Journal Pre-proof

Creating nature-based solutions where they are needed – A spatial

ecosystem service-based decision analysis of green roofs in
Barcelona

Johannes Langemeyer, Diego Wedgwood, Timon McPhearson,

Francesc Baró, Anders L. Madsen, David Barton

PII: S0048-9697(19)35480-4
DOI: https://doi.org/10.1016/j.scitotenv.2019.135487
Reference: STOTEN 135487

To appear in: Science of the Total Environment

Received date: 28 June 2019

Revised date: 8 November 2019
Accepted date: 10 November 2019

Please cite this article as: J. Langemeyer, D. Wedgwood, T. McPhearson, et al., Creating
nature-based solutions where they are needed – A spatial ecosystem service-based
decision analysis of green roofs in Barcelona, Science of the Total Environment (2019),
https://doi.org/10.1016/j.scitotenv.2019.135487

This is a PDF file of an article that has undergone enhancements after acceptance, such
as the addition of a cover page and metadata, and formatting for readability, but it is
not yet the definitive version of record. This version will undergo additional copyediting,
typesetting and review before it is published in its final form, but we are providing this
version to give early visibility of the article. Please note that, during the production
process, errors may be discovered which could affect the content, and all legal disclaimers
that apply to the journal pertain.

© 2019 Published by Elsevier.

 Journal Pre-proof

CREATING NATURE-BASED SOLUTIONS WHERE THEY ARE NEEDED – A

SPATIAL ECOSYSTEM SERVICE-BASED DECISION ANALYSIS OF GREEN
ROOFS IN BARCELONA

Johannes Langemeyer1,2
Diego Wedgwood1
Timon McPhearson3
Francesc Baró1,2
Anders L. Madsen4,5
David Barton6

of
ro
1
Institute of Environmental Science and Technology (ICTA), Universitat Autònoma de

-p
Barcelona (UAB), Edifici Z (ICTA-ICP), Carrer de les Columnes s/n, Campus de la UAB, 08193
Cerdanyola del Vallès, Spain
re
2
Hospital del Mar Medical Research Institute (IMIM), Carrer Doctor Aiguader 88, 08003
Barcelona, Spain
lP

3
Urban Systems Lab, The New School, USA
4
HUGIN EXPERT A/S, DK-9000 Aalborg, Denmark
5
Aalborg University, DK-9220 Aalborg, Denmark
na

6
The Norwegian Institute for Nature Research, Norway
ur

Journal: Science of the total environment

Jo
 Journal Pre-proof

CREATING URBAN GREEN INFRASTRUCTURE WHERE IT IS NEEDED

A SPATIAL ECOSYSTEM SERVICE-BASED DECISION ANALYSIS OF GREEN ROOFS

IN BARCELONA

ABSTRACT:

As cities face increasing pressure from densification trends, green roofs represent a valuable source of

of
ecosystem services for residents of compact metropolises where available green space is scarce. However,

ro
to date little research has been conducted regarding the holistic benefits of green roofs at a citywide scale,

-p
with local policymakers lacking practical guidance to inform expansion of green roofs coverage. The
re
study addresses this issue by developing a spatial multi-criteria screening tool applied in Barcelona, Spain

to determine: 1) where green roofs should be prioritized in Barcelona based on expert elicited demand for
lP

a wide range of ecosystem services and 2) what type of design of potential green roofs would optimize the

ecosystem service provision. As inputs to the model, fifteen spatial indicators were selected as proxies for
na

ecosystem service deficits and demands (thermal regulation, runoff control, habitat and pollination, food
ur

production, recreation, and social cohesion) along with five decision alternatives for green roof design

(extensive, semi-intensive, intensive, naturalized, and allotment). These indicators and alternatives were
Jo

analyzed probabilistically and spatially, then weighted according to feedback from local experts. Results

of the assessment indicate that there is high demand across Barcelona for the ecosystem services that

green roofs potentially might provide, particularly in dense residential neighborhoods and the industrial

south. Experts identified habitat, pollination and thermal regulation as the most needed ES with runoff

control and food production as the least demanded. Naturalized roofs generated the highest potential

ecosystem service provision levels for 87.5% of rooftop area, apart from smaller areas of central

Barcelona where intensive rooftops were identified as the preferable green roof design. Overall, the

spatial model developed in this study offers a flexible screening based on spatial multi-criteria decision
 Journal Pre-proof

analysis that can be easily adjusted to guide municipal policy in other cities considering the effectiveness

of green infrastructure as source of ecosystem services.

KEYWORDS: Cities, Urban, Nature-Based Solutions (NBS), Green Infrastructure (GI), Bayesian Belief

Networks (BBN), Multi-criteria Decision Analysis (MCDA)

of
ro
-p
re
lP
na
ur
Jo
 Journal Pre-proof

1. INTRODUCTION

As cities across the globe rapidly grow and densify, urban greenery plays an increasingly vital role

as nature-based solutions (NBS) to the sustainability challenges associated with urbanization (EC, 2015;

Kabisch et al. 2017). For metropolitan populations who otherwise lack access to nature, green

infrastructure (GI) represent primary local sources of ecosystem services (ES). GI can be defined as

strategically planned network of green and blue spaces such as parks, gardens and ponds; whereas ES

include air purification, recreation, and food supply and are defined as the contribution of ecosystems to

of
human well-being like (Gómez-Baggethun et al., 2013; Keeler et al. 2019). However, the expansion of GI

ro
is frequently constrained by new and existing urban development, often at high density, forcing

municipalities to seek its integration with other urban infrastructures such as buildings. In this context,

-p
green roofs (GRs) – the vegetated coverage of building rooftops – are gaining momentum as a solution
re
for densely populated metropolises to ensure adequate supply of regulating ES like stormwater
lP

management, thermal regulation, ecological habitat (Oberndorfer et al., 2007), to enhance the local

provision of food (Buehler and Junge, 2016), but also to provide less tangible non-material and cultural
na

ES like relaxation and social integration (Mesimäki et al., 2017)

Worldwide, GRs are sprouting as architects and planners embrace rooftop gardens as aesthetically
ur

intriguing yet functional spaces. Municipalities are actively promoting GR installation through policy
Jo

instruments like construction regulations and economic incentives (Carter and Fowler, 2008; Berardi et

al., 2014), while academic study of the subject has grown exponentially (Shafique et al., 2018; Sutton and

Lambrinos, 2015). Despite this rising popularity, to date GRs have not been significantly studied from the

perspective of urban planning and land use policy in the face of an enhanced provision of ES. GR

research has often been limited in scope to technical models or comparisons of individual test rooftops

(MacIvor and Lundholm, 2010; Abualfaraj et al. 2018), offering little insight to planners seeking

guidance for informed GR expansion at a citywide scale. Moreover, existing assessments tend, with some

exceptions (Nurmi et al., 2016), to focus on quantifiable material ES (Lundholm and Williams, 2015)
 Journal Pre-proof

without broader consideration of the cultural benefits of GRs (Czemiel Berndtsson, 2010). In particular,

the lack of investigation into the cultural ES of GRs discounts a number of non-material benefits that are

often seen the most important ES to be provided in cities (Camps-Calvet et al., 2016).

Further, while the ES of GRs are inherently dependent on specific design choices like plant species,

substrate depth, and accessibility (Dvorak and Volder, 2010), many studies neglect to account for the

variability of these factors. Indeed, the common GR nomenclature is fairly reductive, often only

differentiating between two typologies: ‘extensive’, low-maintenance groundcover like grass and sedum

of
with shallow substrates, and ‘intensive’, full rooftop gardens with deep substrates and active usage

ro
comparable to ground-level parks (Rowe, 2011). Such binary classification is limited in accounting for

the true complexity of design wherein two nominally ‘extensive’ roofs can have drastically dissimilar

-p
construction, species heterogeneity, and/or intended use (Mahdiyar et al., 2018), thereby differentiating
re
the ES they can provide. By neglecting to account for the wider breadth of GR designs and associated
lP

barriers (economic, structural, and institutional) as well as potentials in ES provision, much GR research

lacks applicability and can stymy development of effective policy (Williams et al., 2010). In the few
na

recent studies that attempt to investigate the large-scale implementation of GRs, the limitations are

progressively reduced. However, there is still some way to go to increase a wider applicability and
ur

comparability of these studies.

Jo

By way of example, Karteris et al. (2016) calculated the potential contribution of GRs to energy

performance, carbon sequestration, and rainwater retention in Thessaloniki, Greece, but did not account

for the synergies and trade-offs between these benefits that could be assessed using the well-established

framework of ES mapping that distinguishes the supply, demand, and flow of ES (Crossman et al., 2013).

Meanwhile, although Grunwald et al. (2017) follows this approach to explore how rooftop greening in

Braunschweig, Germany could provide four ES – thermal climate, air quality regulation, water retention,

and habitat for biodiversity – the policy-making implications of their findings are relatively limited. This

could be added by means of a decision support tool like Multicriteria Decision Analysis (MCDA)
 Journal Pre-proof

allowing for the comparison of multiple alternatives through the performance and weighting of disparate

variables (e.g. Langemeyer et al., 2016). Finally, while Velázquez et al. (2018) incorporated some expert

feedback in basic MCDA to prioritize potential GR locations in Madrid, Spain based on air pollution,

traffic, existing greenery, and population density, their study – like the others listed – does not account for

non-material ES, nor adequately differentiates between different GR designs.

