in the centralized scenario. In the centralized scenario, 0.39 km² (31.5% impervious) is captured by the wetland. In the decentralized scenario, 0.13 km² (23% impervious) is captured. For the second decentralized wetlands scenario, all impervious areas that were upstream of the centralized wetland are moved upstream of the decentralized wetlands, therefore keeping the area of imperviousness treated the same. This results in an upstream imperviousness of 92.1%. For the green roof, parameters were chosen to match those in Rossman (2010) and typical values found in green roof design and construction resources (Table 6). The model was also run using SWMM green roof input parameters as described in Hamouz and Muthanna (2018), Russwurm et al. (2018) and Worthen and Davidson (2018). There were no observable differences in the system outflow quantity and quality when these parameters were used.

Table 5 SWMM model scenario descriptions.

Name	Description	Details
Business as usual (BAU)	The study catchment as is with one, large centralized wetland.	NA
No wetland (control)	The catchment with no wetland.	Removed the large centralized wetland and replaced it with a node and link.
Decentralized wetlands	Instead of one large wetland, the catchment with a number of smaller wetlands distributed throughout the watershed	Removed the large centralized wetland as in the No Wetland scenario and added 9 wetland storage units (each 11.11% volume of centralized wetland).
Decentralized wetlands (upstream imperviousness)	Instead of one large wetland, the catchment with a number of smaller wetlands distributed throughout the watershed capturing the same impervious area as the BAU Scenario.	Removed the large centralized wetland as in the No Wetland scenario and added 9 wetland storage units (each 11.11% volume of centralized wetland). Modified imperviousness so an equivalent area of impervious surface would be treated by these 9 decentralized wetlands as the large centralized wetland.
Decentralized green roofs	Instead of one large wetland, a number smaller green roofs distributed throughout the catchment capturing the same impervious area as the BAU Scenario.	Removed the large centralized wetland as in the No Wetland scenario and added a green roof to each subcatchment upstream of the former wetland. Each green roof covers 100% of the subcatchment impervious area.
Decentralized soil management	Instead of one large wetland, better managed soil throughout the catchent.	Removed the large centralized wetland as in the No Wetland scenario and increased conductivity of all pervious areas by a factor of 10 (from 4.76 mm/h to 47.6 mm/h)

Table 6 SWMM input parameters for green roof simulation.

Parameter	Value	Source
Surface
Berm Height (cm)	10.16	Assumed
Vegetation Volume Fraction	0.25	Assumed
Surface Roughness	0.1	Rossman (2010)
Surface Slope (%)	1	Rossman (2010)
Soil
Thickness (cm)	15.24	Assumed
Porosity	0.5	Rossman (2010)
Field Capacity	0.2	Rossman (2010)
Wilting Point	0.1	Rossman (2010)
Conductivity	0.5	Rossman (2010)
Conductivity Slope	10	Rossman (2010)
Suction Head (cm)	8.89	Rossman (2010)
Drainage Mat
Thickness (cm)	7.62	Rossman (2010)
Void Fraction	0.5	Rossman (2010)
Roughness	0.1	Rossman (2010)

3 Results

Results were analyzed quantitatively by examining the impact of each model scenario compared to the BAU scenario on peak flooding, flood volumes, and pollutant outflow concentration and load. To compare water quantity volumes, the runoff ratio was calculated as surface runoff divided by total precipitation. Peak discharge was determined from the system outflow hydrograph and time to peak was computed as the time between peak rainfall and peak discharge. System outflow refers to the total outflow from the main study area catchment as well as the additional subareas shown in Figure 1. System outflow was presented so that areas of the study catchment not controlled by the centralized wetland would be included in the results. For water quality, external outflow pollutant load (in lbs) and external outflow concentration (outflow pollutant load in lbs divided by external outflow flow volume) were examined.

System outflow hydrographs for each model scenario using the 60 min 1 y, 10 y, and 100 y design storms are shown in Figure 3. For the 1 y storm, the BAU scenario (large, centralized wetland) resulted in the smallest peak discharge and longest time to peak (Figure 3, Table 7). Removing the large, centralized wetland increased peak discharge by 129.7% and decreased the time to peak by 36.0%, while the decentralized wetlands also led to a 36.0% decrease in time to peak and a 109.4% increase in peak discharge. The decentralized wetlands (upstream imperviousness), decentralized green roofs and improved soil management scenarios all resulted in similar increases in peak discharge (respectively 62.4%, 70.5% and 64.7%). The decentralized wetlands (upstream imperviousness) only slightly increased the time to peak (8%), while the decentralized green roofs and soil management scenarios decreased time to peak respectively by 32.0% and 72.0%.

