Content uploaded by Theodore C Lim
Author content
All content in this area was uploaded by Theodore C Lim on Dec 04, 2018
Content may be subject to copyright.
Available via license: CC BY
Content may be subject to copyright.
ORIGINAL RESEARCH
published: 04 December 2018
doi: 10.3389/fbuil.2018.00071
Frontiers in Built Environment | www.frontiersin.org 1December 2018 | Volume 4 | Article 71
Edited by:
Franco Montalto,
Drexel University, United States
Reviewed by:
Ahmed M. ElKenawy,
Mansoura University, Egypt
Ioannis N. Daliakopoulos,
Technological Educational Institute of
Crete, Greece
*Correspondence:
Theodore C. Lim
tclim@vt.edu
Specialty section:
This article was submitted to
Hydrosphere,
a section of the journal
Frontiers in Built Environment
Received: 31 July 2018
Accepted: 14 November 2018
Published: 04 December 2018
Citation:
Lim TC and Welty C (2018) Assessing
Variability and Uncertainty in Green
Infrastructure Planning Using a
High-Resolution Surface-Subsurface
Hydrological Model and
Site-Monitored Flow Data.
Front. Built Environ. 4:71.
doi: 10.3389/fbuil.2018.00071
Assessing Variability and Uncertainty
in Green Infrastructure Planning
Using a High-Resolution
Surface-Subsurface Hydrological
Model and Site-Monitored Flow Data
Theodore C. Lim 1,2
*and Claire Welty 3,4
1Virginia Tech, Blacksburg, VA, United States, 2Department of Urban Affairs and Planning, School of Public and International
Affairs, Blacksburg, VA, United States, 3Chemical, Biochemical, and Environmental Engineering, University of Maryland,
Baltimore, MD, United States, 4Center for Urban Environment Research and Education, Baltimore, MD, United States
Green infrastructure (GI) is increasingly being used in urban areas to supplement the
function of conventional drainage infrastructure. GI relies on the “natural” hydrological
processes of infiltration and evapotranspiration to treat surface runoff close to where
it is generated, alleviating loading on the conventional infrastructure systems. This
research addresses growing interest in identification and quantification of uncertainties
with distributed, infiltration-based stormwater control measures, retrofitted on private
and public properties and in right-of-ways in existing urban areas. We identify four
major sources of variability and uncertainty in cumulative performance of systems that
rely on extensive implementation of distributed GI: non-additive effects of individual
best management practices (BMPs) at the catchment scale; the spatial configuration
of fine-scale land use and land cover changes; performance changes due to
climate change; and noise levels present in urban flow monitoring programs. Using
a three-dimensional coupled surface-subsurface hydrological model of a residential
sewershed in Washington DC, we find that prolonged, large-magnitude rain events affect
various spatial configurations of GI networks differently. Runoff peaks and volumes can
both be influenced by the spatial permutations of infiltration opportunities in addition
to the absolute magnitude of treated area. However, the magnitude of the last source
of uncertainty—noise levels in urban flow monitoring programs—may be larger than
sources of variability associated with spatial changes in fine-scale land use and land
cover. Changes associated with climate change– more frequent and larger rainfall
events– will likely intensify performance differences between spatial configurations of GI
but also increase noise levels in urban flow monitoring programs.
Keywords: Green infrastructure, variability, uncertainty, fine-scale land cover, climate change, ParFlow
Lim and Welty Uncertainty in Green Infrastructure Planning
INTRODUCTION
The use of Green Infrastructure (GI) for storm
water management has been steadily gaining traction in
US cities, where the US Environmental Protection Agency
(EPA) has formally recognized its role in supplementing aging
conventional sewer and stormwater drainage infrastructure.
Instead of channeling stormwater runoff from development
into conveyance structures (pipes) away from development as
quickly as possible, the purpose of GI is to intercept runoff close
to where it is generated. Using amended soils and vegetation
to slowly infiltrate water into the subsurface, evapotranspire
it through vegetation back into the atmosphere, or at least
provide storage and retention enough to mitigate storm
hydrograph peaks, GI supplements the designed capacity of
the conventional infrastructure to prevent drainage system
surcharging and combined sewer overflows (CSOs). GI Best
Management Practices (BMPs) are also referred to as Low
Impact Development (LID) BMPs or Sustainable Urban
Drainage Systems (SUDS). Examples of BMPs include: rain
gardens, bioswales, pervious pavement, tree wells/trenches, and
rain barrels. Such BMPs are typically engineered or professionally
designed to various extents–ranging from engineer-stamped
construction documents, to “Do-It-Yourself” residential
installations. Below, we review the literature including all
Stormwater Control Measures (SCMs) that incorporate
functions such as infiltration, detention, and retention, and
spatially are either centralized (e.g., one detention pond serving
a whole subdivision) or distributed throughout the landscape.
We refer to “GI BMPs” as a subset of SCMs that are typically
smaller-scale interventions distributed throughout the landscape
(e.g., one or multiple installed per property).
Although the concept of GI is well-accepted, both from
engineering and community development perspectives, there still
exists considerable uncertainty in whether extensive GI plans can
achieve the regulatory goals for which they are being deployed
in a measurable way. In this study, we develop exploratory and
anticipatory scenarios to explore four major areas of uncertainty
in widespread GI network planning: network effectiveness, fine-
scale land use and land cover changes, climate uncertainty, and
signal detection in noisy urban hydrologic datasets. We use a
residential sewershed in Washington DC retrofitted extensively
with GI to explore the potential non-additive cumulative effects
of infiltration and its measurement through two commonly
used urban surface runoff metrics: total event runoff volume
and peak flow. Non-additive cumulative effects of infiltration
within a drainage area could occur when surface conditions
interact with subsurface conditions. An example of this would
be when pervious areas are saturated and contribute runoff to
impervious areas instead of infiltrating runoff. Monitoring data
from the study residential sewershed, collected before and after
GI construction between 2009 and 2015, were used to calibrate
a three dimensional coupled surface-subsurface hydrological
model, ParFlow.CLM. In the following section, we review four
areas of uncertainty in GI planning. While our review focuses
mostly on the US conditions relevant to our study area, the broad
areas of uncertainty—network effectiveness, sub-parcel scale land
cover change, climate, and monitoring—can be applied to many
other geographies of similar climate and density of development.
BACKGROUND: SOURCES OF
UNCERTAINTY IN GREEN
INFRASTRUCTURE PLANNING
Network Effectiveness
There is extensive research that uses inflow and outflow
monitoring and before-after, control-treatment showing that GI
is effective at the site scale in reducing peak flows and runoff
volumes and improving water quality from rainfall events (Davis,
2007, 2008; Emerson and Traver, 2008; Li et al., 2009; Driscoll
et al., 2015; Page et al., 2015). There are several reasons why
the sub-catchment-scale may exhibit non-additive cumulative
effects of individual SCMs. First, SCMs are usually designed to
meet specific criteria, for example, matching a theoretical pre-
development peak runoff or volume generated from the site.
However, they may not take into consideration other changes to
the hydrologic response, such as changes in evapotranspiration,
and overall catchment storage (Li et al., 2017). Second, the
chosen metric usually corresponds to a prescribed “design
storm.” For example, Washington DC requires stormwater SCMs
to maintain peak discharge from the 2-year storm to pre-
development conditions, and therefore SCMs are not designed
to mitigate all events equally. Third, SCMs or BMPs are
often designed and constructed in a decentralized way, site-
by-site, as opportunities arise. Because they are distributed
throughout the landscape and implemented incrementally over
long periods of time, they may not take into consideration how
the timing of runoff hydrographs from multiple sites interact
with each other synergistically in a cumulative downstream
response (Emerson et al., 2005; Voter and Loheide, 2018).
Lastly, individual site interventions can usually be chosen from
a palette of acceptable BMPs, which may be intended to perform
different kinds of runoff mitigation, for example, infiltration
or detention, that will have various effects on the hydrologic
regime.
Jefferson et al. (2017) and Li et al. (2017) published thorough
reviews and analyses of the literature on the network effectiveness
of Stormwater Control Measures (SCMs), which can include
more conventional, centralized retention and detention facilities,
function at the catchment scale. They reveal that many empirical
studies confirm that SCMs increase the minimum rainfall depth
required to produce runoff (Hood et al., 2007; Loperfido et al.,
2014; Fanelli et al., 2017) and decrease peak flows from urban
development (Booth and Jackson, 1997; Meierdiercks et al., 2010;
Smith et al., 2015). In addition, although modeling studies have
shown that both peak volumetric flow rate and total volumes
have been found to decrease with SCMs (e.g., Perez-Pedini et al.,
2005; Avellaneda et al., 2017), there is less empirical consensus
on the network effect of GI on total runoff volumes. Empirical
studies evaluating the capability of SCMs to control total runoff
volumes to nearby streams have shown little to no effect (Booth
and Jackson, 1997; Dietz and Clausen, 2008; Meierdiercks et al.,
2010; Shuster and Rhea, 2013), and that limitations in volume
Frontiers in Built Environment | www.frontiersin.org 2December 2018 | Volume 4 | Article 71
Lim and Welty Uncertainty in Green Infrastructure Planning
mitigation are particularly apparent during larger rain events
(Woznicki et al., 2018).