Here, we seek to unify and improve upon the disparate approaches of these studies by offering a

spatial policy screening tool that integrates the techniques of MCDA-supported ES mapping with the

of
flexibility of the mixed data non-parametric approach of Bayesian Belief Networks (BBNs). BBNs are

ro
graphical models (Nielsen and Jensen, 2009) that are particularly useful to support decision analysis. As

MCDA, they can incorporate a wide range of both qualitative and quantitative data (Chen and Pollino,

2012). -p
re
The spatial screening tool developed for the study was applied to the case study city of Barcelona,
lP

Spain with the specific objectives of: (1) to identify city-wide ES deficit areas where GRs have the

greatest potential to fill an existing lack of supply of material and non-material ES; and (2) to identify the
na

most suitable GR design for optimized ES provision potential in those areas. Further, the tool is designed

to be adaptable to support the effective implementation of other NBS in cities, offering a methodology
ur

that can be tailored to variations in available data and the needs of planners.
Jo

2. MATERIAL & METHODS

2.1. Description of the study area

The chosen study area for this assessment is the municipality of Barcelona, the capital of Catalonia,

Spain. Administratively divided into ten districts and 73 neighborhoods (see Appendix A), Barcelona is

home to 1.62 million inhabitants within its 102 km2 area, making it one of the densest and more compact

municipalities in Europe with very limited green space per capita, mounting to 7 m2/inhabitant in the city

centre (17,62 m2/inhabitant when including the peri-urban park Collserola), which is very low in
 Journal Pre-proof

European comparison (Baró et al., 2015). Located on the northeastern coast of the Iberian Peninsula, the

region is characterized by a Mediterranean climate averaging approximately 600 mm of annual rainfall

and typical yearly temperatures ranging between 9˚C and 24˚C. The city’s iconic rooftop terraces – long

utilized as elevated social spaces (Contreras and Castillo, 2015) – offer accessible and mostly structurally

sound sites for rooftop greening (Fig. 1). According to the Municipal Urban Ecology Agency, Barcelona

had installed 115 GRs by 2014 (BCNecologia, 2014) and current plan are to expand this cover by 5,431

m2 in 2019 and up to 22,000 m2 in 2030 as part of its citywide stimulus program to expand nature-based

of
solutions (BCN, 2017b). However, this represents a tiny fraction of the GR expansion possibilities in the

city, as approximately 65 hectares of suitable rooftops have been identified on publicly owned buildings

ro
alone (BCNecologia, 2010). In order to promote widespread adoption of GRs in Barcelona, this study’s

-p
spatial MCDA framework may assist future GR decision-making and prioritization processes.
re
- INSERT FIGURE 1 ABOUT HERE -
lP
na

2.2. Green Roof Design Alternatives

ur

Five GR design alternatives were selected based on Barcelona’s guidelines (Contreras and Castillo,
Jo

2015), including three standard industry typologies – extensive, semi-intensive, and intensive – that are

differentiated by substrate depth, associated vegetative capacity, and maintenance requirements (FLL,

2002; NTJ, 2012). Additionally, two specialized use categories – naturalized and allotment – were used to

evaluate the effect of intended usage on ES provision. Naturalized roofs are typically planted with

endemic species emulating natural habitats like meadows, while allotment roofs are explicitly designed

for rooftop agriculture. These five alternatives represent a wide range of economic, structural, and

maintenance requirements, as indicated in Table 1.

 Journal Pre-proof

- INSERT TABLE 1 ABOUT HERE –

A workshop was conducted on 5th June 2018 exploring GRs as NBS for Barcelona (See Appendix

B for workshop materials). The participants of the workshop (n=31) included academics, municipal

officials, NGO representatives, and private sector GR experts (see Appendix C for a listing of the

experts). Prior to the workshop, attendees viewed several presentations on public and private GR

initiatives in Barcelona. The participants represent broadly the local expertise on NBS in Barcelona.

of
Following an explanation of the study objectives and model criteria, workshop participants were split into

ro
three moderated groups. To allow for differentiated debates, split-out groups were formed

-p
heterogeneously, making sure each group included experts from academia, city planning (from different

scales), NGOs and private sector representatives. The experts were then asked to evaluate (1), which ES
re
should be prioritized with regard to the given deficits in Barcelona, (2) the capacity of different GR types
lP

to provide ES, and (3) the feasibility to implement different design alternatives.
na

(1) In order to determine which ES should be prioritized in Barcelona, a collective weighting approach

was applied, consisting in the distribution of 30 ‘pebbles’ between six categories of ES (see section 2.3
ur

for their selection). The results of this exercise are shown in Fig. 2 and Table 2.
Jo

- INSERT FIGURE 2 ABOUT HERE (together with Table 2) -

- INSERT TABLE 2 ABOUT HERE (together with Figure 2)-

 Journal Pre-proof

(2) To define the extent each of the five design alternatives are capable to provide ES, a group exercise

requested the unanimous grading of ES under each design alternative from 1 to 3 (‘little to no’,

‘intermediate’, and ‘strong’ provision). Average score are shown in Table 3.

- INSERT TABLE 3 ABOUT HERE -

(3) The general feasibility of installing the five design alternatives considered economic, structural, and

of
institutional barriers, which were ranked individually by workshop participants on a Likert scale of 1 to 5

(‘very low’ to ‘very high’). Results from the ranking exercise are presented in Table 4.

ro
-p
- INSERT TABLE 4 ABOUT HERE -
re
lP

2.3. Ecosystem Services

na

Six ES attributed to GRs were chosen as most relevant in the study area: thermal regulation (micro and

regional climate regulation), stormwater runoff control, habitats for pollinators, food production,
ur

recreational opportunities, and the facilitation of social cohesion (Berardi et al., 2014; Lundholm and
Jo

Williams, 2015; Mesimäki et al., 2017). While not an exhaustive list of GR-related ES – other ES

provided by GRs include aesthetics, mediation of noise, and protection of the building structure – these

services make GR relevant NBS for a variety of urban problems (urban heat island, flash flooding, social

inequities, etc.), allowing for a holistic assessment of GRs’ benefits. Moreover, demand for these ES can

be estimated using spatial proxies available at the citywide scale, unlike other services associated with

GRs. For instance, building-specific GR benefits like noise insulation, roof longevity, and thermal

insulation (Rowe, 2011) were not included due to a lack of data on individualized structural envelopes

and energy usage.

 Journal Pre-proof

This section presents theoretical justifications for the studied ES and their associated demand

indicators (spatially determining the deficit of each of these ES across the study area), along with the

calculations and classifications made in GIS and HUGIN used to create and weight the initial input

rasters. An overview of the ES indicators is given in Appendix D. The relative weights of the indicators

were established by the analysts, based on the respective literature and personal knowledge about their

relevance in the case study city. We acknowledge that the selection and weighting of indicators is

somewhat arbitrary, a more sophisticated approach for example based on a Delphi consultation of experts

of
was beyond the scope of this study. However, the resulting models and sub-models can be easily adapted

in order to incorporate different weights, indicators or even additional ES.

ro
2.3.1. Thermal Regulation

-p
One ES frequently associated with GRs is the regulation of urban temperatures, both via passive
re
building insulation and active mitigation of the Urban Heat Island effect, or the anthropogenic warming of
lP

urban settlements above their rural surroundings (Li et al., 2014). GRs have been proposed as an effective

method for addressing urban heat by increasing albedo (solar reflectance), insulating rooftop membranes,
na

and cooling the rooftop surface directly via vegetative evapotranspiration. Indeed, experimental and

modeling research indicates GRs can reduce surrounding temperatures by as much as 3˚C (Santamouris,
ur

2014). Such mitigation is particularly important in warm climates like Barcelona where the heat island
Jo

can elevate urban temperatures by up to 8˚C (Moreno‐ Garcia, 1994), posing a significant health threat to

vulnerable populations during heat waves, including elderly, children and poorer parts of the population

who often lack access to air conditioning (Harlan et al., 2006). Therefore, to evaluate the demand for

thermal regulation, two components of heat risk were selected as equally-weighted indicators for this

model: UHI intensity and heat stress vulnerability.

Urban Heat Island Intensity

Although UHI can be estimated by models and/or remote sensing of surface temperature, direct

measurement of air temperature was selected as a more representative input for this model. Martin-Vide
 Journal Pre-proof

et al. (2015) evaluated UHI across Barcelona from October 2014 to March 2015 by measuring air

temperature from vehicles moving along three transects of Barcelona. Based on their analysis, using

mapped temperature isolines from three dates with low, intermediate, and high UHI, a single raster was

created in ArcGIS (version 10.6.2) depicting average difference in air temperature across the region.

Unsurprisingly, UHI was highest in the densely built central districts of Eixample and Gràcia and lowest

around the outskirts of the city near the mountains and coastline. For uniform distribution, the raster was

divided into ten classes by ½ standard of deviation (STD) which were given a positive, linear correlation

of
with cooling demand.

ro
Heat Vulnerability

Demographic heat risk has previously been evaluated in multiple ways, using income, age, and race
-p
as mediators (Aubrecht and Ozceylan, 2013). In this study, a pre-existing heat vulnerability map was
re
obtained from the City of Barcelona that evaluated risk via the following indicators, selected based on
lP

input from Barcelona Public Health Agency: elderly (75+ years) population density, building energy

performance, vegetation, and low educational attainment (BCN, 2018a). The resulting map classified
na

Barcelona into five vulnerability categories (very low to very high), which were scaled positively and

linearly to ES demand in HUGIN. The most vulnerable areas were identified in the
ur

North along the Besòs River extending into Nou Barris and Horta, and south in Sants-Montjuïc (Fig. 3).
Jo

- INSERT FIGURE 3 ABOUT HERE -

2.3.2. Runoff Control

Attenuating and delaying the release of stormwater runoff is another major ES provided by GRs

(Lundholm and Williams, 2015). During intense rainstorms, many urban areas are vulnerable to flash

flooding due to the prevalence of impermeable surfaces and insufficient retention capacity of
 Journal Pre-proof

conventional drainage systems. Flooding of this kind in Barcelona can cause millions of Euros in property

damage within flood-prone neighborhoods (Velasco et al., 2013). Only in 2018 – the year this study was

conducted - there were seven extreme precipitation events with over 25mm rainfall within less than 24h,

and local peaks of over 70 l/m² in less than an hour. GRs can reduce the risk of flooding by retaining up to

85% of excess stormwater and delay the runoff release (Czemiel Berndtsson, 2010).

In order to evaluate the potential demand for this attenuation, the current potential for runoff across

Barcelona was estimated. Stormwater runoff is most often expressed in terms of runoff coefficients,

of
which represent the percent runoff resulting from a rainstorm. These coefficients are highly affected by

ro
land cover and soil permeability, but also by slope and rainfall intensity (Sriwongsitanon and Taesombat,

2011). As such, a land use-based estimation of runoff coefficients was selected.

Runoff Coefficient -p
re
In lieu of calculating local runoff coefficients, general values were derived from land use classes
lP

akin to the methodology of Puccinelli et al. (2012). Runoff coefficients were assigned in line with

European Corine Land Cover (Corine) classes according to four soil permeability levels. To improve
na

upon the coarse resolution of Corine data, a detailed Land Cover Map of Catalonia (LCMC, 2009) was
ur

obtained, and the base land cover matched to equivalent Corine classes. Updated land use covers were

then assigned corresponding runoff coefficients averaged from the four Corine permeability levels. The
Jo

resultant map shows Barcelona to have generally high coefficients due to the density of the built

environment, with the greener Collserola and Montjuïc areas absorbing more runoff. Coefficients were

scaled in the range of 0 to 1, with deciles scaled linearly and positively with demand (Fig. 4).