The impacts on peak discharge and time to peak compared to the BAU scenario were similar as the storm size increased to the 10 y and 100 y storms (Table 7). The exception is the impact of the better soil management scenario on peak discharge. For the 1 y storm, the better soil management scenario led to a 64.7% increase in peak discharge compared to the BAU scenario. However, for the 10 y storm, the peak discharge decreased by 8.2% and for the 100 y storm, it increased by 5.1%. The control, decentralized wetlands, decentralized wetlands (upstream imperviousness), and decentralized green roofs scenarios increased peak discharge by 91.5%, 74.8%, 62.4% and 58.3%, respectively, for the 10 y storm and by 97.1%, 77.2%, 69.8%, and 71.1%, respectively, for the 100 y storm. For time to peak, the control, decentralized wetlands, decentralized wetlands (upstream imperviousness), decentralized green roofs, and soil management scenarios increased time to peak by 19.2%, 19.2%, 11.5%, 19.2%, and 65.4% (10 y storm), and 62.8%, 39.5%, 53.5%, 62.8%, and 60.5% (100 y storm) compared to the BAU scenario.

Figure 3 System outflow hydrographs for the 1 y (top), 10 y (middle), and 100 y (bottom) storm SWMM model scenarios.

Table 7 Results of SWMM water quantity modeling.

Model Scenario	Peak Discharge (m3/s/km)	% Change	Time to Peak (min)	% Change	Runoff Ratio	% Change
1 y storm
Business as Usual (large, central wetland)	0.9	–	25	–	0.27	-
Control (no wetland, no GI)	2.1	129.7%	16	−36.0%	0.29	6.9%
Decentralized wetlands	2.0	109.4%	16	−36.0%	0.28	1.5%
Decentralized wetlands (upstream imperviousness)	1.5	62.4%	23	−8.0%	0.24	−12.2%
Decentralized green roofs	1.6	70.5%	17	−32.0%	0.21	−21.2%
Soil management	1.5	64.7%	7	−72.0%	0.16	−42.6%
10 y storm
Business as Usual (large, central wetland)	2.6	-	26	-	0.47	-
Control (no wetland, no GI)	5.0	91.5%	21	−19.2%	0.48	1.1%
Decentralized wetlands	4.6	74.8%	21	−19.2%	0.44	−7.5%
Decentralized wetlands (upstream imperviousness)	4.3	62.4%	23	−11.5%	0.44	−6.8%
Decentralized green roofs	4.2	58.3%	21	−19.2%	0.38	−19.7%
Soil management	2.4	−8.2%	9	−65.4%	0.16	−66.1%
100 y storm
Business as Usual (large, central wetland)	4.7	-	43	-	0.58	-
Control (no wetland, no GI)	9.2	97.1%	16	−62.8%	0.59	1.9%
Decentralized wetlands	8.3	77.2%	26	−39.5%	0.53	−8.6%
Decentralized wetlands (upstream imperviousness)	7.9	69.8%	20	−53.5%	0.53	−8.2%
Decentralized green roofs	8.0	71.1%	16	−62.8%	0.48	−18.0%
Soil management	4.9	5.1%	17	−60.5%	0.25	−57.4%

The runoff ratios for each model scenario are shown in Table 7. While the BAU scenario resulted in the smallest peak discharges (with the exception of the 10 y soil management scenario) and longest time to peak, it did not result in the smallest runoff ratio. In terms of runoff volumes, for the 1 y storm, the soil management scenario led to the greatest reduction of runoff volume, followed by the decentralized green roofs, and decentralized wetlands (upstream imperviousness) (respectively 42.6%, 21.2%, 12.2%). The decentralized wetlands and control scenarios both led to increases in runoff volumes compared to the BAU scenario (respectively 1.5% and 6.9%).

As storm size increased to the 10 y and 100 y storm, the soil management followed by the decentralized green roof scenarios reduced the greatest volume of runoff (respectively 66.1% and 19.7% for the 10 y storm, and 57.4% and 18.0% for the 100 y storm). There was less of a difference in runoff ratio between BAU and control scenarios (respectively 1.1% and 1.9%, compared to 6.9 % for the 1 y storm). For the 1 y storm, the decentralized wetlands increased runoff volumes by 1.5%, but reduced volumes by 12.2% when the decentralized wetlands were paired with impervious surfaces concentrated upstream of the wetlands. For the 10 y and 100 y storms, both decentralized wetlands scenarios resulted in similar reductions of runoff volume: 7.5% and 6.8% for the 10 y storm, and 8.6% and 8.2% for the 100 y storm.

