Green roof effectiveness for minimising storm water runoff in high density futures: Retrofitting potential and urban resilience using SWMM, aerial remote sensing, and UKCP09 weather generator scenarios

ticketdonkeyAI and Robotics

Nov 25, 2013 (3 years and 6 months ago)

142 views

1

Copyright © Environmental Science Society (ESS). All rights reserved 2012.


G
reen roof effectiveness for m
inimising storm water runoff in high density futures:
R
etrofitting potential
and urban
resilience using

SWMM
, aerial remote sensing, and UKCP09

w
eather generator scenarios

J
.P
. Biddulph
. (2012).
SCHOOL OF GEOGRAPHY, EARTH AND ENVIRONMENTAL SCIENCES.
University of Birmingham.



Abstract

This review evaluates climate change, urban infill and urban densification processes in a
localised urban area in Poole, Dorset, where current on
-
going legislative and public
orientated management is on
-
going to control pluvial flooding and associated floo
d risk.
Roof
space has potential in more sustainable urban growth, when considering stormwater runoff
and the problem of impervious. In addition to this, uncertainty in climatic change and within
the context of future higher density urban forms, new techno
logies, enabling more sustainable
urban development, is important.


The effectiveness of a roof
-
based, at source, Sustainable Urban Development System (SUDs)
stormwater control, the Extensive green roof (E
GR
), is investigated in 7 high and medium
density r
esidential and commercial urban areas.

Also, potential problems of urban creep were
investigated, in these areas, using
an object
-
based feature extraction methodology
(ENVI
EX)
and

a temporal (
2000, 2005, & 2007)

high
-
resolution aerial data sets. Probabilistic weather
generator (UKCP09) outputs, USEPA Storm Water Management Model (SWMM) and
LIDAR (0.5m) data were used to simulate a range of outcomes and possible climatic futures,
using
high emissions scenario

(
A1FI),

co
mbined to produce a multimodel ensemble and
subsequent derivation of E
GR

effectiveness at
99.9
th
,90
th
, 50
th

and 10
th

percentile precipitation
events, compared to baseline (1976
-
2011) daily runoff data.
Impervious landcover types did
increase (2
-
5%), howev
er, shadows were a problem so upperbound shadow values were used
which encapsulated shadow pixels.

Green roofs systems were
more effective at mitigating stormwater flows in higher density
residential landuses with more impervious area. In the commercial la
nduse category E
GR

system response to the 90
th

percentile event was found to lower runoff to baseline levels. For
more extreme percentile values (99.9
th
) relative runoff reductions were reduced, and
subsequent green roof systems less effective in all
scenarios
.


2

Copyright © Environmental Science Society (ESS). All rights reserved 2012.



Contents page


Introductory Section
................................
................................
................................
...................

3

Research Exercise
Overview

................................
................................
................................
..

3

Summary

................................
................................
................................
................................

8

Aim

................................
................................
................................
................................
.........

8

Hypothesis 1(H
1
)

................................
................................
................................
....................

8

Hypothesis 2(H
2
)

................................
................................
................................
....................

8

Alternative
Hypothesis 1(H
a1
)

................................
................................
................................

9

Alternative Hypothesis 2(H
a2
)

................................
................................
................................

9

Objectives

................................
................................
................................
...............................

9

Research Gap

................................
................................
................................
.............................

9

Literature Revie
w
................................
................................
................................
.......................

9

Urban Drainage, imperviousness and Hydrological Impacts

................................
.................

9

Climate Change Projection
................................
................................
................................
...

19

The Potential of Green Roofs: A ‘Roof
-
based’ SUDs

................................
.........................

21

Methodology

................................
................................
................................
............................

28

Introduction

................................
................................
................................
..........................

28

Detection and Analysis

................................
................................
................................
............

28

Study Area

................................
................................
................................
............................

28

Scenarios

................................
................................
................................
................................
..

37

Scenario 1: Baseline Climate (B)

................................
................................
.........................

37

Scenario 2: High Emission precipitation intensity (H)

................................
........................

37

Modelling using SWMM

................................
................................
................................
.........

39

Soils
................................
................................
................................
................................
..........

39

Slope

................................
................................
................................
................................
........

43

Scenario 3: The Extens
ive Green Roof System (Gr)

................................
...........................

45

Evaporation Rate

................................
................................
................................
......................

47

Results

................................
................................
................................
................................
......

48

Trends in Impervious
ness
................................
................................
................................
.....

48

Green Roof Scenarios
................................
................................
................................
...........

56

Stormwater Runoff Response
................................
................................
...............................

58

3

Copyright © Environmental Science Society (ESS). All rights reserved 2012.


Co
nclusion, Discussion & Further Research

................................
................................
...........

62

Appendix Section

................................
................................
................................
.....................

64




Introductory Section

This section introduces the main topics and purpose of this study, followed by
an

overview of
the study area, and then finally the aims and objectives.

Research Exercise
Overview


The universal trend of
paving the planet can be considered
one of the primary anthropogenic
modifications of the
environment (
Sutton et

al., 2009).
Watersheds essentially need to be
conserved and, where possible, enhanced with
their natural capacity to promote healthy
ecosystem functioning exploited (
Petts et al. 2002
).

I
mperviousness
, and synonymous

terms,
sealed surfaces
,

hardscapes
and non
-
infiltrating land covers, accompanying urban
development, have

emerged as
significant e
nvironmental and hydrological performance
indicator
s

(Schueler 1994; Whitford

et al.,

2001;
Poelmans

et al., 2010;
Schroll et al., 2011;
Wu & Yuan, 2011
;

Yang 2011;
Weng 2012
)
. Being
sensitive parameters in hydrological
modelling
,

their

accurate

quantification
,


using
remote sensing methodologies
,

have

emerged
i
n environmental analyses
with
the release of higher resolution satellite imagery, such as
IKONOS (1999) and Quickbird (2001)

(Schueler 1994; Canters et al., 2011; Goldstein 2011)
.
This
has c
oincided with research into more automated and t
herefore efficient methods

to
estimate urban functions,
using higher spatial interrelated elements,
processes and responses
,
to environmental and
socioeconomic trends (Wu & Yuan, 2011).
In addition, prev
iously
prohibitive and i
nsufficient resources in image processing

has

allowed the above increments
and more advanced landcover/use classification and change
algorithms to

be
developed
(Schneider and Woodcock 2008)
.

The recent Flood and Water Management
Act 2010 promotes SUDs as essential component
s
for reducing urban flood risk from ‘heavy rainfall’

to ‘adjacent or downstream properties’
and legislates local authorities responsible for more sustainable local drainage performance
(Defra, 2010 pp.1). Biore
tention
,

and more sustainable management
,

aim to rea
ch
4

Copyright © Environmental Science Society (ESS). All rights reserved 2012.


pre
development runoff rates, or green field rates
,
and quality criteria
(
Faulkner

1999; He

&
Davis 2011)
.


The
effectiveness

of

SUDs
,

at mitigating stormwater
surface runoff
,

is an integral component

for more sustainable

urban development (
Clary et al., 2011).

The combination of
urban

hydrologic
al modelling, vector and object
-
based system analysis, and more detailed
representation of system dynamics
,

at higher spatial scales
,

allow
more
detailed ‘what
-
if’
scenarios to be
investigated
,

such as large
retrofit

green roof projects in different

urban
settings

(Amaguchi et al., 2012)
.

In addition, l
and availability for higher density or city
stormwater control are important factors in decision
-
making (
Villarreal, et al., 2004
),
especially as UK density statistics, between
2000 and 2009, have approximately doubled
(from 25 dwellings per hectare to 43 dwellings per hectare) (DCLG 2010a)
. Therefore, a

specific Sustainable Urban Development System
(SUDs), the extensive green roof, is
evaluated as an adaptation method in a local urban area.


Negative environmental performance measures in urban areas affect the water balance and
hydraulic system response which influences wider ecological system
geomorphic and
biogeochemical cycling processes (Whitford

et al.,
2001;
Tillinghast

et al., 2011; Verbeeck et
al., 2011
).


