Environmental Modelling & Software 95 (2017) 246e257

Contents lists available at ScienceDirect

Environmental Modelling & Software

journal homepage: www.elsevier.com/locate/envsoft

A model for simulating the performance and irrigation of green

stormwater facilities at residential scales in semiarid and
Mediterranean regions
Josefina Herrera a, b, Carlos A. Bonilla a, b, Lina Castro c, Sergio Vera b, d, Rodolfo Reyes a, b,

s a, b, e, f, *
Jorge Girona
a
Departamento de Ingeniería Hidra
ulica y Ambiental, Pontificia Universidad Cato
lica de Chile, Avenida Vicun~ a Mackenna 4860, Santiago, Chile
b
Centro de Desarrollo Urbano Sustentable CONICYT/FONDAP/15110020, Avenida Vicun ~ a Mackenna 4860, Santiago, Chile
c
Escuela de Ingeniería Civil, Pontificia Universidad Cato
lica de Valparaíso, Avenida Brasil 2147, Valparaíso, Chile
d
Departamento de Ingeniería y Gestio
n de la Construccio
n, Pontificia Universidad Cato
lica de Chile, Avenida Vicun
~ a Mackenna 4860, Santiago, Chile
e
Centro de Investigacio
n para la Gestio
n Integrada de Desastres Naturales CONICYT/FONDAP/15110017, Avenida Vicun ~ a Mackenna 4860, Santiago, Chile
f
Centro Interdisciplinario de Cambio Global UC, Pontificia Universidad Cato
lica de Chile, Avenida Vicun ~ a Mackenna 4860, Santiago, Chile

a r t i c l e i n f o a b s t r a c t

Article history: Impervious areas change hydrological processes, reducing infiltration and evapotranspiration, and
Received 23 September 2016 increasing direct runoff. Stormwater practices using green infrastructure are implemented locally to
Received in revised form control runoff and preserve the hydrological cycle. Applying these techniques in semiarid and Medi-
10 April 2017
terranean regions requires accounting for aspects related to the maintenance of green areas. This study
Accepted 15 June 2017
Available online 26 June 2017
develops the Integrated Hydrological Model at Residential Scale, a continuous model for representing the
performance and irrigation of green stormwater facilities at residential scales. Among other relevant
process, the model simulates evaporation from bare soil and redistribution between soil layers. Different
Keywords:
Continuous simulation
components of the model were tested using laboratory and numerical experiments, and then an
Soil water content application to a case study and a sensitivity analysis were carried out. The model identifies significant
Green infrastructure differences in the performance of a rain garden with different vegetation, climate and irrigation practices
Hydrological modeling and provides good insight for the maintenance needs of green infrastructure for runoff control.
Irrigation © 2017 Elsevier Ltd. All rights reserved.
Global sensitivity analysis

Software availability Software required: MS Excel (the MATLAB Runtime compiler is

Name of software: IHMORS
Developer: Josefina Herrera and Jorge Girona
s
Contact address: Jorge Girona
s, Departamento de Ingeniería
Hidra
ulica y Ambiental and Centro de Desarrollo Urbano
Sustentable (CEDEUS), Pontificia Universidad Cato
lica de
~ a Mackenna 4860, Santiago, Chile. Email:
Chile, Av. Vicun
jgironas@ing.puc.cl
Availability: Free download at http://jorge-gironas.weebly.com/
software.html (personal website)
First available: 2016
* Corresponding author. Av. Vicun~ a Mackenna 4860, Santiago, Chile.

s).
E-mail address: jgironas@ing.puc.cl (J. Girona
provided). Matlab codes can be obtained from the
developers upon request
1. Introduction
Urban development reduces infiltration rates and surface stor-
age capacity, producing higher direct runoff volumes and peak flow
discharges (Xiao et al., 2007; Freni and Oliveri, 2007). To reduce
these effects, different practices known as sustainable urban
drainage systems (SUDS), low impact development (LID) and best
management practices (BMP) have been developed (Fletcher et al.,
2014). These practices replicate natural processes of capture,
retention and infiltration to preserve the hydrological cycle and
control runoff quantity and quality from frequent events (Huang
et al., 2014; Fletcher et al., 2014; Walsh et al., 2014; Everett et al.,

http://dx.doi.org/10.1016/j.envsoft.2017.06.020
1364-8152/© 2017 Elsevier Ltd. All rights reserved.
 J. Herrera et al. / Environmental Modelling & Software 95 (2017) 246e257
2015; Houdeshel et al., 2015).
SUDS, BMP and LID typically consider green areas that receive
precipitation and runoff from impervious areas (Sample and
Heaney, 2006; Xiao et al., 2007), which irrigate them to some
extent. Such property may be very relevant in semiarid and Med-
iterranean regions as it can lead to lower maintenance costs
(Sample and Heaney, 2006; Houdeshel and Pomeroy, 2013; Sample
et al., 2014). Moreover, considering this supplemental irrigation can
drastically change the design in these climates as compared to
humid areas (Ascione et al., 2013), as well as the selection of the
vegetation (Houdeshel and Pomeroy, 2013; Houdeshel et al., 2015).
Finally, and despite the results of these drainage practices can vary
substantially due to different climate conditions (Huang et al.,
2014), their performance and effectiveness have been less often
tested in semiarid and Mediterranean climates (Houdeshel et al.,
2015).
Hydrological models are tools for assessing the performance of
stormwater facilities. Models like SWMS-2D, HYDRUS and SWAP
simulate water and/or solutes and heat transport to model runoff
control (Li and Babcock, 2014). Nonetheless, despite their complete
treatment of the soil-vegetation-atmosphere processes, none of
them are suitable for simulating complex systems with storage
structures, drainages or connections among areas which are able to
alter the water movement (Li and Babcock, 2014). SWMM v5.1
(Rossman, 2010) is a more suitable model for stormwater facilities
thas has been implemented in semiarid and Mediterranean areas
(e.g., Huang et al., 2014; Walsh et al., 2014). Its LID module to
simulate LID practices (e.g., rain gardens, green roofs, infiltration
trenches) has been reported to produce both unsatisfactory results
when simulating stormwater runoff hydrographs (Burszta-
Adamiak and Mrowiec, 2013; Li and Babcock, 2014; Carson et al.,
2017) as well as successful results after calibrated using observa-
tions (Palla and Gnecco, 2015; Rosa et al., 2015; Yang et al., 2015).
Nonetheless, the model calculates Evapotranspiration (ET) on a
daily basis using temperature data, and thus neither the type of
plant nor the available soil moisture control the process (Rossman,
2010; Carson et al. 2017). Moreover, SWMM does not explicitly
compute bare soil evaporation, which requires certain particular
considerations (Allen et al., 1998, 2005). Furthermore, it is not
possible to capture and visualize the soil water dynamics within
multiple soil layers in different LIDs, nor in contributing sub-
catchments. Finally it is neither possible to explicitly enter an irri-
gation schedule, nor to design one using the dynamics of ET or the
soil moisture. Overall, models able to simulate the surface/subsur-
face processes and the continuous dynamics of the soil water
content controlled by ET and irrigation dynamics are essential
when studying the performance of green infrastructure in semiarid
and Mediterranean regions (Sample and Heaney, 2006; Houdeshel
et al., 2015). These capabilities permit the evaluation of plant sur-
vival and the analysis of the sustainability of drainage techniques.
From our literature review of available models, only the SPAW
model (Saxton and Willey, 2005), which simulates water transport
with an agricultural focus (Li and Babcock, 2014), is able to
explicitly consider the irrigation needs and has been used in urban
settings (Lucas, 2006). However some of its characteristics affect its
suitability for urban areas, as it runs on a daily basis and explicit
connections among areas are not possible. Nonetheless, other
studies have developed original models to deal explicitly with the
issue of irrigation of drainage control practices. Sample and Heaney
(2006) compared and integrated different irrigation management
options within the context of LIDs modeling to perform an eco-
nomic analysis in Boulder, USA. Alternatively, Xiao et al. (2007)
developed and assessed a numerical model on an hourly basis to
simulate hydrological processes at residential scales. They analyzed
rain gutters, cisterns, law retention basins and driveway
247
interceptors, and studied runoff reduction and their efficient use of
irrigation water. Despite these studies successfully simulated the
dynamics of soil water content, they did not focus on the soil
moisture regime so as to determine percentages of time in which
soil water content reaches critical levels for vegetation survival, or
to assist decision making in irrigation. Such characterization would
allow a better quantification of the time involved in irrigation,
which translates into maintenance costs that are time dependent.
This paper presents the Integrated Hydrological Model at Resi-
dential Scale (IHMORS), a software to evaluate, in a continuous
manner, the rainfall-runoff processes and stormwater control at
residential scales, together with the irrigation of green areas by
means of an irrigation module. Thus, the model is particularly
suitable for Mediterranean and semiarid areas in which irrigation is
essential for the vegetation's survival through the year. The model
simulates surface and subsurface processes and accounts for the
water content dynamics in different soil layers. Its different com-
ponents were first tested using laboratory and numerical experi-
ments, and then an application to a case study was carried out. In
this application we assess the long-term performance in terms of
runoff control and irrigation needs of rain gardens with different
vegetation, under different climates and irrigation practices. Finally,
a global sensitivity analysis of the model parameters is also
presented.
2. Methodology
IHMORS is a physically-based continuous hydrological model
for simulating rainfall-runoff processes in urban areas, which fo-
cuses on the performance of stormwater runoff control facilities, as
well as irrigation practices at a residential scale. The model was
developed in MATLAB and uses a MS Excel spreadsheet for data
input. Common SUDS techniques like rain gardens, green roofs,
surface retention areas and driveway interceptors can be simulated
by combining and connecting different subareas, each with
different properties. Input data include: (1) meteorological infor-
mation, (2) time step information, (3) subareas’ spatial configura-
tion, (4) physical properties of subareas, and (5) an optional user
defined irrigation program, although IHMORS also computes irri-
gation programs based on ET demands or a required minimum soil
water content.
The model partially builds on the framework proposed by Xiao
et al. (2007) for the representation of both surface and subsurface
processes together with watering needs. Nonetheless, IHMORS
implements several changes and improvements including: (1) the
explicit and flexible representation of the connectivity among
subareas, (2) the simulation of water redistribution among soil
layers during dry-weather, (3) the evaporation from bare soil linked
to subsurface processes to correctly simulate the soil moisture in
each layer, and (4) the simulation of storage and subsurface runoff
transport through conduit elements.
IHMORS considers a cascade of permeable and/or impermeable
subareas with one or more soil layers each, which are conceived as
rectangular planes interconnected through horizontal runoff flows.
These flows are distributed uniformly over the downstream sub-
areas as an additional form of precipitation. Fig. 1 shows all of the
hydrological processes the model can simulate at each time step Dt
(h) defined by the user. Water entering each subarea in the form of
rainfall, run-on and/or irrigation, can be intercepted by vegetation
or stored by the surface storage capacity. The water that reaches the
surface can infiltrate or return to the atmosphere by evaporation
from bare soil or ET from vegetated soil. Water moves through the
soil layers by percolation and/or redistribution during dry weather
days. Water reaching the last soil layer, can either go to the drainage
system, and/or become deep percolation. Note that a free boundary
248 J. Herrera et al. / Environmental Modelling & Software 95 (2017) 246e257