Additional calculations were made to compare runoff volumes generated as part of each of the three wetland scenarios. A detailed discussion of these results is provided in the next section. The runoff ratios for each wetland scenario, computed as the volume of outflow from the wetland divided by the volume of rainfall upstream of the wetland, are shown in Table 8. Each scenario resulted in a runoff ratio that was greater than the system outflow runoff ratio and that was close to or exceeded the impervious percentage upstream of the wetland or wetlands. The final stored volume (defined as the total volume of water remaining in storage units, pipes, and nodes at the end of the simulation) plus the losses due to flooding (defined as the water that lost to the system when nodes flood more than the 4.6 m² allowable ponded area) as a fraction of the total rainfall was computed for each wetland scenarios. This is the percentage of rainfall that neither infiltrates nor leaves the study area as outflow. For the BAU (centralized wetland) scenario, these values are: 10.1%, 10.2%, and 15.1% (respectively 1 y, 10 y, 100 y storms); for the decentralized wetlands scenario: 10.2%, 13.9%, and 20.2% (respectively 1 y, 10 y, 100 y storms); and for the decentralized wetlands scenario where imperviousness surfaces are moved upstream: 13.1%, 12.7%, 19.8% (respectively 1 y, 10 y, 100 y storms). For comparison, for the control scenario, the percentages of rainfall remaining in storage and lost to flooding are: 8.1%, 9.7%, and 14.0% (respectively 1 y, 10 y, 100 y storms). For all scenarios, the percentages of rainfall remaining in storage and lost to flooding increased as storm sizes increases. The centralized wetlands scenario resulted in more rainfall remaining in storage or lost to flooding than the control scenario, but less than the decentralized wetlands scenarios.

Table 8 Runoff ratios for wetland model scenarios; runoff ratio is computed as the storm total rainfall divided by the runoff generated downstream of the centralized wetland or representative decentralized wetland.

		Runoff ratio
Model Scenario	Upstream imperviousness	1 y storm	10 y storm	100 y storm
Business as Usual (large, central wetland)	31.5%	0.43	0.64	0.76
Decentralized wetlands	23.3%	0.7	0.8	0.9
Decentralized wetlands (upstream imperviousness)	92.1%	0.9	0.9	1

The outflow concentration and loading for each model scenario for the 1 y storm are shown in Table 9. Removing the large centralized wetland increased both the pollutant concentration and load. Replacing the large, centralized wetlands with smaller decentralized ones, when the distribution of impervious surfaces was kept constant, resulted in TN and TDS concentration and loads slightly decreasing (−4.0% and −2.6% for TN, −5.0% and −3.6% for TDS), but NO₃⁻ and TP concentration and loads increasing considerably (78.7% and 81.3% for NO₃⁻, 29.5% and 31.4% for TP). Water quality improvements were similar for the decentralized wetlands with the impervious surfaces concentrated upstream of the wetlands: TN and TDS concentration increased and loads decreased slightly (6.7% and −6.3% for TN, 6.5% and −6.5% for TDS), but NO₃⁻ and TP concentration and loads considerably increased (84.7% and 62.2% for NO₃⁻, 39.3% and 22.3% for TP). The infiltration based scenarios, decentralized green roofs and soil management, led to water quality improvements for TN and TDS concentration and load, but poorer water quality for NO₃⁻ concentration and load. For TP, load improved, but water quality concentration declined compared to the BAU scenario.

Table 9 Results of SWMM water quality modeling for the 1 y design storm; negative percentage change values indicate water quality improvements.

Model Scenario	TN	TDS	NO3−	TP
Outflow concentration (mg/L)
Business as Usual (large, central wetland)	5.85	485.88	1.87	0.48
Control (no wetland, no GI)	5.90	490.21	3.99	0.72
Decentralized wetlands	5.62	461.52	3.35	0.62
Decentralized wetlands (upstream imperviousness)	6.24	517.52	3.46	0.67
Decentralized green roofs	4.62	372.57	3.02	0.54
Soil management	5.67	480.85	3.94	0.71
Outflow concentration % change
Business as Usual (large, central wetland)	–	–	–	–
Control (no wetland, no GI)	0.8%	0.9%	112.7%	48.9%
Decentralized wetlands	−4.0%	−5.0%	78.7%	29.5%
Decentralized wetlands (upstream imperviousness)	6.7%	6.5%	84.7%	39.3%
Decentralized green roofs	−21.1%	−23.3%	61.2%	12.8%
Soil management	−3.1%	−1.0%	110.3%	47.5%
Outflow loading (lbs)
Business as Usual (large, central wetland)	25.0	2075.2	8.0	2.1
Control (no wetland, no GI)	26.9	2238.8	18.2	3.3
Decentralized wetlands	24.3	1999.8	14.5	2.7
Decentralized wetlands (upstream imperviousness)	23.4	1940.2	13.0	2.5
Decentralized green roofs	15.5	1253.7	10.2	1.8
Soil management	13.9	1178.1	9.7	1.7
Outflow loading % change
Business as Usual (large, central wetland)s	–	–	–	–
Control (no wetland, no GI)	7.8%	7.9%	127.4%	59.2%
Decentralized wetlands	−2.6%	−3.6%	81.3%	31.4%
Decentralized wetlands (upstream imperviousness)	−6.3%	−6.5%	62.2%	22.3%
Decentralized green roofs	−37.8%	−39.6%	27.0%	−11.1%
Soil management	−44.4%	−43.2%	20.6%	−15.4%