The particular processes within urbanized catchments that
could lead to differences in cumulative effect have also been
a focus of much research. In many of these studies, both
conventional drainage infrastructure and green (infiltration-
based) infrastructure play a mediating role in determining the
dominant hydrological response to rainfall. For example water
leakages into and out of urban infrastructures, and changes
in evapotranspiration due to vegetation changes, can have
major effects on changes in stream baseflows (Bhaskar et al.,
2016). Infiltration could impact the function of conventional
drainage infrastructure to effectively convey runoff away from
development or cause combined sewer overflows of untreated
sewage into natural water bodies (Endreny and Collins, 2009;
Maimone et al., 2011). The kinds of interactions between
conventional infrastructure and surface-subsurface hydrologic
dynamics are also dependent on the size of the rainfall event.
For example, one study showed how disconnecting storm drains
under lower magnitude events results in the expected decrease in
peak flow, but that antecedent wetness of soil due to increased
infiltration can also result in runoff production in areas not
served by storm drains, a counter-intuitive hydrologic response
(Tague and Pohl-Costello, 2008).
Modeling of urban runoff has been dominated by the
Hortonian concept of runoff generation. In this model
framework, runoff is formed when infiltration rates are exceeded
by rainfall rates. However, the above examples highlight why
a more flexible conceptual model that incorporates surface-
subsurface and infrastructure interactions may be necessary.
The Urban Variable Source Area (UVSA) and watershed
capacitance conceptual model can help organize the conditions
when specific properties of the urbanized catchment, including
soil permeability, slope, depth to groundwater, land use and
land cover, and availability and placement of infiltration
opportunities, will influence how runoff is generated under
different meteorological and morphological conditions, as is
shown in Figure 1 (Miles and Band, 2015; Lim, 2016). The above
examples and the UVSA conceptual model suggests that GI
may exhibit a trade-off in cumulative effectiveness during very
wet conditions, multiday events, when watershed capacitance is
limited, or when infiltration opportunities are clustered together
in high flow accumulation areas.
In this research, we take a simulation approach to modeling
the complex dynamics that could result in spatially and
temporally variable runoff generation in an urbanized
catchment. In order to capture times when infiltration of
runoff to groundwater could result in runoff onto impervious
surfaces, we chose the simulation model ParFlow.CLM, for
its capability to represent negative feedbacks and surface-
subsurface and lateral subsurface interactions among infiltration
opportunities. Figure 2 conceptually shows how ParFlow.CLM
has the capability to model potential surface-subsurface
interactions, compared to two other models, the US EPA’s Storm
Water Management Model, and the Regional Hydro-Ecological
Simulation System (RHESSys) model.
Variability Due To Sub-parcel-Scale Land
Use Change/Cover Change
The urban landscape is constantly undergoing physical and
social change. While many studies investigate the effects of
conversion of agricultural or forested land covers to urban
land cover, changes in the landscape also occur within the
urban boundary. Since the 1990s, US cities show measurable
signs of infill and redevelopment (Schneider and Woodcock,
2008; Nowak and Greenfield, 2012). In addition, residential
and commercial landowners’ landscape preferences may change
over time. Research has shown that landscape aesthetics are
often mimicked among neighbors. Yard landscaping practices,
for example, choice in vegetation types (grass, native plants,
xeriscaping, etc.), are expressions of personal preferences
as well as functions of historical and social norms. There
is evidence that front-yard landscaping is susceptible to
social influence among neighbors and has a distinct spatial
structure (Zmyslony and Gagnon, 1998). Landscaping decisions
have also been correlated with income levels, suggesting
that as neighborhood socioeconomic status (or developers’
anticipation of socioeconomic status) changes, so too might
the landscape (Larsen and Harlan, 2006; Troy et al., 2007).
Abandonment and vacancy also provide opportunities for land
cover change in urban areas, as more cities consider the co-
benefits that could be provided on vacant parcels, including
urban green space provision, urban farming, and community
revitalization, though sometimes such uses are not permanent
(Tzoulas et al., 2007; Schilling and Logan, 2008; Heckert and
Mennis, 2012; Drake and Lawson, 2014; Schifman et al.,
2017).
Influencing private property owner behavior to promote
environmental sustainability is a goal of urban and
environmental planners and stormwater infrastructure
managers. In contrast to conventional stormwater drainage
infrastructure– such as pipes and cisterns– that is located
in the public right-of-way (ROW) or on public property,
GI can be implemented on both public or private property.
Since most land in US cities is privately owned, stormwater
infrastructure managers tend to view implementation of green
infrastructure on private land as an opportunity to integrate
stormwater management goals with landscaping practices,
potentially providing environmental amenity and infrastructure
improvement goals at lower costs to the city (Montalto et al.,
2007; Roy et al., 2008; Keeley et al., 2013).
As a top-down measure, stormwater utility managers use
subdivision and land development ordinances (SALDOs) to
compel private property owners to adopt BMPs on their
properties, or they can encourage adoption through voluntary
means. SALDOs can typically only be applied to properties
undergoing construction or significant reconstruction. SALDOs
stipulate that the owners must implement stormwater control
measures on their properties to meet some design standard (for
example, store and treat runoff resulting from the properties
impervious surfaces for the 1-year, 24-h rain event). However,
in order to speed the adoption of green infrastructure practices
on private property above the redevelopment rate, some
Frontiers in Built Environment | www.frontiersin.org 3December 2018 | Volume 4 | Article 71
Lim and Welty Uncertainty in Green Infrastructure Planning
FIGURE 1 | Conceptual model of factors influencing areas contributing runoff in an urbanized catchment area (adapted from Lim, 2016). Conventional and green
infrastructure placement and density mediates balance between dominance of Hortonian flow, saturation overland flow, and subsurface stormflow processes at the
catchment scale. Red dots indicate processes that are hypothesized to be dependent on antecedent wetness conditions and multiday events.
FIGURE 2 | Conceptualization of various hydrological models’ treatment of overland flow routing and groundwater and example testable hypotheses.
infrastructure managers are trying more innovative means of
promoting land use and land cover change. Such examples
include the use of stormwater fee systems where residents receive
credits to their bills for constructing green infrastructure on
their properties, subsidies for GI installation, and partnerships
with non-profit groups, education and community development
programs (Kertesz et al., 2014; Valderrama and Davis, 2015).
Such programs speed the adoption of GI on private properties by
providing economic incentives for action, increasing knowledge
about environmental impacts, or promoting the flow of
Frontiers in Built Environment | www.frontiersin.org 4December 2018 | Volume 4 | Article 71
Lim and Welty Uncertainty in Green Infrastructure Planning
information about subsidy programs (Ando and Freitas, 2011;
Green et al., 2012; Londoño Cadavid and Ando, 2013;
Montalto et al., 2013; Lim, 2017). The intent of distributed
BMPs is to change hydraulic and hydrologic properties of
landscapes, by improving the permeability and storage capacity
of soils, reducing connectivity of impervious surfaces, and
increasing evapotranspiration rates from vegetation (Fletcher
et al., 2013). The extent to which such changes, based
on policy, social and physical change within urban areas,
can result in detectable improvements to various hydrologic
indicators therefore not only depends on the physical capacitance
of the watershed, but also on its social capacitance (Lim,
2017).
Climate Uncertainty
Global climate change impacts are expected to result in both
increases in total annual precipitation and intensity of extreme
rainfall events in the Northeast United States. In the past century
the total amount of precipitation across the continental US
has increased by 7%, with the largest 1% of events increasing
in frequency by 20% (US Global Change Research Program,
2009). Precipitation has increased particularly in the Northeast
US, which has seen a 58% increase in precipitation volume in
historical trends (Groisman et al., 2005). Downscaled models’
predictions of future changes in precipitation in the Mid-Atlantic
US have been shown to vary dramatically between models.
Uncertainty in precipitation projections are especially high in
summer and fall (Najjar et al., 2009, 2010).
Typically, engineered GI facilities are designed for a particular
size or frequency storm, described by a depth of precipitation
occurring within a given period of time (e.g., 24 h). If this
storm has a 10% probability of occurring in any given year it
is called the “10-year” rain event. The design of stormwater
management systems according to historical frequencies of
extreme events will become less useful under a changing climate
regime (Mailhot and Duchesne, 2010). In parts of North America,
today’s 50-year event will become the 10-year event by the
2090s (Waters et al., 2003). Sometimes, GI design requirements
are expressed as a particular depth of rainfall, or “first flush”
that BMPs must be able to handle. Acknowledging the non-
stationarity of these probabilities under conditions of climate
change, engineers apply a percentage based “safety factor” to the
required design depth that make their designs more conservative
for future conditions (Milly et al., 2008).
