Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
b
and Kh; z K
sat
zh=h
S
32b
15
where h
S
is the water content at natural saturation, W
e
the suction head at air entry, and b a dimensionless
parameter that denes the shape of the water retention
and conductivity curve. These parameters may be esti-
mated from soil texture (Clapp and Hornberger, 1978).
The BrooksCorey Law has been adapted to take into ac-
count the variation in hydraulic conductivity at saturation
with depth:
Ft K
V
1
Whhti
z
S
t z
V
t
_ _
16
The transpiration ux from vegetation, including trees,
serves to empty the vadose zone proceeding upwards. This
ux has been determined by introducing both the depth of
the root zone and the wilting point water content (Feddes
et al., 2001):
z
s
Vadose zone
Capillarity fringe
z
s
-z
cf
z
v
0
z
Saturated zone
Figure 5 Detail of the representation of the vadose zone.
A distributed hydrological model for urbanized areas Model development and application to case studies 273
Tt
h
root
t Dt h
wp
h
S
h
wp
PETt
for the natural soil type of land use
Tt
h
root
t Dt h
wp
h
S
h
wp
PETtf
str
tree
for the street type
Tt 0 for the house type
17
where h
root
is the mean water content within the root zone
(depending on the model parameter z
root
), h
wp
the wilting
point water content (i.e. the point where the suction head
W(h) = 150 m in Eq. (15)), and f
str
tree
the fraction of street sur-
face area covered by trees.
h
root
t Dt ht Dt if z
root
6 z
S
t Dt Dz
cf
;
h
root
t Dt
z
S
t Dt Dz
cf
z
root
_ _
ht Dt
1
z
S
t Dt Dz
cf
z
root
_ _
h
S
otherwise
18
A critical analysis of vadose parameterization is addressed
in Appendix by means of a comparison between this simpli-
ed layout and a more rigorous approach based on Richards
equation.
The saturated zone
The saturated zone is characterized by the saturation level;
it exchanges water with the vadose zone and is drained by
the sewer pipe contiguous with the UHE. Moreover, the sat-
urated zone is used to represent horizontal water exchanges
between the three types of surfaces.
The saturation level is assumed to be uniform for a given
UHE. This mean saturation level z
S
is expressed as the aver-
age saturation level of each surface type weighted by the
land use surface proportion.
If this saturation level is higher than the depth of the
drainage network z
net
, soil water may inltrate into the
drainage network via groundwater drainage due to defects
in the tightness, which leads to network inltration and
empties the saturated zone. This inltration ux has been
estimated by considering that the pipe behaves like an ideal
drain (Cassan, 1986; Gustafsson et al., 1996):
I
drain
t K
nat
S
e
z
S
t=M
k
L
z
soil
z
net
z
S
t
_ _
l
19
where I
drain
denotes the ux of water from the saturated
zone into the drainage network, L, z
soil
and z
net
are geomet-
rical features of the UHE, k and l parameters that depend
on the type and state of both the drain and trench contain-
ing the drain. This groundwater drainage modies the mean
decit and saturation levels according to:
z
S
t z
S
t Dt
I
drain
t
h
S
Dt 20
UHE water budget
Runoff at the outlet of a UHE equals the sum of the various
ow contributions, which encompass the runoff ows from
each land use type and the groundwater drainage, i.e.:
R
tot
t f
hou
A
hou
R
hou
t f
str
A
str
R
str
t f
nat
A
nat
R
nat
t
f
drainrw
A
tot
I
drain
t 21
where f
hou
, f
str
and f
nat
represent the fractions of each land
use type surface connected to the drainage system, and
f
drainrw
represents the fraction of soil inltration water
draining into the rainwater drainage system.
The surface runoff that does not contribute to direct run-
off because of the absence of direct connection to the
drainage network is allocated to the surface reservoir of
the uncovered land use type, hence modifying the water
budget of S
nat
surf
(see Eq. (4)) as follows:
S
nat
surf
t S
nat
surf
t 1 f
hou
R
hou
t 1 f
str
R
str
t
1 f
nat
R
nat
t 22
The 1D modeling representation does not include lateral
exchanges between the three types of land use. These lat-
eral uxes within the vadose zone are indirectly introduced
by averaging the water stored in the land use, weighted by
the land use surface area proportion.
Summary of the model running at the UHE scale
Model inputs at the scale of an urban hydrological element
consist of: the meteorological forcing, which includes
precipitation and potential evapotranspiration; and the
initial saturation depth. The model has been designed to
simulate the evolution in hydrological variables over a
long time series with short time steps of just a few min-
utes to remain consistent with the dynamics of urban
catchments. Model outputs contain the various water
uxes computed for each time step, which can then be
separated into: (i) transfer of water from soil to atmo-
sphere; (ii) rainwater inltration into the vadose zone I
(iii) ow rate in the rainfall drainage system including
the surface runoff on each land use type, and inltration
ux into the sewer network I
drain
; (iv) water storage within
the various compartments, including mean moisture con-
tent of the vadose zone and mean saturation level of
the UHE.
The model requires both morphological features and
physical parameters. The morphological features are listed
in Table 1 and encompass, for each UHE, the surface areas
of each land use type, the fraction of surface covered by
trees f
tree
, and the sewer depth z
net
. These features have
been deduced from urban databanks, but could still benet
from an analysis of satellite imagery and aerial photogra-
phy. The factors f
hou
, f
str
and f
nat
, which represent the di-
rect connection of each land use type to the drainage
network, may be deduced from either local surveys or sta-
tistical studies, and the factor f
drainrw
, depicting the sepa-
ration between direct inltration of soil water into the
rainwater network and into the wastewater network, has
been estimated from the relative drainage densities of each
network.
The model parameters are listed in Table 2; most entries
should have been deduced from physical considerations,
either literature reviews or eld measurements. This
parameterization has also resulted from a more detailed
modeling layout that served to solve the reference Rich-
ards equation (Berthier et al., 2004).
274 F. Rodriguez et al.
Modeling of hydrological processes at the
catchment scale
Flow routing
Rodriguez et al. (2003) have shown that the geometric
information available from UDBs allows constructing the
detailed vector map of water ow paths along UHEs, street
gutters and inside the sewer network (hereafter referred
to as the runoff branching structure and denoted RBS)
and to derive the transfer function of urban catchments.
The present study extends the paper of Rodriguez et al.