The impacts on water quality concentration and loading are similar for the 10 y and 100 y storms (see Tables A1 and A2). Scenarios that led to a difference of more than 20 percentage points between the 1 y and 10 y storms include the decentralized wetland scenario, where the percentage change in NO₃⁻ concentration increased from 78.7% to 111.0% (less water quality improvement), and the soil management scenario, where the percentage change in TN, TDS, NO₃⁻, and TP load decreased from −44.4%, −43.2%, 20.6%,−15.4% to −65.4%, −65.1%, −22.5%, −47.7% (greater quality improvements). Scenarios that led to a difference of more than 20 percentage points between the 1y and 100 y storm include the decentralized wetland scenario (upstream imperviousness), where the percentage change in NO₃⁻ and TP concentrations decreased from 84.7% to 63.1% for NO₃⁻, and from 39.3% to 14.6% for TP, and the soil management scenario, where the percentage change in NO₃⁻ and TP load decreased from 20.6% and −15.4% to 9.8% and −38.2%.

4 Discussion

Jefferson et al. 2017 note that traditionally large centralized greywater stormwater management strategies such as detention ponds were used to manage stormwater, but more recently decentralized or distributed infiltration based GI such as rain gardens and green roofs have grown in popularity. However, there are relatively few studies that have examined the cumulative impact of centralized versus decentralized GI, particularly the impact on water quality, at the catchment scale. In the following section, we discuss the results of this study, specifically the cumulative impacts on catchment scale water quantity and quality of (1) centralized versus decentralized wetlands, and (2) detention versus retention based decentralized GI.

4.1 Centralized versus decentralized wetlands

In the study catchment, the large centralized wetland is specifically designed to slow and detain runoff from large storms. Results suggest that the wetland is functioning as designed. For peak flows in the study catchment, the centralized wetland reduced and slowed peak flows for the 1 y, 10 y, and 100 y storms, more than the decentralized wetlands scenarios, even when the area of impervious treated and storage capacity is kept constant. This result is consistent with Lim and Welty 2017 who found that centralized GI performed better in catchments where storage capacity or watershed capacitance was not exceeded. In this study and ours, centralized stormwater management was most effective when located in higher flow accumulation areas.

In terms of flow volumes, for the 1 y storm, the runoff ratios for the centralized and decentralized wetland scenarios were very similar (only a 1.5% difference). But when the distribution of impervious surfaces is modified and moved upstream of the decentralized wetlands so that the area of impervious surfaces treated remains constant, the runoff ratio decreased by 12.2% compared to the centralized wetland scenario. As storm size increased from the 1 y to the 10 y and 100 y storms, both decentralized wetland scenarios resulted in smaller runoff ratios (between 6.8 and 8.6%) compared to the centralized wetland scenario. However, the smaller runoff ratios associated with the decentralized wetland scenarios were not due to added storage capacity or increased infiltration: the total volume of the 9 decentralized wetlands is equivalent to that of the large, centralized one and the decentralized wetlands did not retain runoff any more effectively than the centralized wetland (Table 8). Instead, the larger runoff ratio of the centralized wetland is due to the long tail of the system outflow hydrograph (Figure 3) as the large centralized wetland slowly drains long after the end of the 1 h duration storm. The smaller, decentralized wetlands drain equally slowly, but that outflow entered into the storm pipe network rather than a surface channel as in the centralized wetland scenario. The pipe network has smaller storage and flow capacities than the surface channel, so outflow from the decentralized wetlands was attenuated and dispersed as it moved through the drainage network. As a result, the final stored volume plus flood losses for the decentralized wetlands scenarios was greater than that of the centralized wetland scenario (i.e. more water was lost to and remains in the drainage network at the end of the simulation compared to the centralized wetland scenario). Results highlight the complexity of hydrologic response in urban watersheds and the importance of considering the impact of drainage network structure in catchment scale urban flood studies (Meierdiercks et al. 2010).

The impacts on pollutant concentration of centralized versus decentralized wetlands can be complicated due to the strong correlation between pollutant concentration and flow volumes. Reductions in pollutant concentration may be a result of reduced pollutant loads or increases in flow volumes (Musolff et al. 2021). For that reason, comparisons between pollutant loads only are discussed here. For all storm sizes, the decentralized wetlands reduce TN and TDS loads about the same as or moderately more efficiently than the centralized wetland scenario. Here the load reductions are tied to reductions in flood volumes: the percentage change in TN and TDS loads between the centralized and decentralized wetland scenarios are similar to the changes in runoff ratios. However, NO₃⁻ and TP loads increase for the decentralized wetlands scenario, even as flood volumes decrease. Because wetlands more efficiently remove NO₃⁻ and TP compared to TN and TDS (Table 3), it is advantageous for runoff to be routed through them. The large centralized wetland treated all of the runoff from a 0.39 km² upstream area, while only runoff from 0.13 km² was treated by the decentralized wetlands. As with peak flows, centralized stormwater management was most effective at removing NO₃⁻ and TP from the catchment when located in higher flow accumulation areas.