Current and future sensitivity to extreme storm events, in Poole, UK, has been highlighted as
a significant problem with potential im
pacts on the local economy and even the economy of
the county (EA, 2009).
The exercise investigates an
urban area of Upton, Dorset (Fig1).




Figure 1.
Left.

An overview map of the urbanised centres of Bournemouth and Poole.
Highlighted is

the study area of Upton (Ordnance Survey EDINA Digimap 2011).
Right.

Is a
BlueSKY aerial digital photo of the study area.


5

Copyright © Environmental Science Society (ESS). All rights reserved 2012.



Stormwater d
rainage systems
,

in areas

of
Bournemouth and Poole
are highly

sensitive to
surcharging (pluvial flooding) from high int
ensity precipitation events (William & Holme
2009; EA 2010
a
).

However, t
here is still uncertainty, not only in future and current urban
pressures, but also in current policy (densification), affecting urban ecological functioning,
and impacting, directly,

the human system (Gray 2005
;
DCLG 2010
b
). Thus, c
ontextual
explorative analyses into, ultimately, more sustainable urban growth, and here stormwater
management
,

is

important
,

whilst sustaining
methodological
accuracy

provision

(Brander et
al., 2004;
Canters 2011)
.


One specific policy, ‘Policy

4
’ is allocated to

managing flood risk when areas are currently in
low, moderate or high flood risk
,

and where surface water is current
ly

effect
ively managed
but where decision making
‘may need to take further
actions to keep pace with climate
change’

(E
A
,
2009

p.11
)
.
U
rban stormwater p
olicy is therefore promoting sustainability in
urban planning, and in Poole, the current preferred option is the retrofitting appraisal of
sustainably managed techniques
,

such as

Sustainable Urban Development Systems (SUDS)
(EA, 2009
;

EA 2010b
).




Housing

growth

p
rojections,
in England
,

are expected to increase by 5.8 million (27%) to
27.5million in 2033 from the 2008 figure

(
DCLG 2010
a
)
, being

an average
annual increase
o
f
232,000 households.
The Regional Spatial Strategy (2006
) has
identified

Poole and
Bournemouth

as
Strategically Significant Cities
, forecasted to be
Main Areas of Growth

in

commercial,
industrial

and
resid
ent
ial

landuses (EA 2010
a).
In addition,

in
low el
evation
costal zones (<10 m above sea level), such as Poole, population densities are high with 60
%

now urbanized
,

therefore resulting in vulnerability to climatic f
actors directly impacting

human health and infrastructure

(Grimmond et al., 2010)
.

F
requency and intensity of precipitation events, at daily (and sub
-
daily) temporal scales (or
time
-
steps), particularly at northern mid
-
high latitudes (Zevenbergen et al., 2011)

may
increase.
In
addition

to climatic
perturbations
, the

planning process is u
nable to regulate,
effectively
, ‘urban creep’ processes at local levels
(Zevenbergen et al., 2011). Defined as,
‘the gradual increase in drained area from patios, driveways, and extension roofs’, urban
creep is a significant contributor to flood risk, due
to sewer capa
city
and condition
problems,
6

Copyright © Environmental Science Society (ESS). All rights reserved 2012.


especially when combined with urban population growth

and climate change (POST 2007;
Ofwat,

2011 p. 30). Integrated modelling of these three drivers increased median sewer flood
volumes by 51%
,

using

projective sc
enario
s

to 2040
,

in England and Wales

(Ofwat,

2011
)
.


This review

synergistically combines and evaluates m
ode
l simulations, using outputs fro
m
the United Kingdom Climate Projections Weather generator (UKCP09WG), baseline climate
data, land cover change variables using aerial urban remote sensing, and
static
hydraulic and
geometric variables of E
GR

systems,
using the modelling
environment

of

the Environmental
Protection
Agency’s

Stormwater Management Model (SWMM), to investigate sensitivity and
effectiveness of scenario retrofitting programme
s

between

different urban landuses

and
climatic
storm event
projections
.

However, it must be noted th
at
t
he urban flood system encompass
es

interconnected
social,
political and economic systems
,

as well as physical elements (Evans et al., 2004a)
.
However,
here, only the
urban system
, impervious land cover trends
, and

hydrological performance

is
7

Copyright © Environmental Science Society (ESS). All rights reserved 2012.


modelled
and taken into account

(Fig.2
).


Figure 2.

the u
rban hydrological system and local flood responses

encapsulated within the
wider p
hysical
ly
-
based

hy
drological flooding system
(Evans et al., 2004a).

In limiting this study and disregarding

the
multifaceted

findings of
social,
economic

and
political

advantages

of urban densification (
Boyko &
Cooper

2011
;

White

& Gatersleben

2011
)
and multidisciplinary nature for
ach
ieving

sustainable development (Flourentzou,
2011),

subsequently
limit
s

the

validity of any conc
l
u
sion
s

made. Thus
,

whilst
high spatial
density

and hydrological processes are

investigated
,
the potential benefits of roof
-
based
SUDs
,

assuming higher density futures is simulated and used to conclude
,

within the

context
of the
higher

density urban form consensus

(
see
Jenks, et al., 2008)
.

In addition,
environmental degradation
, both w
ithin the urban and interrelated aquatic systems,

and

pollutant behaviour associated with urban dynamics and social trends is not considered
(
Boyko &
Cooper

2011
)
.

8

Copyright © Environmental Science Society (ESS). All rights reserved 2012.


Lastly, t
raditional l
ocal
ised manual digitalisation, or
photointerpretation
, can achieve high
accuracy in studying urban systems, their dynamics and associated risk factors
,

but is time
consuming, expensive and subjective (Bolstad et al., 1990 cited in

Rozenstein & Karnieli
2011;

Cleve et al., 2008;
Wu & Yuan 2011
).

Therefore,

temporal landco
ver change
of

imperviousness,

using

a
semi
-
automated
object
-
based maximum
likelihood

classification

approach

is
also
investigated
,

using high

spatial resolution three
-
channel multispectral images.
Derived temporal statistics on sealed surface
s

are combined with high spatial (5x5km)
UKCP09WG outputs and compared with baseline precipitation

and runoff responses,

to
investigate

possible intensity increments
and
quantify any perturbations
,

to
evaluate
effecti
veness of roof
-
based systems

on runoff processes between different urban landuses.

Summary

At source

systems, utilising roof space, ar
e hypothesised to have a greater potential at higher
densities. In addition, relative offsetting and resilience potential is investigated with

possible,

exogenous
,

past and potential urban pressures of urban infill and more intense storm events.
Model simp
licity and effectiveness of automated
extraction to identify

any
infill
, utilising
RGB multispectral aerial imagery is
also
evaluated. Objectives are detailed and are
elaborated upon in subsequent sections.


Aim

To investigate

and model, relative
effectiveness of
localised
extensive green roofs

systems at
mitigating any observable change in
the

range (99.9
th
,90
th
,50
th
,10
th
) of meteorological
(UKCP09WG

and Baseline
) inputs
and any additional pressures (urban infill) between high
and medium density r
esidential and commercial landuse areas.

Hypothesis

1
(H
1
)

Green roof retrofitting will offset

any observable climate

change
forcing
and urban creep
process.

Hypothesis

2
(H
2
)

Green roof retrofitting will be more effective at
mitigating stormwater flows in

higher density
lan
duses with more impervious area.

9

Copyright © Environmental Science Society (ESS). All rights reserved 2012.


Alternative Hypothesis

1
(H
a1
)


Green roof retrofitting will be unable to regulate any
observable increment
in
any
c
limate
change
forcing
or urban creep process
.

Alternative Hypothesis
2
(H
a2
)

Green roof retrofitting will show no trends
at mitigating stormwater flows
between different
landuse or density sites

Objectives

1.

To
evaluated
an object
-
based
semi
-
automated

landc
o
ver extraction
using limited

spectral (RGB)

aerial imagery
, for

deriving

sensitive hydrological parameters

(
impervious data
),

directly for hydrological modelling

input.

2.

To

accurately

quantify
and
evaluate

any negative temporal landcover trends (urban
creep)

3.

Evaluate uncertainty

and the possible range
of future precipitation

scenarios

at lower
(min), median and
upper (
max) recommended
values (
OFWAT 2011)

( 99.9
th
, 90
th
,
50
th

and 10
th

percentile) and compare these outputs to a climatic baseline for the area.