(2003). The vector map of ow paths provides a basis for
routing water from each UHE to the catchment outlet
and, more generally, for calculating the ow at any point
of the catchment drainage system. The hydraulic routing
conguration has been modied to contain two modeling
stages: (i) routing of surface runoff on streets from UHEs
up to the sewer inlets, as represented by the travel time
routing procedure used in URBS-UH (Rodriguez et al.,
2003); and (ii) hydraulic routing inside sewer networks,
as represented by using the classical MuskingumCunge
scheme, which offers an approximate solution to the diffu-
sive wave equation (Cunge, 1969).
The routing conguration is given for each segment of
the RBS, i.e.:
Q
j1
t Dt C
0
Q
j
t Dt C
1
Q
j
t C
2
Q
j1
t 23
with Q denoting the discharge at any node, j and j + 1 the up
and down nodes of the RBS segments, and C
0
, C
1
and C
2
the
routing coefcients whose sum equals 1:
C
0
kx 0:5Dt
k 1x 0:5Dt
;
C
1
kx 0:5Dt
k 1x 0:5Dt
and C
2
k 1x 0:5Dt
k 1x 0:5Dt
24
with k and x denoting the routing parameters. Parameter k
is interpreted as the time taken by the wave to propagate
along the network segment, and parameter x is a dimension-
less coefcient weighting the effect of inow and outow
on the water stored in the network segment. These param-
eters may be estimated from the following equations (Mont-
aldo et al., 2004):
k L
j
=x and x
1
2
1
Q
j
B
j
xS
j
L
j
_ _
25
where x is the kinematic wave celerity, L
j
and S
j
the length
and slope of the considered jth segment respectively, B
j
the
segment width, and Q
j
the current discharge.
Model owchart
The initial conditions are rst determined to be homogenous
over the entire catchment: the storage reservoirs are zer-
oed and the saturation level initialized to an acceptable va-
lue, corresponding to the actual saturation level on the
catchment at this initial date (possibly measured in number
of wells or piezometers). If such saturation level is not avail-
able, a warm-up period will be needed. A sensitivity analysis
showed that an error of 43% on the initial saturation level
leads to a necessary 3 months warm up period. The tree
interception module is to be implemented rst; the rainwa-
ter falling on surfaces covered by trees gets subtracted from
the intercepted water quantity. The modeling approach is
then applied to each UHE independently one from the other.
The entire water budget is allocated with the surface pro-
cesses and soil processes modules and generates the ow
rate contribution of any UHE. The routing conguration is
next applied to each stream of the RBS for calculating the
ow rate in any stream, especially that at the catchment
outlet. All other hydrological variables introduced in Model-
ing of hydrological processes at the UHE scale section re-
main available for each time step on any UHE of the
catchment (see Table 3).
Table 2 Model parameters of one UHE
Parameter Unit Description
S
tree,min
mm Minimum value of the tree interception
reservoir
a min
1
Drainage law coefcient for tree
interception
S
surf,max
mm Maximum capacity of the surface
reservoir (for each land use)
K
S
m/s Hydraulic conductivity at natural
saturation(for each land use)
h
S
Water content at natural saturation
W
e
Suction head at air entry
b Retention curve exponent
M Scaling parameter of the hydraulic
conductivity
a
V
Representative position of the vadose
zone
Dz
cf
m Thickness of the capillary fringe
z
root
m Root depth
k Groundwater drainage coefcient
l Groundwater drainage exponent
Table 1 Morphological features of one UHE
Feature Unit Description
A
*
m
2
Surface area of each land use type: house
(hou), street (str), non-covered surface
(nat)
A
tot
m
2
Total surface area of UHE:
A
tot
= A
hou
+ A
str
+ A
nat
f
str
tree
Street surface area covered by trees
f
nat
tree
Natural soil surface area covered by trees
L m Length of the UHE
z
net
m Depth of the rainwater drainage network
z
soil
m Altitude of the UHE
f
drainrw
Proportion of rainwater drainage network
f
hou
Fraction of house surface connected to
the rainwater drainage network
f
str
Fraction of house street connected to the
rainwater drainage network
f
nat
Fraction of natural soil connected to the
rainwater drainage network
A distributed hydrological model for urbanized areas Model development and application to case studies 275
The minimum data required to run the model are hydro-
logical data of precipitation and PET (Potential evapotrans-
piration) taken from the closest meteorological station. An
initial value of saturation level is welcome, as discussed
above. In the end, some eld experiments could help to
estimate soil hydrodynamic properties, as discussed in Ini-
tial URBS-MO validation on the Reze catchment: Focus on
water budget restitution section.
Presentation of the case studies
Two urban catchments within the Nantes metropolitan area
(Western France) have been considered for our evaluation
of URBS-MO. The rst one is located in Reze; the drainage
network associated with this catchment is composed of sep-
arate sewers and its land use is homogenous residential.
Impervious surfaces account for 37% of drainage network
connections and the soil has been assessed as a silty clay.
This catchment was monitored over a 10-year period (Ber-
thier et al., 1999) and is currently being used to conduct
hydrological modeling tests (Rodriguez et al., 2000; Rodri-
guez et al., 2003; Berthier et al., 2004). The data measured
comprise: (i) rainfall intensity and discharge at the outlet
from 1991 until 2002; (ii) the water table level through
two piezometers installed at two locations of the catch-
ment, from September 1995 until December 1998; (iii) the
groundwater drainage runoff measured during three winter
periods in the wastewater drainage network (Belhadj
et al., 1995), which provides an estimation of both the rain-
water and wastewater components in groundwater drainage
(Berthier et al., 2004); and (iv) meteorological data, such as
Penman evapotranspiration, collected from the meteoro-
logical station located 5 km from this catchment. Fig. 6 dis-
plays the experimental set-up and catchment morphology.
The second catchment is located within the Nantes city
limits and is called Gohards (180 ha) and has already been
used to identify hydrological response of urban catchments
(Rodriguez et al., 2005). The drainage network of this catch-
ment is composed of separate sewers, and the basin contains
single- and multi-family housing, commercial areas and indus-
trial zones. The data derived comprise: (i) rainfall intensity
and discharge at the outlet measured from 1998 until 2002;
(ii) meteorological data, such as Penman evapotranspiration,
collected fromthe meteorological station located 13 kmfrom
the catchment outlet. The Gohards catchment is shown in
Fig. 7, along with the various measurements points.
