Landscape and U ban Planning 245 (2024) 105009
A ailable online 3 Feb ua y 2024
0169-2046/© 2024 The Au ho (s). Published by Else ie B.V. This is an open access a icle unde he CC BY license (h p://c ea i ecommons.o g/licenses/by/4.0/).
How well do NDVI and OpenS ee Map da a cap u e people’s isual
pe cep ions o u ban g eenspace?
Roos Teeuwen
*
, Vasileios Milias, Alessand o Bozzon, Achilleas Psyllidis
Facul y o Indus ial Design Enginee ing, Del Uni e si y o Technology, Landbe gs aa 15, 2628CE Del , he Ne he lands
HIGHLIGHTS
•We compa ed NDVI and OpenS ee Map da a o people’s isual pe cep ions o g eenness.
•NDVI and OpenS ee Map da a o en di e ge om human pe cep ions o g eenness.
•OpenS ee Map cap u es g eenness bes in sho dis ance, NDVI bes in longe dis ance.
•Vege a ion con igu a ion, a ie y, and na u al elemen s enhance pe cei ed g eenness.
•Vege a ed space domina ed by buil -en i onmen elemen s may no be pe cei ed as g een.
ARTICLE INFO
Keywo ds:
U ban g eenspace
Visual pe cep ion
OpenS ee Map
NDVI
C owdsou cing
ABSTRACT
The s udy o u ban g eenspaces ypically elies on h ee ypes o da a: people’s subjec i e pe cep ions collec ed
ia ques ionnai es, ege a ion indices de i ed om sa elli e image y, such as he No malized Di e ence Vege a ion
Index (NDVI), and Land Use o Land Co e maps, such as OpenS ee Map (OSM). Da a on people’s pe cep ions a e
essen ial when esea ching human ac i i ies, ye hey scale poo ly. NDVI and OSM da a, on he o he hand, a e
eely a ailable wo ldwide, hus aluable o assessing ci ies a scale o p io i izing loca ions o in e en ions.
Howe e , i is unclea how e ec i ely NDVI and OSM da a cap u e people’s isual pe cep ions o u ban
g eenspaces. In his wo k, we collec people’s isual pe cep ions o public spaces in h ee majo Eu opean ci ies
h ough c owdsou cing, quan i a i ely compa e hem o NDVI and OSM da a, and quali a i ely in es iga e
dispa i ies. We ound ha NDVI mode a ely co ela es wi h pe cei ed g eenness and ha no only OSM
g eenspaces bu also pocke pa ks and play spaces a e o en conside ed g een. Fu he mo e, we ound ha
people’s pe cep ions co espond bes o OSM da a in small adius dis ances and NDVI da a in la ge adius
dis ances and ha combining NDVI and OSM da a can imp o e iden i ica ion o places in OSM ha a e
commonly conside ed g een. Ou quali a i e analysis e ealed ha con igu a ion and a ie y o ege a ion, and
p esence o o he na u al o buil -up ea u es, in luence people’s pe cep ions o g eenspace. Wi h ou indings we
aim o help esea che s and p ac i ione s make mo e in o med decisions when collec ing g eenspace da a o
hei speci ic con ex , ul ima ely con ibu ing o g een u ban en i onmen s ha e lec people’s pe spec i es.
1. In oduc ion
U ban g eenspaces a e widely associa ed wi h posi i e e ec s on
human heal h and well-being. Depending on he discipline, pa hway,
and con ex , hey a e ypically examined using one o h ee ypes o da a
sou ces (Labib e al., 2020; Nieuwenhuijsen e al., 2017; Ma ke ych
e al., 2017; Zhang e al., 2021). Fi s , da a collec ed h ough la ge-scale
ques ionnai es ha e lec indi idual people’s pe cep ions o g eenspace,
o example o hei esiden ial neighbo hood, a e ypical o
en i onmen al psychology esea ch (K uize e al., 2020; Zhang e al.,
2021). Second, ege a ion indices de i ed om sa elli e image y, such
as he No malized Di e ence Vege a ion Index (NDVI), a e commonly
employed in epidemiological s udies o s udy he abundance o ege-
a ion a ound people’s homes (La kin and Hys ad, 2019; Helbich e al.,
2019; Robe s and Helbich, 2021; Dad and e al., 2015). Finally, Land
Use / Land Co e (LULC) maps ha desc ibe he land su ace in dis inc
ca ego ies, such as OpenS ee Map (OSM), a e equen ly used in ci y
planning o policy assessmen o quan i y he a ailabili y, accessibili y,
* Co esponding au ho .