However, the concept of design storms, even with safety
factors does not include simulation of the continuous
meteorological conditions under which GI is intended
to perform. In the US, many municipalities/stormwater
management districts now recommend hydrological simulation
of continuous representative rainfall time series, in order to
more realistically represent antecedent wetness conditions
between successive events and multiday precipitation events.
Incorporating future climate change-influenced continuous
precipitation patterns requires a model for constructing what a
likely scenario might be.
Applying downscaling methods to the General Circulation
Models (GCMs) is one way to produce a realistic continuous
climate-change influenced precipitation record. GCMs produce
climate change forecasts given a range of future emissions
scenarios; however, they are typically produced at a resolution
too coarse to be used at the urban catchment scale. The
major methods of producing downscaled rainfall events from
climate model simulations for use in local urban drainage
system planning include: dynamic downscaling (using physical
models), empirical transfer function based methods, historical
re-sampling methods, and stochastic rainfall models (Willems
et al., 2012; Wilby et al., 2014). Downscaled inputs from different
models were shown to generate a wide range of estimates for
the local scale. The typical suggested solution to the wide range
of estimates is to conduct simulations with multiple downscaled
methods to represent the variability of the scenarios. This
solution, however, can be computationally costly and complex.
Therefore, one approach to mimic future increase in multiday
wet periods is to extract pieces of historical records that reflect
this condition (Catalano de Sousa et al., 2016). Although this
approach may be considered rather simplistic, in this case,
it was considered sufficient to explore differences between
spatial scenarios, to identify whether additional precipitation
scenarios would be necessary. Combining precipitation scenarios
that would incorporate the range of uncertainty in projected
precipitation for each tested spatial scenario would result in a
much larger number of simulations and could result in effects
that would be more difficult to tease apart.
Increased intensity and frequency of large rain events would
be expected to decrease effectiveness of infiltration-based SCMs,
since infiltration is a slow response, and requires to time recover
capacity between events (Kristvik et al., 2018). Previous modeling
studies have found that GI is typically more effective during
smaller events than larger events, and that differences associated
with GI are more apparent for peak volumetric flow reduction
than for total volume reduction (Palla and Gnecco, 2015; Fry and
Maxwell, 2017).
Measurement
Despite the increasing sophistication and accuracy of
hydrological models, empirical data collection for calibration
and validation of models is still necessary (Maheepala et al.,
2001; Silberstein, 2006). However, there is uncertainty associated
with monitoring stormwater runoff flows in urban areas.
Stormwater runoff monitoring data in urban areas can contain
high levels of noise from errors in data logger software and flow
monitoring equipment as well as from unexpected disturbances
of “experimental conditions” (Liefting and Langeveld, 2008).
Before-and-after Control-Intervention (BACI) experimental
designs are common in hydrological experiments but can be
subject to disruptions (Shuster and Rhea, 2013). Because urban
catchments are inherently full of human activities, confounding
effects such as lawn watering, car-washing or pipe and fire
hydrant leakages may disturb even the most well-planned
experiments. For ongoing flow monitoring necessary for
permitting purposes (for both separated and combined sewer
systems in the US) and demonstrating the effectiveness of GI
networks, it is important to show that hydrologic effects in GI
interventions can be expected to overcome levels of noise present
Frontiers in Built Environment | www.frontiersin.org 5December 2018 | Volume 4 | Article 71
Lim and Welty Uncertainty in Green Infrastructure Planning
in urban stormwater flow monitoring data (Lim and Welty,
2017).
Hypotheses
In this study we address the four types of uncertainty described
above explicitly through the choice of hydrological model and
processes represented, development of scenarios, design of the
evaluation criteria, and comparison with empirical monitoring
data. Table 1 summarizes relevant questions that exist in the
literature regarding widespread planning of stormwater GI:
In a previous study (Lim and Welty, 2017), data generated
from the model applied in the present study showed that,
on average, potential infiltration located in areas of high
accumulation were more effective at mitigating runoff volume
than those located in upslope areas. Lim and Welty also found
weaker evidence of a counterintuitive result that infiltration sites
in downslope areas were also more effective than those in upslope
areas soon after a previous precipitation event. In the present
study, we take a closer examination of one continuous series of
four rain events of both high magnitude and short inter-event
period.
We hypothesize the following:
•Non-additive effects of GI can be observed between
scenarios of spatially explicit impervious surface and
green infrastructure networks and differences in these non-
additive effects will be observable between scenarios, and over
multiday rain events.
•Periods of prolonged wetness and large rainfall events will
increase the localized saturation of infiltration areas within the
study sewershed. This will result in infiltration areas located
in high flow accumulation areas having decreased capability to
mitigate surface runoff, eventually becoming sources of surface
runoff.
•Clearly separable differences between scenarios will be more
apparent in the evaluation of peak flow mitigation than they
will for total runoff volume mitigation.
STUDY SITE DESCRIPTION–RIVERSMART
WASHINGTON
In this research we partnered with Washington DC’s Department
of Energy and the Environment (DOEE) on a project called
RiverSmart Washington that evaluated a monitored urban
sewershed before and after GI installation. DC’s RiverSmart
programs were established to help reduce stormwater runoff
from entering the District’s waterways and the Chesapeake
Bay and to restore ecological function to the landscape. In
2015, Washington DC’s water and wastewater utility provider,
DC Water, revised its Combined Sewer Overflow (CSO) Long
Term Control Plan (LTCP) to include GI components that
allowed it to dramatically downsize two previously planned
underground tunnels (eliminating one, and reducing the planned
capacity of the other from 220,000 cubic meters to 114,000 cubic
meters). This increased regulatory and institutional support to
better understand the physical function of GI configurations
and the effects of alternative site development morphologies
TABLE 1 | Sources of planning uncertainty and incorporation in this study.
Source of Planning
Uncertainty
Incorporation into This Study
Green infrastructure network
effectiveness
Choice of model: 3D coupled
surface-subsurface hydrological model,
ParFlow.CLM
Relative impacts of spatially
explicit fine-scale land use and
land cover changes
Development of modeled scenarios
Performance under increased
wet conditions (climate change)
Selection of simulation period and design
of scenario evaluation/comparison criteria
Role of monitoring Comparisons and study coordination with
empirical monitoring from experimental
sites
at the sewershed scale (DC Water, 2015). In particular, city-
wide initiatives to promote voluntary residential adoption of
subsidized rain gardens and permeable pavement installations
motivated a need to better understand how resulting spatial
configurations may perform compared to facilities in the right-
of-way (ROW), which may be more costly to the city.
Made possible through $4M in joint funding from the U.S.
Fish and Wildlife Service, DOEE, and DC Water, DOEE began
the RiverSmart Washington monitoring program in 2009. The
project first monitored in-pipe flows for the base case, pre-GI
condition for 6 months (from July 2010 to December 2010) as
well as local precipitation monitoring. This initial monitoring
period was followed by extensive construction of GI within
several sewersheds in DC. At the Lafayette demonstration site
(0.05 km2, and originally 34% impervious, with 15% building
footprint and 19% pavement), the District Department of
Transportation (DDOT) oversaw installation of bioretention
bump-outs and permeable pavements designed to treat nearly all
of the public ROW. In total, a total of 2340.2 m2of GI BMPs
were installed in the public ROW, treating total contributing
area of 2945.4 m2(Lim and Welty, 2017). Figure 3 shows site
photographs of BMPs constructed in the public ROW during a
rain event.
GI retrofits were also constructed on private properties by
willing residents. Residential GI BMPS included: permeable
pavers, rain gardens, native landscaping, and rain barrels. Native
landscaping and rain garden installations included amended
soils, adjusted site grading and planted native vegetation.
Permeable pavements increased permeability of impervious
surfaces and provided storage in an underlying gravel layer.
Of the 74 households within the sewershed, 25 agreed to
install subsidized GI on their properties. Private installations
disconnected over 1,400 m2of residential rooftop and over
550 m2of private paths and driveways from the stormwater
drainage system. Prior to rooftop disconnection, all rooftops
were directly connected to the pipe system in the ROW via
buried PVC pipes that drained either into the street of the
adjacent sidewalk (Lim and Welty, 2017). Additional detail on
the hydraulic properties of both the local geology (which was
measured in geotechnical reports as part of the RiverSmart
Frontiers in Built Environment | www.frontiersin.org 6December 2018 | Volume 4 | Article 71
Lim and Welty Uncertainty in Green Infrastructure Planning
FIGURE 3 | Site Photographs of BMPs treating the sewershed’s public ROWs. (A) Permeable asphalt: surface runoff is visible, indicating lower than expected
infiltration performance. (B) Bioswale with a flush curb cut extending beyond the ROW into adjacent grass strip. (C) Permeable concrete installed in the center of a
reverse crowned alley. (D) Permeable concrete spanning the full width of the ROW. (E) Foreground shows permeable rubber sidewalk adjacent to bioswale.