The rainfall event sample for Reze is constituted of 850
events (exceeding 2 mm) over 10 years, while the rainfall
event sample for Gohards contains 330 events over 4 years.:
the mean cumulative rainfall depth is around 9 mm for both
rainfall event samples. The primary morphological charac-
teristics of both catchments have been listed in Table 4,
which shows how the two catchments differ. In terms of
land use, the Reze catchment is more homogenous than Go-
hards (mainly due to its much smaller size); the mean size of
the Reze parcels is three times smaller than that of the Go-
hards parcels, with the standard deviation of all character-
istics being higher in the Gohards data.
Initial URBS-MO validation on the Reze
catchment: Focus on water budget restitution
The Reze catchment has been chosen for the validation of
water budget processes, according to URBS-MO depiction.
Three reasons explain this choice. First, the catchment is
small and may be represented in its entirety by a set of sim-
ilar hydrological elements. Second, The hydrological data
available (ow rates at the outlet, piezometric measure-
ments, groundwater drainage runoff data), make it possible
to evaluate both the soils contribution to total ow rate
and the simulation of saturation level variation. Third, sev-
eral in situ investigations have been performed to estimate
the hydrodynamic characteristic conditions within the soil.
Three comparison criteria were adopted for the model
evaluation: the Nash criterion (denoted C
Nash
), a bias crite-
rion (denoted C
b
), and the determination coefcient R
2
:
C
Nash
1
tn
tt
1
V
2
t V
1
t
2
tn
tt
1
V
1
t
V
1
2
C
b
tn
tt
1
V
2
t V
1
t
tn
tt
1
V
1
t
R
2
tn
tt
1
V
2
t
V
2
V
2
t
V
1
_ _
2
tn
tt
1
V
1
t
V
2
tn
tt
1
V
1
t
V
1
2
26
where V
1
is the reference variable,
V
1
the temporal average
of this variable, and V
2
the simulation variable.
This work has focused on two variables representative of
the general model behavior: ow rate at the catchment out-
let Q
outlet
, to be denoted Q for the sake of convenience; and
the saturated zone level z
S
, considered to be uniform over
this small catchment.
Table 3 Hydrological variables (storages and ux) of the
model Dt is the model time increment
Variable Unit Description
S
tree
mm Tree reservoir storage
S
surf
mm Surface reservoir storage of each land
use type
S
vad
mm Vadose zone storage of each land use
type
h % Mean moisture content in the vadose
zone of each land use type
z
S
mm Depth of the saturated zone of each
land use type
z
V
mm Representative depth of the vadose
zone
E
tree
mm Evaporation ux from trees
P mm/Dt Precipitation
O
tree
mm/Dt Non-intercepted precipitation
ETP mm/Dt Potential evapotranspiration
I mm/Dt Inltration ux from surface to vadose
zone
E
surf
mm/Dt Evaporation ux from surface
R mm/Dt Surface runoff ux
T mm/Dt Transpiration ux
F mm/Dt Flux between vadose zone and
saturated zone
I
drain
mm/Dt Groundwater drainage
Q m
3
/Dt Runoff ow at the UHEs outlet
276 F. Rodriguez et al.
Figure 7 Gohards catchment: triangles show the pluviometers, the star shows both the outlet and the discharge measurement in
the stormwater drainage system.
Figure 6 Reze catchment: circles show the piezometers location, triangles the pluviometers, the star shows both the outlet and
the discharge measurement in the stormwater drainage system, and the diamond the discharge in the wastewater drainage system.
A distributed hydrological model for urbanized areas Model development and application to case studies 277
Model implementation principle
Given the homogenous type of housing stock (Rodriguez
et al., 2003), the 70 component parcels have been repre-
sented by a single UHE, whose morphological features (as
deduced from UDBs) are listed in Table 4. Continuous simu-
lation runs have been carried out over a 10-year period using
a 5-min time step. The base run could be initialized thanks
to a realistic set of parameters (listed in Table 5) whose val-
ues are explained below.
A sensitivity analysis has been conducted and led to the
following conclusions: two parameters may cause major
modications to hydrological behavior within URBS-MO:
the hydraulic conductivity at natural saturation of the
natural soil type K
nat
S
, and the groundwater drainage coef-
cient k. Regarding the hydraulic conductivity of the natural
soil type, its estimation should be deduced from eld mea-
surements as well as from a hydrodynamic study of the tar-
get soil. This concern remains fairly infrequent within the
urban environment, where soil processes tend to get ne-
glected, yet becomes prominent once it has been accepted
that the urban soil plays a role in the hydrological processes
for urban areas. The saturated hydraulic conductivity of ur-
ban soils however is difcult to estimate, due both to the
heterogeneity of this soil, which encompasses different
composite materials, and to the presence of various under-
ground sewers (De Kimpe and Morel, 2000). An average rep-
resentative value is not easy to estimate. To apply the
model to the Reze catchment, the value of K
nat
S
has been ex-
tracted from previous eld measurements on this catch-
ment (see below).
Estimating the groundwater drainage coefcient k is a
more ambiguous undertaking. This parameter could a priori
be deduced from physical considerations, given its impact
on sewer pipe inltration capacity and its dependence on
sewer network conditions: age, pipe materials, etc. This
type of information should be available within urban data-
banks in the future and could be examined for specic stud-
ies. Moreover, since groundwater drainage uxes depend on
this parameter to a great extent, the parameter value could
be deduced from sewer base ow measurements. The meth-
ods developed to quantify clear parasitic water (or inltra-
tion/inow) within wastewater systems (Joannis, 1994)
could prove useful in quantifying the base ow of rainwater
systems. To proceed with estimation of this factor and pro-
vided observed ow rates at the catchment outlet are avail-
able, we would suggest: (i) estimation adjustment in order
to minimize the discrepancy between observed and simu-
lated volumes, and (ii) the use of groundwater ow rate
observations if available. Further discussion will be provided
on this topic, which will necessitate further research inves-
tigations to be able to effectively model and measure soil
water inltration into pipes.