4.

To
investigate relative
hydrological
performance of

simplified extensive green roof
systems to projected and historic extreme events and the potential role of these
systems in different urban forms.


Research Gap

No previous research has used UKCP09 weather generator scenarios and simulated the
possible
range and maximum potential of green roofs at local
,

highly urbanised locations

using

recent modelling capabilities (more explicit account of bioretention systems) in
Computational Hydraulics International PCSWMM (CHI, 2011)(
v.4.4.1037)

and USEPA
Storm Wat
er Management Model (v.
5.0.022)
(SWMM)(USEPA, 2011).


Literature Review

Urban
Drainage, imperviousness and
Hydrological Impacts


Surface runoff, quick
-
response flooding ultimately occurs when

rainfall rate is greater than
the infiltration rate of the surface (Mockus 1972; NRCS 2004). Unfortunately, accompanying
urban development are materials, classified
hydrologically
, as impervious landcover types,
10

Copyright © Environmental Science Society (ESS). All rights reserved 2012.


resulting in a higher proportion of impervi
ous built surfaces, or
sealed surfaces. This
,

coupled
with tradition drainage practices, conveys stormwater from
upstream
urban systems to

drainage,

river and tributary systems at increased rates and volumes (
Niemczynowicz 1999;
EPA, 2006;
Silva et al., 20
06;

Gill et al., 2008
;
Wheater

& Evans 2009).



The
schematic
flow diagram
(Fig3
)

represents the

fundamental runoff

responses from

rural

and urban surfaces
,

highlighting

the urban

impervious effect
on flux

processes
.
Overland
flow fluxes (
effective
precipitation (Pe))

(Eq.1)

are the processes from rainfall inputs minus
abstractions (rainfall excess) (Yen 1986).
These abstractions, or losses, include depression
storage, infiltration, interception (by vegetation and subsequent evaporation) and
evapotra
nspiration
(Viessman & Lewis 1996; Butler & Davies 2004
)


Pe = P
-



losses







Figure 3.
(A)
-

Rural surface hydrological response to precipitation. (B) Urban surface
hydrological
response

to precipitation

(Whitford et al., 2001).


Predominantly, the

most significant physical process is infiltration
,

with abstractions from
evapotranspiration
,

relatively negligible

(
Yen 1986
)
.
However, w
hen both, imperviousness
and rainfall intensity
increase this relative significance decreases (Yen 1986
). In addition,
e
vapotranspiration processes
may
become more dominant during summer
seasons for

green
roof system
s

(
Berghage
et al.,
2009
)
(
Fig.4).




(EQ.1
)

11

Copyright © Environmental Science Society (ESS). All rights reserved 2012.




F
igure 4.

Averaged

monthly precipitation
re
t
ention

values
from thr
ee extensive green roof
systems

(
Berghage

et al.,
2009
).



Also, u
rban microclimates may influence evaporation rates significantly
. As e
vaporation

rates

vary
significantly

between areas
and can account for a large output over long term
water
balances

(
Mitchell

et al., 2001; Berghage et al., 2009)
. However, other studies have
disregarded evaporation
,

as
rate
s

approximately
accounted

for 0.5% of total loss over a 3
-
day
flood event (Apirumanekul,2001 cited in
Chen et al., 2009).

Pyke et al., (2011)
used change factors
,

where baseline
multipliers

are applied to annual
volume and

upper (95
th
) intensity events
, to simulate the
response

to changing
imperviousness
.
Impervious surfaces were found to be the most significant factor being the
most sensitive parameters influencing runoff responses, at these annual temporal scales
(Fig.5). In addition, impervious affects and subsequent low impa
ct simulation of areas with
reduced impervious, when
isolating precipitation volume,
and event intensity, a reduction
of
25
-
16%
in

impervious cover
age

had the greatest potential to significantly mitigate
stormwater runoff volume (and pollutant loads) (
Pyk
e et al., 2011).

12

Copyright © Environmental Science Society (ESS). All rights reserved 2012.



Figure 5. Simulated sensitivity of annual runoff volume to intensity, volume and impervious
change factors (
Pyke et al.,
2011)
.


At source stormwater control to manage the influx and associated negative environmental
performance,
associated, have therefore gained popularity as an alternative solution for managing
stormwater in urban areas (Martin, et al., 2007)


P
ost
-
development gross (%) increases in impervious cover can exacerbate

flooding problems
(DCLG, 2010b
).
In
Pauleit (et
al., 2005) and Whitford (et al., 2001) surface runoff
performance measures were transferable to indicating floodin
g potential in a catchment.
Utilising
flood hydrographs in design storm events
,
hydrological processes and
responses
are

estimated

with imperv
iousness trends

and findings

include runoff volume
increase
(Hollis,
1975), time of concentration

(Thompson 1999)

decrease (
system
surface flow
& river
regime
and

flux

response increase) and more
rapid and

severe peak discharge (flow rate)
,

influencing
directly different flood frequency curves (Moscrip and Montgomery, 1997 & Moon et al.,
2004 cited in Huang et al., 2008)
.

Theoretically, any increase in annual and/or peak

runoff is
likely to have a significant effect on the magnitude and freq
uency of flooding in the area of
13

Copyright © Environmental Science Society (ESS). All rights reserved 2012.


concern (
Hollis 1975;
Perry & Nawaz 2008)
, depending on the site
-
specific and catchment
descriptor characteristics

(Bayliss 1999).

In
addition
,

to higher
flood peaks
,

groundwater
recharge
(basef
low
flux) is reduced locally
(
Wheater

& Evans 2009)
.


Whitford (et al., 2001)
concluded and suggested the following main points:
-



Ecological performance was greatly influenced by the proportion of green space



Social
-
economic deprivation may impact density and therefore exert pressure
s on
ecological performance



Recommend research on ways to alleviate negative ecological trends associated with
densification

Higher density urban areas have increased impervious surfaces that negatively influence
hydrological parameters at these local le
vels
(Ferguson, 1998). In addition,
higher density
development limits

the availability of land for
green space thus affecting environmental
performance
(
Mentens et al., 2006;

Petts, et al. 2002
).
Gill et al., (2007)

investi
gated
environmental performance,
at the city scale in Greater Manchester
, and importance of green
space using
high and low UKCIP02 emission scenarios (Hulme et al., 2002 cited in

Gill et al.,
2007),
using a simple

empirically
-
based model
,
the SCS runoff curve number

to produce
relative measures of
runoff reduction from green roofing scenarios,
between different urban
areas

(Fig.8)
.


The
SCS runoff curve number method incorporates antecedent
moisture conditions (AMCs),
soil type, and land cover in

predicting the
runoff volume generated from a storm event (
Mack,
1995;
Pandit & Gopalakrishnan 1996
)
.
The model is empirically
-
based and
formulated

for
easy application scenarios

to derive estimates of
direct runoff

from storm
events using
watershed data (Mockus 1972).
W
eighted
-
average runoff curve numbers (CN) are calculated
for the individual areas or sub
-
parcels

(see

Carter & Jackson 2007) and a weighted CN is
derived from soil type (hydrologic soil group
-

HSG),


antecedent

m
oisture and percentage
cover of each land
use (Chow, 1988 cited in Whitford, et al., 2001). Experimental studies on
minimum infiltration

rate of soil types classify HSGs into four categories

A, B, C, and D

(Pandit & Gopalakrishnan 1996
)
.

Classifications from A
-
D refer to decreasing infiltration an
d
water transmission rates. Thus
, c
lass D
(
Clay soils
)

impede flow rates dramatically. In
contrast, class A
,

generally
consist of deep,
well
-
drained

sandy deposits and
coarse gravels

14

Copyright © Environmental Science Society (ESS). All rights reserved 2012.


with low runoff potential and high infiltrati
on and water transmission rates
(Mack 1995;
Mockus 1972).


Specifically
, ‘
Roof
Greening’
scenarios

were applied, to simulate the addition of green roofs
computing the relative performances between study areas and
scenarios (Fig 7 & 8
)

(pp.10).
Both high density residential areas and town centres, with more built areas, had greater runoff
reductions and

proportionally larger reductions in runoff for high emission scenarios. It was
noted, however, that
antecedent moisture conditions may become ‘w
etter

with climate change’
(
Gill et al., 2007 pp. 7).