E-mail add esses: [email p o ec ed] (R. Teeuwen), [email p o ec ed] (V. Milias), [email p o ec ed] (A. Bozzon), [email p o ec ed] (A. Psyllidis).
Con en s lis s a ailable a ScienceDi ec
Landscape and U ban Planning
jou nal homepage: www.else ie .com/loca e/landu bplan
h ps://doi.o g/10.1016/j.landu bplan.2024.105009
Recei ed 15 Sep embe 2023; Recei ed in e ised o m 6 Decembe 2023; Accep ed 15 Janua y 2024
Landscape and U ban Planning 245 (2024) 105009
2
o size o o mal g eenspaces (La kin and Hys ad, 2019; Kabisch and
Haase, 2014; Wood e al., 2017; Zhang e al., 2021; Taubenb¨
ock e al.,
2021).
While en i onmen al pe cep ions om ques ionnai es a e essen ial
when in es iga ing human ac i i ies (Flowe s e al., 2016; Fonga e al.,
2019), such da a scale poo ly due o ime, cos s, geog aphical co e age,
and spa ial ex en cons ain s. NDVI and LULC maps such as OSM, on he
o he hand, a e eely a ailable wo ldwide and ou inely upda ed o e
ime, making hem aluable o assess he a ailabili y o accessibili y o
g eenspaces a scale, o o p io i ize loca ions o in e en ions (La kin
and Hys ad, 2019; Nieuwenhuijsen e al., 2017; Ma ke ych e al., 2017;
Labib e al., 2020). E en hough LULC maps o en only ep esen o mal
g eenspaces such as pa ks and u ban o es s, in o mal and small-scale
g eenspaces o g een s ee s ha e also been ound o bene i people
(Luo and Pa uano, 2023; Wolch e al., 2014; Rao e al., 2007; Ma ke ych
e al., 2017). Fu he mo e, while bo h NDVI and LULC maps a e
essen ial in g eenspace s udies, hey may no mi o subjec i e human
pe cep ions (Leslie e al., 2010; Ko hencz and Blaschke, 2017; Zhang
e al., 2021; Labib e al., 2020; Ma ke ych e al., 2017). Al hough ecen
wo ks ha e a emp ed o inco po a e he human pe spec i e in o la ge-
scale g eenspace s udies using s ee -le el image y (Rzo kiewicz e al.,
2018; La kin and Hys ad, 2019; Biljecki and I o, 2021; Lu, 2019), o he
bes o ou knowledge, hese ha e p ima ily employed compu e ision
echniques such as au oma ed objec de ec ion and scene ecogni ion,
which do no acknowledge he subjec i i y o human pe cep ion.
We aim o con ibu e o a be e unde s anding o how well la ge-
scale open da ase s, speci ically NDVI and OSM LULC maps, cap u e
people’s isual pe cep ions o u ban g eenspace. Ou goal is o e alua e
how well hese da ase s ma ch wi h each o he , and whe e hey de ia e,
and o explain such dispa i ies using he spa ial ea u es o in es iga ed
loca ions, o in o m he design o u u e g eenspace s udies.