The values of tree interception parameters were ex-
tracted from Calder (1977) and Grimmond and Oke (1991);
they are representative of deciduous trees. The minimum
value of tree interception reservoir S
min
is 1 mm, except
during the winter season when it equals 0.3 mm due to
the lack of leaf cover. The maximum reservoir capacities
have been extracted from Hollis and Ovenden (1988) for
impervious land uses and from Grimmond and Oke (1991)
for the green areas; these reveal a large storage capacity
of green areas, in comparison with the small storage capac-
ities of roofs and streets (Zech et al., 1994). The hydraulic
conductivity of the street has been extracted from Hassan
and White (1997). The hydraulic characteristics of the soil
were mainly derived from eld measurements taken on soil
samples from two different depths (0.2 m and 1.5 m below
the surface) in Reze (Berthier et al., 2004). This sample was
characterized by a h
S
of 0.43, a w
e
of 0.2 m and b equal to 5.
These values are characteristic of a loamy or sandy loam soil
type, according to the classication of Cosby et al. (1984);
the natural hydraulic conductivity K
nat
S
equals
1.3 10
5
m s
1
. M was deduced from tting of Eq. (8) for
a 1.5-m soil depth. The point assumed to be representative
of the vadose zone denitely constitutes one of the compu-
tational parameters; a value of a
v
= 0.5 m has been adopted
a priori, in accordance with the sensitivity analysis
Table 4 Morphological features for Reze and Gohards
catchment
Catchment Reze Gohards
Mean Surface area (m
2
) 709 1948
Standard deviation 139 5405
Mean fraction of house land use type f
hou
0.16 0.19
Standard deviation 0.05 0.14
Mean fraction of street land use type f
str
0.19 0.27
Standard deviation 0.08 0.12
Mean fraction of natural land use type f
nat
0.65 0.54
Standard deviation 0.08 0.19
Mean fraction of tree surface f
tree
0.14 0.16
Standard deviation 0.06
Mean UHE length L (m) 43.3 62.5
Standard deviation 15.4 53.2
Mean sewer network depth z
net
(m) 1.2 3.55
Standard deviation 0.7 1.4
Table 5 Physical parameters for Reze catchment
Parameter Unit Value Variation range
(sensitivity analysis)
S
min
mm 1/(0.3 in winter) 0.0015
a min
1
0.04 00.2
S
nat
S max
mm 5 0.00140
S
hou
S max
mm 0.5 0.0012.5
S
str
S max
mm 3.5 0.00117.5
K
nat
S
m/s 1.3e
5
1.e
8
1e
4
K
str
S
m/s 7.5e
8
1.e
9
1e
6
h
S
/ 0.43 0.300.50
W
e
/ 0.20 0.050.70
b / 5 212
M / 0.2 0.0510
a
V
/ 0.5
z
root
m 1.3 0.14
k / 4 0100
l / 2 0.15
278 F. Rodriguez et al.
presented in Appendix. The depth of root zone has been
used as a mean representative value of 1.3 m was chosen,
with respect to a depth of 0.1 m for grass and 3 m for small
deciduous trees (Feddes et al., 2001) and to the respective
surface areas of these vegetation land use. The groundwa-
ter drainage law parameters may be correlated with the
state of the sewer system: a small value of k characterizes
a sewer network with few defects, while a high k value is
more typical of a sewer network in poor condition, with
the potential for considerable water inltration. The values
selected for this base run are those proposed by Cassan
(1986) for an agricultural drain, although this parameter
would still need to be calibrated.
As mentioned above, the high sensitivity of the model to
the value of groundwater drainage coefcient k and its lack
of physical meaning impose adjusting this value with re-
spect to the discrepancy between simulated and observed
runoff volumes. Fig. 8a shows the comparison between run-
off depths simulated for each rainfall event, using the refer-
ence value, and the observed runoff depths. The ow rate
has been overestimated and the bias criterion on runoff
depths equals 46%. This scatter plot indicates a discrepancy
due to winter rainfall events: for these events, groundwater
drainage is very low, since a small value has been assigned
to drainage coefcient k, which consequently results in a
very high saturation level. This situation induces a large
contribution from natural soil to runoff, which amounts to
approximately 10% of total rainfall (see Table 6), i.e. equiv-
alent to the house or street runoff (not a realistic result
within this urban catchment). Moreover, in this case, the
simulated groundwater drainage ux has been underesti-
mated, in comparison with observed inltration ow rates.
A trial-and-error calibration of this k coefcient has been
carried out with respect to the bias criterion estimated on
the total ow rate and has led to a value of k = 40. With this
value, both the agreement between observed and simulated
groundwater drainage uxes and the bias criterion on rain-
fall events are improved (Fig. 8b). Increasing groundwater
drainage then contributes to decreasing the saturation level
and facilitates inltration of surface water into the vadose
zone. Runoff produced by streets and natural soils therefore
gets reduced (Table 6) and the contribution of natural soil is
signicantly reduced, reading close to zero.
Water budget evaluation
The availability of a very long and continuous data series of
ow rates and saturation levels enables focusing the model
evaluation step on the simulation of runoff production and
on the evolution in the soils hydric state. Model perfor-
mance is to be illustrated through its capacity to simulate
the various contributions to runoff.
How rainfall contributes to runoff?
The model provides a comparison of the various hydrological
uxes occurring over the entire simulation period. Fig. 9 as-
sesses simulated uxes on the representative UHE. Values
are expressed as a percentage of rainfall and depend on
the chosen reference parameter conguration. A value of
100% rainfall transforms into 97% non-intercepted rainfall,
which gets split into 60% on natural soils, 17% on roofs and
20% on streets, due to the relative proportions of these land
use types on the Reze catchment. Evaporation and transpi-
ration uxes amount to more than 40% of the total rainfall.
The tree interception rate remains small: this result is cor-
related with the low level of tree coverage on this catch-
ment (0.17% of street surfaces and 24% of natural soil
surfaces). The main contribution to ow rate stems from
covered surfaces, with 15% from houses and 7.5% from
streets; the natural soil contribution lies close to zero.
The soil contribution is due solely to groundwater drainage,
which represents about 30% of the total rainfall, as divided
between the wastewater (18.4) and stormwater sewer
(10.7) networks. This result, unfortunately, cannot be vali-
dated at the present time, yet it is consistent with other
experimental results observed on urban catchments (Hollis
and Ovenden, 1988).