Figure 6. Cityscape runoff trends to climate induced precipitation flux increments (Low &
High) (
Gill

et al., 2007
)





15

Copyright © Environmental Science Society (ESS). All rights reserved 2012.



Fig
ure 7
Sensitivity of high density residential
areas to precipitation inputs with relative
remedial green roof and urban tree scenarios at

normal antecedent moisture conditions
(
adapted from
Gill et al., 2007
)




16

Copyright © Environmental Science Society (ESS). All rights reserved 2012.


Figure 8.
Relative performance between
urban area
and green roof greening scenarios
using
incremental
precipitation

forcing parameters,

at

normal antecedent moisture conditions
(
adapted from
Gill

et al., 2007).


L
and avail
ability, therefore, for higher density or city stormwater control are important
factors in decision making processes
and unsustainable trends at higher spatial resolutions is
required, as limited information is available on the extent, distribution and change of
impervious surfaces

(Verbeeck et al, 2011;
Villarreal, et al., 2004
).
In addition,
‘Impervious
surface’ now has

been classified, in remote sensing methods, as a category of land cover
or/and land use

(Weng et al., 2012
)
.

Dallimer

(et

al.,

2011
) u
sed

relative measures of gridded vegetation indices to estimate
Enhanced Vegetation Index (EVI) difference and temporal
trends since 2000 (to 2008),
between 13 urban study areas, using rural areas as control groups. The coarse resolution
spectroradiometer showed negative trends of overall green space after 2000, interestingly, the
year policy guidelines were introduced to e
ncourage densification as opposed to urban
expansion. In addition, more significant growth in urban population and densification
indicators is apparent in these southern cities (
including Bournmouth) (
Fuller & Gaston 2010
cited in
Dallimer

et

al.,

2011
).


To accurately quantify land cover change, remote sensing requires temporal considerations
from
ph
eno
logical

and anthropogenic
factors (Irannejad & Shao 2002). T
emperate climatic
regions will have a significant seasonal effect therefore
satellite

imagery f
rom the same
season and annual time steps (Mather & Koch 2011)
are required thus mimising

temporal
mismatches (Foody, 2010)
. In addition, a
n integral knowledge of the area being classified is
often a perquisite
for

accurate supervised classification using
t
he
maximum likelihood
classifier

(Rozenstein & Karnieli 2011).

In

heterogeneous areas the spectral complexity
,
variety and the
subsequent
difficulty in training classifiers with representative signatures,

unsupervised classifications, such as the ISODATA

c
lustering algorithm have been found to
derive better classification estimates
(Rozenstein & Karnieli 2011).

Past high resolution
urban environmental
change approaches

have generally used manual digitisation techniques
(Verbeeck et al, 2011; Perry and Nawaz

2008
)

.

17

Copyright © Environmental Science Society (ESS). All rights reserved 2012.


Transport related infrastructure is often the most significant land use contributing to total
impervious area
(
Schueler 1994)

(Fig.9
a
).



Fig
ure 9
a
.

Theoretical total roof area and transport related impervious (hatched line) areas in
residential (right
) and commercial (left)
traditional
land use types

(
Schueler 1994)
.


Planning Policy Statement 3 (PPS3) (
D
CLG, 2011), now, does not class private front
or back
gardens as ‘Brownfield’. Thus, it is reasonable to assume that any past urban infill processes
now will subside

(Kaźmierczak & Cavan 2011)
.

As discussed, however, Urban Creep
processes may still be occurring at local urban scales (
Zevenbergen et al
., 2011
).

Verbeeck
(
et al, 2011
)

investigated original housing and
impervious
area
characteristics

between housing projects
built between
1923
-
1962
, using
building plans combined with
d
istance and parcel variables

to investigate proximity
probability
of
impervious change

at this
scale
(Fig.9
b
)


Transition probabilities were greatest close to
housing
area plot

and f
ront garden
areas in all
of the three built forms. All
parcels

digitized had garden areas and o
verall impervious
increments of
38% to 56%

occurred in the
housing projects
.

18

Copyright © Environmental Science Society (ESS). All rights reserved 2012.


Fig
ure
9
b
.
(A)
Impervious change proba
bi
lities using function of distance to the street (l) and distance to the house (r). (B) simulated street
neighbourhood (
Verbeeck
et al, 2011
).
19

Copyright © Environmental Science Society (ESS). All rights reserved 2012.



Land use and hydrological soil group have a direct effect on interception,
surface retention,
evapotranspiration, and

resistance to overland flow

(Thompson
1999
;

Yen 1986
)
.

Perry and
Nawaz (2008)

measured impervious garden statistics and used an empirically
-
based
model
,

the
long
-
term hydrologic impact assessment (L
-
THIA)
,

where hydrological soil group and
the
CN

method is applied
, allowing

temporally derived
surface input and output metrics to be
easily

compared
.
During
aerial interpretation
,

two problems persisted in landcover
classification. These were gravel drive ways

which closely resembled other impervious
features

(
identified

using wheel line markings)
,

and shadows. Shadows were classified as
pervious and therefore impervious garden statistics
were assumed

to be un
derestimated.
Between 1971
-
2004
garden paving was fo
und to
increase

impervious
ness

by
12.6%
and
annual runoff increased by 12% and
was

likely to have a significant effect on the ‘frequency
and magnitude of flooding

in the study area’ (pp.10).



Climate Change

Projection


Catchment scale studies indicate
that, globally, areas

at

high latitude
will have an
increased
frequency of intense precipitation (Arnell, 1999; IPCC 2008

Berndtsson, 2009
).

The relative
impact
,

of which
,

is dependent upon the
climate change scenario proposed
. H
owever, the
underlying
assumption is that flooding events are likely to increase (
Fedeski & Gwilliam
2007) due to increased
frequency and intensity of rainfall events.


P
rojecting possible perturbations in precipitation intensity
,

at urban drainage
scales

require
higher
temporal

and spatial resolution

climate models

(Sanderson 2010)
.

There are current limitations in statistical downscaling dependent upon regional and global
climatic processes and, in addition, extreme precipitation values undergo the non
-
stationary
phenomenon, where inconstancies in methodological data collection, such

as observational
and micro
-
climatic factors are combined with the assumption that dependent variables will be
constant in a future global climate
(
Willems et al., 2012)
.


The
United Kingdom Climate Projections (UKCP09) use statistical Bayesian trends, to
produce
probabilistic

outcomes
encompassing

uncertainty, as global climate models require
parameterisation within and between the atmospheric, oceanic, and cryospheric system
s,

20

Copyright © Environmental Science Society (ESS). All rights reserved 2012.


and the internal bi
ogeochemical cycles,
whilst taking into account variable

a
nd uncertain
feedbacks and

external processes (Murphy et al., 2009).


UKCP09

projections

show
decr
eased daily summer precipitation and for standardised
northern climatological seasons (winter (DJF) and summer (JJA)), in

the localised region of
south Dorset (Bournemouth), the frequency of winters storm return periods are expected to
decrease but summer events are more uncertain (Perry, 2006;
Sanderson

2010).

The UKCP09 (V2
) and

the probabilistic climate change projections methods estimates
statistical climatic outcomes generated from the likelihood of ensemble UKCP09 climate
models and their relative accuracy to reproduce past
climatic runs (
UKCP 2010).

The
Bayesian probability
method to create statistical ensembles of climatic model outputs allow
cumulative Distribution Function

(CDF) and

Probability Density Functions

(PDF) to be
produced. Crucially, the CDF provides a threshold where an output probability is provided
(%) to d
isseminate relative probabilities of a measure being below or exceeding a certain
threshold (Murphy et al., 2009). The range of possible outcomes, and their probabilities, is
due to the level of
parameterisation

and therefore the overall model structure be
tween
projections based upon the
known

dynamics of feedback processes and
constituent
parameters

(Murphy et al., 2009).
This conceptual and methodological design does provide
relative confidences of possible outcomes, thus assisting and providing decisi
on making in
adaptation impact studies, a quantitative and confidence measure (UKCP 2010).