To ha end, we ollow a wo-s ep app oach. Fi s , we collec la ge-
scale open-sou ce NDVI and OSM da a o h ee majo Eu opean ci -
ies. We apply a c owdsou cing app oach o ob ain people’s isual
g eenness pe cep ions o a ious so s o public spaces. Second, we
assess how well hese isual pe cep ions co espond o o di e ge om
he in o ma ion included in NDVI and OSM da a. We hypo hesize ha :
(H1) he e is a s ong posi i e co ela ion be ween NDVI alues and
pe cei ed g eenness, as hese a e bo h commonly used o quan i y
en i onmen al g eenness, (H2) pe cei ed g eenness is highe o
egula -size OSM g eenspaces han o pocke pa ks, play spaces, open
public spaces, and s ee s, as de ini ions o g eenspace a e o en limi ed
o g eenspaces la ge han a ce ain h eshold a ea, whe eas pocke
pa ks, play spaces, open public spaces, and s ee s can also be pe cei ed
as (in o mal) g eenspaces, as o he s udies sugges , and (H3) pe cei ed
g eenspaces a e be e e lec ed in da a when hey a e selec ed using a
combina ion o OSM ca ego ies and NDVI alues, a he han simply
OSM ca ego ies o jus NDVI alues. To es ou hypo heses, we employ
s a is ical analyses, ollowed by a quali a i e hema ic analysis o
disco e which spa ial quali ies explain di e ences be ween NDVI and
OSM da a and people’s isual pe cep ions.
In he emainde o his pape , we explo e g eenspace da a sou ces
used in ela ed wo k, de ail wha da a sou ces we collec and how we
analyze hem, p esen and discuss ou indings and hei implica ions,
and conclude wi h ou key conclusions and u u e lines o esea ch.
2. G eenspace da a sou ces
In his wo k, we adap he de ini ion o g eenspace by he WHO
Regional O ice o Eu ope (2017) o “u ban space cha ac e ized by
ege a ion o any kind”, including s ee ees and oadside ege a ion,
g een oo s and acades, g eenspace on p i a e g ounds, and pa ks,
playg ounds, o g eenways. We na ow ou ocus o g eenspaces ha a e
publicly accessible, he eby allowing people o engage in ou doo
ac i i ies.
To s udy g eenspaces, esea che s use da a o a ying ypes and
scales, depending on hypo heses and ou comes o in e es (Ma ke ych
e al., 2017). Examples include measu es o a ailabili y, accessibili y,
isibili y, and use o g eenspace (Labib e al., 2020; Ma ke ych e al.,
2017). La ge-scale da a a e essen ial o in o ming policy, measu ing
how well ci ies adhe e o such ules, and s udying he epidemiological
consequences o g eenspace (Ma ke ych e al., 2017; La kin and Hys ad,
2019; Kabisch and Haase, 2014).
O he s udies collec da a on people’s pe cep ions ega ding g een-
spaces, which a e c i ical when s udying people’s beha io in g een-
spaces (Ma ke ych e al., 2017), o ins ance h ough ques ionnai es
among esiden s o in e iews wi h pa k isi o s (Sunde all and Jans-
son, 2020; Talal and San elmann, 2021; Kabisch and Haase, 2014). In
his wo k, we ocus on people’s isual subjec i e pe cep ions, which we
de ine as pe cep ions gene a ed by isual s imuli, such as a pho o o a
place, and u he in luenced by he indi idual’s expe iences, p e e -
ences, emo ions, and con ex .
The ollowing sec ions go o e a ious da a sou ces and collec ion
me hods and discuss hei di e ences and simila i ies ha mo i a e ou
s udy.