0 5 10 15 20 25 30 35
0
5
10
15
20
25
30
35
Observed runoff depth (mm)
S
i
m
u
l
a
t
e
d
r
u
n
o
f
f
d
e
p
t
h
(
m
m
)
0 5 10 15 20 25 30 35
0
5
10
15
20
25
30
35
Observed runoff depth (mm)
S
i
m
u
l
a
t
e
d
r
u
n
o
f
f
d
e
p
t
h
(
m
m
)
(a) =4 C
Nash
= -0.30 and C
b
= 0.46 (b) =40 C
Nash
= 0.95 and C
b
= 0.1
Figure 8 Comparison of the runoff volume produced by each rain event, observed in x-coordinate and simulated by URBS-MO in y-
coordinate Simulation realized with two different values of k.
Table 6 Various runoff contributions to the total ow rate,
expressed in percentage of the gross rainfall for Reze
catchment
k Q
hou
Q
str
Q
nat
Q
drain
Q
tot
P
4 15.1 8.8 10.4 5 45.26 100
40 15.1 7.5 0 10.7 33.3 100
A distributed hydrological model for urbanized areas Model development and application to case studies 279
Flow rate simulation
Three types of results have been compared (see Table 7).
First, the instantaneous ow rates Q during rainfall events
were compared with the observed rates. For each rainfall
event, the ow rate volumes V and ow coefcient C
F
were
then compared with observations: a very strong agreement
was found between the simulated ow rate volumes V,
which proves the ability of the model to simulate the water
budget for the given catchment. The URBS-MO model repro-
duces more than half of the observed variability in the ow
coefcient (R
2
= 54%) of the series of rain events; this per-
formance is similar to that of a highly detailed 2D model
(UHE model) applied to the same catchment over the same
period (Berthier et al., 2004).
Saturation level simulation
The simulated saturation level is assumed to be uniform;
this assumption admittedly is not fully realistic, as illus-
trated in Fig. 10, which shows the temporal evolution in
the piezometric measurements at two points on the catch-
ment. The comparison will focus on the dynamic evolution
T
r
e
e
i
n
t
e
r
c
e
p
t
i
o
n
e
v
a
p
o
r
a
t
i
o
n
97.3
100
2.7
R
a
i
n
f
a
l
l
N
o
n
i
n
t
e
r
c
e
p
t
e
d
R
a
i
n
f
a
l
l
0
storage = -0.9
0.02
0.04
2.0
15.1
7.5
0.9
60.1 43.5
11.6
10.7
60.2
17.1
20.0
R
a
i
n
f
a
l
l
S
u
r
f
a
c
e
e
v
a
p
o
r
a
t
i
o
n
Surface runoff
T
r
a
n
s
p
i
r
a
t
i
o
n
I
n
f
i
l
t
r
a
t
i
o
n
Groundwater drainage into
rainwater sewer system
wastewater sewer system
18.4
T
r
e
e
i
n
t
e
r
c
e
p
t
i
o
n
e
v
a
p
o
r
a
t
i
o
n
R
a
i
n
f
a
l
l
N
o
n
i
n
t
e
r
c
e
p
t
e
d
R
a
i
n
f
a
l
l
T
r
e
e
i
n
t
e
r
c
e
p
t
i
o
n
e
v
a
p
o
r
a
t
i
o
n
R
a
i
n
f
a
l
l
N
o
n
i
n
t
e
r
c
e
p
t
e
d
R
a
i
n
f
a
l
l
R
a
i
n
f
a
l
l
N
o
n
i
n
t
e
r
c
e
p
t
e
d
R
a
i
n
f
a
l
l
storage =
R
a
i
n
f
a
l
l
S
u
r
f
a
c
e
e
v
a
p
o
r
a
t
i
o
n
Surface runoff
T
r
a
n
s
p
i
r
a
t
i
o
n
I
n
f
i
l
t
r
a
t
i
o
n
Groundwater drainage into
rainwater sewer system
wastewater sewer system
storage =
R
a
i
n
f
a
l
l
S
u
r
f
a
c
e
e
v
a
p
o
r
a
t
i
o
n
Surface runoff
T
r
a
n
s
p
i
r
a
t
i
o
n
I
n
f
i
l
t
r
a
t
i
o
n
Groundwater drainage into
rainwater sewer system
wastewater sewer system
trees
Figure 9 Hydrological variables on the UHE for the whole simulation period (01/01/199131/12/2000), expressed in percentage
of the gross rainfall. Water uxes are detailed for natural soils, and are omitted fro street and roof land uses, for a sake of
conveniency. Values indicated in italic are valid for the whole UHE, other values are only valid for each land use type.
Table 7 Comparison criteria for simulations of runoff
volumes V, ow coefcient C
F
and instantaneous ow rates
Q for the whole period (Reze)
C
b
R
2
C
Nash
V +1.4 % 0.95 0.95
C
F
0.5% 0.54 0.50
Q 0.58% 0.59 0.51
0 200 400 600 800 1000
3
2.5
2
1.5
1
0.5
0
Time (days)
s
a
t
u
r
a
t
i
o
n
l
e
v
e
l
(
m
)
PZ1
PZ2
simulation
Figure 10 Saturation level evolution on catchment Reze
between 1996 and 1998, observed in 2 piezometers and
simulated by URBS-MO.
280 F. Rodriguez et al.
of saturation levels over a three-year period of available
data. The variation in simulated saturation level agrees with
observations over the entire simulation period, except for
summer 1998, for which the model is unable to reproduce
the particularly sharp and rapid decrease in the saturation
level. Two arguments can be forwarded to explain this dif-
ference: either the model cannot successfully simulate in-
tensely dry meteorological periods, or this decrease is
caused by human activity (e.g. pumping in the vicinity of
the catchment). In the latter case, this kind of hydrological
behavior cannot be reproduced by URBS-MO due to the set
of assumptions concerning the saturated zone of the catch-
ment. Potential exchanges between the saturated zone of
the catchment and the surrounding saturated zone were
not taken into account herein. This assumption would be
reasonable in an urban catchment, but might be debatable
should the soil contribution appear to be sizable.
URBS-MO validation on the Gohards catchment
The Gohards catchment has been selected for the validation
of the complete model, including the water budget and
transfer processes. Due to its size and land use heterogene-
ity, this catchment serves to illustrate the value of the
spatially distributed modeling approach inherent in
URBS-MO.