However, the stochastic time series generated from a baseline
(1961
-
1995


for
UKCP09WG),
used

to calibrated cli
m
a
to
log
i
cal

data,
will disregard possible outliers outside
the baseline data range. In addition, the number of iterations and outputs is not able to
compensate for this limitation (Sanderson 2010;
Andersson & Chapman
2011;
Ofwat 2011).
Also,

whilst some global biogeochemica
l feedback systems are now included
,

such as the
carbon and sulphate aerosol cycles and their dynamics, less understood environmental
feedbacks

such as
,

potentially significant
,

methane sources from terrestrial wetland
,
thawing

permafrost and
clathrate

sy
stems are not included

(
Schmidt

2004; Murphy et al., 2009).
Another external factor is the future behaviour of solar irradiance which has,
likely,

perturbed the past climate

(Stott et al. 2003 cited in Murphy et al., 2009), but is presumed
minimal in
UKCP09 as recent stability has coincided with continued climate warming
(Murphy

et al., 2009).

21

Copyright © Environmental Science Society (ESS). All rights reserved 2012.


General

Circulation models (GCM) are too coarse for local hydrological and eco
-
hydrological impact assessments, where, finer temporal and spatial
resolutions

a
re required

(
Fatichi

et al., 2011)
. T
he orographic processes incorporated into Regional Climate models
(RCM) (Christensen and Christensen, 2007 cited in Chun 2010)

and
downscaling and higher
resolution data sets (5x5km)

IN WG

provide further detailed
resolution phenomena, taking

into account localised physical thermodynamic proces
ses by incorporating topography and
costal influences using

more representative climatic data for the locality

(UKCP 2009).
However, the WG

uses the more coarse resolution UKC
P09 probabilistic

projection data (
25x
25 km) and
uses this data in the 5 by 5 km grids. Uncertainty in downscaling is incorporated
into the probabilistic projections (Murphy et al., 2009). Change factors are then applied to the
WG, based on the computed v
ariance measures between
ensemble

outputs,
UKCP09
probabilistic projections and baseline data (UKCP 2009)
,

using perturbed physics ensembles
(PPEs) (Sexton & Murphy 2011)
.

Dynamic, short
-
term, temporal trends in storm duration may change for specific
return
periods
,

but
the
current sub
-
daily modelling is
highly
uncertain and would significantly limit
the validity and confidence in any impact model (
Sanderson 2010)
.

However,
sub
-
daily
convective
storm processes

and subsequent
WG

outputs at hourly scales

is unable to take this
into account. Thus, UKCP09 is unable
to simulate
hourly trends
confidently
(UKCP 2009
;
UKCP 2011
).

Probabilities

of any mitigation and its effects using socioeconomic and
technology

scenarios

are not included as
UKCP09 uses IPCC’s high, medium and low emission scenarios to
perturb the future climate
simulations (
Murphy et al., 2009). A minimum of 100 30
-
year
daily simulations (3000 years) is
recommended

to incorporate the
probabilistic

projections
PDF and model va
riants (UKCP 2009).


The Potential of Green Roofs: A ‘Roof
-
based’ SUDs


SUDs approaches mimic a more natural runoff response and t
he potential of green roofs to
present predevelopment hy
drographs through interception
,
infiltration,
storage and
evaporati
ve loss functions may
particularly

be effective
in highly urbanised watersheds
,

where opportunities for
other
storm
water
regulative measures
are lacking (Carter & Jackson
2007
; Swan 2010).

22

Copyright © Environmental Science Society (ESS). All rights reserved 2012.



Typically, a green roof
consists

of three layers

(
Mentens

et al.,

2006) :



vegetation layer,



substrate layer



drainage layer

Extensive green roofs have thin growing mediums (
25
-
150mm)

(Fig.9
c
)
,

consequently
,

limiting plant availability to certain grasses, herbaceous perennials, annuals, and drought
tolerant succulents such as
Sedum

(
Mentens et al.
,

2006;

Alfredo et al., 2010;

Rowe 2010
;

Stovin et al., 2012).


Fig
ure 9
c
.

Vertical profile of a modular green roof system (
Oberndorfer et al., 2007).



Experimental

performance of Green infrastructure (GI
) and different SUDs practices
are
relatively varied and inconclusive
,

due to different methodological design and site condit
ion
s

(Alfredo et al., 2010).
Volumetric

retention (VR) and maximum field capacity of extensive
green roof systems may depend on climatic temperature, rainfall intensity and antecedent dry
weather periods (ADWP)
,

combined with structural component characteristics (Palla et al.,
2009
; Gregoire


& Clausen 2011;

Stovin et al., 2012)
.




In addition
,

for
the
seasonal

effect on evaporative flux rates, winter
mean values have ranged
from

<0.5 mm/day to 3 mm/day in

w
inter and

summer
, respectively

(Kasmin et al. 2010
cited in
Stovin et al., 2012)
.


23

Copyright © Environmental Science Society (ESS). All rights reserved 2012.



W
hen considering, flood response the catchment

is dependent on

the

previous cumulative
processes of rainfall
inputs
.
This ADWP and subsequent s
torage show
significant
dependent

relationships in rainfall
-
discharge
relationships (
Marshall 2009). Probabilities of a flood
response occurring from an influx of precipitation is greatly increased when soils are at field
capacity (when the soil moisture deficit (SMD) = 0) (Bayli
ss, 1999).


Regionally, SMD vary seasonally and on annual scales (
Fig.11
)
.


F
igure 1
0
.
Met Office Rainfall and Evaporation Calculation System

(MORECS) soil moisture
deficit, annual and seasonal trends,
for South West England (EA, 2011
a
)


M
oisture
retention capacity

is finite and is
dependent

upon slope and substrate.
I
nterception

processes,

concerning abstractions, is significantly dependent
upon landuse and seasons

(Yen
1986).

Thus, before runoff occurs an upper limit of retention, combined with any additional
storage in the drainage layer has to be reached
(
Stovin
et al., 2012).

A simplified green roof
hydrological system flow diagram (Fig.11) details the main flux and storage

processes.

24

Copyright © Environmental Science Society (ESS). All rights reserved 2012.



Figure 11.
Green roof hydrological sys
tem flow diagram (
Stovin et al., 2012
).

Green roof systems are not so dependent on the
infiltration capacity of the urban soils

and
have an immediate effect on disconnecting roof systems
(
Brander

et al., 2004
;
Carter &
Jackson 2007
)
.
Green roofs decrease in efficiency when rainfall depths are greater than the
re
te
ntion capacity of the soil (Hilten, et al., 2008). This
trend
relates to the soil columns field
capacity
(saturated hydraulic conductivity),
when
saturated
outflow equals inflow,
hydrographs subsequently mimic impervious roof

systems
. However, detention is provided
temporally
, attenuating quick
-
flow regimes associated with D
irectly
C
onnected
I
mpervious
A
reas
(DCIA)
, such as roof
s

(VanWoert, et al. 2005;

Hilten, et al., 2008
)
.
DCIA have
a
propo
r
t
ion
ally

greater
effects

on runoff

events. Lee and Heaney (2003
cited in
Rivas

2009)

found that
DCIA’s contributed

to 70% of total runoff from the urban area whilst only
constituting 44
% of the spatial area.


A recent meta
-
analysis found an

av
erage value of retained storm
water data of 56%
(
Gregoire

& Clausen

2011)

(Fig.1
2
).
H
owever
,
these
accumulative

retention values
lack

explicit
account of
ADWP and
,

more relevant for impact assessment,
larger storm event retention

(
Stovin
et al., 2012).



25

Copyright © Environmental Science Society (ESS). All rights reserved 2012.



Fig
ure 12
. Average extensive g
reen
-
roof precipitation retention

values

data values (
adapted
from
Gregoire

&
Clausen 2011)

using a

water balance [(precipitation


runoff)/precipitation)]
×
100 equation
.



Both,
structural

properties

such as
stratigraphic width
, slope

and depth

and biological factors,
such as, vegetation typology,
assemblages and density measures

also affect volume retention

(Palla et al., 2009
; Gregoire


& Clausen 2011
)

(Table.
1).