Permeable pavers in parking lane are visible in the background.
Washington monitoring project), and the GI retrofits can be
found in Lim and Welty (2017).
After the construction of all public ROW and voluntary
residential GI retrofits, in-pipe flows and precipitation were
monitored for the site for 6 months. Like the Base case (pre-
GI construction) monitoring, in-pipe flows were measured
using an ADS Flowshard meter that used four ultrasonic level
sensors to record stage data, a low-profile Doppler velocity
monitor and a pressure sensor. The monitoring equipment
transmitted data via a cellular communications-enabled data
logger. Figure 4 shows the land cover types present within the
sewershed boundary after the GI construction. The sewershed
is defined as the area draining to the point where the in-pipe
flow monitoring both before and after GI construction was
located.
METHODS
Model Selection
The ParFlow.CLM model was chosen for its capability to
model potential hypothesized feedbacks between locally
saturated groundwater conditions and overland flow that
is key to the UVSA model (Miles and Band, 2015). Unlike
other ecohydrological models that include feedbacks between
groundwater and surface runoff generation (e.g., RHESSys),
ParFlow.CLM is three-dimensional, and therefore can capture
spatially explicit effects of event-based local saturation. Unlike
lumped-parameter models such as the US EPA’s SWMM model,
which requires the total modeled area to be represented as a series
of subcatchments, ParFlow.CLM solves the Richards equation
via finite differences over a regular gridded discretization.
The numerical solution of Richards equation allows the user
to apply ParFlow.CLM over grids as fine as <1 m ×1 m, to
much larger grids (e.g., 500 ×500 m in regional models).
Further, ParFlow.CLM is optimized to perform on massively
parallelized high performance computers, which makes it very
efficient (Ashby and Falgout, 1996; Jones and Woodward, 2001;
Kollet and Maxwell, 2006; Maxwell, 2013). It has also been
dynamically linked to a land-atmosphere model CLM that
allows for coupled simulation between the land surface and
groundwater models (Maxwell and Miller, 2005; Kollet and
Maxwell, 2008).
Frontiers in Built Environment | www.frontiersin.org 7December 2018 | Volume 4 | Article 71
Lim and Welty Uncertainty in Green Infrastructure Planning
FIGURE 4 | Domain of the study sewershed with public and private installations of GI and monitoring locations. Building footprints and sidewalks are in brown; streets
are in gray; pervious areas are indicated by pale green; dark blue indicates public GI projects; dark green indicates private GI projects; the red arrow points to the
monitoring location.
The model was calibrated by adjusting Manning’s n for
several events during the monitored pre-GI period (2009),
comparing the modeled output with in-pipe monitored flows
from the site. More detail on calibration and model spinup and
parameterization of the ParFlow model application is provided in
Lim and Welty (2017).
Scenario Development
Scenario analyses can be divided into two major types:
exploratory scenarios, and anticipatory scenarios. Exploratory
scenarios are used to test known processes of change and past
extrapolations to define future scenarios. Anticipatory scenarios
start with desired or feared visions of the future. Anticipatory
scenarios may incorporate potential policy responses, and
expert and stakeholder-defined assumptions to frame the
subjectivity of anticipated future states (Mahmoud et al., 2009).
Neither exploratory nor anticipatory scenarios are meant to be
predictions of future states. Rather, they are meant in this study to
represent physical sensitivity of the site to realistic parameters (in
the case of the exploratory scenarios), and the physical sensitivity
of the site to potential policy and social-process driven change. In
addition to a Base scenario, eight additional scenarios were tested,
organized below into the exploratory and anticipatory types of
scenario development:
Exploratory
•GI_DRY: Treat runoff from rooftops on low flow accumulation
properties with GI
•GI_WET: Treat runoff from rooftops on high flow
accumulation properties with GI
•IS_DRY:Remove impervious surface areas on low flow
accumulation properties
•IS_WET: Remove impervious surface areas on high flow
accumulation properties
Anticipatory
•GI_ROOF: Treat runoff from all rooftops on private property
•GI_ROW: Treat the public ROW surface (equal treated area to
GI_ROOF)
•IS_DISC: Disconnect all roofs from the storm drain (drain
onto grass)
•IS_MAX: Allow maximum impervious surface area per
property according to zoning regulations
More detail and maps of scenarios can be found in Lim and Welty
(2017).
Selection of Simulation Period
After model spinup, all scenarios were initialized using a
common pressure boundary layer and simulated using a 6-
month period of meteorologic input data from 1 March 2015–1
Frontiers in Built Environment | www.frontiersin.org 8December 2018 | Volume 4 | Article 71
Lim and Welty Uncertainty in Green Infrastructure Planning
September 2015. From this 6-month period, one 10-day window
was chosen for further analysis for this study. Although the
10-day window is not derived from any downscaled GCM
scenario, it was chosen as a period of rainfall that could become
more characteristic with future climate change. This window
represented the wettest 10-day period (total cumulative rainfall)
during the summer of 2015 and contained the single largest
rain event in the modeled period (Figure 5). The total rainfall
accumulated during the four distinct rainfall events during this
period was 110 mm (34.0, 21.6, 7.6, and 47.0 mm).
Regional climate change scenarios indicate that the Mid-
Atlantic US will likely experience both increases in total rainfall
and increases in rainfall intensity (Easterling, 2000; Najjar et al.,
2009). The use of an especially wet period is similar to the
approach of used in the historical re-sampling method of testing
how future climate change scenarios might affect local hydrology
(Willems et al., 2012). In addition, Global Climate Change
models indicate that prolonged increase wetness in the Mid-
Atlantic US is likely to occur in winter months. Here, we have
chosen a wet interval from summer months hypothesizing that
summer-months infiltration capacity will be optimally recovered
between events due to increased evapotranspiration from the site.
This is meant to represent “the best” that distributed stormwater
management practices on the site can do.
Scenario Evaluation Criteria
Two convenient measures of stormwater runoff response
commonly used in land development standards are total runoff
volume and peak flow. Both these measures are derived from
event-based overland flow time series output from the ParFlow
simulations for each scenario. The method for calculating
overland flow at any point within the domain is based on
Manning’s equation:
Q=VA =1.00
nAR 2
3S1
2) (1)
where Qis volumetric flowrate (L3T−1), Vis flow velocity (LT−1),
Ais cross-sectional area (L2), nis Manning’s roughness coefficient
(TL−1/3), Ris the hydraulic radius (L), and Sis bed slope. Within
ParFlow, Manning’s equation (above) is adapted to use pressure
head calculated at any surface grid cell, so that the equation for
overland flow at that point is:
Q=dx
nP5
3S1
2(2)
where dx (L) is the horizontal resolution of the domain, and Pis
the pressure head (L) output from the three dimensional array at
the time tat the location of the grid cell. The ParFlow application
of Manning’s equation assumes that for wide channels, the
hydraulic radius can be replaced by depth, which is equivalent
to pressure head (Maxwell et al., 2016). The grid cell that was
chosen to calculate overland flow was the outlet of the sewershed,
where flow monitoring was carried out, pre- and post-installation
of GI, also referred to the sewershed’s “pour point” (red arrow
in Figure 4). All overland flow from the sewershed flows past
this point, therefore overland flow at this point is an integrated
measure of flow heterogeneity within the sewershed. Overland
flow was calculated for the entire simulation period for all nine
scenarios.
Two measures of effect of GI configuration on surface runoff
were chosen to compare the scenarios: the total volume of
runoff resulting from an event, and peak volumetric flow rate
of the event. From an infrastructure management perspective,
lower overall volumes and lower peak flows are both desirable
outcomes. This may differ from urban stream restoration
goals that may seek to restore pre-development baseflows
while mitigating flashiness (storm runoff peaks). From an
infrastructure-centric perspective, infrastructure managers are
typically trying to reduce loading on centralized drainage
infrastructures, especially those that are shared with domestic
wastewater conveyance.
Contextualizing Site Sensitivity to Noise
Levels in Monitoring Data
In order to contextualize the variation in site runoff (total
runoff volume) between each of the four events in the 10-
day period, we compared the differences in runoff volumes
between pairs of modeled scenarios to the monitored data.