as evaporation (mm):

ES ¼ kET0 t (4)

where k is the leaf area index, and t is the time from the beginning
of the storm (h) identified after defining a minimum dry inter-
event time. Because the interception acts homogeneously over
rainfall and irrigation, the potential dependency of interception on
the irrigation method should be considered when defining the
irrigation plan. During dry periods the intercepted water evapo-
rates, freeing up storage for future storms.

2.1.3. Infiltration
The model assumes infiltration excess overland flow to generate
surface runoff. Nonetheless, when soil reaches saturation, the
simulated infiltration rate reduces to also represent the saturation
excess overland flow mechanism (Beven, 2012). The infiltration rate
f (mmh
1) and cumulative infiltration F (mm) are computed using
the Green and Ampt equation (Green and Ampt, 1911):
Fig. 1. Conceptual representation of the physical processes at a residential scale
 
simulated in IHMORS. jDq
f ¼ Ks þ1 (5)
F
condition for the last soil layer is used for this purpose. In parallel,
where Ks is the saturated hydraulic conductivity (mmh
1) and Dq
non-infiltrated water becomes runoff contributing to the drainage
(m3m
3) and j (mm) are the soil water content variation and
system or another subarea.
suction head at the wetting front respectively. Note that F ¼ 0
before irrigation or a precipitation event begins. j varies with
2.1. Hydrological processes moisture content and is calculated using the van Genuchten
equation (van Genuchten, 1980):
2.1.1. Evapotranspiration  
Evapotranspiration ET (mmh
1) from vegetated soils is simu- 
n n
1
n
1
lated after computing the reference evapotranspiration ET0 j ¼ jb Se
1 (6)
(mmh
1) using the hourly Penman-Monteith equation:

 where jb is the bubbling pressure (mm), n is the curve shape
0:408DðRn
GÞ þ g Tþ273
37 u e*
e
2 a
ET0 ¼ (1) parameter and Se is the relative saturation:
D þ gð1 þ 0:34u2 Þ
q
qr
where Rn is the net radiation (MJ m
2h
1), G is the soil heat flow Se ¼ (7)
qs
qr
(MJ m
2h
1), g is the psychometric constant (kPa C
1), T ( C) and
u2 (ms
1) are the hourly air temperature and wind speed at 2 m where qs and qr are the saturated and residual soil water contents
height respectively, e* and ea (kPa) are the saturation and actual respectively (m3m
3).
vapor pressure, and D is the slope vapor pressure curve (kPa C
1).
Then, ET is given as (Allen et al., 1998):
2.1.4. Bare soil evaporation
8 Evaporation E (mmh
1) from bare soil is simulated as a two
>
< ET0 Kc q1 > qFC1
stage-process (Allen et al., 2005). In the first stage E occurs at a rate
ET ¼ q1 (2)
>
: ET0 Kc q1  qFC1 limited only by energy availability at the surface. Then E decreases
qFC1 and depends on the soil hydraulic properties and surface moisture.
IHMORS simulates E for the first stage with the equation proposed
where q1 (m3m
3) is the average soil water content q (m3m
3) of by Allen et al. (2005), whereas an extra parameter b is used to
the first layer, qFC1 is the soil water content at field capacity of the simulate the second one (Snyder et al., 2000):
first soil layer (m3m
3), and Kc is the crop coefficient that depends

on the plant date and the duration of the different stages of crop Kr Kc;max ET0 Kr ¼ 1
growth. Note that q1 =qFC1 is a form of the water-stress coefficient
E¼ (8)
bKr Kc;max ET0 Kr < 1
defined by Allen et al. (1998).
where Kc;max is the maximum Kc value representing an upper limit
2.1.2. Interception on evaporation due to natural constraints on available energy, and
Interception I (mm) by vegetation and surface storage is esti- Kr is the daily reduction coefficient given by:
mated for each Dt using a modified Merriam equation (Merriam, 8
>
> 1 qd  qREW
1960): >
>
<
  0:5qWP
qd
Kr ¼ qREW > qd > 0:5qWP (9)
I ¼ S 1
e
S þ ES > qWP
qREW
P
(3) > 0:5
>
>
:
0 qd < 0:5qWP
where S is the maximum interception capacity (mm), P is cumu-
lative precipitation and/or irrigation (mm) and ES is the water lost where qd is q at the end of the previous day (m3m
3), qWP is q at
 J. Herrera et al. / Environmental Modelling & Software 95 (2017) 246e257 249