Table
1
.

Precipitation re
tention (%) response of a extensive green roof system to varying
event
preci
pi
tation

intensity

and slope increments for dry initial conditions (Villarreal &

Bengtsson 2005)

Intensity (mm/min)

Slope (◦)

Total Retention (%)

0.4

2

8

14

62

43

39

0.8

2

8

14

54

30

21

1.3

2

n/a

14

21


10


Mentens (et al., 2006) used an extensive green roof system (at
20◦) and compared
accumulative runoff to normal roof

(Fig.13)
.

26

Copyright © Environmental Science Society (ESS). All rights reserved 2012.



Figure 13. 14.6 m 24 hr cumulative rainfall runoff response comparison between traditional
and extensive

green roof system (slope of 20

)

(
Mentens

et al., 2006)


Different modelling approaches have been used to estimate the hydrological
effecti
v
eness

of
green roofs for urban stormwater management.
Carter & Jackson
(
2007)

used the SCS runoff
cu
rve number (CN) method
to model green roof potential at parcel, zone and subwatershed
scales.
Th
e potential of green roofs
were investigated
using a ratio expression of
total
roof
space in
the
area
. Both, h
ighly urbanised watersheds
benefit
ed

more significantly from
such manageme
nt
(Carter & Jackson 2007)

but

decreasing efficacy,
at reaching
predevelopment

hydrographs

occurred,
with increasing storm volumes (Carter & Jackson
2007).

VanWoert, et al. (2005)
fou
nd that over a 14 month period

a roof
-
based system
retained 82.8% of the rainfall that fell, and detained 60.6%.
A gravel

roof

system

reta
ined
47.7% of the
total
rainfall.


SWMM
(v.
5.0.022)

(
USEPA, 2011)

now incorporates storm water management (LID; SUDs;

BMPs) tools
enabling
quantitative

simulation
,

modelling explicit and distributed
management
approaches
(
EPA, 2006).


Assuming

soils
are
not saturated
and therefore

not contributing to excess rainfall, high
spatial assessment of drainage elements in urban areas consist of rainfall collecting surfaces,
such as imperious surfaces (roofs, front driveways) in cadastral parcels contributing to
laminar or multi
-
channelled

flows
,

and concentrated flows in gutters and undergrounded
27

Copyright © Environmental Science Society (ESS). All rights reserved 2012.


pipes

(Rodriguez et al, 2003)
. This concentrated flux is limited by inlets of the underground
sewer system

(Rodriguez et al,

2003) and
has
potential of blockages (Defra 2008)
.

Goldstein (2011) inve
stigated two spatial models for estimating

event

peak ru
noff
and
cumulative
volume from an urban area.
Two levels of
discretisation

were investigated,

a
lumped and parcel
scale (or unit scale) approach,
requiring a
lumped

aggregated

representation and a spatially distributed approach, respectively (Fig.14).
The u
nit
scale
requires more spatial d
etail for each hydrologic unit, incorporating

spatial

heterogeneity and
connectivity (surface flow)

between the individual areas. However, th
is

technique was more

data intensive,
time consuming
and subsequently complex.




Fig
ure 14.

A com
parison of parcel
-
scale
and s
ubcatc
hment aggregation scale catchmen
t

modelling techniques (Goldstein 2011)
.

The dependence upon model outputs and the degree of visualisation of constituent parts
provides a somewhat unsystematic method of investigation. However, the
degree of
discretisation can be dependent on the objective

of the

model
(Zaghloul
1981)
. For examp
le,
coarse scale
s are

applicable when only outflow from subareas is required. Whereas, conduit
flow, generally,
requires higher levels of
discretisation
,

particularly when calibration and
performance modelling is required.
The
lumped model

simplifies

flow
paths
,

but
was
found
to be
effective

and efficient
at
applying ‘wide
-
spread’ retrofit
LID
models (Goldstein 2011).
The low resolution model overestimated peak and volume reductions
(T
able.2
).



28

Copyright © Environmental Science Society (ESS). All rights reserved 2012.


Table 2
.
Relative

p
eak and
cumulative

flow reduction from LID between high and low
(Goldstein 2011)


Methodology

Introduction


Explicit assumptions and the level of parameterisation is detailed and combined with
UKCP09WG probabilistic outputs to derive a simple urban system hydrological
model and
evaluative framework based
-
upon a static roof
-
based extensive green roof system. The
modelling approach uses Computational Hydraulics International PCSWMM (CHI, 2011)

(
v.4.4.1037),

the USEPA Storm Water Management Model (v.
5.0.022)

(SWMM)(USEPA,
2011) and the
image processing and analysis software ENVI (EX) feature extraction.


Model control parameters were estimated using high spatial urban remote sensing methods.
These temporal and spatial parameters were imputed into an information hydroinform
atic
database and a catchment modelling system was used within the latest version of SWMM to
simulate responses (
Choi & Ball 2002).


Social aspects and economic constraints combined with prospective governance are difficult
to predict (Evans et al., 2004b)

and thus are not considered in any scenario
-
based outcomes.


Detection and Analysis

Study Area


A ground survey was implemented to gather various components to better aid the landcover
classification and model parameters required. Roads, pavement and parking areas were
29

Copyright © Environmental Science Society (ESS). All rights reserved 2012.


asphalt in the groundtruthed areas and therefore were categorized into impervious.

A number
of properties had red/paving front drives.

Area
-
based analyses calculating roof area statistics using UTM projected coordinates to give
answers in square meters (M
2
) (Longley et al., 2011).

Roof areas were found to be Directly
Connected Impervi
ous Area (DCIA). Roof area and house density (ha) were derived using
Ordnance Survey

(OS, 2011) MasterMap data,

by dividing house point counts by
subcatchment area (OFWAT 2011). In addition, spatial green roof area (%) statistics were
calculated (Appendix 1), which encompassed adjacent roofing areas, such as garages.

Medium house density (16
-
25 houses/ha)(MedRes)
subareas were selected by using
OS

Code
-
Point Polygons

and a random point
generator in ENVI. Five

medium residential

(MedRes)

parcels were then selected. High (>39houses/ha)
(HighRes)

and commercial areas
(CommercialA)
were not selected
randomly (Fig.15).



Figure 15. Aerial

image
(from BI 2011)
of Upton with high, medium and
commercial

landuse
sites modelled as lumped subcatchment sub
-
sets.

30

Copyright © Environmental Science Society (ESS). All rights reserved 2012.



Pixel
-
based methods were first used to investigate the accuracy and subsequent spatial
resolution required to
derive

a

min
im
um recommended classification accuracy of 85 percent

(
Anderson, 1971 cited in Anderson et al., 1976) .

Three raster images were obtained (table 3) from 2000, 2005 and 2007. Images were cloud
free.

Imagery was
then interpolated

and resampled

to 0.5m.


Table 3.

A table detailing the date, sources, cloud cover and resolution of raster data used in
the study with ancillary data


Date

Pixel Resolution
(m)

Total Cloud
Cover

Distributor

Spectral
band width

11
-
03
-
2000

0.14

0%

LandMapp

(RGB)
colour
composite

10
-
07
-
2005

0.06


0%

GeoInformation
Group



Cities
Revealed


(RGB) colour
composite

21
-
04
-
2007

0.50


0%

GeoPerspectives
-

BlueSky


(RGB) colour
composite


A per
-
pixel approach was first used. A medium smoothing low pass filter was chosen
(kernel
3x3) for reducing noise effects (PV outlier/extremes) and, as the method preserves edges
more effectively than other methods such as moving
-
average filters

(
Ratti and

Richens 1999;
Mather & Koch 2011).
In addition, a linear stretch was applied to utilise full dynamic range
and lower

and upper (5
th

and 95
th
percentiles) and to cancel pixel (outlier) values. This
Resulted in an increased

contrast and detail between spectral PV

improving visual
interpretabi
lity between zones (
Thapa & Murayama 2009
; Cleve et al., 2008; Mather & Koch
2011
)
(Fig.15).

Transformed divergence probabilistic values provided an indicator of the

separability
between classifications. Values above 1.6 indicate good separability (Richards and Jia 1999
cited in
Metternicht, 2003
).
31

Copyright © Environmental Science Society (ESS). All rights reserved 2012.