- INSERT FIGURE 4 ABOUT HERE -

 Journal Pre-proof

2.3.3. Pollinator Habitat

GRs play an important role in promoting urban biodiversity as habitats for local fauna, particularly

insects and some birds, that pollinate urban flora and regulate invasive pests (MacIvor and Lundholm,

2010). Indeed, GRs can contribute to ‘green corridors’ that allow these beneficial species to circumvent

urban barriers to movement (Orsini et al., 2014). To identify the deficit in such connectivity, this study

simplified the ESTIMAP pollination model (Zulian et al., 2013) that uses land cover to estimate two

indicators of pollinator habitat potential: floral availability and nesting suitability. These indices were

of
both given negative linear correlation to demand, weighted 3:1 towards floral availability as nesting sites

ro
are less prevalent within urban settings (Stange et al., 2017).

Floral Availability & Nesting Suitability

-p
The original ESTIMAP framework modelled the relative pollination potential of wild insects across
re
Europe by assigning two habitat suitability scores between 0 and 1 to Corine classes, adjusted for
lP

agricultural crop type and proximity to roads, water bodies, and forest edges (Zulian et al., 2013). As with

runoff control, Corine scores were translated to the urban scale by correlating to the LCMC (2009),
na

averaging both base cover and composite land use classes to account for industrial and port areas. Due to
ur

the compact scope and lack of typical agriculture and forest edges, adjustment of this base score was

deemed unnecessary. Both maps show Barcelona to be widely unsuitable for pollinators, apart from
Jo

predominantly green areas like Collserola and some larger parks (Fig. 5).

- INSERT FIGURE 5 ABOUT HERE -

 Journal Pre-proof

2.3.4. Food Production

As GRs increasingly emerges as an auxiliary source of healthy food production and food security

(Whittinghill and Rowe, 2011), GRs offer significant potential for food production, particularly in cities

like Barcelona where agricultural land is inexistent and existing urban gardens for the production of food

are scarce (Camps-Calvet et al., 2016). At maximum capacity, rooftop farming is estimated to be able to

supply large parts of cities’ fruits and vegetables demands (Orsini et al., 2014). To evaluate demand for

such production, three indicators were selected, including walking distance to existing urban gardens

of
(weighted 60%), population density (weighted of 30%), and grocery store count per neighborhood

ro
(weighted 10%).

Garden Network Distance to Community Gardens

-p
While participation in urban gardens is highly variable, evidence suggests that most community
re
gardens are predominantly utilized by residents of its immediate neighborhood (Meenar and Hoover,
lP

2012). Thus, walking distance to existing urban gardens plays an important role in assessing the demand

for new rooftop gardens. Excluding private and school gardens, 44 extant urban garden sites were
na

obtained from Camps-Calvet et al. (2016) and the crowd-sourced Barcelona-Sostenible Map (BCN,
ur

2018b). Utilizing a similar methodology to Meenar and Hoover (2012), the network walking distance
Jo

around each garden site was calculated in ArcGIS as 300 m buffer service areas using a map of walkable

Barcelona streets, created from the open-source OpenStreetMap (OSM) base layer (OSM, 2018). The

resulting map shows Barcelona to be generally relatively well serviced by urban gardens, other than city

outskirts and areas within the neighborhood of Sarrià-Sant Gervasi, although urban gardens are generally

undersized and can only serve a relatively small number of beneficiaries, which is why the surrounding

population density is critical to be considered.

 Journal Pre-proof

To evaluate relative demand in HUGIN, a simple distance decay function was adapted from the

formula for facility accessibility found in Giles-Corti and Donovan (2002), which was given an inverse

relationship to demand and normalized to a 0 to 1 scale:

𝑑0 𝛽
𝑞(𝑑) = 1 − ( ) (eq. 4)
𝑑

where q is the relative demand for a new facility, d is the distance to an existing facility, and β is the

facility-specific distance decay factor, assumed to be 1 for urban gardens.

of
Population Density

ro
While ES are inherently defined by their benefit to humans, certain services are felt more directly

than others; for instance, a garden’s vegetable harvest is more tangibly beneficial to a local user than its

-p
runoff absorption. Population density was therefore considered a mediator of demand for food production
re
(as well as for recreation), which provide explicit and localized benefits to individual residents. Heat
lP

vulnerability and ethnic heterogeneity indicators already account for population distribution.

To calculate population density, a map of Catalonia was obtained from IDESCAT with the 2016
na

population distribution visualized in a multiresolution grid of 62.5, 125, and 250 meters (IDESCAT,
ur

2016), sized according to the quadtree method of Lagonigro et al. (2017) to ensure that each pixel

contained a threshold of 17 residents. This grid was clipped to Barcelona city limits and converted to a
Jo

2x2 m resolution raster displaying people/hectare (a lower resolution could have been chosen but we

aimed at matching the resolution of the indicators with that of the rooftop layer, further described under

2.4.1). Population in Barcelona is centrally concentrated with the densest areas located in historical

settlements like Ciutat Vella and Sant Andreu. This population density raster was classified into nine

groups with a ½ standard deviation (STD) range and a null class with zero inhabitants.

Neighborhood Grocery Store Count

Urban residents generally obtain their vegetables from markets and groceries, so the presence or

absence of these facilities affects the need for alternate produce sources like GRs (Walker et al., 2010).
 Journal Pre-proof

Food accessibility has been studied extensively, often via GIS assessment of store density (Caspi et al.,

2012). For this study, 2,028 properties were selected from a geocoded list of all businesses in Barcelona

(BCN, 2016b), labelled as ‘Fruit and Vegetables’ or manually identified using known supermarket chain

names (e.g. ‘Condis’ or ‘Dia’). When intersected with local neighborhoods in ArcGIS, grocery stores are

found to be located predominantly within high-traffic areas such as Gràcia and Raval. The count of

grocery stores was then classified into ten groups using the Jenks natural breaks method and scaled

negatively and linearly to demand (Fig. 6).

of
ro
- INSERT FIGURE 6 ABOUT HERE –

-p
re
2.3.5. Recreational Opportunities
lP

GI generally assumed to offer many opportunities for recreational activity (Bancroft et al., 2015)

which, in turn, are associated with numerous health benefits (Sugiyama et al., 2014). The association
na

between GRs and recreation, while lacking empirical quantification, is frequently cited in both academic

and grey literature that mention potential for physical recreation through gardening activities, walking,
ur

and other forms of physical activity, e.g. outdoor gym and yoga (Spala et al., 2008). We assume GR to
Jo

partly compensate the lack of opportunities for "ground-based" recreation, given by walkability of streets,

fitness facilities, and the availability of parks (Holliday et al., 2017). Therefore, neighborhood walkability

was selected as primary indicator with 35% model weight, distance to existing sport facilities and

neighborhood greenness were weighted 25% each, and population density (see Section 2.3.4) was

attributed 15% weight.

Neighborhood Walkability

Neighborhood walkability is a well-established indicator of physical activity within the built

environment (McCormack and Shiell, 2011). Often estimated using proxy-based models with variables
 Journal Pre-proof

such as land use and network form (Lefebvre-Ropars et al., 2017), walkability can be accurately assessed

via tabulating ‘proximity journeys’ of under 10 minutes that are typically local and non-motorized. Using

a data set of 24,000 telephone interviews conducted in Barcelona as part of the 2006 regional Everyday

Mobility Inquiry, Marquet and Miralles-Guasch (2015) analyzed the weekday travel of residents over 16

years of age and mapped the relative frequency of proximity trips at a neighborhood level. While most of

the city was relatively uniform (22-27% of all travel), older, denser areas like the Old Town, Poble Sec,

and San Andreu showed more (27-30%) proximity trips while lower walkability (19-22%) was observed

of
in the districts of Sarrià-Sant Gervasi and Sant Martí. For the study model, these three classes were scaled

linearly and negatively with demand.

ro
Sports Facility Distance

-p
The usage of sports facilities like gyms and sports fields for physical activity is greatly affected by
re
distance to users’ homes (Giles-Corti and Donovan, 2002). Network walking distance was calculated in
lP

ArcGIS using 100 m and 500 m service area buffers around sport facility locations, compiled from 466

addresses labeled as ‘Sports’ on the city commercial properties list (BCN, 2016b), along with 364 sports
na

fields labeled as ‘sport zones’ on the Barcelona sub-parcel map (BCN, 2012) or with the composite land

use ‘sport areas’ on the LCMC (2009). The resultant map shows sports facilities distributed generally
ur

evenly, with a slight lack of coverage in el Barri Gòtic and other non-settled outskirt areas. Demand was
Jo

calculated using the distance decay formula (Eq. 4), with β = 1.16 per Giles-Corti and Donovan (2002).

Neighborhood Greenness

The location and ES supply of existing GI is assumed to lower the demand for new green spaces

like GRs. While such provision is implicitly assessed in the model using land use-based indicators for

some ES, we deemed an additional measure of greenness was necessary for recreation (as well as for

social cohesion), as multiple studies suggest that GI mediates physical and psychological health through

these mechanisms (Maas et al., 2009). Surrounding greenness was used as the chosen indicator over

objective proximity to GI, per the findings of Dadvand et al. (2016). To quantify surrounding greenness,
 Journal Pre-proof

the normalized difference vegetation index (NDVI) of the city was obtained from the city, classified such

that all NDVI values above 0.18 were considered as ‘green’ (Barcelona Regional, 2015). Subsequently,

percent greenness was calculated for each census tract by dividing the ‘green’ area by total tract area. To

account for proximity, this percentage was then added to the average percent greenness of all neighboring

tracts, and the result normalized to a maximum of 1 (Fig. 7).

– INSERT FIGURE 7 ABOUT HERE –

of
ro
2.3.6. Social Cohesion

-p
Broadly characterized as interpersonal relationships that facilitate cooperation and trust (Chan et al.,
re
2006), social cohesion is an intangible ES that is difficult to quantify yet forms a key component of urban
lP

life. GRs offer great potential to provide this as communal spaces that promote social interaction and a

unique sense of place (Mesimäki et al., 2017), in line with numerous studies that suggest that GI
na

potentially facilitates social cohesion (e.g. Maas et al., 2009; Markevych et al., 2017) and strengthens

social ties (Kaźmierczak, 2013).

ur

Further, allotment gardens can foster shared values and community identity (Langemeyer et al.,
Jo

2018), suggesting a similar potential for rooftop gardens in particular. This study used two established

indirect proxies to evaluate demand for social cohesion: income inequality and ethnic heterogeneity

(Easterly et al., 2006). These indicators were each weighted 40% in HUGIN, with an additional 20%

assigned to neighborhood greenness (see Section 2.3.5) to account for the potential mediating effect of on

the ground GI.

 Journal Pre-proof

Income Differential

Income inequality may be correlated with decreased social trust, particularly in poorer

neighborhoods (Kawachi et al., 1997). To represent income inequality in Barcelona, this study used the

city’s Available Family Income (RFD) statistic, which combines Gross Family Income with education

level, employment, car ownership, and real estate prices (BCN, 2016a). This indicator is calculated at the

neighborhood level and compared against the average income of all Barcelona residents. As of 2016, the

wealthiest areas were in eastern Barcelona around the Sarrià-Sant Gervasi and Les Corts districts, with the

of
poorest located in the north around Nou Barris and south in Sants-Montjuïc. Data was then classified

ro
using ½ STD with demand increasing linearly and positively away from the mean.