2.1. Objec i e measu es o g eenspace using spa ial da a
Among all ege a ion indices de i ed om sa elli e image y, he
No malized Di e ence Vege a ion Index (NDVI) is he mos widely used
(Ma ke ych e al., 2017). NDVI is an objec i e emo e sensing index ha
cap u es ege a ion by calcula ing he di e ence be ween ed and nea -
in a ed ligh e lec ed by he land su ace. NDVI maps a e o en ob-
ained om Landsa o Sen inel sa elli e missions. Bo h o hese missions
p o ide open da a a egula in e als wo ldwide, wi h he Eu opean
Sen inel-2 mission p o iding da a a a high esolu ion o 10 m (Labib
e al., 2020; Ma ke ych e al., 2017). Al e na i es o NDVI include he
G een Ra io Vege a ion Index (GVRI) (S ipada e al., 2006), Soil-
Adjus ed Vege a ion Index (SAVI) (Hue e, 1988), and Enhanced Vege-
a ion Index (EVI) (Hue e e al., 2002). Indices such as NDVI a e
pa icula ly ele an o s udying he p esence o a ailabili y o g een-
space, o ins ance a ound people’s home loca ions o along he ou es
hey ake as cap u ed in GPS acks (Ma ke ych e al., 2017; Robinson
e al., 2018; Spo swood e al., 2021; Robe s and Helbich, 2021).
Land Use / Land Co e (LULC) maps ep esen he land su ace in
dis inc classes, such as buildings, oads, pa ks, and o es s, allowing o
s udy he size, shape, kind, accessibili y, o spa ial layou o designa ed
g eenspaces (Ma ke ych e al., 2017; Nieuwenhuijsen e al., 2017).
LULC maps a e commonly u ilized o g eenspace accessibili y s udies;
hey accoun o a la ge sha e o objec i e s udies on g eenspace o
human ac i i ies (Labib e al., 2020). OSM, in pa icula , is a ype o
LULC map ha is inc easingly being used in academic s udies as an
open-sou ce and global al e na i e o local comme cial o au ho i a i e
LULC da ase s (Ba ing on-Leigh and Milla d-Ball, 2017) and is an
e ec i e al e na i e o local da a in e ms o i s accu acy and p ecision
(Liao e al., 2021). Al e na i e LULC maps include, o example, he
U ban A las in Eu ope (used by Tu unen e al., (2023)) and local da a
egis ies (e.g., municipal canopy co e and s ee ee da a used by Ba ´
o
e al. (2021)).
Geo-loca ed s ee -le el image y is gaining impo ance o u ban an-
alyses, including s udies on u ban g eene y (Biljecki and I o, 2021; Labib
e al., 2020). Examples include measu ing he G een View Index in
images (Li e al., 2015; Lu, 2019), de ec ing ege a ion objec s h ough
compu e ision (Song e al., 2022; Chen and Biljecki, 2023), o me ging
s ee -le el image y wi h LULC da a (Zhang e al., 2021). Las ly, a ious
s udies make use o social media da a, such as he equency o Flick
pho os and Twee s pos ed pe loca ion (Hams ead e al., 2018), he
con en s o Twee s (Robe s, 2017), and he ca ego ies o objec s
de ec ed in Ins ag am pho os (Song e al., 2022).
R. Teeuwen e al.
Landscape and U ban Planning 245 (2024) 105009
3
2.2. Cap u ing subjec i e pe cep ions h ough in e iews, ques ionnai es,
and audi s
People’s subjec i e en i onmen al pe cep ions a e ypically ob ained
h ough in e iews o ques ionnai es (Nieuwenhuijsen e al., 2017;
Ma ke ych e al., 2017). Examples include in-si u in e iews wi h pa k
isi o s. Fo ins ance, Talal and San elmann (2021) conduc in e iews
wi h pa k isi o s o unde s and hei mo i a ions o isi ing, expe i-
ences, pe cep ions o accessibili y, and sugges ions o imp o emen s,
and Sunde all and Jansson (2020) conduc walking in e iews wi h
g eenspace use s o lea n abou hei desi ed use, con en , a mosphe e,
inclusi i y, and managemen o a g eenspace.