Fig 15. Image pre
-
processing step, using a median kernel filter (3 by 3) and contrast
stretching: Comparison of histog
ram for RGB PV after 5% linear stretch. Note the 5% white
line.


Significant spectral confusion occurred in 2007 and 2000 classifications, using the
supervised
classification (Fuzzy convolution
-

3x3) approach (Table 4)(fig.16
a
), probably being
significantly due to the limited spectral bands available and therefore the per
-
pixel approach
not being able to distinguish sufficiently
spectrally similar
land uses

in a highly
heterogene
ous
urban surface
(
Cleve

et al., 2008;
Long &


Srihann 2004
).

32

Copyright © Environmental Science Society (ESS). All rights reserved 2012.



Figure 16
a
. Classification results using per
-
pixel maximum
-
likelihood supervised
classification (median c
onvolution filter, 3x3) results.
Note the spectral confusion
(highlighted) between two roofs and

adjacent
pavement (P) and road (R)

Table 4. Jeffries
-
Matusita transformed divergence values for supervised classification of
landcover (2007).


Tree/bush/sh
rub
[Green3]

Garden/gr
ass

Building/R
oof [Red2]

Road
[Yellow3
]

Pavement/dr
ive
s

Shadow
[Black]

Tree/bush/sh
rub [Green3]

N/A

1.218059
64

1.9878086
4

2.000000
00

1.99999978

1.999981
55

Garden/grass

[Green1]

1.21805964

N/A

1.9693064
9

1.999999
99

1.99994640

2.000000
00

Building/Ro
of [Red2]

1.98780864

1.969306
49

N/A

1.535583
39

0.95460015

1.819626
74

Road
[Yellow3]

2.00000000

1.999999
99

1.5355833
9

N/A

1.57691714

1.997745
35

33

Copyright © Environmental Science Society (ESS). All rights reserved 2012.


Pavement/dr
ive
s

1.99999978

1.999946
40

0.9546001
5

1.576917
14

N/A

1.999996
07

Shadow
[Black]

1.99998155

2.000000
00

1.8196267
4

1.997745
35

1.99999607

N/A


Thus, more detailed land cover data concerning the urban areas could not be distinguished,
due to the lack of
separability
, between the different landcovers. Therefore, an alternative
more coarse classification (impervious & pervious), using object
-
based e
xtraction, was
investigated.

An
object
-
based
(OB)
method using ENVI Feature Extraction was

then

used.

OB
classification models take into account spatial as well as spectral information and therefore
methods have been developed with significantly higher re
lative classification accuracy in
urban remote sensing (Benz et al., 2004; Wang et al., 2004 cited in
Wu & Yuan 2011).

A

supervised

object
-
based workflow
, using the
feature extraction tool in ENVI
EX

(Fig 16
b
),
was used

to
produce landcover classification

images

VIS, 2009).

34

Copyright © Environmental Science Society (ESS). All rights reserved 2012.



Figure 16
b
. A flow diagram illustrating the Feature extraction workflow using object
-
based
classification made available in ENVI EX (adapted from VIS, 2009).

Imagery was

resampled to a universal spatial scale (0.5m) and 3 images were used in the
classification from 2000, 2005 and 2007.

This classification method will simplify detailed and complex landuses by categorising into
three distinct classification endmembers of i
mpervious, pervious and shadow. In addition,
object
-
based methods may have increased accuracy compared to per
-
pixel classifiers which
depend solely on spectral characteristics (
Myint,et al., 2011).

Nearest neighbour classifiers
were used. Figure 16
(c)

illu
strates subarea ‘MEDRes1’ and segmentation

step

results
,

based
on the scale parameter.
Initially,

s
egmentation level C was chosen
,

as scale 50 encapsulated
objects and effectively segmented objects. However
, to

prevent
over
-
segmentation and

under
-
35

Copyright © Environmental Science Society (ESS). All rights reserved 2012.


segmentation rule sets
(Marpu 2010)

iterative visual interpretation

was required
,

and a
subsequent scale range (45
-
55)

was chosen.


A



B

C


D



Figure 16
c
. Segmentation level (A) 1 (scale 10), (B) 2 (scale 25), (C) 3 (scale 50) and (D) 0
(scale 0).


The scale parameter (λ) is effectively the degree of heterogeneity permitted when merging
adjacent objects (Wu and Yuan 2011). The
Full
λ
-
Schedule algorithm is used in ENVI where
a decision
-
rule
-
based segmentation is applied (EQ.2) based on the neighbourin
g region pairs
(Oi, Oj) being

ranked and when ti,j equals or is less than

≤ λ, then the cells are merged
(Robinson et al., 2002; VIS, 2009).


(EQ.2)

36

Copyright © Environmental Science Society (ESS). All rights reserved 2012.


A pre
-
processing convolution median filter (Kernal 5x5) step was undertaken to reduce noise
effects (Ratti & Richens 1999), and spatial subsets were used for each sub
-
area.

A
population error matrix

method was used. Where a reference population is assumed

and
compared to the classified image (
Stehman

1997
). This approach can result in some
landcovers receiving small samples (Im et al., 2011). However, a random proportional
stratification function in ENVI was used and therefore 50 points were generated for
each
class. The a
ccuracy

assessment requires
accurate ground reference data (Pontius & Lippitt
2006; Foody 2010).
Here, a
ccuracy provision is likely to be high, due to the small number of
landcover categories
. In addition, t
he reference data sets
, used to
assess accuracy,

were
assumed to be 100% accurate (Foody 2010)

and confusion matrix error tables were
produced
(Appendix 6
).

Gravel is pervious, however, spectrally, is very similar to impervious land cover types.
Automated extraction classified Gravel as
impervious due to the spectral confusion (or
similar spectral signature) (Perry and Nawaz 2008). Gravel shown in (Fig.17), where wheel
lines are visible. In addition, spectrally similar pervious alternatives (permeable paving ect.)
cannot be determined and

thus all impervious surfaces are assumed not to be of these
alternate systems (Perry & Nawaz 2008).
Shadow pixels were combined with impervious and
pervious classifications to estimate an upperbound and lower value of impervious land cover
.


Figure 17.
Image file, during object
-
based supervised classification, with ‘gravel’ land cover
type hi
ghlighted. Note the wheel lines
.

37

Copyright © Environmental Science Society (ESS). All rights reserved 2012.


Scenarios

Three scenarios were used, based upon green roof application, climate change projections and
historic daily precipitation

intensity values. Static parameters, concerning soil and geometric
hydrological properties are provided in addition to spatially distributed subcatchment
parameters.

Scenario 1: Baseline Climate (B)

39 years (1973
-
2011) of daily precipitation data (mm/d)

were obtained from the National
Environmental Satellite, Data and Information Service and the National Climate Data Centre
(NCDC), for the weather station at Hurn (Table 5), 11km away from the study area of Upton.

The data was categorised into seasonal da
ta sets of
standard northern seasons of 3 month
winter (DJF) and summer (JJA) (Perry, 2006)
and daily total percentile (
10
th
, 50
th
, 90
th
, 99.9
th
)
values were calculated. Only wet weather days were taken into account.


Table 5. Summary table of weather station and daily precipitation data record obtained from
NCDC.


Weather Station

Latitude/Longitude

Elevation (aod)

Length of
Record

Bournemouth/Hurn

+50.783
-
001.833

+0010.0

1973/01/01
-
2011/12/30


Scenario 2: High Emission precipitation intensity (H)

Using the
UKCP09 Weather Generator (WG) allowed statistical projections

to be made at
da
ily time steps utilising the higher resolution data sets (5x5km
) (Fig.18
a
).

This

resolution
allowed simulations to

be made
u
s
ing SWMM software.

38

Copyright © Environmental Science Society (ESS). All rights reserved 2012.



Figure 18
a
. (left) An overview map with superimposed 25 by 25 km gridded weather data
used by UKCP09 Cliamte projections and the selected downscaled 5 by 5 km data grid used
by the UKCP09 Weather Generator (Map adapted from UKCPUI 2011).