wilting point (m3m
3), and qREW (m3m
3) is q corresponding to the unsaturated flow for short periods of time.
readily evaporated water level (Allen et al., 2005):
2.1.7. Irrigation
dREW
qREW ¼ qFC
(10) Irrigation typically corresponds to a daily amount of water
ze applied at a certain frequency within the day (this amount and/or
frequency may change throughout the year). This approach can be
where dREW is the maximum water depth than can evaporate in the
quiet inefficient when the occurrence of rainfall is not considered
first stage (mm) and ze is the soil depth subjected to drying through
(Stewart and Musick, 1982). IHMORS addresses this issue by
evaporation, with values ranging between 0.10 and 0.15 m (Allen
considering 3 irrigation plans proposed by Sample and Heaney
et al., 2005). Experimental values of b and dREW are shown in sec-
(2006): (1) An irrigation plan provided by the user (i.e. a depth
tion 3.1.
vs. time table), (2) the use of the previous plan as long as the field
capacity is not exceeded (i.e. the availability of a soil moisture
2.1.5. Percolation sensor is assumed), and (3) A daily plan defined by a certain per-
The percolation rate pej (mm h
1) through each soil layer j when centage of the previous 24 h ET0 , which can vary for different time
q > qFC is not instantaneously simulated but attenuated by a delay steps and are provided by the user. Alternatively, IHMORS can
factor, as proposed by Savabi and Williams (1995): compute the irrigation needs so that q in the top layer is always
8 above a certain desirable value defined by the user.
>
>
>
>  
Dt

dj
>
< qj
qFC 1
e tj qj > qFCj 2.2. Outputs of the model
pej ¼
j
Dt (11)
>
>
>
> 0 qj  qFCj 2.2.1. Soil water content
>
:
A single but time-variant soil water content q is assumed for
each layer j. Its rate of change in time wj (mm) is estimated using the
where dj is the thickness of layer j (mm) and tj is the travel time mass balance equation:
through layer j (h): 
dwj f
vc ET
ð1
vc ÞE
pe1 þ r1 j ¼ 1 ðtop soil layerÞ
qj
qFCj ¼
dt pej
1
pej þ rj
rj
1 j>1
tj ¼ dj (12)
Kj (15)

where Kj is the unsaturated hydraulic conductivity of layer j where vc is the portion of the area covered with vegetation. Then, q
(mmh
1) given by van Genuchten (1980) and Mualem (1976): in each layer j, qj , at each Dt is estimated as:
   2
n=ðn
1Þ ðn
1Þ=n dwj Dt
Kj ¼ Ks SLej 1
1
Sej (13) qj ¼ qi;j þ (16)
dt dj

where L is the pore tortuosity, whose typical value is 0.5.
2.1.6. Redistribution
Redistribution is the water movement in unsaturated soils, due
to capillarity and the gravitational gradient, that occurs during dry
weather conditions or when infiltration is null (Smith et al., 2002).
Redistribution is simulated with a simplified version of the
Richards' equation that uses the vertical component of the Darcy's
law (Corradini et al., 2000; Guo and Luu, 2015), and a delay factor:
! 2.2.2. Rainfall excess and surface runoff
 
dj
DtK

Dq dj
rj ¼ Kj þ cosa 1
e j j
dz

1
where rj is the redistribution rate from layer j (mmh ) to the layer
j±1 (above or below depending on the flow direction) located at a
distance dz (mm), dj is the suction head difference between layer j
and layer j±1 (mm), a is the angle between the flow direction and
the vertical axis (i.e. for a positive upward flow a ¼ p), and Dqj is the
variation of q in layer j due to redistribution. Hence, water flows
from the layer with the lower absolute suction head to the adjacent
one. For 3 or more layers, rj is split according to a factor pr , given by
the dj values between adjacent layers and that from where redis-
tribution flow is established. Thus, more redistribution occurs be-
tween layers with larger values of dj.
Note that IHMORS simplifies the simulation of the unsaturated
flow (percolation and distribution) in order to model continuously
other various surface processes taking place in urban environ-
ments. Hence, we recommend a model such as HYDRUS (Simunek
et al., 2013) in case the simulation focuses mainly on the
where qi;j is the soil water content of layer j at the beginning of Dt
(m3m
3).
IHMORS first solves the mass balance without considering the
redistribution. q is lower and upper limited by qr and qs respec-
tively. Thus, the infiltration rate is iteratively reduced if q > qs ,
whereas if q < qr , the evaporation and/or the percolation rate is
iteratively lowered. Only after these adjustments can the redistri-
bution be calculated for the same Dt.
The rainfall excess ie (ms
1) in each Dt is given by
(14)
ie ¼ p
f (17)
where p combines precipitation, irrigation and run-on from up-
stream subareas. Note that p accounts for ET from intercepted
water (Fig. 1), whereas ET from readily available excess rainfall is
neglected in Equation (17). Nonetheless ET and its effect on the
available water content are continuously simulated for the top soil
layer (Equation (15)). Surface runoff Qsp (m3s
1) is calculated using
the non-linear reservoir equations for a rectangular plane with
storage depth hp (m) (Huber et al., 2010):
dh Qsp
¼ ie
(18)
dt A
w
5 1
Qsp ¼ h
hp 3 s2 (19)
nm
where h is the flow depth (m), t is the time (s), w is the subarea
250 J. Herrera et al. / Environmental Modelling & Software 95 (2017) 246e257
width (m2), nm is the Manning coefficient and s is the surface slope
(m m
1).
2.2.3. Subsuperficial hydrograph
The output of percolated water from the last layer is modeled
combining a linear channel with lag-time t (h) and a linear reser-
voir with parameter k (h). The corresponding instantaneous unit
hydrograph u(t) is (Bras, 1990):
1 ðt
tÞ
uðtÞ ¼ e
k
k small MAE values lower than the observed s). For all the cases, the
Parameters t and k are representative of the residence time of
percolated water before reaching the point where the hydrograph
is simulated. These parameters can be defined to simulate a variety
of situations typical of drainage systems, such as contribution to
base flow and subsurface drainage through perforated pipes. The
subsurface outflow Qsb (m3s
1) corresponds to the convolution
between uðtÞ and pe from the layer, which is solved numerically
utilizing a time step finer than Dt.
3. Model testing
We used three experiments to test critical components of the
model not commonly considered explicitly by other hydrological
models for urban settings: bare soil evaporation, subsurface runoff
hydrograph and soil moisture redistribution. In particular, imple-
menting bare soil evaporation and redistribution in the model
required a novel approach, whereas the analysis of subsurface
runoff facilitated testing the simulation of the downstream
contribution of percolated flows. Moreover, this testing process was
used to estimate values for some unknown relevant parameter,
which allowed an implementation with rain gardens described
later.
As goodness-of-fit measures between observed and predicted N
data (Ot and Pt respectively), we adopted the Mean Absolute Error
(MAE) and the Modified Coefficient of Efficiency (MCE) (Legates
and McCabe, 1999; Bennett et al., 2013):
X