Because the monitored data was taken from a period different
from the modeled scenarios, we captured the variation of the
runoff volume conditional on total event depth by calculating the
absolute width of the confidence percentile intervals estimated
from the regression of the total event volume on the total event
precipitation from the monitored precipitation and flow data
from the summer months of the pre-GI period (March–August
2010). We also included two important controls to capture the
effects of antecendent wetness on local saturation: the interevent
period (length of time between each rain event and the previous
rain event), and the interaction between the event precipitation
depth and the interevent period. Equation 3 shows the regression
specification:
runoff volumeb,t=β0,b+β1,bprcpt+β2,bintertimet
+β3,bprcpt∗intertimet+et,b(3)
where runoff volume{b,t}is the runoff volume calculated from
the empiricall monitoring data from the pre-GI conditions (Base
case) during time t(m3)prcptis the total depth of precipitation
during event t(mm), intertimetis the inter-event period in hours
between the start of event tand the end of previous event t-
1,betabare the coefficients estimated through linear regression
and e{t,b}is the error. After obtaining the coefficients through
regression, the estimated model was used to predict the linear
relationship between precipitation depth and runoff volume,
given four different interevent periods: 23, 42, 47, and 69 h
(the mean interevent period in the empirical data was 57.3 h).
These four interevent periods correspond the interevent periods
for Events 1, 4, 2, and 3 in the 10-day simulation window,
respectively.
The confidence interval of these predicted runoff volumes
represents the area in which the “true” mean runoff volume is
likely to reside, taking into account the amount of variation and
Frontiers in Built Environment | www.frontiersin.org 9December 2018 | Volume 4 | Article 71
Lim and Welty Uncertainty in Green Infrastructure Planning
FIGURE 5 | June 20, 2015–June 30, 2015. Inset shows the selected window for overland flow examination in the context of the entire simulation period, from March
1, 2015–September 1, 2015. This window includes the highest-intensity rainfall event as well as the wettest 10-day period.
number of observations in the empirical data. The width of the
confidence interval was calculated by differencing the upper and
lower confidence interval limits for the 90% confidence level.
Each of the scenarios’ runoff volumes, were then subtracted from
to runoff volumes of the base case simulation for in each event
during the 10-day period. If the event-based differences exceeded
the width of the confidence interval, then the effect of that
scenario (compared to the Base case) exceeds the boundaries of
confidence present in the monitored data, and may be noticeable
over the ordinary levels of noise in the data.
RESULTS AND DISCUSSION
Event-Based Comparisons of Peak Runoff
and Total Volume
A comparison of the overall 10-day rankings shows that IMP2
(maximum allowable imperviousness per-parcel for every parcel)
had the highest magnitude values for both max peak flow over
the 10-day period and for total volume of runoff over the 10-
day period. The max peak flow for IS_MAX (0.070 cms) is
19% greater than the max peak flows for Base (0.059 cms) and
23% greater than the IS_DISC, the scenario where all roofs
are disconnected from the ROW. During the 10-day period,
disconnecting roofs from the ROW decreased the max peak flow
by 3%, compared to Base. In the first two rain events, IS_DISC
mitigated peak flows compared to Base. However, by the third
rain event, the peak flow from IS_DISC marginally exceeded the
peak flow from Base. This suggests that the mere disconnection of
rooftop imperviousness with no provision of additional storage
in the receiving lawn area may do little to mitigate flow peaks
during multiday events, after the initial soil storage is exhausted.
A comparison between total runoff volumes between Base and
IS_DISC even show that disconnected roofs resulted in about 4%
more total runoff volume than Base, suggesting that additional
volume capture is necessary (for example through rain barrels
or rain gardens) in order for downspout disconnection to have
the desired effect on flow mitigation and illustrating how trends
for peak flow reduction do not necessarily translate into trends
for total volume reduction. The measures and rankings for each
scenario were calculated (Table 2). A visualization of changes in
ranking between events is shown in Figure 6.
Figure 6 illustrates all changes in rankings in flow peak
and total volume magnitudes that occur over the 10-day
analysis period. Examining rank crossovers (RCs) between paired
scenarios across the four events in the 10-day period allows us to
explore potential thresholds for changes in hydrological response,
and to anticipate outcomes of potential policy responses.
For the exploratory scenarios (GI_DRY/WET, IS_DRY/WET),
RCs highlight when and how previous events begin to affect
the performance of the site. For the anticipatory scenarios
(GI_ROOF/ GI_ROW, IS_DISC, and IS_MAX), RCs highlight
how an intended policy compares to an alternative under a
multiday scenario.
Peak Flow RCs and Magnitude Comparisons
Comparing paired scenario peak flow rankings (Figure 6A), the
only RC between paired scenarios occurs between Base and
IMP_DISC (discussed above). Other paired scenarios GI_ROW
and GI_ROOF, GI_DRY and GI_WET, and IS_DRY and IS_WET
all maintain consistent relative rankings: GI_ROW has lower flow
peaks than GI_ROOFS in all four events; GI_WET has lower flow
peaks than GI_DRY in all four events, and IS_DRY has lower
flow peaks than IS_WET in all four events. The RC exhibited
by IS_DISC and Base on the third event may reflect that the
subsurface storage difference between the two scenarios has been
exhausted and that delayed runoff response from previous events
in the IS_DISC scenario may also have contributed to raising the
third event peak flow.
Frontiers in Built Environment | www.frontiersin.org 10 December 2018 | Volume 4 | Article 71
Lim and Welty Uncertainty in Green Infrastructure Planning
TABLE 2 | Scenario rankings for peak flows and total event volumes for four consecutive events in 10-day window.
Event 1: 6/21/2015 Event 2: 6/24/2015 Event 3: 6/25/2015 Event 4: 6/28/2015 10-day max
cms rank cms rank cms rank cms rank cms rank
PEAK FLOWS
Base 0.059 2 0.034 2 0.019 3 0.056 2 0.059 2
GI_DRY 0.054 4 0.028 4 0.017 4 0.050 4 0.054 4
GI_WET 0.050 6 0.026 5 0.016 5 0.048 7 0.050 6
IS_DRY 0.050 7 0.026 7 0.015 7 0.048 6 0.050 7
IS_WET 0.050 5 0.026 6 0.015 6 0.048 5 0.050 5
GI_ROOF 0.048 8 0.025 8 0.015 8 0.045 8 0.048 8
GI_ROW 0.030 9 0.016 9 0.009 9 0.030 9 0.030 9
IS_DISC 0.057 3 0.032 3 0.019 2 0.056 3 0.057 3
IS_MAX 0.070 1 0.039 1 0.022 1 0.066 1 0.070 1
TOTAL RUNOFF VOLUMES
m3 rank m3 rank m3 rank m3 rank m3 rank
Base 603 3 376 3 105 3 1,043 3 2,127 3
GI_DRY 478 5 256 6 70 5 884 5 1,688 4
GI_WET 469 6 253 7 66 7 869 6 1,657 6
IS_DRY 478 4 243 8 66 6 893 4 1,680 5
IS_WET 460 7 257 5 65 8 857 7 1,639 7
GI_ROOF 437 8 227 9 58 9 822 8 1,543 8
GI_ROW 354 9 272 4 72 4 728 9 1,425 9
IS_DISC 641 2 386 2 112 2 1,085 2 2,225 2
IS_MAX 749 1 441 1 136 1 1,209 1 2,534 1
FIGURE 6 | Comparisons of magnitude rankings for (A) peak volumetric flow rates, and (B) event runoff volumes, between nine simulated scenarios, over four rainfall
events in a 10-day window.
The differences in magnitude between peaks between IS_DRY
and IS_WET were negligible (about 1% in all four events),
indicating that spatial configuration of imperviousness when
no additional storage volume is provided has limited effect
on peak flow mitigation. In contrast, differences in spatial
configuration in placement of GI treatment areas (GI_ROW vs.
GI_ROOF and GI_DRY vs. GI_WET) were as high as 66%.
Larger differences in attributed to spatial configuration were
Frontiers in Built Environment | www.frontiersin.org 11 December 2018 | Volume 4 | Article 71
Lim and Welty Uncertainty in Green Infrastructure Planning
observed between the GI_ROW and GI_ROOF scenarios, which
had 14.2 and 15.6% of the area within the sewershed treated with
GI, respectively. GI_DRY and GI_WET had smaller proportions
of their total contributing area retrofit with GI (7.3 and 8.2%,
respectively), about half of the total treated area in the GI_ROW
and GI_ROOF scenarios. This result implies that differences
in peak flow mitigation associated with spatial configuration
and placement of GI become more apparent as the total area
treated with GI increases. In Miles (2014), no differences in
streamflow were found when upslope vs. downslope roofs in a
low-medium density neighborhood were treated with GI. In that
study, residential rooftops comprised only about 7% of the total
watershed area.
Total Runoff Volume RCs and Magnitude
Comparisons
Total runoff volumes for each event exhibited several RCs
between paired spatial configuration scenarios (Figure 6B).