4.2 Decentralized detention based versus infiltration based GI

For the study catchment, when comparing the decentralized wetlands (detention) versus decentralized green roofs (retention or infiltration), as expected, the green roof scenario reduced volumes more than either of the decentralized wetlands scenarios for all storm sizes. With the reductions in volume, for all storm sizes, the decentralized green roof scenario resulted in similar or smaller peak discharges and similar or longer times to peak compared to the decentralized wetlands scenarios. For water quality, the retention based GI, which promotes infiltration and volume reductions, resulted in similar or better water quality outcomes than the decentralized wetlands for TN, TDS, and TP. However, NO₃⁻ loads were more efficiently removed by the decentralized wetlands scenarios. Similar to above, wetlands are particularly efficient at removing NO₃⁻ (Table 3, Leisenring et al. 2014; Hansen et al. 2018) thus a greater load of NO₃⁻ can be removed from the system when runoff flows through wetlands rather than infiltrating.

A second model scenario examines distributed infiltration based stormwater management. The soil management scenario keeps the percentage of impervious surfaces the same as the other scenarios, but infiltration is promoted in the pervious portions of the catchment by increasing hydraulic conductivity by a factor of 10. The time to peak for the soil management scenario was shorter than any of the other model scenarios. The impact on peak flows was comparable to the decentralized wetlands and green roof scenarios for the 1 y storm, but peaks were reduced more as the storm size increased. The soil management scenario also resulted in the greatest reduction in runoff volumes compared to all other scenarios. The impact on water quality for the 1 y storm is similar to that of the green roof scenario. But as storm size increased from the 1 y to 10 y and 100 y storms, the soil management scenario resulted in the greatest water quality improvements for all pollutants, including NO₃⁻. Results suggest that, in the study catchment, urban soils can impact hydrologic response and that even when opportunities to treat urban impervious areas with GI are limited, there may be similar or greater benefits gained by improving infiltration in the pervious portions of a catchment.

4.3 Study limitations and future work

One limitation of this study is that a single concentration was used to represent each pollutant, when in reality pollutant concentrations change over time and over the course of a single storm (Jefferson et al. 2017). As high frequency sensor data for water pollutants become more readily available (Kaushal et al. 2021), it will become more feasible to better model water quality characteristics at finer time scales. While the focus of this study was on the storm event response due to stormwater wetlands and green roofs in the Kromma Kill subwatershed, results would likely vary if different time scales, pollutant types, hydrological parameters, GI type, wetland design characteristics, and development patterns were examined. Given the gaps in understanding of the impacts of GI at the catchment scale, particularly on water quality, future work should examine catchment level impacts of GI across a wide range of study areas and model scenarios.

5 Conclusion

While there has been a lot of research into the performance of individual GI projects, there are still many uncertainties about their performance at the catchment scale (Jefferson et al. 2017; Golden and Hoghooghi 2018). The objectives of this study were (1) to use a stormwater model to examine the impacts of GI (green roofs, wetlands and soil management best practices) on water quality and flooding in the study catchment; and (2) to identify performance differences between centralized versus decentralized GI in the study catchment. A US EPA SWMM v5.1 stormwater model was built from field observations and geospatial datasets. Field observations of rainfall were used as model forcing, and streamflow and water quality measurements were used to calibrate the model. The model was then modified to represent several design scenarios where a GI strategy was added to or removed from the catchment and the impact on water quantity and quality was examined.

Results suggest that, in the study catchment, the large centralized wetland best reduced and slowed flood peaks compared to the decentralized GI scenarios. But for flood volumes, infiltration based decentralized GI performed best. Reductions in flood volumes are tied to increases in infiltration from impervious surfaces but also changes in drainage network structure and modifications to the pervious portions of the catchment. Even when storage capacity and the area of impervious surfaces treated were kept constant, decentralized wetlands reduced flood volumes more than a large centralized one due to attenuation and dispersal of decentralized wetland outflow within the storm pipe network. And when infiltration in the pervious portion of the catchment was promoted, runoff volumes were reduced in the catchment even more than in scenarios with smaller effective impervious percentages. For water quality, while reductions in flood volumes do generally result in water quality benefits, for pollutants that are particularly efficiently removed by wetlands, such as NO₃⁻, it was more advantageous to route runoff through a large centralized wetland rather than letting it infiltrate where it fell.

As the trend to favour distributed GI continues (Jefferson et al. 2017), it will be important for watershed managers to recognize that for multi-objective management plans, distributed GI may not be able to fully replace centralized stormwater management practices. Furthermore, while minimizing effective imperviousness my be promoted as a one-size-fits-all solution to reducing flooding and improving water quality (Meierdiercks et al. 2017), drainage network structure and infiltration in pervious portions of the catchment should also be considered.

Acknowledgments

The authors would like to thank three anonymous reviewers whose suggestions greatly improved this paper. The authors would also like to thank Hailey Beers, Daniel Winters, and Manuel Pena for assistance with field data collection, and Daniel Winters for help with model development. The research was supported by the New York State Water Resources Institute (a portion of the manuscript was submitted to the NYSWRI as a technical report associated with this funding) and the New York State Department of Environmental Conservation Hudson River Estuary Program in cooperation with the New England Interstate Water Pollution Control Commission.