1
MAE ¼ N Ot
Pt
t
P theless, modeling more soil layers can eventually improve the
Ot
P t representation of this variability.
MCE ¼ 1
Pi (22)
t Ot
O
where O is the observed mean. MAE is residual measure that
quantify the goodness-of-fit in the units of the variable (Bennett
et al., 2013). Additionally, the MCE is a relative error measure to
test the model's ability to preserve the pattern of observed data
(Bennett et al., 2013), which combines the correlation coefficient
and observed and simulated means (m) and standard deviations (s)
(Legates and McCabe, 1999; Bennett et al., 2013).
3.1. Evaporation testing
To test the evaporation model, five substrates used in green
roofs were dried under ambient conditions and weighed daily to
measure evaporative water loss. qWP (Equation (9)) was measured
in the laboratory, whereas dREW (Equation (10)) was defined so that
the extension of the first stage in evaporation were 2e3 days (Allen
et al., 1998). IHMORS calculated ET, and then evaporation parame-
ters from Equations (8)e(10) were calibrated (Table 1). The mois-
ture dynamics of the samples simulated every 5 min (Fig. 2) closely
resembles that of the experimental data (MCE values > 0.8 and
Table 1
Evaporation parameters for five soil samples.
Unit Sample
A b c D e
qWP 3
m m
3
0.026 0.120 0.004 0.110 0.170
dREW mm 22 27 16 16 6
Kc;max 2.368 2.658 1.627 1.489 0.428
b 0.208 0.462 0.523 0.800 1.934
(20)
first stage in evaporation lasted for 2 days. Note that the irregu-
larities of the simulated curves are due to the Dt adopted.
3.2. Subsurface runoff hydrograph testing
Using the Armfield Ltd. S12Mkll Advanced Hydrology Study Sys-
tem, we carried out two experiments with sample a soil (Table 1) to
test the model's capability to simulate subsurface flow and the
outflow hydrograph. Using nozzles we applied rain pulses of 139
and 128 mm h
1 during 15 min over 6.3 cm depth of soil, which was
packed in a square box 50 cm wide draining through a lateral
orifice. Initial q values were 0.32 and 0.33 m3 m-3 respectively. Both
q in the center at 3 cm depth and the outflow from the box were
measured every 5 min. Percolation parameters from Equations
(11)e(13) were measured in the laboratory (Table 2), and the pa-
rameters to calculate the subsurface hydrograph (Equation (20))
were calibrated and validated with the first and second experiment
respectively.
Fig. 3a and Fig. 3b show the dynamics of q for both experiments.
Observed data were very constant over time and thus the model
simulates the observations as good as their average in both cases
(i.e. MCE values ~0 and MAE values ~ s). The subsurface flow is well
simulated, and both the calibration and validation hydrographs
produced by IHMORS match the observations fairly well (Fig. 3c
and d). Likely because IHMORS treats each soil layer as a lumped
system, the mean of observed q is better reproduced than its vari-
ability in each layer. Additionally, part of the model error is due to
(21) the assumption of a homogeneous discharge from the horizontal
bottom, whereas in the experiment the drainage is lateral. None-
3.3. Redistribution testing
The redistribution component of IHMORS was tested against
HYDRUS-1D, a popular software package that solves the Richards’
equation to simulate unsaturated flow in porous media. We simu-
lated a 0.6 m depth soil composed of a 0.2 m depth top layer (layer
1) and a 0.4 m depth bottom layer (layer 2) with the same soil
parameters (Table 2). Three cases with different initial water con-
tents for layers 1 and 2 were defined (i.e. 0.14 and 0.15 for case 1,
0.15 and 0.15 for case 2, and 0.16 and 0.15 for case 3). Fig. 4 compares
the evolution of q in both layers just due to redistribution calculated
with HYDRUS-1D (H) and IHMORS (I). In all the cases, MCE
values > 0.5, small MAE values and similar values of m and s
demonstrate that the dynamics of q simulated by both models are
similar, despite the simpler approach of IHMORS. Note that values
tend to be constant as the hydraulic head becomes the same for
both layers. For an even more robust assessment, the redistribution
routine can be further evaluated using other initial moisture con-
tents in each layer, or observed data in a multilayer soil system.
 J. Herrera et al. / Environmental Modelling & Software 95 (2017) 246e257 251

Fig. 2. Time evolution of experimental (EXP) and simulated (IHMORS) soil water content under evaporation for soil samples described in Table 1. m and s for observed and simulated
data are also presented.

Table 2 control and dynamics of q and irrigation needs, under different

Soil parameters of the substrate used in laboratory experiments. climates and irrigation practices. A rain garden is a depression in
Unit Value the ground filled with plants and a surface mulch layer, which al-
lows storing, infiltrating and evapotranspiring stormwater runoff
Percolation parameters
qr m3m
3 0.01
(Cahill, 2012).
qs m3m
3 0.540 Two rain gardens were designed and continuously simulated
n 1.8 over years 2012 and 2013 as if they were implemented in the
Ks mmh
1 1250 Chilean cities of Santiago (33 260 S 70 390 W) and Temuco (38 460 S
qFC m3m
3 0.230
72 380 W), which are representative of the dry (semi-arid) and wet
L 0.5
Hydrograph parameters
extremes of a Mediterranean climate. Santiago has a warm
t h 10
7 temperate climate with dry summers, a mean annual precipitation
k 6$10
6 of 313 mm, and ~25e30 rainy days per year. Temuco has a warm
temperate humid climate with warm summers, a mean annual
precipitation of 1157 mm, and ~160 rainy days per year (Peel et al.,
2007; DGAC, 2016). In both cases, the rain gardens are fully covered
with vegetation, have an area A ¼ 2 m2, and are designed to control
the runoff from a 100% impervious area Aimp ¼ 10 m2. The simu-
lations considered a minimum dry weather time of 6 h to separate
rainfall events, and a simulation time step of Dt ¼ 15 min. Hourly
records of precipitation, temperature, wind velocity, relative hu-
midity and net radiation from INIA (2012) were used. Annual pre-
cipitation during 2012 and 2013 were 295 and 156 mm in Santiago,
and 582 and 623 mm in Temuco.