Ethnic Heterogeneity
-p
Although subject to some debate among experts, ethnic heterogeneity has generally been shown to
re
negatively affect social cohesion (Laurence, 2009). To calculate this heterogeneity in Barcelona,
lP

nationality statistics were used in lieu of ethnicity data, which is not collected in Spain (BCN, 2017a).

The 178 nationalities present in Barcelona were grouped into ten cultural clusters per the GLOBE study
na

(House et al., 2004; Mensah and Chen, 2013). Diversity between these groups was then calculated for
ur

each census tract using Theil’s entropy score (Iceland, 2004):

𝑟
Jo

1
𝐸𝑖 = ∑ 𝛱𝑟𝑖 × 𝑙𝑛( ) (eq. 5)
𝛱𝑟𝑖
𝑟=1

where E is the entropy of a tract, i, and Π signifies the population of a particular ethnic group, r. Results

of this calculation determined that Ciutat Vella and parts of Sant Martí are the most diverse areas of

Barcelona, with less heterogeneity being observed further away from the coast. Entropy scores were

classified by ½ STD and scaled linearly and positively with demand (Fig. 8).

– INSERT FIGURE 8 ABOUT HERE –

 Journal Pre-proof

2.4. Model Construction

The spatial MCDA framework of this study encompasses two complementary BBN models. Firstly,

an ES demand model assessed the deficit (or need) for six ES across Barcelona in order to identify where

the implementation of GRs would have the most benefit. A second ES supply model evaluated the

potential ES provision of each rooftop in the city under five GR design alternatives that best match the ES

demand in a specific location.

of
2.4.1. Model Structures

ro
The study models were constructed using HUGIN Researcher v8.6, the original commercial BBN

-p
modelling software (Andersen et al., 1989). BBN modelling is grounded in fundamental probability
re
theory dating from the 18th century, BBNs have been used since the 1980s for a vast array of applications,
lP

ranging from epidemiology to development of artificial intelligence (Barton et al., 2012). BBNs are a

particularly useful tool for decision-support analysis as they can incorporate a wide range of both
na

qualitative and quantitative data (i.e. expert opinion and experimental outputs), are easily updated as new

information becomes available, and allow for both inductive and deductive reasoning (Chen and Pollino,
ur

2012). For more in-depth information on how to build and evaluate BBN models see for example
Jo

Kjærulff and Madsen (2013), Marcot et al., (2006), or Jensen (2001).

In a BBN, variables are graphically represented by nodes linked together within a non-looping

causal network. Each node can exist in a number of possible states (i.e. nesting suitability on an index

between ‘0’ and ‘1’). The latest version of HUGIN can also spatially integrate with Geographic

Information Systems (GIS) using a plug-in for the open-source mapping software QGIS. This tool links

GIS raster layers to HUGIN nodes in a BBN, which are then calculated for each individual pixel creating

a new output map.

 Journal Pre-proof

Thus, the general modeling approach for both BBNs was to first obtain or create spatial indicators

of ES demand using ArcGIS Desktop 10.6 and QGIS v 2.18.15 (for further details see Appendix E). All

input rasters were standardized by resolution (2x2m), extent (Barcelona municipal limits), and projection

(ETRS 1989 UTM Zone 31N). Layers were then assigned input nodes in HUGIN using numeric interval

states (i.e. ‘0-0.5’ and ‘0.5-1’) corresponding to raster classifications appropriate to the data type. ES with

different units were scaled in HUGIN to a standard index for direct comparison, as required by multi-

attribute value functions in MCDA (Kremer et al., 2016). Additionally, results from the expert workshop

were added as weighting nodes with numeric or labelled states (i.e. ‘extensive’ or ‘intensive’). Finally,

of
BBNs culminated in output utility nodes that used model-specific formulae to evaluate the demand of

ro
each ES. This utility was first calculated solely in HUGIN for baseline results, and then mapped in QGIS

in order to visualize ES demand.

-p
re
Demand Model
lP

For the demand model (Appendix F) input nodes, representative spatial indicators were selected

for each ES (Appendix D). Indicator raster classifications were scaled to a standard index of potential
na

demand between 0 and 1 using individual scaling nodes with decile intervals. Overall scaling was either

positive or negative (i.e. demand rises with temperature but decreases with greenery) and followed a
ur

linear trend or distance decay curve.

Jo

Scaling nodes were then combined into a single aggregate node for each ES, weighted according to

the relative weight of each indicator. This raw ES demand was further adjusted according to expert

feedback on Barcelona’s ES deficits and needs (Table 2) to evaluate expected utility.

The expected utility for each potential GR location is computed conditional on the ecosystem services

provided at each location (L). The identification of a location 𝛿(𝐿) in the supply model determines the

ecosystem services at that location before GR implementation. The expected utility associated with utility

of ecosystem service (j) (Uj) is computed by summing the parent configuration over the product of the
 Journal Pre-proof

utility function and the joint probability distribution of the parent variables determining the ecosystem

service (conditional on the choice of location):

𝐸𝑈(𝑈𝑗|𝛿(𝐿)) = ∑𝑊,𝑆𝑗 𝑢𝑈𝑗 (𝑊, 𝑆𝑗) ∗ 𝑝(𝑊, 𝑆𝑗|𝛿(𝐿)) (Eq.1)

where

p()=marginal probability

W= stakeholder criteria weights

of
Sj = scaled ecosystem service j (each ecosystem service is scaled/normalized in order to be comparable)

The total expected utility at each location is the sum of the expected utilities (EU) of each service.

ro
HUGIN software QGIS plug-in can be used to compute the Maximum Expected Utility (MEU) in all

-p
rasters which was mapped in QGIS to form an aggregate ES demand raster, along with the individual
re
utilities of each ES.
lP

Supply Model

To create the spatial inputs of the supply model (Appendix G), the individual ES output rasters
na

from the demand model were then clipped in ArcGIS using a masking layer depicting all rooftops in

Barcelona, created from maps of Barcelona sub-parcels (BCN, 2012) and potential green roofs
ur

(BCNecologia, 2014). The resulting layers represent the modeled ES deficit at the location of the city’s
Jo

rooftops before GR are implemented. In HUGIN, these layers were given input nodes with twenty 0.01

interval states between 0 and 0.2.

Additionally, a decision node (D) was created with the five design alternatives as labelled states.

This node was linked to six weighting nodes representing percent ES contribution, each with three 0.33

intervals between 0 and 1, matching the expert groups’ evaluation of ES provision by different GR design

alternatives (Table 3). Similarly, the decision node was linked to a feasibility weighting node (F),

weighting the GR design alternatives with the individual Likert scale feasibility responses of the experts
 Journal Pre-proof

(Table 4), translated to 0.2 numeric interval states between 0 and 1. Expected utility nodes for each

ecosystem service (j) produced by the GR design, were defined by the following expression:

𝐸𝑈(𝑈𝑗|𝛿(𝐷)) = ∑𝐹,𝑆𝑗 𝑢𝑈𝑗 (𝐹, 𝑆𝑗) ∗ 𝑝(𝐹, 𝑆𝑗|𝛿(𝐷)) = ∑𝐹,𝑆𝑗 𝑢𝑈𝑗 (𝐹, 𝑆𝑗) ∗ 𝑃(𝑆𝑗|𝛿(𝐷)) ∗ 𝑃(𝐹) (Eq.2)

where

p()=marginal probability

of
F = feasibility weighting of each green roof design

ro
Sj = scaled unmet ecosystem service demand j potentially met by GR design

-p
Only (𝑃(𝑆𝑗|𝛿(𝐷)) depends on the choice of GR design 𝛿(𝐷). This distribution is computed by

inference in the BBN model. The total expected utility is the sum of the expected utilities of each GR
re
design alternative.
lP
na

Utility nodes for each ES were defined by the following expression:

𝑃𝑠 (𝑎) = 𝐹(𝑎) × 𝐸𝑠 (𝑎) × 𝑄(𝑠) (Eq. 3)

ur
Jo

where P signifies the expected utility of ES provided by GR, s, that each GR alternative, a, provides; F

and E represent the expert-evaluated feasibility and extent of ES provision for each alternative,

respectively; and Q indicates the full provision of each ES.

Five output maps were then created in QGIS representing the MEU for each of the design alternative

decisions. These maps were analyzed and synthesized in ArcGIS to produce a single map depicting which

GR design provides the highest potential contribution to ES provision.

The spatial outputs of the BBN demand and supply models depict aggregate potential ES deficit

and provision across Barcelona, using standardized indices for each ES ranging between 0 and 1. These
 Journal Pre-proof

rasters were analyzed in ArcGIS (version 10.6.2) using Zonal Statistics to identify the rasters’ maximum,

minimum, mean, and standard of deviation at the city, district, and neighborhood scales. Supplementary

results of the two models are presented in Appendix H.

of
ro
-p
re
lP
na
ur
Jo
 Journal Pre-proof

3. RESULTS

3.1. Demand Model Results

The output of the demand model, representing the weighted, aggregated ES demand across

Barcelona, highlights clear areas where GR development should be encouraged. ES demand in Barcelona

was lowest in the Collserola Natural Park in northwest, increasing dramatically as the built environment

densifies. Notably, both the minimum (0.306) and maximum (0.722) demand values were found in the

of
Horta-Guinardó district, highlighting the extreme differences between Collserola and the urbanized

ro
neighborhoods adjacent to them. Indeed, potential demand was concentrated in the densely populated

-p
residential neighborhoods, generally located centrally (Raval, Barri Gòtic, and Eixample) but with several

isolated areas of high demand present in the East (Besòs i Maresme) and North (Verdun, Teixonera, and
re
Carmel). The largest mean district demand was observed in Sants-Montjuïc, which was responsible for
lP

almost a quarter of Barcelona’s total demand and exceeded what would be ‘expected’ based on its area

relative to the rest of the city by 2.2% (Fig. 9).

na
ur

Disaggregated into individual ES, habitat for pollinators was identified as the most needed ES.

Demand for thermal regulation was also high for all districts apart from Sarrià-Sant Gervasi and Les
Jo

Corts, where it was surpassed by recreation and social cohesion. Demand for these cultural ES was

important across the city, averaging 16.1% and 17.1% respectively of the Barcelona mean. Social

cohesion was predicted to be the second most demanded ES, reflecting the high importance given to this

ES by experts.