Ques ionnai es o en employ Like scales o ob ain quan i ied sub-
jec i e measu emen s, o ins ance asking esponden s o a e he
pe cei ed quali y and amoun o g eenspace in hei su oundings (Zij-
lema e al., 2018), he pe cei ed amoun o g eenness and how sa is ied
people a e wi h i s quali y, amoun , main enance, and sa e y (K uize
e al., 2020), o he pe cei ed quan i y and usage quali y o g eenspaces
nea hei homes (Zhang e al., 2021).
Al e na i ely, esea che s conduc audi s o measu e he quan i y
and quali y o g eenspace (Nieuwenhuijsen e al., 2017). A subse o
g eenspace s udies employs s ee -le el image y o elici pe cep ions o
o conduc audi s, such as assessing he exis ence o ea u es in g een-
spaces h ough s ee -le el image y (Rzo kiewicz e al., 2018). O he
examples include wo k by Du e al. (2021), who p o ided pa k isi o s
wi h pho os o pa k scenes o help hem ecall hei isi ing expe ience
while answe ing a ques ionnai e abou hei heal h and well-being, o
Van Vlie e al. (2021) who conduc ed a ideo-based choice expe imen
on pa k a ibu es such as ees, u ni u e, cleanliness, acili ies, and
biodi e si y.
Al hough subjec i e da a p o e impo an when s udying use o
g eenspace (Flowe s e al., 2016; Fonga e al., 2019), hei collec ion is
ypically cons ained by ime, money, and geog aphical ex en .
2.3. Cap u ing subjec i e pe cep ions h ough c owdsou cing campaigns
To add ess hese empo al, mone a y o geog aphic limi a ions, e-
sea che s collec people’s isual en i onmen al pe cep ions om many
people h ough c owdsou cing campaigns, ypically elici ed h ough
s ee -le el image y. C owdsou cing is a me hod o ec ui ing a g oup o
pa icipan s o execu e a ask online, i.e., ex-si u o emo ely, whe eas
s ee -le el image y allows o emo ely mimic a scale wha pedes ians
may obse e (La kin and Hys ad, 2019; Lu, 2019; Ma ke ych e al.,
2017). As such, s ee -le el image y-based c owdsou cing campaigns
enable esea che s o sou ce pe cep ions om a as and di e se numbe
o places and indi iduals wo ldwide in a ime- and labo -e icien
manne (Milias e al., 2023).
Examples include s udies ha collec pe cep ions by asking hei
pa icipan s in ques ionnai es o choose which loca ion hey p e e o o
a e places on a Like scale and hen in i ing hem o explain hei e-
sponses by selec ing op ions om a lis o inpu ing keywo ds. Examples
include asking people o choose he mos sa e, uppe -class, o unique-
looking place ou o wo places p esen ed in image y (Salesses e al.,
2013); o he mos happy, beau i ul, o quie place (Que cia e al., 2014);
le ing pa icipan s selec he leas and mos sa e o a ac i e looking
place ou o ou images (Candeia e al., 2017); and by asking people o
i ually na iga e ci y s ee s while a ing how sa e and a ac i e hey
pe cei e hei pa h in a ious places (Milias e al., 2023).
2.4. Di e ences and simila i ies be ween subjec i e and objec i e
g eenspace da a
Few s udies ha e in es iga ed he ex en o which la ge-scale spa ial
da a and people’s pe cep ions o g eenspaces ma ch. These s udies
sugges , howe e , ha consis ency is limi ed. Leslie e al. (2010)
disco e ed a lack o ag eemen wi h o e all pe cei ed g eenness and a
signi ican bu modes co ela ion only o g eenness expanse and no o
s ee g eenness, g een spo s acili ies, and g een ameni ies when
compa ing NDVI maps wi h people’s pe cep ions o hei esiden ial
su oundings cap u ed in ou g eenspace componen s. Zhang e al.