Principle of model implementation
This catchment encompasses 911 cadastral parcels, which
combined represent as many individual UHEs. The morpho-
logical features of each UHE have been estimated thanks
to the urban databanks available for the Nantes metropoli-
tan area (France) and reveal a broad range of surface areas,
and natural and impervious fractions (see Table 4). The
model has been implemented according to the owchart de-
scribed in modeling of hydrological processes at the catch-
ment scale section, with special emphasis on applying the
water budget allocation to every UHE, and has generated
the simulation of spatially-distributed ow rates and satura-
tion levels. Rainfall intensities have been considered
homogenous over the entire catchment, as deduced from
the average rainfall intensity observed on the two pluviom-
eters. With rainfall intensities and ow rates available at 5-
min time steps, a continuous simulation could be run for 4
full years using these 5-min interval data records.
As for the estimation of water budget parameters, the
following assumptions have been adopted. Due to proximity
of the Reze and Gohards catchments and the similarity of
their surface and soil properties, the parameter set ac-
cepted for Reze has been transposed to Gohards, except
for the groundwater drainage parameter k, which is specic
to the sewer network. These parameters were assumed to
be identical for all UHEs in the Gohards catchment. The
parameter k was calibratedby making an adjustment with
respect to the discrepancy between simulated and observed
runoff volumes (Fig. 11a). The bias criterion has its mini-
mum value for k = 17, which might indicate that the Go-
hards sewer pipes and trenches system displays different
characteristics than those of Reze; this statement however
cannot be veried, the only available data for that is the age
of the sewer pipe implementation, and it was similar on
both catchments.
Transfer parameters associated with the Muskingum
routing conguration have been estimated in applying the
following assumptions. The routing parameter x is assumed
equal to 0.2, which is a mean current value adopted for ur-
ban hydraulic routing (Chocat, 1997, p. 658). The mean
time k has been estimated by considering that wave veloc-
ity x is spatially variable. We have chosen the following
formulation, as deduced from Mannings Equation and sug-
gested by Semsar (1995): x
j
= 0.8V
j
, where V
j
is computed
from Mannings Equation applied to the jth segment of the
drainage network (Rodriguez et al., 2003). The calculation
of V
j
involves the pipe lling rate h. Semsar (1995) sug-
gested a lling rate corresponding to 80% of the maximum
ow rate through the pipe. We tested model sensitivity to
this parameter. Year 2000 data served as inputs for this
test, and the Nash criterion on instantaneous ow rates
was considered (Fig. 11b). This routing parameter proves
inuential for very small values of the lling rate. The
model performs better when the lling rate ranges be-
tween 50% and 100%. We ultimately decided on a value
of 50% for this parameter.
Flow rate evaluation
A strong level of agreement has been found between simu-
lated ow rate volumes V and observations (see Table 8),
-20
-10
0
10
20
30
40
0 10 20 30 40
C
b
(
H
)
-0.1
0
0.1
0.2
0.3
0.4
0.5
0.6
0 0.2 0.4 0.6 0.8 1
Filling rate
C
N
a
s
h
(
Q
)
a
b
Figure 11 Application of the model to catchment Les Gohards Sensitivity (a) to the parameter k (groundwater drainage
coefcient) with regards to the bias criterion and (b) to the parameter h (lling rate of sewer pipes) with regards to the Nash
criterion.
A distributed hydrological model for urbanized areas Model development and application to case studies 281
although the comparison criteria are not as effective as for
the Reze catchment simulation. On the one hand, results
show a marginal overestimation of runoff volumes with
URBS-MO, while on the other, the mean simulated ow
coefcient (0.33) is slightly smaller than the mean observed
ow coefcient (0.35) and the bias criterion on ow coef-
cients is negative. The simulation of instantaneous ow
rates displays a good correlation coefcient (R
2
= 0.74)
and a correct Nash coefcient (C
Nash
= 0.54). At this point,
it must be emphasized that the Gohards catchment simula-
tion extends over a continuous four-year time period and
comprises 330 rainfall events, spanning the entire range of
hydrological behavior of the catchment. The results ob-
tained must be assessed in light of this array of evaluation
conditions, which makes for a rather different test than
most hydrological models.
The models ability to simulate peak ow rates has been
analyzed by displaying the scatter diagram of simulated vs.
observed peak rates for all 330 rainfall events occurring dur-
ing the 4-year simulation period. This plot is characterized
by a high proportion of small peak ow values (Fig. 12). A
good agreement has been found between the two peak val-
ues, as assessed by a correlation coefcient R
2
of 77%.
The contributions to outow from the various land uses
are summarized in Table 9. It appears that the runoff pro-
duced by pervious surfaces is signicant on this catchment
(about 6% of the rainfall, i.e. 16% of the total ow rate dur-
ing rainfall events). Such a contribution varies considerably
from one rainfall event to the next and depends on the par-
ticular season. The runoff produced by pervious surfaces
during rainfall events is very small during the summer (1%)
and increases up to 20% of the total ow rate during winter-
time (Table 10). This contribution seems to be overesti-
mated, compared to the results obtained both by Boyd
et al. (1993) on several urban catchments and by Berthier
et al. (2004) on the Reze catchment. This result however
may be explained by the way the model functions. It is pres-
ently assumed that the rainwater in excess of the inltra-
tion capacity of UHE natural surfaces actually adds to
runoff. In many cases, surrounding walls or hedges prevent
runoff from occurring, with the excess water being stored
in the parcel and then inltrating after the rainfall event.
This process has not been built into in the model due to a
lack of sufcient information concerning those UHEs on
which the natural component cannot contribute to runoff.
Distributed nature of the results
First of all, the simulations conducted with the 911 UHEs of
the catchment have been compared to those performed
with just a single representative UHE, in assuming a homog-
enous catchment. The morphological characteristics of this
UHE have been deduced by averaging characteristics over
the entire catchment; the process parameters turn out to
be the same. As indicated in Table 10, the distributed re-
sults are consistently better than the homogenous ones,
which conrm the benet of a spatially distributed descrip-
tion of hydrological processes for urban catchments. The
simulation performed with the homogenous conguration
shows that the model underestimates ow rate volumes,
as a consequence of the small contribution of natural soil
to runoff. The saturation level, in this case, is much lower
Table 8 Comparison criteria for simulations of runoff
volumes V, ow coefcient C
F
and instantaneous ow rates
Q for the whole period (Gohards) values in brackets
represents the results with one sole UHE for the whole
catchment
C
b
R
2
C
Nash
V +7.1% (1.3%) 0.89 (0.86) 0.79 (0.69)
C
F
5.8% (15.6) 0.43 (0.35) 0.20 (0.22)
Q +2.2% (7.5) 0.74 (0.65) 0.54 (0.37)
0 2000 4000 6000 8000
0
1000
2000
3000
4000
5000
6000
7000
8000
Observed peak flowrate (l/s)
S
i
m
u
l
a
t
e
d
p
e
a
k
f
l
o
w
r
a
t
e
(
l
/
s
)
Figure 12 Comparison of observed and simulated peak ows
of Les Gohards catchment for the whole simulation period.