– INSERT FIGURE 9 ABOUT HERE –

 Journal Pre-proof

3.2. Screening tool results

The output of the screening tool estimated the potential ES provision potential of the five design

alternatives for every rooftop in Barcelona and identified the highest contributing design for that location.

The estimated potential ES provision of GRs in Barcelona was found to be relatively small, ranging from

0.12 to 0.28 on the index. By nature of the model design, the potential ES provision was relative to the

aggregate demand, so Sants-Montjuïc was correspondingly identified as the district with the largest

portion (20%) of Barcelona’s aggregate ES provision. As an industrial zone, this district has several

of
factories with large rooftops that collectively offer a mean potential ES provision of 0.23. This district

ro
mean is just surpassed by that of Ciutat Vella, which had the greatest mean provision due to its dense

concentration of rooftops in an area of high need, similar to Eixample which had the third highest mean

-p
(Appendix H.4). At the neighborhood level, roofs with the highest ES provision generally correlated with
re
the ‘hotspots’ of demand, although differences in rooftop density identified additional areas (Ciutat
lP

Meridiana, Roquetes, and Font de la Guatlla) that could be suitable for GRs (Fig. 10).
na

– INSERT FIGURE 10 ABOUT HERE –

ur
Jo

For the decision component, naturalized roofs were selected as the optimal GR design for the

majority of Barcelona, accounting for 87.5% of rooftop area (Appendix H.2). Intensive roofs were picked

for the bulk of the remainder and were deemed most effective in the neighborhood of Gràcia, where they

were the chosen design for nearly half of the rooftops. Semi-intensive roofs accounted for only 0.05% of

the rooftop area, despite a comparable city-wide mean provision to intensive roofs. Neither extensive nor

allotment roofs were selected.

 Journal Pre-proof

4. DISCUSSION

4.1. Green Roof Prioritization

Results from the city-wide models operationalized in this study offer a number of findings that

support future land use policy in Barcelona yet are also applicable to GI development elsewhere. From a

prioritization perspective, the model identified numerous neighborhoods across Barcelona where GRs

could offer important NBS to the city’s environmental challenges. As expected, the areas that are

of
identified as priorities for GR tended to be densely populated urbanized neighborhoods, often with

ro
diverse residential populations. GR development in these areas could be directly beneficial to local

-p
communities, provided that design and implementation are undertaken with input from residents to

preserve the character of the neighborhood (Anguelovski et al., 2017).

re
While the greatest potential GR benefits were concentrated in the continuous urban fabric of
lP

Barcelona’s center, numerous areas of high demand of ES were also identified in communities located on

the outskirts of the city, often adjacent to expansive areas of greenery like the Collserola mountain range.
na

This suggests that proximity to green areas may have less of an impact on aggregate ES demand than
ur

more immediate factors like urban form or parcel land cover, as well as specific vulnerabilities and needs

of the population. As there is considerable debate concerning the mechanisms by which green space
Jo

services the local community (Triguero-Mas et al., 2015; Markevych et al., 2017), planners should not

preclude GR installation near existing GI nor assume the extent of ES provision based purely on level of

‘greenness.’

In addition to the aforementioned residential areas, particularly high potential for GR

implementation was found in the industrial park to the South of the city. Although only 20% of the land

within this zone is covered by structures, the large area of individual factory rooftops offer significant

potential for GR development, particularly if using low maintenance designs that provide regulating ES

like thermal regulation and runoff control that are especially useful when considering impermeable
 Journal Pre-proof

surfaces common to industrial parks (Snodgrass and Snodgrass, 2006). Moreover, industrial roofs have

been suggested as ideal sites for rooftop farming, although structural limitations like sloped or thin metal

roofs must be factored into the planning process (Sanyé-Mengual et al., 2015). Nevertheless, this study’s

findings reinforce the notion that policymakers should consider all possible building types and future

usages for GR initiatives.

We see potential in our modeling to inform spatially tailored GI policies. In Barcelona, it might

help steering financial incentives provided by the City Council for the implementation of GRs, in terms of

of
location and GR type. Applied to Paris, where the creation of green roofs is mandatory (in new buildings),

ro
it might help refining the requirements for GR designs; this might also be the case in New York City were

a similar legislation is underway. Yet, the model was run with relative large amounts of input data, which

-p
might not necessarily be available in other smaller or global South cities; however, the selection of
re
indicators embeds some flexibility; it can be adapted to local data availabilities and complemented by
lP

expert judgment, where formal data is missing, as shown here for example for the capacity of different

GR types to provide ES (Table 3), although expert judgment involves its own limitations as further
na

discussed in the following. It is also worth mentioning that the ES demand model is not limited to

informing GR policies but might be applied to any other GI planning.

ur

4.2. Green Roof Design Considerations

Jo

The results of the supply model clearly indicate the primacy of naturalized roofs over other design

alternatives in Barcelona. A possible explanation is that the high importance placed on habitat and

pollination by the demand model translated to the naturalized roof alternative dominating its competition

in the supply model. These results suggest that biodiversity and green connectivity should be prioritized

by policymakers. Yet, another conclusion might be a need for recalibration of the model and the selection

of additional indicators beyond the ESTIMAP model. However, some experts did indicate they consider

naturalized roofs were essentially more effective extensive roofs, due to their similar design

characteristics (see Table 1).

 Journal Pre-proof

Such expert bias may also explain why allotment roofs rated so well for feasibility and estimated

ES provision, despite a disproportionately low assessed need (by the experts) for Barcelona. Workshop

participants toured a functioning allotment garden immediately before the workshop and were lectured

about their dietary and social benefits, which certainly could have influenced their subsequent discussion.

As these instances suggest, over-emphasis on qualitative expert inputs is an understandable critique of

this study, as qualitative data in general can be perceived as suffering from subjectivity and generalization

(Landuyt et al., 2013). However, this drawback can easily be addressed via additional workshops and

of
future model calibration, i.e. explicitly incorporating probabilities and uncertainty measures related to

qualitative data. The latter correction represents a strength of BBN-based analysis.

ro
A more severe limitation in our study is given by the lacking availability of data on housing

-p
structure and the actual capacity of roofs to carry diverse types of GRs, including naturalized and
re
intensive GR, which rendered the highest potentials with regard to ES deficits in the city of Barcelona,
lP

but which might also require important structural investments to be implemented and which cannot be

created on all types of buildings. Similarly, the relatively equal weighting of intensive and semi-intensive
na

roofs could be attributed to underestimating the true economic and structural differences between the two

alternatives. These considerations underscore that the provision of ES cannot be the only relevant criteria
ur

for the implementation of GRs. Economic, structural, and institutional barriers have only been treated
Jo

superficially in this study, other relevant aspects, such as labor, water, energy, fertilizer needs have not

been considered in this study but would most likely influence the final selection of GR alternatives

considerably.

4.3. Model Improvement

The spatial BDA framework proposed in this study represents a proof of concept, with significant

opportunities for future refinement and expansion. For instance, the current BBN models would benefit

from continued iterative development allowing for improved scaling via expert calibration or expanded

scope based on stakeholder feedback (Marcot et al., 2006). Indeed, the expert workshop identified several
 Journal Pre-proof

priority ES for Barcelona –namely, air filtration, noise reduction, and environmental education– that may

prove capable of improving the BBN’s applicability to local policy goals. Additionally, the spatial

indicators of the modeled ES could be expanded by more precise information (e.g. the size of urban

gardens for food production would provide additional information on the deficit in food supply), updated

with more recent data or augmented using emerging technological advances that allow for the remote

identification of rooftop materials (Nadal et al., 2017). With such technology, a third BBN module could

well be added to the framework for the purpose of predicting suitable GR by synthesizing a variety of

of
structural, economic, and institutional factors.

ro
Clearly there is significant room for development, particularly if the framework fully embraces the

extensive functionality provided by HUGIN, such as the upcoming ability to visualize uncertainty over a

-p
geographic area following Landuyt et al. (2015). The model is highly sensitive to expert assumptions, the
re
software’s core probabilistic calculations offer a way of capturing uncertainty in expert judgement and
lP

identifying uncertainty of unobserved variables (e.g. roof and building loading capacity, subsoil

characteristics). By way of example, one possible scenario could be to implement quantitative ES

na

modeling of GRs, integrating model uncertainty (e.g. following Marcot, 2012). As this was not the focus

of this investigation, the potential ES provision of GRs was estimated in a fairly simplistic manner using
ur

exclusively qualitative inputs derived from expert preferences. While this method has been used to
Jo

estimate non-material ES like GR aesthetics (Lee, 2014; Loder, 2014), it could be improved with the

introduction of experimental values from multiple sources. To account for spatial and experimental

variability, uncertainty between this data could first be evaluated in HUGIN and then the merged with the

expert preferences in an integrated and novel approach to assessing a GR’s ES provision.

5. CONCLUSIONS

The novelty of this study lies in its multi-faceted approach to policy screening that is within the first

of the field of GR research to incorporate cultural ES as well as a plurality of alternate design scenarios.
 Journal Pre-proof

This approach allows for holistic analysis of the myriad factors that affect the ES provision of GRs,

resulting in an assessment of both potential ES demand across Barcelona (or any city) and the potential

ES provision of five unique design alternatives. The study shows important insights to inform Barcelona’s

GR Strategy. The overarching model structure developed in this study is applicable to other locations and

research questions. Indeed, parallel applications are ongoing for Oslo and New York City, which will

allow for comparison between GR policy and provision across different cities. However, the application is

constraint by large data availability and the possibility to collaborate with a diverse group of qualified

of
experts, which might hamper its application elsewhere.

ro
Although focused specifically on GRs, this study establishes a spatial multi-criteria screening

approach that can address several pressing issues facing urban planners when seeking to prioritize any

-p
type of GI development or investment, for example to understand where investments in NBS are most
re
effective and to understand what type of design goals should be emphasized to maximize ES benefit to
lP

local residents. The multi-criteria screening framework encompasses both material and non-material ES

inputs and is powered by a modular BBN architecture that allows for easy modification and updating.
na

Further, by integrating both quantitative and qualitative inputs, the framework offers results that are

transparent, scientifically robust, and immediately relevant to stakeholder concerns.

ur
Jo

ACKNOWLEDGEMENTS

This research was funded through Horizon 2020 Program of the European Commission project

'NATURVATION' (code 730243) and the 2015-2016 BiodivERsA COFUND call for research proposals,

with the national funders the Swedish Research Council for Environment, Agricultural Sciences, and

Spatial Planning; Swedish Environmental Protection Agency; German Aerospace Center (DLR); National

Science Centre (Poland); the Research Council of Norway; and the Spanish Ministry of Science,

Innovation and Universities. This research was also supported through the Maria de Maetzu Unit of
 Journal Pre-proof

Excellence grant (MDM-2015-0552) at ICTA-UAB. TM’s participation was additionally supported by the

U.S. National Science Foundation through the Urban Resilience to Extreme Weather-Related Events

Sustainability Research Network (URExSRN; NSF grant no. SES 1444755). The manuscript is based on

a study conducted as an MSc thesis by DW, which was funded through an Erasmus Mundus scholarship

by the European Commission contract number JEMES CiSu UAB 2016/No.1. Thanks to all collaborators

namely Cynthia Echave (and the Agencia de Ecologia Urbá), Nicholas Martin, Zofie Cimburova, Alaitz

Zabala Torres, Martin Karlsen, Luís Campos, Marta Conde, Isabelle Anguelovski and to all the

of
participants who attended the workshop as well as the Barcelona Municipal Institute for People with

Disabilities (IMPD) for hosting the event. We would finally like to thank two unknown reviewers and the

ro
editor for their valuable remarks and deep engagement with the manuscript.