During the first rainfall event (34 mm) GI_ROW reduced total
runoff volumes more than GI_ROOF (by 23%), and IMP_WET
reduced total runoff volumes more than IMP_DRY (by 4%). Both
these comparisons provide evidence that spatial configuration of
GI and imperviousness matter: when run-on opportunities and
storage areas are located in more downslope areas, more runoff
volume is intercepted. However, after the first event, during the
second (21 mm) and third (7.6 mm) events, the scenarios that
provide upslope infiltration and storage opportunities mitigate
more total volumes than the scenarios that provide downslope
infiltration and storage opportunities. For GI_ROW/GI_ROOF
these differences are by 17 and 19% for events 2 and 3,
respectively. For IS_DRY/IS_WET, these differences are much
smaller: 5 and 0.4% for events 2 and 3, respectively. After
these two events, capacity is “recovered” in downslope areas,
and maximum infiltration opportunities in the downslope
configurations again realizes its advantage in intercepting more
subsurface flow during the fourth event (1.85 mm).
While peak flows were lower for the disconnected roof
scenario (IS_DISC) compared to Base in three out of four rain
events, Base had lower total volumes of runoff compared to
IS_DISC in all four rain events. The increased volume of total
runoff in each event for IS_DISC ranged between 2.7 and 6.5%
higher than the volume of total runoff for the Base scenario. No
RCs were observed for GI_DRY and GI_WET: the total runoff
volume from GI_DRY was slightly higher than the total runoff
volume from GI_WET in each rainfall event during the multiday
period (ranging between 1.6 and 6.3% larger volumes).
Event-Based Comparisons to Noise Levels
in Monitoring Data
The 90% confidence levels of estimated runoff volumes, given
total precipitation depth and length of interevent period
(according to Equation 3), are shown as lines in Figure 7. The
figure shows that for shorter interevent periods (e.g., 23 h), the
amount of noise in the monitored data was greater, requiring
a greater difference between scenarios before we would be able
to confidently effectiveness of the scenario attributable to sub-
parcel-scale land use changes.
FIGURE 7 | Differences in runoff volume required to overcome noise in
monitoring data, given depth of rainfall event and interevent period (colored
lines) compared to runoff volume differences between two scenarios for each
event (colored boxes). Colored Boxes represent differences in runoff volume
simulated between two most different micro land use scenarios in this study:
IS_MAX and GI_ROW. None of the colored boxes exceed the predicted
magnitude of noise associated with that event.
None of the scenarios exhibit large enough differences from
the Base case to surpass the level of noise at the 90% confidence
level, in any of the four events during the 10-day simulation
period. Of all the combinations of scenarios in this study,
the maximum differences in total event runoff volume were
between IS_MAX (the maximum allowable impervious surface
per property as per zoning regulations) and GI_ROW (all of
the public ROW treated with GI). The differences in event-
based runoff volume for these two scenarios during the four
events is shown in Figure 7. Even between the most and least
optimal scenarios tested in this study, during a multi-day wet
period however, Figure 3 indicates that we would not be able to
confidently distinguish the effects of micro land cover changes
from the noise in the monitoring data. This is a different result
from what was found in Lim and Welty (2017), where the
differences between these two scenarios were expected to be
distinguishable from the noise, when relationships between all
rainfall events were analyzed together (not just the events during
a particularly wet period).
CONCLUSIONS
This study used a three-dimensional surface-subsurface coupled
model, ParFlow.CLM to examine four main areas of uncertainty
relevant to GI Planning: network effects, sub-parcel-scale land
use and land cover changes, climate, and urban pipeflow
monitoring noise. Isolating a 10-day simulation window
containing four rain events, we showed evidence of non-additive
effects of GI networks. Non-additive effects become apparent
during multi-day events when subsurface storage capacity,
especially in watershed capacitance-constrained scenarios, had
Frontiers in Built Environment | www.frontiersin.org 12 December 2018 | Volume 4 | Article 71
Lim and Welty Uncertainty in Green Infrastructure Planning
not yet been recovered. While in general, downslope network
spatial configurations intercept more runoff volume than
upslope configurations, the opposite is true during multiday
events. This implies that GI in downslope locations may be
more susceptible to obsolescence (given equal design storm
specifications) than GI in upslope locations under changing
climate conditions.
Previous studies have found that GI networks have empirically
decreased peak flows, but there is less evidence that GI networks
reduce runoff volumes associated with urbanization (Jefferson
et al., 2017). The results of this study in a medium-density
residential neighborhood retrofit with GI show that peak flows
behave in a more consistent way, and peak flow mitigation
is not sensitive GI in differently configured spatial networks,
even under multi-day conditions. Total runoff volume, however
does exhibit evidence of being more responsive to spatial
configuration. In empirical studies, the difficulty of quantifying
spatial configurations, and therefore not controlling for them,
may be one reason why the effects of GI networks on total runoff
generation have been weaker. Other reasons may be that design
standards either do not specify a particular GI facility to meet
a total runoff-based performance standard, or that GI designs
typically do not consider the performance of upslope/downslope
contributions of other infiltration opportunities that could
affect a particular BMP or be affected by a particular
BMP.
Lastly, the problem of detectable change (Lim and Welty,
2017) has implications for municipalities to be able to adapt
their infrastructure under changing conditions. In this study,
we demonstrated that an increase in noise/variability associated
with rain events happening quickly following a previous rain
event makes it more difficult to confidently attribute changes
attributable to GI or micro land cover implementation. Non-
stationary climate conditions, when rain events in many areas
are expected to increase in severity or frequency, will likely
make it more difficult to detect unaddressed needs and adapt the
management of the infrastructure.
AUTHOR CONTRIBUTIONS
TL contributed the project concept, research design,
collaboration with partners, ParFlow simulations, analysis,
and manuscript preparation. CW contributed to simulations and
manuscript preparation.
ACKNOWLEDGMENTS
We would like to thank Steve Saari (Washington DC Department
of Energy and the Environment) for providing data from
the RiverSmart Washington Program. The study utilized
computational resources provided through the NSF’s Extreme
Science and Engineering Discovery Environment (XSEDE)
through the project Multiscale surface-subsurface modeling of
the Baltimore region (TG-EAR130027) to CW. Simulations were
conducted on Stampede at the Texas Advanced Computing
Center. TL time was supported by a doctoral fellowship from
the University of Pennsylvania’s Department of City and Regional
Planning. CW time was supported in part by US EPA assistance
agreement R835555 and NSF cooperative agreements CBET
1444758 and BCS 1444755.
REFERENCES
Ando, A. W., and Freitas, L. P. C. (2011). Consumer demand for green stormwater
management technology in an urban setting: the case of Chicago rain barrels.
Water Resour. Res. 47:W12501. doi: 10.1029/2011WR011070
Ashby, S. F., and Falgout, R. D. (1996). A parallel multigrid preconditioned
conjugate gradient algorithm for groundwater flow simulations. NSE 124,
145–159.
Avellaneda, P. M., Jefferson, A. J., Grieser, J. M., and Bush, S. A. (2017). Simulation
of the cumulative hydrological response to green infrastructure. Water Resour.
Res. 53, 3087–3101. doi: 10.1002/2016WR019836
Bhaskar, A. S., Hogan, D. M., and Archfield, S. A. (2016). Urban base flow with low
impact development. Hydrol. Process. 30, 3156–3171. doi: 10.1002/hyp.10808
Booth, D. B., and Jackson, C. R. (1997). Urbanization of aquatic
systems: degradation thresholds, stormwater detection, and the
limits of mitigation1. J. Am. Water Resour. Assoc. 33, 1077–1090.
doi: 10.1111/j.1752-1688.1997.tb04126.x
Catalano de Sousa, M. R., Montalto, F. A., and Gurian, P. (2016). Evaluating green
infrastructure stormwater capture performance under extreme precipitation. J.
Extr. Even. 03:1650006. doi: 10.1142/S2345737616500068
Davis, A. P. (2007). Field performance of bioretention: water quality. Environ. Eng.
Sci. 24, 1048–1064. doi: 10.1089/ees.2006.0190
Davis, A. P. (2008). Field performance of bioretention: hydrology impacts. J.
Hydrol. Eng. 13, 90–95. doi: 10.1061/(ASCE)1084-0699(2008)13:2(90)
Dietz, M. E., and Clausen, J. C. (2008). Stormwater runoff and export changes with
development in a traditional and low impact subdivision. J. Environ. Manage.
87, 560–566. doi: 10.1016/j.jenvman.2007.03.026
Drake, L., and Lawson, L. J. (2014). Validating verdancy or vacancy? The
relationship of community gardens and vacant lands in the U.S. Cities 40,
133–142. doi: 10.1016/j.cities.2013.07.008
Driscoll, C. T., Eger, C. G., Chandler, D. G., Davidson, C. I., Roodsari, B. K., Flynn,
C. D., et al. (2015). Green Infrastructure: Lessons from Science and Practice.
Syracuse, NY: Science Policy Exchange.