A range of emission scenari
os
(High, medium and Low)
are recommend
ed
to
inform possible
management strategies, encompassing a range of outcomes (Murphy et al., 2009). Here,
medium (
A1B

(balanced
)) and

high emissions were used to compare outputs. For the runoff
model, more specifically, the high Special Report on Emissions Scenarios (SRES) fossil
intensive (A1FI) (2040
-
2059) emission scenario was investigated (
IPCC 2000;
UKCIP 2001;
Arnell et al., 2004)
.

Similar to the baseline data, scenarios
were aggregated into Winter
(DJF) and

Summer (JJA)
seasons using probabilistic sub
-
sampling of high and medium scenarios and percentile values
were determined (10
th
, 50
th
, 90
th
, 99.9
th
).

UKCP09WG produced 30 annual model outputs and a total of
51 30
-
year iterations were
used, in each of the four climatic sample outputs, resulting in a total of 6120 years of daily
precipitation data being simulated (Appendix 2).

More appropriately for the
probabilistic nature of the projections
, the 90th percentile value
(and other outputs), for example, are hypothesised as the
90th

percentile of
climate

variability

(
Eames

et al., 2011)
.

The past and future
parameters are assumed to be representative
of
a r
ange of historic and
future precipitation events
.

In addition, the station point measurement is assumed
representative
,

and
that
downstream flow measures receive
uniform
input
,

in

each subarea

(
Farahmand

et al., 2007)
.

39

Copyright © Environmental Science Society (ESS). All rights reserved 2012.


Modelling using SWMM

Soils

The main geology in the local borough are sedimentary Bracklesham Group sands and clays
(EA 2011b).
Drift deposits, in the region, consist of predominantly fine grained (
sandy) and
sandy clay loam textures accompanying
more clayey soils, present with wet
land features, to
the south (Fig.19). The underlain bedrock consists of semiconsolidated sequential

sands and
clays units (Edmunds et al., 2002;Heath

2004;EA 2011a;
EA, 2011b) (Fig.18
b
). Further
hydrogeological, and borehole data, is not obtained here.

Di
sturbed and compacted urban soils, if ignored can create errors when estimating infiltration
rates (USEPA 1999)
. However, here soil texture and hydrological parameters are assumed
universal in all urban subareas. The surface soil texture type for the regio
n is assumed to be
sandy loam (
NSRI 2012)

and homogenous.

Soil map units (soil associations) were overlain using the area
-
weighted average function in
PCSWMM and parameters were then computed using the Green
-
Ampt infiltration model
(EA, 2006) used in SWMM.
Parameters for hydraulic routing were ignored

and s
tead
y flow
routing was used, as drainage components are not included (
Rossman

2004
).
Parameters
require estimated values of capillary
suction head
(Ψ)(mm), saturated hydraulic c
onductivity
(K) (mm/hr) and
initial moisture

d
eficit

(
IMD
)
.


The permeability of the soil (K) at saturation and the
intergranular adhesive forces
affecting,
directly, holding capacity

(
Ψ
)

and the

permeability of the soils
(Huber, 1992
;

Heaney et al., 1999)

provide a continuous physically based water balance model requiring initial moisture
deficit
(IMD)
.
IMD is a fractional expression between soil

porosity and

moisture content

(
Huber 1992
)
. M
aximum
SMD (wilting point
)

IMD value from
Huber
(
1992
) was used,
thus, antecedent soil moisture
conditions can be classed as very dry and that, in all

scenarios, no

rainfall has occurred in the
past 24hours

(
Choi

& Ball

2002).

Summary hydrological parameters are provided (table 6).

40

Copyright © Environmental Science Society (ESS). All rights reserved 2012.



Figure 18
b
. Bedrock geological map
from
E
DINA Geology Digimap (BGS, 2011).
41

Copyright © Environmental Science Society (ESS). All rights reserved 2012.


Figure 19. Soil distribution map

for study region with soil
map units

and vertical profiles for locality (NSRI 2012).
42

Copyright © Environmental Science Society (ESS). All rights reserved 2012.


Table 6. Soil infiltration

parameters for sandy loam
texture classification



Parameter

units

Value

saturated hydraulic conductivity
(K)

Mm/hr

26
**


Suction/Capillary head ψ (mm
F



ㄱN
G

f䵄潩獴畲s⁤e晩f楴⁵獩湧
浡m業畭up䵄j


B

〮㌳⨪M

*

Appendix 3 ** Appendix 4

***Appendix
5

Precipitation Intensity and area effects
are

assumed uniform throughout the catchment areas
(Evans et al., 2004b
;
Alfredo et al., 2010). Runoff is modelled using an initial
-
continuous
runoff loss model

(Ramier et al., 2011)
. When
cumulative rainfall is lo
wer (or equal)
to

initial losses (L
i
), and continuous losses (
Lc
)(
Equation (EQ.3 & 4)) no

effective rainfall (W
Pe
)

occurs :



Alternatively, if rainfall flux (W
p
) into the system is greater than initial losses
and summed
continuous losses,
runoff occurs (
Equation (EQ.4 & 5)).



Overland flow and storage simulation is computed using Manning's equation and the
continuity equation (Torno 1975).


(EQ.3)

(EQ.5)

(EQ.4)

43

Copyright © Environmental Science Society (ESS). All rights reserved 2012.


The overland flow path (width parameter) represents the r
esidence o
f flow fluxes in a
subcatchment (Park et al., 2008), thus,

having an effect on peaks but minimal effect on runoff
volume. As an aerial weighted average is used
, it is assumed that

flux processes are
essentially in equilibrium when considering
out
flow and therefore sensitivity

to changes in
discretization
will be
minimal (Huber 1992).

Urban watersheds ge
n
er
ally have shorter width parameters and therefore greater time of
concentrations
(Gironás et al.,
2009). Here, overland flow length
(R
o
)

was estimated by
computing an average of 3 length measures

from
back
housing
parcel
s to the
centre

of the
adjacent
street (EQ.6) (
Gironás et al., 2009).

W = A / R
o

In each density subarea this value was then
divided by
total area
to derive the
subcatchm
e
nt
width parameter used in SWMM

(Rivas,
2009) (Appendix. 7). The density value width
values were then used for other subgroups in the equivalent density sub
-
group
.

Slope

LiDAR data (0.5m) was obtained from
Geomatics Group
(2011).
Raster and LIDA
R data was
overlain and upper bound mean slopes were calculated for each subarea. Maximum heights
for each subarea
, using a 50 layer multicolour level contouring approach in ENVI were used
to derive average slope data using the transect function in
PCSWMM
(Figure 20).

(EQ.6)

44

Copyright © Environmental Science Society (ESS). All rights reserved 2012.


Figure 20.
Example slope derivation using LIDAR and elevation contours to select maximum and minimum elevation points in study areas.
(A)(MedRes1) Slope = 0.255 (B) (MedRes2) Slope= 0.28.

(A)

(B)

45

Copyright © Environmental Science Society (ESS). All rights reserved 2012.



Subcatchme
nt outflow (Q) was computed using n, slope (s), depression storage (d
p
),
precipitation volume (d) and width (W) parameters in each hydrological unit (
or
subcatchme
nt) (Park et al., 2008) (EQ.7).


The associated parameterised values that were not explicitly investigated here are
summarised in Table 7.

Table 7. Subcatchment surface parameters

Parameter

Value

Manning’s roughness for impervious and (n)

〮〱㈪M

Manning’s roughness for pervious lan
搠d潶o爠
⡮E

〮ㄵ⨪

䑥灲p獳s潮⁳o潲o来渠䥭灥牶楯畳慮搠r潶o爠
⡭洩

ㄮ㤰㔪

䑥灲p獳s潮⁳o潲o来渠灥牶楯畳慮搠r潶o爠
⡭洩

㌮㠱P

f浰m牶楯畳⁁rea⁷楴栠 e牯⁤r灲p獳s潮o
獴潲s来
 F

㈵
Oe晡畬u

癡汵e
F
⡒潳獭a渠㈰㄰n

⩁灰G湤楸‸‪⨠䅰灥湤楸‹


pce湡物漠㌺⁔桥

䕸瑥湳楶t⁇ ee渠n潯映oy獴敭s⡇rF


A semi
-