-p
re
lP
na
ur
Jo
 Journal Pre-proof

FIGURES

of
ro
-p
re
lP
na

Fig. 1. Barcelona Rooftops seen from Parc del Turó del Putget.

Picture by Börkur Sigurbjörnsson derived from Flickr.com under CC BY 2.0.

ur
Jo
 Journal Pre-proof

Runoff Control
30%
25%
20%
Social Cohesion 15% Food Production
10%
5%
0%

of
Recreation &
Habitat & Pollination
Relaxation
Group 1

ro
Group 2
Group 3
Thermal Regulation

-p Average

Figure 2. Ecosystem Service Prioritization Weights. Based on a group evaluation (Pebble-distribution method)
re
embedded within an expert workshop (n=31), conducted in Barcelona (Spain), 5th June 2018.
lP
na
ur
Jo
 Journal Pre-proof

of
ro
-p
re
Fig. 3. Demand for Thermal Regulation. (a) Urban heat island (UHI) effect (Martin-Vide et al., 2015), (b)
lP

demographic vulnerability to heat (BCN, 2018a); STD - Standard deviation.

na
ur
Jo
 Journal Pre-proof

of
ro
-p
re
Fig. 4. Demand for Runoff Control. Land-use based runoff coefficients (LCMC, 2009; Puccinelli et al., 2012).
lP
na
ur
Jo
 Journal Pre-proof

of
ro
-p
re
Fig. 5. Demand for Pollinator Habitat. (a) Floral availability and (b) nesting suitability for pollinators (ESTIMAP,
lP

LCMC, 2009; Zulian et al., 2013).

na
ur
Jo
 Journal Pre-proof

of
ro
-p
re
lP

Fig. 6. Demand for Food Production. (a) Neighborhood grocery count, (b) walking distance to urban gardens.
na
ur
Jo
 Journal Pre-proof

of
ro
-p
re
Fig. 7. Demand for Recreational Opportunities. (a) Walking distance to sport facilities, (b) relative frequency of
lP

proximity walks (Marquet and Miralles-Guasch, 2015).

na
ur
Jo
 Journal Pre-proof

of
ro
-p
re
Fig. 8. Demand for Social Cohesion. (a) Relative income difference assessed by Available Family Income (RFD),
lP

(b) ethnic heterogeneity assessed by entropy of nationalities. STD - Standard deviation.

na
ur
Jo
 Journal Pre-proof

of
ro
-p
re
lP
na
ur
Jo

Fig. 9. Demand Model Output. The aggregated demand across Barcelona’s ten districts for six ecosystem services

(ES) calculated using fourteen spatial indicators in HUGIN Researcher and QGIS. ES demand is indexed between 0

and 1, and each district’s portion of total city demand is displayed along with the percent difference of this value

from what would be expected based on district area.

 Journal Pre-proof

of
ro
-p
re
lP
na
ur
Jo

Fig. 10. Screening Tool Output. The maximum estimated ES provision of all rooftops in Barcelona if GRs were

implemented, based on ES demand and expert weighting of five design alternatives. In each raster cell, the design

alternative with the maximum ES provision was selected for mapping.

 Journal Pre-proof

TABLES

Substrate
Weight Cost
Alternative Depth Maintenance Vegetation
(kg/m2) (€/m2)
(cm)

Succulents, perennial herbs, grasses,

Extensive 10-15+ 120-225 70-90 Low
ornamentals, underground perennials

Grasses, herbs, aromatics, bulbs,

Semi-Intensive 15-30+ 150-450 90-130 Med
creepers, small shrubs

of
Above, with medium to large shrubs,

ro
Intensive 30-100+ 650+ 150+ High small to large conifers, palms, or other

-p trees

Predominantly indigenous species,

re
Naturalized 15-30+ 200-450 70-130 Low
wildflowers, and shrubs
lP

Garden vegetables, aromatic and

Allotment 30-40+ 450+ 120+ High
medicinal plants, fruit trees
na

Table 1. GR design alternatives (Contreras and Castillo, 2015)

ur
Jo
 Journal Pre-proof

Group 1 Group 2 Group 3 Average

Ecosystem Service n=8 n=7 n=9 n=31

Thermal Regulation 26.67% 23.33% 23.33% 24.44%

Runoff Control 13.33% 10.00% 13.33% 12.22%

Habitat & Pollination 23.33% 13.33% 20.00% 18.89%

Food Production 6.67% 10.00% 6.67% 7.78%

Recreation & Relaxation 10.00% 20.00% 16.67% 15.56%

of
Social Cohesion 20.00% 23.33% 20.00% 21.11%

ro
Table 2. Expert prioritization of Ecosystem Service Needs. Based on a group evaluation (Pebble-distribution

method) embedded within an expert workshop (n=31), conducted in Barcelona (Spain), 5th June 2018.

-p
re
lP
na
ur
Jo
 Journal Pre-proof

Green Roof Design Alternative

Extensive Semi-Intensive Intensive Naturalized Allotment

Ecosystem Service Average Score (Group 1|2|3)

0.50 0.83 1.00 0.83 0.33

Thermal Regulation
(0|1|0.5) (0.5|1|1) (1|1|1) (1|0.5|1) (0.5|0|0.5)

0.33 0.75 0.75 0.67 0.67

Runoff Control
(0|0.5|0.5) (1|0.5|0.75) (1|0.5|0.75) (0.5|0.5|1) (0.5|0.5|1)

of
0.42 0.50 0.67 1.00 0.58
Habitat & Pollination

ro
(0|0.75|0.5) (0|0.5|1) (0.5|0.5|1) (1|1|1) (0.5|0.25|1)

0.00 0.17 0.67 0.33 1.00

Food Production
(0|0|0)
-p
(0|0.5|0) (0.5|1|0.5) (0.5|0.5|0) (1|1|1)
re
0.33 0.83 1.00 0.67 1.00
Recreation & Relaxation
lP

(0|0.5|0.5) (0.5|1|1) (1|1|1) (0.5|1|0.5) (1|1|1)

0.17 0.83 1.00 0.50 1.00

na

Social Cohesion
(0|0.5|0) (0.5|1|1) (1|1|1) (0.5|0.5|0.5) (1|1|1)
ur

Total Contribution (Max

1.75 3.92 5.08 4.00 4.58
6)
Jo

Table 3. Estimated Ecosystem Service Provision by Different Green Roof Design Alternatives. Based on a

group evaluation (Likert-scale) embedded within an expert workshop (n=31), conducted in Barcelona (Spain), 5th

June 2018.
 Journal Pre-proof

Green Roof Design Alternative

Extensive Semi-Intensive Intensive Naturalized Allotment

Feasibility Score No. of Votes (Percent)

2 0 4 2 0
Very Low = 1
(8.33%) (0.00%) (16.67%) (8.33%) (0.00%)

1 7 6 2 2

of
Low = 2
(4.17%) (29.17%) (25.00%) (8.33%) (8.33%)

ro
5 8 6 10 11
Medium = 3
(20.83%)

12
(33.33%)

7
-p (25.00%)

5
(41.67%)

10
(45.83%)

9
re
High = 4
(50.00%) (29.17%) (20.83%) (41.67%) (37.50%)
lP

4 2 3 0 2
Very High = 5
(16.67%) (8.33%) (12.50%) (0.00%) (8.33%)
na

Aggregate
3.63 3.17 2.88 3.17 3.46
ur

(Max 5)
Jo

Normalized (Min-Max) 1.00 0.39 0.00 0.39 0.78

Table 4. Feasibility of Different Green Roof Design Alternatives Considering Economic, Structural, and

Institutional Barriers. Based on a individual evaluations (Likert-scale ranking) embedded within an expert

workshop (n=31), conducted in Barcelona (Spain), 5th June.

 Journal Pre-proof

REFERENCES

1. Abualfaraj, N., Cataldo, J., Elborolosy, Y., Fagan, D., Woerdeman, S., Carson, T., & Montalto, F. (2018).

Monitoring and Modeling the Long-Term Rainfall-Runoff Response of the Jacob K. Javits Center Green

Roof. Water, 10(11), 1494.

2. Agència d’Ecologia Urbana de Barcelona (BCNecologia), 2010. Cobertes i Murs Verds a Barcelona: Estudi

sobre les existents, el potencial i les estratègies d'implantació. pp.

3. Agència d’Ecologia Urbana de Barcelona (BCNecologia), 2014. Treball de camp (Maig 2014).

4. Andersen, S.K., Olesen, K.G., Jensen, F.V., Jensen, F., 1989. HUGIN-A Shell for Building Bayesian Belief

of
Universes for Expert Systems, IJCAI, pp. 1080-1085.

ro
5. Anguelovski, I., Connolly, J.J.T., Masip, L., Pearsall, H., 2017. Assessing green gentrification in

-p
historically disenfranchised neighborhoods: a longitudinal and spatial analysis of Barcelona. Urban

Geography 39, 458-491.

re
6. Aubrecht, C., Ozceylan, D., 2013. Identification of heat risk patterns in the U.S. National Capital Region by
lP

integrating heat stress and related vulnerability. Environment international 56, 65-77.

7. Bancroft, C., Joshi, S., Rundle, A., Hutson, M., Chong, C., Weiss, C.C., Genkinger, J., Neckerman, K.,
na

Lovasi, G., 2015. Association of proximity and density of parks and objectively measured physical activity

in the United States: A systematic review. Social science & medicine 138, 22-30.
ur

8. Barcelona Regional, 2015. Primera Identificació De La Infraestructura Verda De Barcelona A Partir De

L’índex NDVI.
Jo

9. Baró, F., Haase, D., Gómez-Baggethun, E., Frantzeskaki, N., 2015. Mismatches between ecosystem

services supply and demand in urban areas: A quantitative assessment in five European cities. Ecol. Indic.