Table 9 Various runoff contributions to the total ow rate,
expressed in percentage of the gross rainfall for Gohards
catchment
k Q
hou
Q
str
Q
nat
Q
drain
Q
tot
P
17 13.3 13.5 6.7 10.1 43.7 100
Table 10 Seasonal variation of the pervious runoff contri-
bution on Gohards catchment during the simulation period
(19982001)
Season Percentage
of the
gross rainfall
Percentage
of the total
ow rate
Considered
ow rate in
the year (%)
Winter 11.4 21 33.7
Spring 2.8 7 14.9
Summer 0.3 1 14.5
Autumn 9.1 19 36.9
282 F. Rodriguez et al.
than in the distributed conguration, and both the natural
soil and groundwater drainage components are inexistent.
It may be observed that in the distributed conguration,
the hydrological behavior may be quite different from one
UHE to the next, and the portion of runoff generated by nat-
ural surfaces and the soil vary considerably. The UHEs with a
high natural soil contribution generally display special mor-
phological features; their mean surface area amounts to
29,000 m
2
since the mean value equals 1948 m
2
for the en-
tire catchment. This large number stems from two factors:
(i) a high pervious surface area leads to a sizable amount of
inltrated water; and (ii) an elongated shape, relative to a
high value of the UHE length L, leads to a small value of
groundwater drainage Q
drain
. For these UHEs, both factors
induce an increase in water content within the vadose zone
and an increase in saturation level during the winter season,
in addition to producing runoff on pervious surfaces. This
high variability in pervious surface runoff however is unreal-
istic and constitutes a limitation to our modeling method.
The simulation result is due mainly to model assumptions,
which consider every elementary unit as a parcel drained
by the adjacent sewer system. Such assumptions might no
longer be valid for parcels whose characteristics vary more
widely, and especially the large natural parcels. Under
these conditions, the natural circulation of groundwater
along the hill slope can no longer be neglected and must
certainly become a signicant contribution to the water ux
during wintertime. This function seems to be well suited to
small parcels but fails for parcels with a greater length L: in
this case, drainage is too slow, which induces an articial in-
crease in saturation level and generates excessive pervious
runoff on specic UHEs. This nding was not observed on
the Reze catchment, where parcels were very homogenous
and not as long, in comparison with the Gohards catchment
(see Table 4).
Moreover, this distributed model allows for the estima-
tion of many hydrological variables for each UHE, such as
the various contribution components to runoff, saturation
level, water content and the actual evapotranspiration.
Fig. 13 illustrates the spatial variability in total evapotrans-
piration for the entire simulation period. The mean evapo-
transpiration value for all UHEs amounts to 33.8% of
rainfall, with this variable ranging from 4% to 61%; it tends
to be higher on those UHEs characterized by a high percent-
age of natural land use. The scatter diagram of this evapo-
transpiration vs. the natural fraction of each UHE (Fig. 14)
250 m
50 - 61
40 - 50
30 - 40
20 - 30
3,7 - 20
Figure 13 Spatial distribution of the cumulated evapotranspiration (in percentage of the total rainfall depth) simulated by
URBS-MO between January, 1999 and December, 2002 on catchment Les Gohards. Dark parcels are characterized by high
evapotranspiration values.
A distributed hydrological model for urbanized areas Model development and application to case studies 283
conrms the obvious proportionality between evapotranspi-
ration and natural soil fraction, but shows the existence of a
signicant dispersion in evapotranspiration for a given frac-
tion of natural soil, a signicant dispersion in evapotranspi-
ration appears on the order of 10% or 15%. Factors other
than the morphological parameters, such as soil water con-
tent and a representation of soil storage capacity in the
model, may serve to explain this variability. It can be re-
marked that these factors are likely to be inuenced by
sewer pipe depth. Fig. 15 shows the distribution in storage
capacity, as can be deduced from the saturation level; the
estimation of this variable could prove valuable when imple-
menting rainwater inltration techniques that help deter-
mine the zones where rainwater inltration could be
benecial. These results constitute an initial illustration of
the possibilities offered by such a distributed urban model-
ing approach and will be subsequently validated by compar-
ison with distributed saturation level observations using an
application on various case studies.
Conclusion
This study has presented a distributed hydrological model
adapted for urban areas. This model has been designed to
furnish a more thorough description of the hydrological
Figure 14 Total simulated cumulated evapotranspiration (in
percentage of the total rainfall depth) versus natural soil
fraction of each UHE on catchment Les Gohards.
250 m
270 - 526
220 - 270
176 - 220
136 - 176
20 - 136
Figure 15 Spatial distribution of the storage capacity of the soil (in mm), simulated by URBS-MO on July 1, 2002 on catchment Les
Gohards. Dark parcels are characterized by high storage capacity.
284 F. Rodriguez et al.
behavior of urban catchments, thanks to an estimation of
the various uxes contributing to the water budget; these
include: evapotranspiration, inltration, runoff, and soil
water drainage through the sewer network. This rst appli-
cation of URBS-MO has conrmed the effectiveness of a
highly detailed distributed hydrological model based on ur-
ban databank records and devoted to urban areas. URBS-MO
takes advantage of the spatial variability of land use, as
characterized by the urban databank content managed by
Geographical Information Systems. The model has been
tested at two different scales: a small and homogenous ur-
ban catchment, representative of an urban hydrological ele-
ment; and a medium-sized and more heterogenous urban
catchment. Accordingly, the relevance of both water bud-
get modeling and the whole distributed modeling chain
could be evaluated by using long and continuous series of
atmospheric forcing spanning several years. The model
shows that evapotranspiration, water inltration into sew-
ers are indeed capable of generating a signicant compo-
nent of the water budget, in agreement with previous
experimental studies (Belhadj et al., 1995; Berthier et al.,
2004).