4.2. Rain garden design

Both rain gardens were designed according to traditional stan-

Fig. 3. Observed (EXP) and simulated (IHMORS) q dynamics and subsurface hydro-
graphs. (a) and (c) show the results for experiment 1 (calibration), whereas (b) and (d)
show results for experiment 2 (validation). m and s for observed and simulated data are
also presented.
4. Application to rain gardens
4.1. Description
To illustrate its capabilities, IHMORS is used to simulate and
assess the long-term performance of rain gardens in terms of runoff
dards (UDFCD, 2010; Cahill, 2012) to control the water quality
capture volume (WQCV), i.e. the volume representative of runoff
from frequent storm events (UDFCD, 2010). The Chilean regulation
defines the WQCV as the runoff volume produced by a represen-
tative frequent precipitation of Pr ¼ 10 mm (Santiago) and
Pr ¼ 12 mm (Temuco) falling over the contributing impervious area
Aimp (m2) (MOP, 2013). For both cases we considered a contributing
rectangular surface with a roughness coefficient nimp ¼ 0.011, slope
simp ¼ 2% and width wimp ¼ 5 m. For the top layer (L1), in both cases
we defined a depth d1 ¼ 0.2 m representative of the depth of the
root zone (Shorten and Pleasants, 2007), whereas for the bottom
layer (L2) a depth d2 ¼ 0.4 m was selected. Furthermore, both rain
gardens have the same slope (sgarden ¼ 0.1%) and width (wgarden ¼
2 m). Only the ponding depth D differs for both locations (D ¼
50 mm for Santiago and D ¼ 60 mm for Temuco), as this design
variable depends on the local WQCV.
To facilitate our analysis, we used the same substrate from the
252 J. Herrera et al. / Environmental Modelling & Software 95 (2017) 246e257
Fig. 4. Soil water content simulated by IHMORS (I) and HYDRUS (H). For both models 1 and 2 correspond to layers 1 and 2. The initial water contents considered for layers 1 and 2
were (a) 0.14 and 0.15, (b) both 0.15, (c) 0.16 and 0.15. m and s for HYDRUS and IHMORS results are also presented.
validation experiments for both rain gardens (i.e. sample a) but with
a wilting point is qWP ¼ 0.15 m3 m-3. Different vegetation for each
city was chosen based on the local climate. Hence, species of Sedum
(Sedum rupestre, Sedum spurium, Sedum kamtschaticum, and Sedum
rubrotinctum) and grass, whose corresponding parameters ac-
cording to Equations (2) and (3) are presented in Table 3, were
adopted for Santiago and Temuco respectively. An initial water
content of 0.15 m3 m-3 in both layers was assumed for each garden.
4.3. Results and discussion
Fig. 5 shows the temporal dynamics of q for both locations. As
expected, q in Santiago is less variable, very responsive to the
seasonal precipitation in the middle of the year (winter), and de-
creases at very constant rates during the dry months of fall and
summer (Fig. 5a). In contrast, q in Temuco is much more variable
throughout the year as precipitation occurs throughout the season.
For both cities, q increases with precipitation events and decreases
quickly after each peak, as reported in previous studies (Xiao et al.,
2007; Houdeshel and Pomeroy, 2013). Because L2 is deeper, the
variability in this layer is much less significant than for L1 in both
cities. For example, in Temuco q tends to a value of 0.23 in L2, except
when it rains or during dry periods in summer.
Annual volume balances in percentages of the total incoming
water (i.e. precipitation plus run-on) for both locations are pre-
sented in Fig. 6. These balances show the behavior of both the rain
garden as a whole (dashed lines) and the layers within them. For
Table 3
Vegetation parameters for each rain garden.
Unit Vegetation
Sedum
k 3
S Mm 1.27
Kca 0.53b
a
Constant value for the year.
b
Sherrard and Jacobs, 2005.
c
Allen et al. (1998).
both cities, most of incoming water becomes infiltration (F) (~97%)
whereas a very minor portion is intercepted (I). Interception is
similar despite the larger interception capacity of the Sedum used in
Santiago, as the annual rainfall in Temuco is larger. Because infil-
tration rates were not exceeded by precipitation during both years
of simulation, no surface runoff was simulated. Around 90% of the
incoming water (87.9% in Santiago and 90.1% in Temuco) leave the
rain gardens through L2.
In Santiago, 70.3% of the incoming water to the rain garden
percolates from L1 to L2, whereas an additional 25% redistributes to
L2 during dry-weather conditions (Fig. 6). The remaining ~10% in L1
evapotranspires. In contrast, in Temuco redistribution to L2 during
the dry season is the most relevant flow acting over the water
incoming to L1 (~48%). This occurs because the high rates of rainfall
raise q in L1 more often (Fig. 5), which in turn increases the suction
head gradient and enhances redistribution. This phenomenon also
increases q of L2 above qFC , which explains the larger percolation
from L2 in Temuco. Proportionally, water loss through ET from L1
was greater in Santiago although the crop coefficient of grass in
Temuco exceeds that of the Sedum used in Santiago. A similar result
is reported by DehghaniSanij et al. (2004), who measured higher
values of ET0 for soils in semiarid regions. Lastly, note that redis-
tribution due to capillarity from L2 to L1 is minor in both cities,
although larger in Santiago (7.7% vs. 4.1%). Overall these results
demonstrate that redistribution during dry-weather days can
relevant, and should be simulated by models describing water
flows in LID's and green areas in general.
To illustrate the effect of the rain gardens on the incoming flow,
we use the flow duration curve, i.e. the curve defining the per-
centage of time a given flow is equaled or exceeded (Cigizoglu and
Bayazit, 2000). Hence, we compare the curves for the input flow
and the output flow (i.e. the percolation from L2) both for Santiago
and Temuco (Fig. 7a). We used values of k ¼ t ¼ 0.0056 h in
Grass Equation (20), which implies the simulation of a conduit draining
2.8 the percolation with an average response time of 40 s. There is an
0.254 overall reduction in the peak flows and an increase of the low flows
0.95c for both locations, although this is more notorious in Temuco,
where precipitation is more frequent. The peak flow attenuation
and the corresponding impact over the recession flows is illustrated
 J. Herrera et al. / Environmental Modelling & Software 95 (2017) 246e257 253

Fig. 5. Precipitation and temporal evolution of the soil water content using continuous simulation during 2012 and 2013 for Santiago (a) and Temuco (b).

in Fig. 7b and c, which compare the response to single events in

both locations. This behavior is typical of stormwater control
practices that reduce and delay peak flows.

4.3.1. Evaluation of different design practices

Fig. 6. Simulated water balance for the rain gardens in Santiago (a) and Temuco (b)
during 2012 and 2013. L1 and L2 are the soil layers 1 and 2, P is the precipitation, F is
infiltrated water, I is the intercepted water, Pe is the percolated water, ET is the
evapotranspired water, and R is the redistributed water during dry weather. Values in
parentheses are percentages of the incoming water.
We use IHMORS to evaluate the performance of the rain garden
under different design practices (i.e. vegetation types and connec-
tivity with upstream contributing areas), although one could also
vary the type of substrate. The analysis considers both grass and
Sedum for each location, as well as bare soil. Likewise, the rain
garden in Temuco was tested after removing the summer precipi-
tation to analyze the impact of a dry summer over the dynamics of
q.
Table 4 compares the principal outflows of the rain gardens with
and without vegetation. All values are expressed as percentages of
the incoming water in each location. Percolation is the lowest in
both cities (and thus evaporation is the largest) when no vegetation
is used. As expected, the water lost by evaporation (E) from soil is
proportionately higher in drier rather than humid areas. Note also
that the difference of ET between grass and Sedum is more signif-
icant in Santiago (5.3 vs 2.8). Thus, it becomes clear that vegetation
should be carefully selected when designing green infrastructure in
semiarid regions.
We now focus our analysis on assessing the temporal dynamics
of q by means of what we call the soil water content duration
(SWCD) curve. This curve represents the percentage of time that a
given value of q is equaled or exceeded, and is similar to the flow
duration curve. The SWCD curve characterizes the overall dynamics
of q, and allows detecting the number of days in which a particular
value (e.g. qFC , qWP ) is exceeded, or in which irrigation becomes

Table 4
Simulated water balance of rain gardens in Santiago and Temuco with Sedum (S),
grass (G) and bare soil (B). Precipitation becomes percolation (Pe) and evapotrans-
piration or evaporation (ET/E). Values are expressed as percentages of the incoming
water.

Santiago Temuco

S G B S G B
Fig. 7. Entering and exiting simulated flow duration curve (a) during 2012 and 2013 for
Pe 87.9 84.9 82.3 92.1 90.1 83.7
Santiago (Q IN
S and Q S
OUT
) and Temuco (Q IN OUT
T and Q T ). Hydrographs for single events
ET/E 10.0 15.3 18.9 3.8 6.6 16.1
with their starting date are presented for Santiago (b) and Temuco (c).
254 J. Herrera et al. / Environmental Modelling & Software 95 (2017) 246e257