55, 146–158. doi:10.1016/j.ecolind.2015.03.013

10. Barton, D.N., Kuikka, S., Varis, O., Uusitalo, L., Henriksen, H.J., Borsuk, M., de la Hera, A., Farmani, R.,

Johnson, S., Linnell, J.D., 2012. Bayesian networks in environmental and resource management. Integrated

environmental assessment and management 8, 418-429.

11. Berardi, U., GhaffarianHoseini, A., GhaffarianHoseini, A., 2014. State-of-the-art analysis of the

environmental benefits of green roofs. Applied Energy 115, 411-428.

 Journal Pre-proof

12. BCN, 2012. Ajuntament de Barcelona. Departament del Pla de la Ciutat Plànol temàtic d’edificis segons

alçades per districte en format CAD.

13. BCN, 2016a. Ajuntament de Barcelona. Departament d'Estadística, Distribución territorial de la renta

familiar en Barcelona.

14. BCN, 2016b. Ajuntament de Barcelona. Direcció de Comerç. Inventory of premises of the city of

Barcelona. Barcelona.

15. BCN, 2017a. Ajuntament de Barcelona. Departament d'Estadística. Nacionalidad de la población por

secciones censales.

of
16. BCN, 2017b. Ajuntament de Barcelona. Àrea d'Ecologia Urbanisme i Mobilitat Programa d'impuls de la

ro
infraestructura verda urbana : mesura de govern.

17. BCN, 2018a. Ajuntament de Barcelona. Àrea d'Ecologia Urbanisme i Mobilitat. Resum Executiu: onades

de calor. Barcelona, 10.

-p
re
18. BCN, 2018b. Ajuntament de Barcelona. El Mapa Barcelona + Sostenible.

19. Botzat, A., Fischer, L.K., Kowarik, I., 2016. Unexploited opportunities in understanding liveable and
lP

biodiverse cities. A review on urban biodiversity perception and valuation. Global Environmental Change

39, 220-233.
na

20. Buehler, D., & Junge, R. (2016). Global trends and current status of commercial urban rooftop farming.

Sustainability, 8(11), 1108.

ur

21. Camps-Calvet, M., Langemeyer, J., Calvet-Mir, L., Gómez-Baggethun, E., 2016. Ecosystem services
Jo

provided by urban gardens in Barcelona, Spain: Insights for policy and planning. Environmental Science &

Policy 62, 14-23.

22. Carter, T., Fowler, L., 2008. Establishing green roof infrastructure through environmental policy

instruments. Environmental management 42, 151-164.

23. Caspi, C.E., Sorensen, G., Subramanian, S.V., Kawachi, I., 2012. The local food environment and diet: a

systematic review. Health & place 18, 1172-1187.

24. Chan, J., To, H.-P., Chan, E., 2006. Reconsidering Social Cohesion: Developing a Definition and

Analytical Framework for Empirical Research. Social Indicators Research 75, 273-302.
 Journal Pre-proof

25. Chen, S.H., Pollino, C.A., 2012. Good practice in Bayesian network modelling. Environmental Modelling

& Software 37, 134-145.

26. Contreras, E., Castillo, I. Àrea d'Ecologia Urbanisme i Mobilitat, 2015. Guia de terrats vius i cobertes

verdes. Barcelona, 74.

27. Crossman, N.D., Burkhard, B., Nedkov, S., Willemen, L., Petz, K., Palomo, I., Drakou, E.G., Martín-

Lopez, B., McPhearson, T., Boyanova, K., Alkemade, R., Egoh, B., Dunbar, M.B., Maes, J., 2013. A

blueprint for mapping and modelling ecosystem services. Ecosystem Services 4, 4-14.

28. Czemiel Berndtsson, J., 2010. Green roof performance towards management of runoff water quantity and

of
quality: A review. Ecological Engineering 36, 351-360.

ro
29. Dadvand, P., Bartoll, X., Basagana, X., Dalmau-Bueno, A., Martinez, D., Ambros, A., Cirach, M.,

Triguero-Mas, M., Gascon, M., Borrell, C., Nieuwenhuijsen, M.J., 2016. Green spaces and General Health:

-p
Roles of mental health status, social support, and physical activity. Environment international 91, 161-167.
re
30. Dvorak, B., Volder, A., 2010. Green roof vegetation for North American ecoregions: A literature review.

Landscape and Urban Planning 96, 197-213.

lP

31. Easterly, W., Ritzen, J., Woolcock, M., 2006. Social cohesion, institutions, and growth. Economics &

Politics 18, 103-120.

na

32. European Commission (EC), 2015. Towards an EU Research and Innovation policy agenda for nature-

based solutions & re-naturing cities. Final report of the Horizon 2020.
ur

33. European Environment Agency (EEA), 2017. Updated CLC illustrated nomenclature guidelines, Austria, p.
Jo

124.

34. Forschungsgesellschaft Landschaftsentwicklung Landschaftsbau (FLL), 2002. Guideline for the Planning,

Execution and Upkeep of Green-Roof Sites: Roof-Greening Guideline. Bonn, p. 95.

35. Giles-Corti, B., Donovan, R.J., 2002. The relative influence of individual, social and physical environment

determinants of physical activity. Social science & medicine 54, 1793-1812.

36. Gómez-Baggethun, E., Gren, Å., Barton, D.N., Langemeyer, J., McPhearson, T., O’Farrell, P., Andersson,

E., Hamstead, Z., Kremer, P., 2013. Urban Ecosystem Services, in: Elmqvist, T., Fragkias, M., Goodness,

J., Güneralp, B., Marcotullio, P.J., McDonald, R.I., Parnell, S., Schewenius, M., Sendstad, M., Seto, K.C.,
 Journal Pre-proof

Wilkinson, C. (Eds.), Urbanization, Biodiversity and Ecosystem Services: Challenges and Opportunities: A

Global Assessment. Springer Netherlands, Dordrecht, pp. 175-251.

37. Grunwald, L., Heusinger, J., Weber, S., 2017. A GIS-based mapping methodology of urban green roof

ecosystem services applied to a Central European city. Urban Forestry & Urban Greening 22, 54-63.

38. Harlan, S.L., Brazel, A.J., Prashad, L., Stefanov, W.L., Larsen, L., 2006. Neighborhood microclimates and

vulnerability to heat stress. Social science & medicine 63, 2847-2863.

39. Holliday, K.M., Howard, A.G., Emch, M., Rodríguez, D.A., Rosamond, W.D., Evenson, K.R., 2017.

Where Are Adults Active? An Examination of Physical Activity Locations Using GPS in Five US Cities.

of
Journal of Urban Health 94, 459-469.

ro
40. House, R.J., Hanges, P.J., Javidan, M., Dorfman, P.W., Gupta, V., 2004. Culture, leadership, and

organizations: The GLOBE study of 62 societies. Sage publications.

-p
41. Iceland, J., 2004. The multigroup entropy index (also known as Theil’s H or the information theory index).
re
US Census Bureau. Retrieved July 31, 2006.

42. Institut d'Estadística de Catalunya (IDESCAT), 2016. Població de Catalunya georeferenciada a 1 de gener
lP

de 2016.

43. Jensen, F. V, 2001. Bayesian networks and decision graphs. Springer Verlag, New York.
na

44. Kabisch, N., Korn, H., Stadler, J., & Bonn, A. (2017). Nature-Based Solutions to Climate Change

Adaptation in Urban Areas. Theory and Practice of Urban Sustainability Transitions.

ur

45. Karteris, M., Theodoridou, I., Mallinis, G., Tsiros, E., Karteris, A., 2016. Towards a green sustainable
Jo

strategy for Mediterranean cities: Assessing the benefits of large-scale green roofs implementation in

Thessaloniki, Northern Greece, using environmental modelling, GIS and very high spatial resolution

remote sensing data. Renewable and Sustainable Energy Reviews 58, 510-525.

46. Kawachi, I., Kennedy, B.P., Lochner, K., Prothrow-Stith, D., 1997. Social capital, income inequality, and

mortality. American journal of public health 87, 1491-1498.

47. Kaźmierczak, A., 2013. The contribution of local parks to neighbourhood social ties. Landscape and Urban

Planning 109, 31-44.

 Journal Pre-proof

48. Keeler, Bonnie L., Perrine Hamel, Timon McPhearson, Maike H. Hamann, Marie L. Donahue, Kelly A.

Meza Prado, Katie K. Arkema et al. Social-ecological and technological factors moderate the value of

urban nature. Nature Sustainability 2, no. 1 (2019): 29.

49. Kjærulff, U.B., Madsen, A.L., 2013. Bayesian Networks and Influence Diagrams: a Guide to Construction

and Analysis. 2nd Ed. Springer, New York, USA. Springer.

50. Kremer, P., Hamstead, Z.A., McPhearson, T., 2016. The value of urban ecosystem services in New York

City: A spatially explicit multicriteria analysis of landscape scale valuation scenarios. Environmental

Science & Policy 62, 57-68.

of
51. Lagonigro, R., Oller, R., Martori, J.C., 2017. A quadtree approach based on European geographic grids:

ro
reconciling data privacy and accuracy. SORT : Statistics and Operations Research Transactions 41, 139-

158.

-p
52. Land Cover Map of Catalonia (LCMC), 2009. Land Cover Map of Catalonia, 4th Edition ed. CREAF and
re
Generalitat de Catalunya.

53. Landuyt, D., Broekx, S., D'Hondt, R., Engelen, G., Aertsens, J., Goethals, P.L.M., 2013. A review of
lP

Bayesian belief networks in ecosystem service modelling. Environmental Modelling & Software 46, 1-11.

54. Landuyt, D., Van der Biest, K., Broekx, S., Staes, J., Meire, P., Goethals, P.L.M., 2015. A GIS plug-in for
na

Bayesian belief networks: Towards a transparent software framework to assess and visualise uncertainties

in ecosystem service mapping. Environmental Modelling & Software 71, 30-38.

ur

55. Langemeyer, J., Camps-Calvet, M., Calvet-Mir, L., Barthel, S., Gómez-Baggethun, E., 2018. Stewardship
Jo

of urban ecosystem services: understanding the value(s) of urban gardens in Barcelona. Landscape and

Urban Planning 170, 79-89.

56. Langemeyer, J., Gómez-Baggethun, E., Haase, D., Scheuer, S., Elmqvist, T., 2016. Bridging the gap

between ecosystem service assessments and land-use planning through Multi-Criteria Decision Analysis

(MCDA). Environmental Science & Policy 62, 45-56.

57. Laurence, J., 2009. The Effect of Ethnic Diversity and Community Disadvantage on Social Cohesion: A

Multi-Level Analysis of Social Capital and Interethnic Relations in UK Communities. European

Sociological Review 27, 70-89.