References

1. Ahiablame, L.M., B.A. Engel, and I. Chaubey. 2012. “Effectiveness of low impact development practices: literature review and suggestions for future research.” Water, Air, & Soil Pollution 223 (7): 4253-4273. https://doi.org/10.1007/s11270-012-1189-2
2. Antolini, F., and E. Tate. 2021. “Location matters: A framework to investigate the spatial characteristics of distributed flood attenuation.” Water 13 (19): 2706.
3. Carvallo Aceves, M., M. Fuamba, and J. Angui. 2017. “Water quality performance of in-series BMPs.” Journal of Water Management Modeling 26: C430. https://doi.org/10.14796/JWMM.C430
4. Center for Watershed Protection (CWP). 2007. “National pollutant removal performance database.“ Technical report, CWP.
5. Chowdhury, R.K., and J.S. Abaya. 2018. “An experimental study of greywater irrigated green roof systems in an arid climate.” Journal of Water Management Modeling 26: C437. https://doi.org/10.14796/JWMM.C437
6. Coutts, C., and M. Hahn. 2015. “Green infrastructure, ecosystem services, and human health.” International Journal of Environmental Research and Public Health 12 (8): 9768–98. https://doi.org/10.3390/ijerph120809768
7. Di Vittorio, D., and L. Ahiablame. 2015. “Spatial translation and scaling up of low impact development designs in an urban watershed.” Journal of Water Management Modeling 23: C388. https://doi.org/10.14796/JWMM.C388
8. Dietz, M.E. 2007. “Low impact development practices: A review of current research and recommendations for future directions.” Water, Air, and Soil Pollution 186 (1): 351–63. https://doi.org/10.1007/s11270-007-9484-z
9. Dietz, M.E., and J.C. Clausen. 2008. “Stormwater runoff and export changes with development in a traditional and low impact subdivision.” Journal of Environmental Management 87 (4): 560–6. https://doi.org/10.1016/j.jenvman.2007.03.026
10. Fahy, B., and H. Chang. 2019. “Effects of stormwater green infrastructure on watershed outflow: Does spatial distribution matter?” International Journal of Geospatial and Environmental Research 6 (1): 5.
11. Fletcher, T.D., W. Shuster, W.F. Hunt, R. Ashley, D. Butler, S. Arthur, S. Trowsdale, S. Barraud, A. Semadeni-Davies, J-L Bertrand-Krajewski, P.S. Mikkelsen, G. Rivard, M. Uhl, D. Dagenais, and M. Viklander. 2015. “SUDS, LID, BMPs, WSUD and more: The evolution and application of terminology surrounding urban drainage.” Urban Water Journal 12 (7): 525–42. https://doi.org/10.1080/1573062X.2014.916314
12. Golden, H.E., and N. Hoghooghi. 2018. “Green infrastructure and its catchment‐scale effects: An emerging science.” Wiley Interdisciplinary Reviews: Water 5 (1): e1254. https://doi.org/10.1002/wat2.1254
13. Hamouz, V., and T.M. Muthanna. 2018. “Modelling of green and grey roofs in cold climates using EPA’s stormwater management model.” In International Conference on Urban Drainage Modelling, 385–91. Springer, Cham. https://doi.org/10.1007/978-3-319-99867-1_65
14. Hansen, A.T., C.L. Dolph, E. Foufoula-Georgiou, and J.C. Finlay. 2018. “Contribution of wetlands to nitrate removal at the watershed scale.” Nature Geoscience 11 (2): 127–32. https://doi.org/10.1038/s41561-017-0056-6
15. Hathaway, A.M., W.F. Hunt, and G.D. Jennings. 2008. “A field study of green roof hydrologic and water quality performance.” Transactions of the ASABE 51 (1): 37–44. https://doi.org/10.13031/2013.24225
16. Hood, M.J., J.C. Clausen, and G.S. Warner. 2007. “Comparison of stormwater lag times for low impact and traditional residential development.” Journal of the American Water Resources Association (JAWRA) 43 (4): 1036–46. https://doi.org/10.1111/j.1752-1688.2007.00085.x
17. James, W., L.A. Rossman, and W.R.C. James. 2010. User’s guide to SWMM 5. Guelph: CHI.
18. Javier, J.R.N., J.A. Smith, K.L. Meierdiercks, M.L. Baeck, and A.J. Miller. 2007. “Flash flood forecasting for small urban watersheds in the Baltimore metropolitan region.” Weather and Forecasting 22 (6): 1331–44. https://doi.org/10.1175/2007WAF2006036.1
19. Jefferson, A.J., A.S. Bhaskar, K.G. Hopkins, R. Fanelli, P.M. Avellaneda, and S.K. McMillan. 2017. “Stormwater management network effectiveness and implications for urban watershed function: A critical review.” Hydrological Processes 31 (23): 4056–80. https://doi.org/10.1002/hyp.11347
20. Kaushal, S.S., G.E. Likens, M.L. Pace, J.E. Reimer, C.M. Maas, J.G. Galella, R.M. Utz, S. Duan, et al.. 2021. “Freshwater salinization syndrome: From emerging global problem to managing risks.” Biogeochemistry 1–38. https://doi.org/10.1007/s10533-021-00784-w