This study has conrmed the importance of urban soil
and soilatmosphere interaction on the hydrological pro-
cesses at work in urbanized areas once focus is placed on
the continuous hydrological modeling over longer periods:
years, seasons. The results obtained indicate that some
uncertainties still remain, regarding both the set of model-
ing principles pertaining to water circulation in the soil and
the estimation of some model parameters, particularly
those controlling how the water budget gets processed.
The following modeling principles need to be further exam-
ined: (i) the assumption according to which water circula-
tion within the saturated zone is fully controlled by the
drainage network; (ii) parameterization of the exchanges
between groundwater and sewer network; (iii) the assump-
tion according to which the water supply system leakage is
neglected. Interactions between the sewer system and the
saturated zone remain a very current topic of research
which deserves detailed investigation Moreover, taking into
account the water supply system leakage could be a new
challenge in hydrological modelling of urban catchments.
This phenomenon can prove signicant (Lerner, 2002; Zhang
and Kennedy, 2006) and should be investigated by introduc-
ing a new source term of water in the soil; it induces new
data requirements, such as the private water consumption.
In the end, channeling all of the runoff produced by natural
surfaces directly into the sewer system may ultimately be
inappropriate in many urban congurations, where built
obstacles block runoff and force the soil surface contribu-
tion to inltrate.
The evaluation of such a distributed model makes it nec-
essary to develop new observation strategies. Observation
of the groundwater drainage component might, for exam-
ple, be pursued in the stormwater or wastewater system
during wet periods, by measuring the low ow rates. An
evaluation of the spatial distribution in saturation level con-
stitutes another area of interest: the model is able to simu-
late a very dense distribution of this variable, yet such
information remains impossible to validate. The ultimate
goal is to evaluate hydrological uxes from the surface to
the atmosphere. At the present time, we are unable to
judge whether or not the simulated evapotranspiration is
representative; nonetheless, a set of well-adapted experi-
mental methods are needed to better estimate both energy
and hydrological budgets in urban areas.
The objectives of urban stormwater management have
evolved over the previous decades and introduction of the
concept of sustainable development has sparked a new
stage of evolution (Delleur, 2003) towards an integrated ap-
proach of urban water management. A detailed modeling
approach, as illustrated in URBS-MO, could contribute to
updating urban models.
Acknowledgement
This study received a nancial support from the French re-
search program in hydrology (PNRH) of the ANR-ECCO.
Appendix
The vadose zone is represented by a simplied parameteri-
zation of the water movement, which assumes that Darcys
law can be applied between the saturation level and sur-
face, respectively, and a point V indicative of the global
behavior of the vadose zone. The validity of this set-up must
be established prior to proceeding with model application.
This validity step can be assessed by comparing the pro-
posed parameterization with Richards equation being con-
sidered as the reference solution. This comparison has been
performed under the following conditions. The comparison
pertains to a natural UHE, covered by vegetation and 5 m
thick. Surface interception is neglected and this UHE is
not drained by a lateral trench. The test then focuses on
the vertical uxes between the atmosphere, the vadose
zone and the saturated zone. The hydraulic conductivity
at UHE saturation is constant with depth, and a high value
has been set for parameter M in Eq. (8) in order to verify this
assumption. The UHE is submitted to atmospheric forcing
that combines rainfall and potential evapotranspiration dur-
ing a one-year period, discretized with a 5-min time step.
The meteorological data have been extracted from a case
study (presented in the accompanying paper) and are typi-
cal of French oceanic climate, with an annual rainfall depth
of 722 mm, composed mainly of low-intensity precipitation
events. The reference water uxes and quantities in the
ground have been calculated by means of the NSAT nite
element code that solves Richards equation and represents
the ux of water in the soil (Berthier et al., 2004). The val-
ues of model parameters adopted for this evaluation exer-
cise are listed in Table 11. The hydrological variables
targeted by the validity test are: saturation level, transpira-
tion ux, inltration ux in the vadose zone, and mean
moisture content of the vadose zone. A sensitivity analysis
of the simplied parameterization has been performed
and reveals that the best accuracy is obtained for values
of a
v
included in the interval [0.5, 0.7]; this nding led us
to adopt a value of a
v
= 0.5. The results obtained with
a
v
= 0.5 are displayed in Fig. 16, along with the comparison
between the reference solution and the simplied parame-
terization. It appears that the proposed parameterization
provides for a good simulation of the transpiration, surface
A distributed hydrological model for urbanized areas Model development and application to case studies 285
inltration and uxes between the vadose and saturated
zones, with coefcients of determination in excess of 80%.
The mean moisture content of the vadose zone has been
rather accurately reproduced, despite a slight bias affecting
the simulation of temporal evolution in the saturated level.
In summary, the comparison between the reference solu-
tion yielded by Richards equation and the simplied pro-
posed in this paper conrms the validity of such a
parameterization in reproducing the water ux in UHEs for
the considered application.
References
Anderton, S., Latron, J., Gallart, F., 2002. Sensitivity analysis and
multi-response, multi-criteria evaluation of physically based
distributed model. Hydrological Processes 16, 333353.
Belhadj, N., Joannis, C., Raimbault, G., 1995. Modelling of rainfall
induced inltration into separate sewerage. Water Science and
Technology 32, 161168.
Berthier, E., Rodriguez, F., Andrieu, H., 1999. The Reze urban
catchments database. Water Resources Research 35, 1915
1919.
Berthier, E., Andrieu, H., Creutin, J.D., 2004. The role of soil in the
generation of urban runoff: development and evaluation of a 2D
model. Journal of Hydrology 299 (34), 252266.
Berthier, E., Dupont, S., Mestayer, P.G., Andrieu, H., 2006.
Comparison of two evapotranspiration schemes on a sub-urban
site. Journal of Hydrology 328, 635646.
Beven, K.J., Kirkby, M.J., 1979. A physically based, variable
contributing area model of basin hydrology. Hydrological Sci-
ences Bulletin 24 (1), 4369.
Boyd, M.J., Bull, M.C., Knee, R.M., 1993. Pervious and impervious
runoff in urban catchments. Hydrological Sciences 38 (6), 463
478.
Brun, S.E., Band, L.E., 2000. Simulating runoff behavior in an
urbanizing watershed. Computers, Environment and Urban
Systems 24, 522.
Calder, I.R., 1977. A model of transpiration and interception losses
from a spruce forest in Plynlimon Central Wales. Journal of
Hydrology 33, 247275.
Cassan, M., 1986. Aide-memoire dhydraulique souterraine. Presses
de lE