(2021) ound no co ela ion o people’s pe cei ed quan i y and usage
quali y o g eenspaces nea hei homes wi h canopy co e and a bes
e y weak co ela ion wi h pa k a ea, ege a ion co e , and G een View
Index. Ko hencz and Blaschke (2017) assessed pa k isi o s’ a ings o
g eenness, accessibili y, and unc ions o pa ks, and ound no co ela-
ions wi h NDVI o pa k a ea, while hey did ind a mode a e co ela ion
o people’s imp ession o g eenness wi h he pe cen age o ege a ed
su ace. Hyam (2017) disco e ed a co ela ion be ween he au ho ’s
a ing o pe cei ed na u alness, and na u al componen s in s ee - iew
image y de ec ed h ough compu e ision; And Helbich e al. (2019)
ound no co ela ion be ween NDVI and deep-lea ning-based me ics o
s ee - iew g eenness.
Ou s udy aims o add o ou unde s anding o he p e iously e-
po ed (lack o ) associa ions be ween la ge-scale g eenspace da a, such
as NDVI and LULC maps, and people’s isual pe cep ions o g eenspaces.
Tha is, we do no necessa ily p esume ha hese da a a e compa able,
bu a he seek o p o ide e idence on hei di e ences and simila i ies,
as well as in which ci cums ances subs an ial di e ences a ise. Th ee
ac o s dis inguish ou wo k. Fi s , we include in ou s udy a di e se
ange o public spaces ha di e in e ms o ype, geog aphical se ing,
and ege a ion le el. Second, we collec mul iple people’s pe cep ions
on he same loca ions. Thi d, we in es iga e po en ial causes o da ase
di e ences by quali a i ely analyzing he easons people gi e o hei
assessmen s and he spa ial cha ac e is ics o each loca ion o u he
s eng hen ou quan i a i e indings.
3. Me hods
We collec ed and analyzed g eenspace da a in h ee Eu opean ci ies:
Ba celona, Ro e dam, and Go henbu g. We used Py hon o collec NDVI
da a, and LULC da a om OSM, and we used a c owdsou cing app oach
wi h Google S ee View (GSV) image y o collec people’s isual pe -
cep ions o g eenspaces. We hen es ed ou hypo heses and conduc ed
addi ional explo a o y and sensi i i y analyses, and quali a i ely
in es iga ed wha spa ial cha ac e is ics explain de ia ions be ween
people’s pe cep ions o g eenspace and wha is cap u ed in he map
da a. Fig. 1 shows a summa y o ou s eps and Fig. 2 depic s he da a we
collec ed o each loca ion: median NDVI alues, OSM ca ego ies, and
people’s isual pe cep ions o g eenness. Links o eposi o ies con ain-
ing ou code and (pseudonymized) da a a e p o ided in he da a a ail-
abili y s a emen a he bo om o his a icle.
3.1. Th ee case-s udy ci ies
We selec ed h ee case-s udy ci ies in Eu ope: Go henbu g (Sweden),
Ro e dam (The Ne he lands), and Ba celona (Spain). OSM da a in
Eu ope is ound o be ela i ely comple e (Zhou e al., 2022). The
selec ed ci ies a e all majo ci ies in hei espec i e coun ies, wi h
Go henbu g and Ro e dam ha ing compa able popula ions o app ox.
583,000 (in 2021) and app ox. 592,000 (in 2022), espec i ely, while
Ba celona has a subs an ially la ge popula ion o o e 1,640,000 (in
2022) (Ajun amen de Ba celona, 2022; CBS, 2022; S a is ikmyndigh-
e en, 2021). All h ee ci ies ha e an impo an ha bo . By selec ing case-
s udy ci ies om No he n, Wes e n, and Sou he n Eu opean egions
(UN, 2019), di e en ege a ion zones (Roy e al., 2012), and di e se
co e age o g een land (Zhou e al., 2022), we accoun o a ying
en i onmen al quali ies. Ba celona is si ua ed be ween a seaside wi h
beaches and a o es ed moun ain ange inland, wi h a a ie y o pa ks,
including his o ic pa ks such as he Mon juïc hill and a chi ec An oni
Gaudí’s Pa k Güell, complemen ed by ees dis ibu ed along i s s ee s.