21. Koch, B.J., C.M. Febria, M. Gevrey, L.A. Wainger, and M.A. Palmer. 2014. “Nitrogen removal by stormwater management structures: A data synthesis.” Journal of the American Water Resources Association Resources Association (JAWRA) 50 (6): 1594–1607. https://doi.org/10.1111/jawr.12223
22. Leisenring, M., J. Clary, and P. Hobson. 2014. “International stormwater best management practices (BMP) database pollutant category statistical summary report: Solids, bacteria, nutrients, and metals.” Technical report, International Stormwater BMP Database.
23. Leopold, L. 1968. Hydrology for urban land planning; A guidebook on the hydrologic effects of urban land use. Reston, VA: U.S. Geological Survey. Circular 554.
24. Lim, T.C., and C. Welty. 2017. “Effects of spatial configuration of imperviousness and green infrastructure networks on hydrologic response in a residential sewershed.” Water Resources Research 53 (9): 8084–104.
25. Loperfido, J.V., G.B. Noe, S.T. Jarnagin, and D.M. Hogan. 2014. “Effects of distributed and centralized stormwater best management practices and land cover on urban stream hydrology at the catchment scale.” Journal of Hydrology 519: 2584–95.
26. Meierdiercks, K.L., M.B. Kolozsvary, K.P. Rhoads, M. Golden, and N.F. McCloskey. 2017. “The role of imperviousness and landscape characteristics in controlling runoff quantity and quality in small urban watersheds.” Hydrological Processes 31: 4384–97. https://doi.org/10.1002/hyp.11367
27. Meierdiercks, K.L, J.A. Smith, M.L. Baeck, and A.J. Miller. 2010. “Analyses of urban drainage network structure and its impact on hydrologic response.” Journal of the American Water Resources Association (JAWRA) 46 (5): 932–43. https://doi.org/10.1111/j.1752-1688.2010.00465.x
28. Musolff, A., Q. Zhan, R. Dupas, C. Minaudo, J.H. Fleckenstein, M. Rode, J. Dehaspe, and K. Rinke. 2021. “Spatial and temporal variability in concentration–discharge relationships at the event scale. Water Resources Research 57 (10): e2020WR029442.
29. Palla, A., and I. Gnecco. 2015. “Hydrologic modeling of low impact development systems at the urban catchment scale.” Journal of Hydrology 528: 361–8.
30. Pennino, M.J., R.I. McDonald, and P.R. Jaffe. 2016. “Watershed scale impacts of stormwater green infrastructure on hydrology, nutrient fluxes, and combined sewer overflows in the mid-Atlantic region.” Science of the Total Environment 565: 1044–53. https://doi.org/10.1016/j.scitotenv.2016.05.101
31. Rossman, L.A. 2010. “Modeling low impact development alternatives with SWMM.” Journal of Water Management Modeling 18: R236-11. https://doi.org/10.14796/JWMM.R236-11
32. Russwurm, I.L., B.G. Johannessen, A.S. Gragne, J. Lohne, T.M. Muthanna, and E.G. Fridthjofsdottir. 2018. “Modelling green roof detention performance in cold climates.” In HIC 2018 (EPiC Series in Engineering, vol. 3) edited by G. La Loggia, G. Freni, V. Puleo and M. De Marchis, 1804–13. https://doi.org/10.29007/4rp8
33. Schueler, T.R. 2003. Impacts of impervious cover on aquatic systems. Ellicott City, MD: Center for Watershed Protection (CWP). Watershed Protections Research Monograph No. 1.
34. Shuster, W.D., R.A. Darner, L.A. Schifman, and D.L. Herrmann. 2017. “Factors contributing to the hydrologic effectiveness of a rain garden network (Cincinnati OH USA).” Infrastructures 2 (3): 11. https://doi.org/10.3390/infrastructures2030011
35. Shuster, W., and L. Rhea. 2013. “Catchment scale hydrologic implications of parcel-level stormwater management (Ohio USA).” Journal of Hydrology 485:177–87. https://doi.org/10.1016/j.jhydrol.2012.10.043
36. Smith, B.K., J.A. Smith, M.L. Baeck, and A.J. Miller. 2015. “Exploring storage and runoff generation processes for urban flooding through a physically based watershed model.” Water Resources Research 51: 1552–1569. https://doi.org/10.1002/2014WR016085
37. Sychterz, A., T. Schill, and B. Verspagen. 2014. “Monitoring performance of low impact development measures implemented at the Conestoga College South Campus.” Journal of Water Management Modeling 22: C371. https://doi.org/10.14796/JWMM.C371
38. United States Environmental Protection Agency (US EPA). 2021. What is green infrastructure? https://www.epa.gov/green-infrastructure/what-green-infrastructure
39. Worthen, L.L., and C.I. Davidson. 2018. “Sensitivity analysis using the SWMM LID Control for an extensive green roof in Syracuse, NY.” International Building Physics Conference 29. https://doi.org/10.14305/ibpc.2018.ps29
40. Yang, B., and S. Li. 2013. “Green infrastructure design for stormwater runoff and water quality: Empirical evidence from large watershed scale community developments.” Water 5 (4): 2038–57. https://doi.org/10.3390/w5042038