essential. Thus, this curve can potentially be used to estimate irri-

gation costs associated with not only the amount of water needed,
but also with the amount of time involved in the process (i.e.
personnel expense, maintenance, transportation, contracts, etc.).
Fig. 8 shows the SWCD curves for Santiago (Fig. 8a) and Temuco
(Fig. 8b) as well as the critical values qFC and qWP . SWCD curves are
compared for three cases in which either Sedum (S), grass (G) or
bare soil (B) is considered for the rain garden. Furthermore, an
additional SWCD curve is plotted for Temuco, which is obtained
after removing summer precipitation. Results show that planting
Sedum instead of grass is better in terms of having overall larger
soil moisture contents. This is much more obvious in Santiago,
where for ~50% of the time the differences in q exceed 0.01 m3 m-3.
On the other hand, the difference is minor in Temuco and is more
noticeable only for the driest 10% of the time. The SWCD curves can
be compared using as a reference qWP , which corresponds to the
theoretical minimum value of q to be satisfied with irrigation. Thus,
irrigation is essential for at least 60% of the time in Santiago,
although different volumes would be needed depending on the
type of vegetation. In Temuco, irrigation is necessary for approxi-
mately 25% of the time, and the water volume involved is very
similar despite the vegetation type. Fig. 9. Soil water content duration curves for the rain garden in Santiago with Sedum
In Santiago bare soil allows having, during certain periods of connected (C) and disconnected (WC) with the contributing impervious area.
time, values of q larger than those simulated when using vegeta-
tion. This condition occurs for ~35% (grass) and ~10% (Sedum) of the
a year), a 0.01 m3 m-3 difference in q between the connected and
time (around the end of summer and beginning of fall). This does
disconnected rain garden is simulated for ~180 days. Overall, results
not occur in Temuco, as using grass or Sedum allows having values
show that capturing rainwater and using it for landscape watering
of q exceeding the ones associated with a bare soil. If Temuco had
needs is a reasonable and realistic way to reduce water used in
no summer precipitation, the q dynamics of a rain garden with
landscape irrigation (Seymour, 2005; Houdeshel et al., 2012),
grass would be more similar to that of semiarid regions like San-
particularly in regions where precipitation and the hot season are
tiago. Indeed Fig. 8b shows that q is lower than values to be ob-
concurrent.
tained with bare soil for approximately the 60 driest days in the
year (i.e. 20% of the time).
Finally, we assess the relevance of the connectivity between the
4.3.2. Evaluation of different irrigation schedules
rain garden and the upstream contributing impervious area. Fig. 9
In semi-arid and Mediterranean regions an irrigation plan and
compares the SWCD curves for Santiago simulated when such
schedule become key. To explore this issue we use IHMORS to test 3
connection is and is not considered (i.e. the garden only receives
irrigation plans for the rain garden with Sedum implemented in
water from the precipitation falling on it). The SWCD curve of the
Santiago: (1) a base user-defined irrigation program shown in
disconnected rain garden falls below that of the connected rain
Table 5 (P1) defined for an experimental green roof with Sedum in
garden, and thus it becomes clear that runoff captured by green
Santiago (Reyes et al., 2016); (2) a constant irrigation of 8.6 mm (i.e.
infrastructure can become available for the vegetation. Despite the
the daily average of P1) during 15 min, which is applied every day at
reduced number of days with precipitation in Santiago (25e30 days
8:00 a.m. only if q < qFC (i.e. the plan simulates the existence of a soil
moisture sensor) (P2); and (3) a variable irrigation program cor-
responding to a proportion of the previous 24 h of ET0 , which is
applied daily at 8:00 a.m. for 15 min (P3). In IHMORS P1 and P2 are
entered by the user, although for P2 the model determines when
the irrigation takes places based on the simulated q. The model
automatically computes P3 using the proportion factor specified by
the user.

Table 5
Irrigation program P1. Irrigation is applied during 15 min.

Month Time of application Depth (mm)

January 6:00e14:00e22:00 10
February 6:00e14:00e22:00 6
March 6:00e14:00e22:00 5
April 6:00e14:00e22:00 2.5
May 6:00e14:00e22:00 2
June 8:00 2
July 8:00 2
August 8:00 2
September 8:00 2
October 8:00 2
Fig. 8. Soil water content duration curve of (a) a rain garden in Santiago with Sedum
November 8:00e18:00 2.5
(S), grass (G) and bare soil (B), (b) a rain garden in Temuco with Sedum, grass, bare soil
December 8:00e13:00e18:00 4
and grass without considering summer precipitation (GS).
 J. Herrera et al. / Environmental Modelling & Software 95 (2017) 246e257 255

Table 6
Sensitivity results for each layer in Santiago (S) and Temuco (T). Also general results
for each layer (G1 and G2) and for both layers (G) are shown. Subscripts 1 and 2
denote the layer to which the parameter belongs. Very important and important
parameters are marked in dark gray and clear gray respectively.

Parameters Range Layer 1 Layer 2 G

S T G1 S T G2

d1 0.1e0.5 3 5 3 13 10 10 3
qi1 qr-qs 15 10 10 5 11 5 5
qr1 0.001e0.205 12 4 4 14 6 6 4
qs1 0.332e0.582 1 6 1 8 7 7 1
n1 1.039e2.09 7 1 1 1 3 1 1
L1 0e2.5 21 12 12 20 21 20 12
Ks1 288e1368 13 13 13 17 18 17 13
jb1 0e1791 11 7 7 15 8 8 7
qFC1 qr-qs 2 3 2 3 4 3 2
d2 0.1e0.5 5 20 5 6 19 6 5
qi2 qr-qs 20 18 18 18 20 18 18
qr2 0.001e0.205 9 11 9 7 9 7 7
qs2 0.332e0.582 10 2 2 2 2 2 2
n2 1.039e2.09 6 9 6 4 1 1 1
L2 0e2.5 19 14 14 21 14 14 14
Ks2 288e1368 16 15 15 11 15 11 11
jb2 0e1791 8 16 8 9 12 9 8
qFC2 qr-qs 4 8 4 12 5 5 4
Fig. 10. Soil water content duration curves for the rain garden in Santiago with Sedum Kc 0.25e1.25 14 19 14 19 17 17 14
and three irrigation schedules: a monthly irrigation program (P1), a constant irrigation S 0.254e1.27 18 17 17 16 16 16 16
program with soil moisture sensor reporting field capacity (P2), and an irrigation plan k 0.001e18 17 21 17 10 13 10 10
that replicates the previous day evapotranspiration (P3). The curve associated with no
irrigation (WI) is also presented.

Additionally, the table reports a general parameter rank for L1 (G1)