Go henbu g is s a egically placed a a i e ou le in o he sea, and i has
se e al g eenspaces wi hin i s bo de s, including pa ks such as he
R. Teeuwen e al.
Landscape and U ban Planning 245 (2024) 105009
4
cen ally loca ed Kungspa ken, na u e ese es such as ¨
Anggå dsbe gen,
and o he ypes o g eenspaces. Ro e dam is dis inguished by mode n
mo phology and a chi ec u e esul ing om he ci y’s econs uc ion
ollowing signi ican bombing du ing Wo ld Wa II. I has se e al well-
known pa ks such as he K alingse Bos o es and lake, and The Pa k
loca ed on he Meuse i e side. Bo h Ro e dam and Go henbu g ha e
empe a e ma i ime clima es, while Ba celona has a wa me Medi e -
anean clima e.
3.2. Collec ing OSM, NDVI, and GSV da a
As candida e loca ions o analysis, we iden i ied u ban public spaces
wi h ele an OSM ca ego ies, NDVI alues, and GSV image y a ailable.
We scoped o public spaces loca ed wi hin walking dis ance om he
u ban cen e s o hese case-s udy ci ies, based on he Eu opean Com-
mission’s Human Se lemen Laye models and guidelines (Schia ina
e al., 2022; Waddell and Ul a sson, 2003).
OSM da a: We collec ed public space and pedes ian s ee ne wo k
da a om OSM using he O e pass API and he Osmnx lib a y (Boeing,
2017). We collec ed a a ie y o public spaces, ep esen ed as polygons:
ege a ed spaces, ypically e e ed o as g eenspace; and o he spaces
ha may — depending on hei cha ac e — be pe cei ed as such ac-
co ding o he WHO de ini ion (WHO Regional O ice o Eu ope, 2017).
We excluded spaces ha a e inaccessible ia he pedes ian s ee
Iden i y u ban cen es
o 3 case s udy ci ies
Collec OSM da a Collec NDVI da a
Collec GSV me ada a
Iden i y loca ions wi h
- OSM ca ego y
- median NDVI
- GSV pano ama
S a i y loca ions
- OSM ca ego y
- NDVI quan ile
Sample loca ions
Rec ui pa icipan s
Collec pe cep ions
a ing
median NDVI
OSM ca ego y
eason
- a ing & eason
- OSM ca ego y
- median NDVI
Tes hypo heses
Pe o m quali a i e
analyses
OSM, NDVI & GSV da a
Candida e loca ions
Visual pe cep ions
Da a analysis
Rese geome y o GSV
poin & gene a e bu e
Check quali y
Pe o m explo a o y &
sensi i i y analyses
Fig. 1. O e iew o me hodological s eps.
R. Teeuwen e al.
Landscape and U ban Planning 245 (2024) 105009
5
ne wo k o ha a e smalle han 200 squa e me e s (i.e., he size o a
ypical ennis cou ). Fo ege a ed spaces, we me ged o e lapping o
adjacen spaces in o one, such as sh ubbe y adjacen o a o es , and
di e en ia ed be ween di e en sizes. As a esul , we ob ained OSM
polygons o i e OSM ca ego ies: egula -size g eenspaces (speci ically
pa ks, na u e ese es, o es s, woods, sc ubs, sh ubbe y, hea h,
meadows, g ass(lands) illage g eene y, and ells, a leas 0.5 ha in size
(Teeuwen, Psyllidis, & Bozzon, 2023; Ambien e I alia, 2003)); pock-
e -size g eenspaces (same ca ego ies, up o 0.5 ha in size (Wood e al.,
2017; Labib e al., 2020; Pescha d e al., 2014)); public open spaces
(speci ically squa es, pedes ian a eas, ma ke places, and common
g ounds); play spaces (speci ically playg ounds and public schoolya ds);
and s ee s accessible o pedes ians (i.e., o walking as de ined by
Boeing (2017)).