Easterling, D. R. (2000). Climate Extremes: observations, modeling, and impacts.
Science 289, 2068–2074. doi: 10.1126/science.289.5487.2068
Emerson, C., Welty, C., and Traver, R. (2005). Watershed-scale evaluation of
a system of storm water detention basins. J. Hydrol. Eng. 10, 237–242.
doi: 10.1061/(ASCE)1084-0699(2005)10:3(237)
Emerson, C. H., and Traver, R. G. (2008). Multiyear and seasonal variation
of infiltration from storm-water best management practices. J. Irrigat.
Drain. Eng. 134, 598–605. doi: 10.1061/(ASCE)0733-9437(2008)134:
5(598)
Endreny, T., and Collins, V. (2009). Implications of bioretention basin spatial
arrangements on stormwater recharge and groundwater mounding. Ecol. Eng.
35, 670–677. doi: 10.1016/j.ecoleng.2008.10.017
Fanelli, R., Prestegaard, K., and Palmer, M. (2017). Evaluation of infiltration-
based stormwater management to restore hydrological processes in urban
headwater streams. Hydrol. Process. 31, 3306–3319. doi: 10.1002/hyp.
11266
Fletcher, T. D., Andrieu, H., and Hamel, P. (2013). Understanding,
management and modelling of urban hydrology and its consequences
for receiving waters: a state of the art. Adv. Water Resour. 51, 261–279.
doi: 10.1016/j.advwatres.2012.09.001
Fry, T. J., and Maxwell, R. M. (2017). Evaluation of distributed BMPs in an urban
watershed—High resolution modeling for stormwater management. Hydrol.
Process. 31, 2700–2712. doi: 10.1002/hyp.11177
Green, O. O., Shuster, W. D., Rhea, L. K., Garmestani, A. S., and Thurston, H.
W. (2012). Identification and induction of human, social, and cultural capitals
through an experimental approach to stormwater management. Sustainability
4, 1669–1682. doi: 10.3390/su4081669
Frontiers in Built Environment | www.frontiersin.org 13 December 2018 | Volume 4 | Article 71
Lim and Welty Uncertainty in Green Infrastructure Planning
Groisman, P. Y., Knight, R. W., Easterling, D. R., Karl, T. R., Hegerl, G. C., and
Razuvaev, V. N. (2005). Trends in intense precipitation in the climate record. J.
Clim. 18, 1326–1350. doi: 10.1175/JCLI3339.1
Heckert, M., and Mennis, J. (2012). The economic impact of greening urban
vacant land: a spatial difference-in-differences analysis. Environ. Plan. A 44,
3010–3027. doi: 10.1068/a4595
Hood, M. J., Clausen, J. C., and Warner, G. S. (2007). Comparison of stormwater
lag times for low impact and traditional residential development. J. Am. Water
Resour. Assoc. 43, 1036–1046. doi: 10.1111/j.1752-1688.2007.00085.x
Jefferson, A., Bhaskar, A. S., Hopkins, K. G., Fanelli, R., Avellaneda, P., and
McMillan, S. K. (2017). Stormwater management network effectiveness and
implications for urban watershed function: a critical review. Hydrol. Process.
31, 4056–4080. doi: 10.1002/hyp.11347
Jones, J. E., and Woodward, C. S. (2001). Newton–Krylov-multigrid solvers for
large-scale, highly heterogeneous, variably saturated flow problems. Adv. Water
Resour. 24, 763–774. doi: 10.1016/S0309-1708(00)00075-0
Keeley, M., Koburger, A., Dolowitz, D. P., Medearis, D., Nickel, D., and Shuster,
W. (2013). Perspectives on the use of green infrastructure for stormwater
management in cleveland and milwaukee. Environ. Manage. 51, 1093–1108.
doi: 10.1007/s00267-013-0032-x
Kertesz, R., Green, O. O., and Shuster, W. D. (2014). Modeling the hydrologic and
economic efficacy of stormwater utility credit programs for US single family
residences. Water Sci. Technol. 70, 1746–1754. doi: 10.2166/wst.2014.255
Kollet, S. J., and Maxwell, R. M. (2006). Integrated surface–groundwater
flow modeling: a free-surface overland flow boundary condition in
a parallel groundwater flow model. Adv. Water Resour. 29, 945–958.
doi: 10.1016/j.advwatres.2005.08.006
Kollet, S. J., and Maxwell, R. M. (2008). Capturing the influence of groundwater
dynamics on land surface processes using an integrated, distributed watershed
model. Water Resour. Res. 44:W02402. doi: 10.1029/2007WR006004
Kristvik, E., Kleiven, G. H., Lohne, J., and Muthanna, T. M. (2018). Assessing
the robustness of raingardens under climate change using SDSM and temporal
downscaling. Water Sci. Tech. 77, 1640–1650. doi: 10.2166/wst.2018.043
Larsen, L., and Harlan, S. L. (2006). Desert dreamscapes: Residential
landscape preference and behavior. Landsc. Urban Plan. 78, 85–100.
doi: 10.1016/j.landurbplan.2005.06.002
Li, C., Fletcher, T. D., Duncan, H. P., and Burns, M. J. (2017). Can stormwater
control measures restore altered urban flow regimes at the catchment scale? J.
Hydrol. 549, 631–653. doi: 10.1016/j.jhydrol.2017.03.037
Li, H., Sharkey, L. J., Hunt, W. F., and Davis, A. P. (2009). Mitigation of impervious
surface hydrology using bioretention in north carolina and maryland. J. Hydrol.
Eng. 14, 407–415. doi: 10.1061/(ASCE)1084-0699(2009)14:4(407)
Liefting, H. J., and Langeveld, J. G. (2008). “Sewer monitoring projects: data
collection, data handling and data quality in arnhem”, in 11th International
Conference on Urban Drainage (Edinburgh).
Lim, T. C. (2016). Predictors of urban variable source area: a cross-section analysis
of urbanized catchments in the united states. Hydrol. Process. 30, 4799–4814.
doi: 10.1002/hyp.10943
Lim, T. C. (2017). An empirical study of spatial-temporal growth patterns of a
voluntary residential green infrastructure program. J. Environ. Plann. Manag.
61, 1363–1382. doi: 10.1080/09640568.2017.1350146
Lim, T. C., and Welty, C. (2017). Effects of spatial configuration of imperviousness
and green infrastructure networks on hydrologic response in a residential
sewershed. Water Resour. Res. 53, 8084–8104. doi: 10.1002/2017WR020631
Londoño Cadavid, C., and Ando, A. W. (2013). Valuing preferences over
stormwater management outcomes including improved hydrologic function.
Water Resour. Res. 49, 4114–4125. doi: 10.1002/wrcr.20317
Loperfido, J. V., Noe, G. B., Jarnagin, S. T., and Hogan, D. M. (2014). Effects
of distributed and centralized stormwater best management practices and
land cover on urban stream hydrology at the catchment scale. J. Hydrol. 519,
2584–2595. doi: 10.1016/j.jhydrol.2014.07.007
Maheepala, U. K., Takyi, A. K., and Perera, B. J. C. (2001). Hydrological data
monitoring for urban stormwater drainage systems. J. Hydrol. 245, 32–47.
doi: 10.1016/S0022-1694(01)00342-0
Mahmoud, M., Liu, Y., Hartmann, H., Stewart, S., Wagener, T., Semmens, D.,
et al. (2009). A formal framework for scenario development in support
of environmental decision-making. Environ. Model. Softw. 24, 798–808.
doi: 10.1016/j.envsoft.2008.11.010
Mailhot, A., and Duchesne, S. (2010). Design criteria of urban drainage
infrastructures under climate change. J. Water Resour. Plan. Manage. 136,
201–208. doi: 10.1061/(ASCE)WR.1943-5452.0000023
Maimone, M., O’Rourke, D. E., Knighton, J. O., and Thomas, C. P. (2011).
Potential impacts of extensive stormwater infiltration in Philadelphia. Environ.
Eng. Appl. Res. Pract. 14, 2–12.
Maxwell, R. M. (2013). A terrain-following grid transform and preconditioner for
parallel, large-scale, integrated hydrologic modeling. Adv. Water Resour. 53,
109–117. doi: 10.1016/j.advwatres.2012.10.001
Maxwell, R. M., Kollet, S. J., Smith, S. G., Woodward, C. S., Falgout, R. D.,
Ferguson, I. M., et al. (2016). ParFlow User’s Manual.Available online at: https://
inside.mines.edu/~rmaxwell/parflow.manual.2-15- 16.pdf
Maxwell, R. M., and Miller, N. L. (2005). Development of a coupled land surface
and groundwater model. J. Hydrometeorol. 6, 233–247. doi: 10.1175/JHM422.1
Meierdiercks, K. L., Smith, J. A., Baeck, M. L., and Miller, A. J. (2010).