Appendix

Table A1 Results of SWMM water quality modeling for the 10 y design storm; negative percentage change values indicate water quality improvements.

Model Scenario	TN	TDS	NO3−	TP
Outflow concentration (mg/L)
Business as Usual (large, central wetland)	5.17	434.22	1.60	0.43
Control (no wetland, no GI)	5.19	436.38	3.54	0.64
Decentralized wetlands	5.80	488.33	3.39	0.63
Decentralized wetlands (upstream imperviousness)	5.33	448.67	2.71	0.51
Decentralized green roofs	4.10	342.61	2.77	0.50
Soil management	5.28	447.75	3.67	0.66
Outflow concentration % change
Business as Usual (large, central wetland)	–	–	–	–
Control (no wetland, no GI)	0.5%	0.5%	120.9%	48.8%
Decentralized wetlands	12.4%	12.5%	111.0%	48.1%
Decentralized wetlands (upstream imperviousness)	3.2%	3.3%	69.1%	19.0%
Decentralized green roofs	−20.6%	−21.1%	72.5%	16.0%
Soil management	2.3%	3.1%	128.8%	54.5%
Outflow loading (lbs)
Business as Usual (large, central wetland)	69.3	5829.3	21.5	5.7
Control (no wetland, no GI)	70.5	5925.2	48.1	8.6
Decentralized wetlands	72.1	6063.1	42.0	7.9
Decentralized wetlands (upstream imperviousness)	66.7	5614.6	34.0	6.4
Decentralized green roofs	44.2	3694.7	29.8	5.4
Soil management	24.0	2035.5	16.7	3.0
Outflow loading % change
Business as Usual (large, central wetland)	–	–	–	–
Control (no wetland, no GI)	1.6%	1.6%	123.4%	50.5%
Decentralized wetlands	3.9%	4.0%	95.2%	37.0%
Decentralized wetlands (upstream imperviousness)	−3.8%	−3.7%	57.6%	10.9%
Decentralized green roofs	−36.2%	−36.6%	38.6%	−6.8%
Soil management	−65.4%	−65.1%	−22.5%	−47.7%

Table A2 Results of SWMM water quality modeling for the 100 y design storm; negative percentage change values indicate water quality improvements.

Model Scenario	TN	TDS	NO3−	TP
Outflow concentration (mg/L)
Business as Usual (large, central wetland)	5.24	441.66	1.69	0.44
Control (no wetland, no GI)	5.28	445.20	3.62	0.65
Decentralized wetlands	5.84	487.31	3.31	0.62
Decentralized wetlands (upstream imperviousness)	5.10	429.17	2.75	0.51
Decentralized green roofs	4.28	360.16	2.92	0.52
Soil management	5.23	438.37	3.58	0.64
Outflow concentration % change
Business as Usual (large, central wetland)	–	–	–	–
Control (no wetland, no GI)	0.7%	0.8%	114.7%	46.9%
Decentralized wetlands	11.5%	10.3%	96.4%	39.6%
Decentralized wetlands (upstream imperviousness)	−2.6%	−2.8%	63.1%	14.6%
Decentralized green roofs	−18.2%	−18.5%	73.1%	18.3%
Soil management	−0.2%	−0.7%	111.9%	45.1%
Outflow loading (lbs)
Business as Usual (large, central wetland)	149.6	12611.6	48.2	12.7
Control (no wetland, no GI)	153.5	12953.0	105.4	19.0
Decentralized wetlands	152.4	12719.8	86.5	16.2
Decentralized wetlands (upstream imperviousness)	133.6	11248.6	72.1	13.3
Decentralized green roofs	100.3	8433.4	68.4	12.3
Soil management	63.5	5330.2	43.5	7.8
Outflow loading % change
Business as Usual (large, central wetland)	–	–	–	–
Control (no wetland, no GI)	2.7%	2.7%	118.8%	49.7%
Decentralized wetlands	1.9%	0.9%	79.5%	27.6%
Decentralized wetlands (upstream imperviousness)	−10.6%	−10.8%	49.7%	5.2%
Decentralized green roofs	−32.9%	−33.1%	41.9%	−3.0%
Soil management	−57.5%	−57.7%	−9.8%	−38.2%