SWCD curves associated with each plan are shown in Fig. 10,
which also presents the curve when no irrigation is implemented.
For the three irrigation programs, q remains close to the wilting
point, although different water volumes are involved. P1 and P2
imply using pretty much the same amount of water (i.e. 12.59 m3
and 12.52 m3), although P2 reduces by 7 the number of days in
which irrigation is provided during the 2 years. In fact, this result
demonstrates the advantages of using the SWCD curve to under-
stand the impacts of irrigation programs beyond the irrigation
volumes involved. P3 consumes a significant less amount of water
(84% less than P1) but vegetation may still wilt during the driest
month of the year (i.e. ~10% of the time) when q < qWP , and thus
irrigation must be intensified. For the sake of comparison, the
amount of irrigation needed to preserve q always above the average
between qFC and qWP (i.e. q > 0.19) is 13.6 m3.
4.3.3. Global sensitivity analysis
A global sensitivity analysis was carried out to identify key pa-
rameters affecting the model performance. We adopted a screening
method known as LH-OAT because of its efficiency and low
computational cost (van Griensven et al., 2006). This method
combines latin hyperbolic (LH) and one factor at-a-time (OAT)
sampling, which allows controlling the total number of simulations
while an optimal representation of the parameter space is estab-
lished (van Griensven et al., 2006). As a result, the method calcu-
lates partial and total effects (in percentage) associated with the 21
parameters considered in the analysis, and ranks them in order of
importance. Following van Griensven et al. (2006) approach, pa-
rameters with ranking positions of 1, 2e6, 7e20, and 21 are cate-
gorized as very important, important, slightly important and not
important respectively.
We carried out 22,000 runs with parameters varying according
to the ranges shown in Table 6. The model performance for each run
was characterized by the MCE (Equation (22)) for the simulated
dynamics of q in each layer, after assuming the results in Fig. 5 to be
the “observations”. Table 6 shows the resulting parameters’ ranking
according to their total effects on q in each layer for each city.
and L2 (G2), equal to the highest rank among the locations. Finally,
an overall general parameter rank (G) given by the highest rank
between G1 and G2 is also reported. The resulting rankings are
different for each city, but the parameters of each layer affect the
behavior of q in it the most (i.e. parameters of L1 affect q1 and pa-
rameters of L2 affect q2 ). The effect of the parameters of L1 over q2 is
slightly more notorious than that of parameters of L2 over q1, likely
because both percolation and redistribution take place from L1 to
L2, but only redistribution occurs from L2 to L1. Overall 3 very
important parameters are identified (last column G, Table 6): the
curve shape parameter of the van Genuchten equation in L1 and L2
(n1 and n2 ), and qs in L1 (qs1 ). However, qs1 is slightly important for
simulations in L2 (column G2) whereas n2 is just important for L1.
Four parameters in L1 (d1 , qi1 , qr1 , qFC1 ) and 3 in L2 (d2 , qs2 and qFC2 )
are important, whereas 11 parameters, including those represen-
tative of the vegetation, are slightly important. Note that together
with n, qFC and qS are consistently sensitive parameter regardless of
the location and the layer. Hence, the percolation process driven by
the [qFC -qS ] range, proves to be very relevant in controlling the
performance of rainfall gardens. Note that qWP was not included in
the sensitivity analysis as neither irrigation nor evaporation from
bare soil are simulated.
5. Conclusion
This study developed and tested IHMORS, a hydrological model
which simulates surface and subsurface processes to analyse the
performance and irrigation of green infrastructure and stormwater
facilities at residential scales. Laboratory and numerical experi-
ments allowed validating the bare soil evaporation, subsurface
runoff hydrograph and soil moisture redistribution routines. These
critical components of IHMORS are rarely simulated explicitly in
other similar models, and become very relevant in semiarid and
Mediterranean environments. Other capabilities of IHMORS
include: (1) it continuously simulates the soil water content dy-
namics and surface and subsurface runoff; (2) it can simulate a
wide range of spatial configurations by combining and connecting
256 J. Herrera et al. / Environmental Modelling & Software 95 (2017) 246e257
different contributing areas and drainage control practices; (3) each
subarea can contain different soil layers interconnected through
vertical flows representing percolation and redistribution.
The model was used to simulate the performance during 2 years
of a 2 m2 rain garden controlling an impervious area of 10 m2. Both
a dry (Santiago) and wet (Temuco) extreme of Chilean Mediterra-
nean climates, as well as different vegetation types and irrigation
practices were simulated. We conclude that:
1 The model provided a comprehensive understanding of the
performance in terms of both runoff control and the irrigation
needs for maintenance. Indeed, surface runoff was not gener-
ated in any of the rain gardens tested, but significant differences
in terms of irrigation needs were simulated.
2 In Temuco irrigation became crucial approximately 25% of the
time, regardless of the type of vegetation. On the other hand,
less water was evapotranspired from the rain garden with
Sedum in Santiago, and differences larger than 0.01 m3 m-3 in
soil water content between rain gardens with Sedum and grass
were simulated during ~50% of the time. Nonetheless, irrigation
is essential for at least 60% of the time in Santiago, regardless of
the vegetation.
3 In Santiago, daily irrigation using the previous day simulated
evapotranspiration reduces by 80%e85% the volume of water
employed by other irrigation programs in which water pulses
are applied throughout the day. Nonetheless more irrigation is
still essential during the driest days. Interestingly, despite
considering the same annual water volume, one irrigation pro-
gram required 7 more days of irrigation than the other during
the study period.
4 In the semiarid climate of Santiago, run-on water entering the
rain gardens from the contributing area decreases its irrigation
needs. Despite rainfall occurring on average 25e30 days per
year in the city, the run-on from impervious areas is associated
with significant differences in the soil water content for ~180
days as compared to those simulated without the run-on
contribution.
5 The model's most sensitive parameters are the curve shape
parameter of the van Genuchten equation and the saturated
water content of the top layer, which affect the percolation.
Potential directions for future research with the model include:
1) improving it by incorporating processes such as subsurface
horizontal flow, and pollutant transport and removal, 2) expanding
its use to evaluate real stormwater facilities in several climate re-
gions, and 3) linking its capabilities with a cost analysis to estimate
irrigation and maintenance costs associated with both water vol-
umes and durations of irrigation involved.
Acknowledgements
This study was funded by FONDECYT through Grant 1131131. We
thank the support from CONICYT/FONDAP Grants 15110020 and
15110017, and the Arturo Cousin ~ o Lyon and CONICYT-PCHA/2014-
22140398 scholarships. Bare soil evaporation data were provided
by grant INNOVA-CORFO 12IDL2-13630. Lastly, we thank the ref-
erees for their constructive suggestions. This work was partially
developed within the framework of the Panta Rhei Research
Initiative of the International Association of Hydrological Sciences.
References
Allen, R.G., Pereira, L.S., Raes, D., Smith, M., 1998. Crop Evapotranspiration: Guide-
lines for Computing Crop Water Requeriments. FAO, Rome (Irrigation and
Drainage paper).
Allen, R.G., Pereira, L.S., Smith, M., Raes, D., Wright, J.L., 2005. FAO-56 Dual crop
coefficient method for estimating evaporation from soil and application ex-
tensions. J. Irrigat. Drain. Eng. 131 (1), 2e13.
Ascione, F., Bianco, N., de’ Rossi, F., Turni, G., Vanoli, G.P., 2013. Green roofs in Eu-
ropean climates. Are effective solutions for the energy savings in air-condi-
tioning? Appl. Energy 104, 845e859.
Bennett, N.D., Croke, G.F.W., Guariso, G., Guillaume, J.H.A., Hamilton, S.H.,
Jakeman, A.J., Marsili-Libelli, S., Newham, L.T.H., Norton, J.P., Perrin, C.,
Pierce, S.A., Robson, B., Seppelt, R., Voinov, A.A., Fath, B.D., Andreassian, V., 2013.
Characterising performance of environmental models. Environ. Model. Softw.
40, 1e20.
Beven, K.J., 2012. Rainfall-runoff Modelling: the Primer, second ed. Wiley-Blackwell.
Bras, R.L., 1990. Hydrology: an Introduction to Hydrologic Science, 1 ed. Adisson-
Wesley, Boston.
Burszta-Adamiak, E., Mrowiec, M., 2013. Modelling of green roofs' hydrologic per-
formance using EPA's SWMM. Water Sci. Technol. 68 (1), 36e42.
Cahill, T.H., 2012. Low Impact Development and Sustainable Stormwater Manage-
ment, 1 ed. Wiley, New Jersey.
Carson, T., Keeley, M., Marasco, D.E., McGillis, W., Culligan, P., 2017. Assessing
methods for predicting green roof rainfall capture: a comparison between full-
scale observations and four hydrologic models. Urban Water J. 14 (6), 589e603.
Cigizoglu, H.K., Bayazit, M., 2000. A generalized seasonal model for flow duration
curve. Hydrol. Process. 14 (6), 1053e1067.
Corradini, C., Melone, F., Smith, R.E., 2000. Modeling local infiltration for a two-
layered soil under complex rainfall patterns. J. Hydrol. 237 (1), 58e73.
DehghaniSanij, H., Yamamoto, T., Rasiah, V., 2004. Assessment of evapotranspiration
estimation models for use in semi-arid environments. Agric. Water Manag. 64
(2), 91e106.
Direccio
n General de Aeron
autica Civil (DGAC), 2016. Direccio
n Meteorolo
gica de
Chile. http://www.meteochile.cl (accedded 15/09/2016).
Everett, G., Lamond, J., Morzillo, A.T., Chan, F.K., Matsler, A.M., 2015. Sustainable
drainage systems: helping people live with water. Proc. ICE-Water Manag. 169
(2), 94e104.
Fletcher, T.D., Shuster, W., Hunt, W., Ashley, R., Butler, D., Arthur, S., Viklander, M.,
2014. SUDS, LID, BMPs, WSUD and more - the evolution and application of
terminology surrounding urban drainage. Urban Water J. 12 (7), 525e542.
Freni, G., Oliveri, E., 2007. Mitigation of urban flooding: a simplified approach for
distributed stormwater management practices selection and planning. Urban
Water J. 2 (4), 215e226.
Green, W.H., Ampt, G.A., 1911. Studies on soil physics, 1. The flow of air and water
through soils. J. Agric. Sci. 4 (1), 1e24.
Guo, J.C., Luu, T.M., 2015. Hydrologic model developed for stormwater infiltration
practices. J. Hydrolog. Eng. 20 (9), 1e8.
Houdeshel, C.D., Pomeroy, C.A., Hultine, K.R., 2012. Bioretention design for xeric
climates based on ecological principles. J. Am. Water Resour. Assoc. 48 (6),
1178e1190.
Houdeshel, C.D., Pomeroy, C.A., 2013. Storm-water bioinfiltration as no-irrigation
landscaping alternative in semiarid climate. J. Irrigat. Drain. Eng. 142 (2), 1e6.
Houdeshel, C.D., Hultine, K.R., Johnson, N.C., Pomeroy, C.A., 2015. Evaluation of three
vegetation treatments in bioretention gardens in a semi-arid climate. Landsc.
Urban Plan. 135, 62e72.
Huang, J.J., Li, Y., Niu, S., Zhou, S.H., 2014. Assessing the performances of low impact
development alternatives by long-term simulation for a semi-arid area in
Tianjin, Northern China. Water Sci. Technol. 70 (11), 1740e1745.
Huber, W.C., Rossman, L.A., Dickinson, R.E., 2010. EPA storm water management
model, SWMM5. In: Singh, V.P., Frevert, D.K. (Eds.), Watershed Models. Taylor &
Francis, Boca Raton, Florida, USA, pp. 343e346.
Instituto de Investigaciones Agropecuarias, 2012. Estaciones. Agromet. Red agro-
meteorolo
gica de Inia. www.agromet.inia.cl (accedded 2015).
Legates, D.R., McCabe, G.J., 1999. Evaluating the use of “goodness-of-fit” measures in
hydrologic and hydroclimatic model validation. Water Resour. Res. 35 (1),
233e241.
Li, Y., Babcock, R.W., 2014. Green roof hydrologic performance and modelling: a
review. Water Sci. Technol. 69 (4), 727e738.
Lucas, W.C., 2006. The SPAW model: application to infiltrating BMP facilities. In:
World Environmental and Water Resource Congress 2006: Examining the
Confluence of Environmental and Water Concerns, pp. 1e10. http://dx.doi.org/
10.1061/40856(200)395.
Merriam, R.A., 1960. A note on the interception loss equation. J. Geophys. Res. 65
(11), 3850e3851.
Ministerio de Obras Públicas (MOP), 2013. Aspectos generales del drenaje Urbano.
In: Manual de drenaje urbano, pp. 142e146. Santiago.
Mualem, Y., 1976. A new model for predicting the hydraulic conductivity of un-
saturated porous media. Water Resour. Res. 12 (3), 513e522.
Palla, A., Gnecco, I., 2015. Hydrologic modeling of Low Impact Development systems
at the urban catchment scale. J. Hydrol. 528, 361e368.
€ppen-
Peel, M.C., Finlayson, B.L., McMahon, T.A., 2007. Updated world map of the Ko
Geiger climate classification. Hydrol. Earth Syst. Sci. 11, 1633e1644.
Reyes, R., Bustamante, W., Giron