Analyses of Urban Drainage Network Structure and Its Impact on Hydrologic
Response1: Analyses of Urban Drainage Network Structure and Its Impact
on Hydrologic Response. JAWRA 46, 932–943. doi: 10.1111/j.1752-1688.2010.
00465.x
Miles, B., and Band, L. E. (2015). Green infrastructure stormwater management
at the watershed scale: urban variable source area and watershed capacitance.
Hydrol. Process. 29, 2268–2274. doi: 10.1002/hyp.10448
Miles, B. C. (2014). Small-Scale Residential Stormwater Management in Urbanized
Watersheds: A Geoinformatics-Driven Ecohydrology Modeling Approach.
Available online at: http://search.proquest.com/docview/1648168622/abstract?
(Accessed September 9, 2015).
Milly, P. C. D., Betancourt, J., Falkenmark, M., Hirsch, R. M., Kundzewicz,
Z. W., Lettenmaier, D. P., et al. (2008). Stationarity is dead: whither water
management? Science 319, 573–574. doi: 10.1126/science.1151915
Montalto, F., Behr, C., Alfredo, K., Wolf, M., Arye, M., and Walsh,
M. (2007). Rapid assessment of the cost-effectiveness of low impact
development for CSO control. Landsc. Urban Plan. 82, 117–131.
doi: 10.1016/j.landurbplan.2007.02.004
Montalto, F. A., Bartrand, T. A., Waldman, A. M., Travaline, K. A., Loomis, C. H.,
McAfee, C., et al. (2013). Decentralised green infrastructure: the importance
of stakeholder behaviour in determining spatial and temporal outcomes. Struc.
Infrastruc. Eng. 9, 1187–1205. doi: 10.1080/15732479.2012.671834
Najjar, R., Patterson, L., and Graham, S. (2009). Climate simulations of major
estuarine watersheds in the Mid-Atlantic region of the US. Clim. Change 95,
139–168. doi: 10.1007/s10584-008-9521-y
Najjar, R. G., Pyke, C. R., Adams, M. B., Breitburg, D., Hershner, C., Kemp, M.,
et al. (2010). Potential climate-change impacts on the Chesapeake Bay. Estuar.
Coast. Shelf Sci. 86, 1–20. doi: 10.1016/j.ecss.2009.09.026
Nowak, D. J., and Greenfield, E. J. (2012). Tree and impervious cover
change in U.S. cities. Urban Forestry Urban Green. 11, 21–30.
doi: 10.1016/j.ufug.2011.11.005
Page, J. L., Winston, R. J., Mayes, D. B., Perrin, C., and Hunt, I. I. I., W. F.
(2015). Retrofitting with innovative stormwater control measures: Hydrologic
mitigation of impervious cover in the municipal right-of-way. J. Hydrol. 527,
923–932. doi: 10.1016/j.jhydrol.2015.04.046
Palla, A., and Gnecco, I. (2015). Hydrologic modeling of Low Impact
Development systems at the urban catchment scale. J. Hydrol. 528, 361–368.
doi: 10.1016/j.jhydrol.2015.06.050
Perez-Pedini, C., Limbrunner, J., and Vogel, R. (2005). Optimal
location of infiltration-based best management practices for storm
water management. J. Water Resour. Plan. Manage. 131, 441–448.
doi: 10.1061/(ASCE)0733-9496(2005)131:6(441)
Roy, A. H., Wenger, S. J., Fletcher, T. D., Walsh, C. J., Ladson, A. R., Shuster, W.
D., et al. (2008). Impediments and solutions to sustainable, watershed-scale
urban stormwater management: lessons from Australia and the United States.
Environ. Manage. 42, 344–359. doi: 10.1007/s00267-008-9119-1
Schifman, L. A., Herrmann, D. L., Shuster, W. D., Ossola, A., Garmestani, A., and
Hopton, M. E. (2017). Situating green infrastructure in context: a framework
for adaptive socio-hydrology in cities. Water Resour. Res. 53, 10139–10154.
doi: 10.1002/2017WR020926
Schilling, J., and Logan, J. (2008). Greening the rust belt: a green infrastructure
model for right sizing America’s Shrinking Cities. J. Am. Plan. Assoc. 74,
451–466. doi: 10.1080/01944360802354956
Frontiers in Built Environment | www.frontiersin.org 14 December 2018 | Volume 4 | Article 71
Lim and Welty Uncertainty in Green Infrastructure Planning
Schneider, A., and Woodcock, C. E. (2008). Compact, dispersed, fragmented,
extensive? A comparison of urban growth in twenty-five global cities using
remotely sensed data, pattern metrics and census information. Urban Stud. 45,
659–692. doi: 10.1177/0042098007087340
Shuster, W. D., and Rhea, L. (2013). Catchment-scale hydrologic implications of
parcel-level stormwater management (Ohio USA). J. Hydrol. 485, 177–187.
doi: 10.1016/j.jhydrol.2012.10.043
Silberstein, R. P. (2006). Hydrological models are so good, do we still need data?
Environ. Model. Softw. 21, 1340–1352. doi: 10.1016/j.envsoft.2005.04.019
Smith, B. K., Smith, J. A., Baeck, M. L., and Miller, A. J. (2015). Exploring
storage and runoff generation processes for urban flooding through a
physically based watershed model. Water Resour. Res. 51, 1552–1569.
doi: 10.1002/2014WR016085
Tague, C., and Pohl-Costello, M. (2008). The potential utility of physically based
hydrologic modeling in ungauged urban streams. Ann. Assoc. Am. Geogr. 98,
818–833. doi: 10.1080/00045600802099055
Troy, A. R., Grove, J. M., O’Neil-Dunne, J. P. M., Pickett, S. T. A., and
Cadenasso, M. L. (2007). Predicting opportunities for greening and patterns
of vegetation on private urban lands. Environ. Manage. 40, 394–412.
doi: 10.1007/s00267-006-0112-2
Tzoulas, K., Korpela, K., Venn, S., Yli-Pelkonen, V., Kazmierczak, A., Niemela, J.,
et al. (2007). Promoting ecosystem and human health in urban areas using
Green Infrastructure: a literature review. Landsc. Urban Plan. 81, 167–178.
doi: 10.1016/j.landurbplan.2007.02.001
US Global Change Research Program (2009). Global Climate Change Impacts in
the United States. New York, NY: Cambridge University Press Available online
at: www.globalchange.gov/usimpact
Valderrama, A., and Davis, P. (2015). Wanted: Green Acres:How Philadelphia’s
Greened Acre Retrofit Program is Catalyzing Low-Cost Green Infrastructure
Retrofits on Private Property. NRDC.
Voter, C. B., and Loheide, S. P. (2018). Urban residential surface and
subsurface hydrology: synergistic effects of low-impact features at the
parcel scale. Water Resour. Res. 54, 8216–8233. doi: 10.1029/2018WR
022534
Waters, D., Watt, W. E., Marsalek, J., and Anderson, B. C. (2003).
Adaptation of a storm drainage system to accommodate increased rainfall
resulting from climate change. J. Environ. Plann. Manage. 46, 755–770.
doi: 10.1080/0964056032000138472
Wilby, R. L., Dawson, C. W., Murphy, C., O’Connor, P., and Hawkins,
E. (2014).The statistical downscaling model-decision centric (SDSM-
DC): conceptual basis and applications. Climate Res. 61, 259–276.
doi: 10.3354/cr01254
Willems, P., Arnbjerg-Nielsen, K., Olsson, J., and Nguyen, V. T. V. (2012).
Climate change impact assessment on urban rainfall extremes and urban
drainage: Methods and shortcomings. Atmospheric Res. 103, 106–118.
doi: 10.1016/j.atmosres.2011.04.003
Woznicki, S. A., Hondula, K. L., and Jarnagin, S. T. (2018). Effectiveness
of landscape-based green infrastructure for stormwater management in
suburban catchments. Hydrol. Process. 32, 2346–2361. doi: 10.1002/hyp.
13144
Zmyslony, J., and Gagnon, D. (1998). Residential management of urban
front-yard landscape: a random process? Land. Urban Plan. 40, 295–307.
doi: 10.1016/S0169-2046(97)00090-X
Conflict of Interest Statement: The authors declare that the research was
conducted in the absence of any commercial or financial relationships that could
be construed as a potential conflict of interest.
Copyright © 2018 Lim and Welty. This is an open-access article distributed under the
terms of the Creative Commons Attribution License (CC BY). The use, distribution
or reproduction in other forums is permitted, provided the original author(s) and
the copyright owner(s) are credited and that the original publication in this journal
is cited, in accordance with accepted academic practice. No use, distribution or
reproduction is permitted which does not comply with these terms.
Frontiers in Built Environment | www.frontiersin.org 15 December 2018 | Volume 4 | Article 71