 Journal Pre-proof

58. Lee, K., 2014. The role of green roof views in attention restoration and work performance. University of

Melbourne, Melbourne, Australia.

59. Lefebvre-Ropars, G., Morency, C., Singleton, P.A., Clifton, K.J., 2017. Spatial transferability assessment

of a composite walkability index: The Pedestrian Index of the Environment (PIE). Transportation Research

Part D: Transport and Environment 57, 378-391.

60. Li, D., Bou-Zeid, E., Oppenheimer, M., 2014. The effectiveness of cool and green roofs as urban heat

island mitigation strategies. Environmental Research Letters 9, 055002.

61. Loder, A., 2014. ‘There's a meadow outside my workplace’: A phenomenological exploration of aesthetics

of
and green roofs in Chicago and Toronto. Landscape and Urban Planning 126, 94-106.

ro
62. Lundholm, J.T., Williams, N.S.G., 2015. Effects of Vegetation on Green Roof Ecosystem Services. 223,

211-232.

-p
63. Maas, J., van Dillen, S.M., Verheij, R.A., Groenewegen, P.P., 2009. Social contacts as a possible
re
mechanism behind the relation between green space and health. Health & place 15, 586-595.

64. MacIvor, J.S., Lundholm, J., 2010. Insect species composition and diversity on intensive green roofs and
lP

adjacent level-ground habitats. Urban Ecosystems 14, 225-241.

65. Mahdiyar, A., Tabatabaee, S., Abdullah, A., Marto, A., 2018. Identifying and assessing the critical criteria
na

affecting decision-making for green roof type selection. Sustainable Cities and Society 39, 772-783.

66. Markevych, I., Schoierer, J., Hartig, T., Chudnovsky, A., Hystad, P., Dzhambov, A.M., de Vries, S.,
ur

Triguero-Mas, M., Brauer, M., Nieuwenhuijsen, M.J., Lupp, G., Richardson, E.A., Astell-Burt, T.,
Jo

Dimitrova, D., Feng, X., Sadeh, M., Standl, M., Heinrich, J., Fuertes, E., 2017. Exploring pathways linking

greenspace to health: Theoretical and methodological guidance. Environmental research 158, 301-317.

67. Marcot, B.G., Steventon, J.D., Sutherland, G.D., McCann, R.K., 2006. Guidelines for developing and

updating Bayesian belief networks applied to ecological modeling and conservation. Can. J. For. Res. 36,

3063–3074.

68. Marcot, B.G., 2012. Metrics for evaluating performance and uncertainty of Bayesian network models. Ecol.

Modell. 230, 50– 62.

69. Marquet, O., Miralles-Guasch, C., 2015. The Walkable city and the importance of the proximity

environments for Barcelona’s everyday mobility. Cities 42, 258-266.

 Journal Pre-proof

70. Martin-Vide, J., Artola, V.M., Cordobilla, J., Moreno, C., 2015. La Isla de Calor en El Área Metropolitana

de Barcelona y La Adaptación al Cambio Climático, in: Direcció de Serveis Ambientals de l’AMB (Ed.),

METROBS 2015. Barcelona Regional, Barcelona, p. 101.

71. McCormack, G.R., Shiell, A., 2011. In search of causality: a systematic review of the relationship between

the built environment and physical activity among adults. International Journal of Behavioral Nutrition and

Physical Activity 8, 125.

72. Meenar, M., Hoover, B., 2012. Community Food Security via Urban Agriculture: Understanding People,

Place, Economy, and Accessibility from a Food Justice Perspective. Journal of Agriculture, Food Systems,

of
and Community Development, 143-160.

ro
73. Mensah, Y., Chen, H.-Y., 2013. Global clustering of countries by culture–an extension of the GLOBE

study.

-p
74. Mesimäki, M., Hauru, K., Kotze, D.J., Lehvävirta, S., 2017. Neo-spaces for urban livability? Urbanites’
re
versatile mental images of green roofs in the Helsinki metropolitan area, Finland. Land Use Policy 61, 587-

600.
lP

75. Moreno‐ Garcia, M.C., 1994. Intensity and form of the urban heat island in Barcelona. International

Journal of Climatology 14, 705-710.

na

76. Nadal, A., Alamus, R., Pipia, L., Ruiz, A., Corbera, J., Cuerva, E., Rieradevall, J., Josa, A., 2017. Urban

planning and agriculture. Methodology for assessing rooftop greenhouse potential of non-residential areas
ur

using airborne sensors. The Science of the total environment 601-602, 493-507.
Jo

77. Nielsen, T.D., Jensen, F.V., 2009. Bayesian networks and decision graphs. Springer Science & Business

Media.

78. Normas Tecnológicas de Jardinería (NTJ), 2012. NTJ 11C: Cubiertas Verdes, Barcelona, p. 90.

79. Nurmi, V., Votsis, A., Perrels, A., Lehvävirta, S., 2016. Green Roof Cost-Benefit Analysis: Special

Emphasis on Scenic Benefits. J. Benefit-Cost Anal. 7, 488–522. doi:10.1017/bca.2016.18

80. Oberndorfer, E., Lundholm, J., Bass, B., Coffman, R.R., Doshi, H., Dunnett, N., Gaffin, S., Köhler, M.,

Liu, K.K.Y., Rowe, B., 2007. Green Roofs as Urban Ecosystems: Ecological Structures, Functions, and

Services. BioScience 57, 823-833.

 Journal Pre-proof

81. OpenStreetMap contributors (OSM), 2018. Retrieved from:

http://download.geofabrik.de/europe/spain.html. Accessed 2018.06.16. .

82. Orsini, F., Gasperi, D., Marchetti, L., Piovene, C., Draghetti, S., Ramazzotti, S., Bazzocchi, G., Gianquinto,

G., 2014. Exploring the production capacity of rooftop gardens (RTGs) in urban agriculture: the potential

impact on food and nutrition security, biodiversity and other ecosystem services in the city of Bologna.

Food Security 6, 781-792.

83. Puccinelli, C., Marcheggiani, S., Munafò, M., Andreani, P., Mancini, L., 2012. Evaluation of Aquatic

Ecosystem Health Using the Potential Non Point Pollution Index (PNPI) Tool, Diversity of Ecosystems.

of
InTech.

ro
84. Rowe, D.B., 2011. Green roofs as a means of pollution abatement. Environmental pollution 159, 2100-

2110.

-p
85. Santamouris, M., 2014. Cooling the cities – A review of reflective and green roof mitigation technologies
re
to fight heat island and improve comfort in urban environments. Solar Energy 103, 682-703.

86. Sanyé-Mengual, E., Cerón-Palma, I., Oliver-Solà, J., Montero, J.I., Rieradevall, J., 2015. Integrating
lP

Horticulture into Cities: A Guide for Assessing the Implementation Potential of Rooftop Greenhouses

(RTGs) in Industrial and Logistics Parks. Journal of Urban Technology 22, 87-111.
na

87. Shafique, Muhammad, Reeho Kim, and Muhammad Rafiq. "Green roof benefits, opportunities and

challenges–A review." Renewable and Sustainable Energy Reviews 90 (2018): 757-773.

ur

88. Snodgrass, E.C., Snodgrass, L.L., 2006. Green roof plants: a resource and planting guide. Timber Press
Jo

Portland.

89. Spala, A., Bagiorgas, H.S., Assimakopoulos, M.N., Kalavrouziotis, J., Matthopoulos, D., Mihalakakou, G.,

2008. On the green roof system. Selection, state of the art and energy potential investigation of a system

installed in an office building in Athens, Greece. Renewable Energy 33, 173-177.

90. Sriwongsitanon, N., Taesombat, W., 2011. Effects of land cover on runoff coefficient. Journal of

Hydrology 410, 226-238.

91. Stange, E., Zulian, G., Rusch, G., Barton, D., Nowell, M., 2017. Ecosystem services mapping for municipal

policy: ESTIMAP and zoning for urban beekeeping. One Ecosystem 2, e14014.
 Journal Pre-proof

92. Sugiyama, T., Cerin, E., Owen, N., Oyeyemi, A.L., Conway, T.L., Van Dyck, D., Schipperijn, J.,

Macfarlane, D.J., Salvo, D., Reis, R.S., Mitas, J., Sarmiento, O.L., Davey, R., Schofield, G., Orzanco-

Garralda, R., Sallis, J.F., 2014. Perceived neighbourhood environmental attributes associated with adults

recreational walking: IPEN Adult study in 12 countries. Health & place 28, 22-30.

93. Sutton, R. K., and John Lambrinos. Green roof ecosystems: Summary and synthesis. In Green Roof

Ecosystems. Sutton, R. K. (ed.). Springer, Cham, 2015. 423-440.

94. Triguero-Mas, M., Dadvand, P., Cirach, M., Martinez, D., Medina, A., Mompart, A., Basagana, X.,

Grazuleviciene, R., Nieuwenhuijsen, M.J., 2015. Natural outdoor environments and mental and physical

of
health: relationships and mechanisms. Environment international 77, 35-41.

ro
95. Velasco, M., Krersting, T., Russo, B., Sunyer, D., 2013. Report of the Urban risk management plan for

Barcelona. PREPARED, pp.

-p
96. Velázquez, J., Anza, P., Gutiérrez, J., Sánchez, B., Hernando, A., García-Abril, A., 2018. Planning and
re
selection of green roofs in large urban areas. Application to Madrid metropolitan area. Urban Forestry &

Urban Greening.
lP

97. Walker, R.E., Keane, C.R., Burke, J.G., 2010. Disparities and access to healthy food in the United States: A

review of food deserts literature. Health & place 16, 876-884.

na

98. Whittinghill, L.J., Rowe, D.B., 2011. The role of green roof technology in urban agriculture. Renewable

Agriculture and Food Systems 27, 314-322.

ur

99. Williams, N.S.G., Rayner, J.P., Raynor, K.J., 2010. Green roofs for a wide brown land: Opportunities and
Jo

barriers for rooftop greening in Australia. Urban Forestry & Urban Greening 9, 245-251.

100. Zulian, G., Maes, J., Paracchini, M., 2013. Linking Land Cover Data and Crop Yields for Mapping and

Assessment of Pollination Services in Europe. Land 2, 472-492.

 Journal Pre-proof

Graphical abstract

of
ro
-p
re
lP
na
ur
Jo
 Journal Pre-proof

Highlights

1. The article addresses the question of where to build green roofs most effectively with
regard to citizen needs
2. A spatial multi-criteria screening tool for the creation of green roofs is developed
3. Ecosystem service deficits are spatially defined by combined social-ecological evaluation
criteria
4. Finally, the optimal green roof design for an effective ecosystem service provision is
determined

of
ro
-p
re
lP
na
ur
Jo