n, P.A., Rojas, V., Suarez, F., Vera, S.,
as, J., Paste
Victorero, F., Bonilla, C.A., 2016. Effect of substrate depth and roof layers on
green roof temperature and water requirements in semi-arid climate. Ecol. Eng.
97, 624e632.
Rosa, D.J., Clausen, J.C., Dietz, M.E., 2015. Calibration and verification of SWMM for
low impact development. J. Am. Water Resour. Assoc. 51 (3), 746e757.
Rossman, L.A., 2010. Storm Water Management Model User's Manual. Version 5.0.
 J. Herrera et al. / Environmental Modelling & Software 95 (2017) 246e257
National Risk Management Research Laboratory, Office of Research and
Development, US Environmental Protection Agency, Cincinnati.
Sample, D.J., Heaney, J.P., 2006. Integrated management of irrigation and urban
stormwater infiltration. J. Water Resour. Plan. Manag. 132 (5), 362e373.
Sample, D., Lucas, W., Janeski, T., Roseen, R., Powers, D., Freeborn, J., Fox, L., 2014.
Greening Richmond, USA: a sustainable urban drainage demonstrations project.
Proc. ICE-Civil Eng. 167 (2), 88e95.
Savabi, M.R., Williams, J.R., 1995. Evaporation and Environment, WEPP Model
Documentation. USDA-ARS National Soil Erosion Research Laboratory
Publication.
Saxton, K.E., Willey, P.H., 2005. The SPAW model for agricultural field and pond
hydrologic simulation. In: Singh, V.P., Frevert, D.K. (Eds.), Mathematical
Modeling of Watershed Hydrology. CRC Press, LLC, Boca Raton, Florida, USA,
pp. 401e435.
Seymour, R. S. (2005). Capturing rainwater to replace irrigation water for land-
scapes: Rain harvesting and rain gardens. Proceedings of the 2005 Georgia
water resources conference; 25e27 Apr. 2005; Athens, GA.
Sherrard, J.A., Jacobs, J.M., 2005. Vegetated roof water-balance model: experimental
and model results. J. Hydrolog. Eng. 17 (8), 858e868.
Shorten, P.R., Pleasants, A.B., 2007. A stochastic model of urinary nitrogen and water
flow in grassland soil in New Zealand. Agric. Ecosyst. Environ. 120 (2), 145e152.
Simunek, J., Sejna, M., Saito, H., Sakai, M., van Genuchten, M.T., 2013. The HYDRUS-
1D Software Package for Simulating the One-dimensional Movement of Water,
Heat, and Multiple Solutes in Variably-saturated Media, Version 4.17. HYDRUS
257
software series 3. Department of Environmental Sciences, University of Cali-
fornia Riverside, Riverside, California, USA.
Smith, R.E., Smettem, K.R., Broadbridge, P., Woolhiser, D.A., 2002. Infiltration Theory
for Hydrologic Applications. American Geophysical Union, Washington DC.
Snyder, R.L., Bali, K., Ventura, F., Gomez-MacPherson, H., 2000. Estimating evapo-
ration from bare or nearly bare soil. J. Irrigat. Drain. Eng. 126 (6), 399e403.
Stewart, B.A., Musick, J.T., 1982. Conjunctive use of rainfall and irrigation in semiarid
regions. In: Hillei, D. (Ed.), Advances in Irrigation, pp. 1e24.
Urban Drainage and Flood Control District (UDFCD), 2010. Stormwater best man-
agement practices. In: Urban Storm Drainage, Criteria Manual. Water resources
publications, Denver, Colorado.
van Genuchten, M.T., 1980. A closed-form equation for predicting the hydraulic
conductivity of unsaturated soils. Soil Sci. Soc. Am. J. 44 (5), 892e898.
van Griensven, A., Meixner, T., Grunwald, S., Bishop, T., Diluzio, M., Srinivasan, R.,
2006. A global sensitivity analysis tool for the parameters of multi-variable
catchment models. J. Hydrol. 324 (1), 10e23.
Walsh, T.C., Pomeroy, C.A., Burian, S.J., 2014. Hydrologic modeling analysis of a
passive, residential rainwater harvesting program in an urbanized, semi-arid
watershed. J. Hydrolog. 508, 240e253.
Xiao, Q., McPherson, E.G., Simpson, J.R., Ustin, S.L., 2007. Hydrologic processes at the
urban residential scale. Hydrol. Process. 21 (16), 2174e2188.
Yang, W., Li, D., Sun, T., Ni, G., 2015. Saturation-excess and infiltration-excess runoff
on green roofs. Ecol. Eng. 74, 327e336.