NDVI da a: We used Google’s Ea h Engine API o collec high-
esolu ion sa elli e ege a ion indices om he Cope nicus Sen inel-2
mission (Ma ke ych e al., 2017; Labib e al., 2020). We used all im-
age y be ween May and Sep embe 2021, i.e., he g owing season o
ege a ion in Eu ope (Robe s and Helbich, 2021), and calcula ed he
a e age NDVI alue pe as e cell. We hen calcula ed he a e age NDVI
alue pe OSM polygon, while igno ing alues less han 0 (i.e., wa e )
(Ma ke ych e al., 2017).
GSV me ada a: Using Google’s S ee View S a ic API, we looked o
he nea es GSV image y o up o 10 andom poin s wi hin each OSM
polygon, wi h a maximum sea ch adius o 15 m (Ami i and C ain,
2019). When we ound an image cap u ed om 2018 o 2022 in May o
Sep embe (i.e., he ege a ion g owing season (Robe s and Helbich,
2021; Tu unen e al., 2023)), we s o ed i s me ada a. We conside ed
image y sou ced by Google, and 360-deg ee pano amas uploaded by
GSV use s, as pa icula ly in g een u ban a eas ha a e inaccessible by
ca , use -con ibu ed image y is a widesp ead al e na i e o image y
sou ced by Google.
Iden i ying candida e loca ions: Ou candida e loca ions a e OSM
polygons o a ious ca ego ies o which we ha e bo h an NDVI alue
and GSV image y a ailable. We hen ook a andom sample o 140
candida e loca ions pe case-s udy ci y, while ensu ing equal sp ead
be ween bo h OSM ca ego ies (i.e., sampling equal numbe s o egula -
size g eenspaces, pocke -size g eenspaces, public open spaces, e c.) and
NDVI- alue qua e s. We manually checked i hei associa ed GSV im-
age y is sui able o collec ing isual pe cep ions: we excluded images
cap u ed indoo s o unde g ound, du ing nigh - ime o e en s, o poo
image quali y, aken om bi d’s o og’s iew pe spec i e, o when
sigh o he loca ion hey we e sampled o was obs uc ed (e.g., by a
wall). We eplaced hese loca ions wi h ano he andomly sampled
candida e om he same OSM ca ego y, NDVI qua ile, and ci y, un il all
sampled loca ions passed he check.
Finally, we ese he geome y o hese sampled loca ions o he poin
om which he GSV image was aken. We ecalcula ed he median NDVI
and de e mined which OSM place ca ego ies we e loca ed wi hin 15 m
bu e zone a ound his poin (Ami i and C ain, 2019). By doing so, we
ensu ed ha people’s pe cep ions, NDVI alues, and OSM ca ego ies all
e e ed o he same loca ion. We also in es iga ed bu e s o 25, 29, 43
and 100 m o de ine a loca ion’s immedia e su oundings o assess he
sensi i i y o ou esul s o he adius dis ance chosen. In ela ed s udies,
25 m we e ound o be ele an o cap u ing g eene y isible om a
loca ion (Kuo e al., 2018); 29 and 43 m o obse ing e en s in u ban
en i onmen s (Ami i and C ain, 2019); and 100 m o ep esen ing he
indi idual human scale in g eenspace heal h esea ch (Labib e al.,
2020).
3.3. Collec ing people’s isual pe cep ions
We hen collec ed people’s isual pe cep ions o he sampled loca-
ions h ough a ques ionnai e on P oli ic: an online c owdsou cing
pla o m designed o academic esea ch (Pee e al., 2017). We
ec ui ed pa icipan s who cu en ly li e in Eu ope and a e p o icien in
he English language. We ensu ed di e si y in age and gende and paid
pa icipan s minimum wage in he Ne he lands, he coun y o he au-
ho s’ a ilia ion. Pa icipan s p o ided in o med consen o pa icipa e
and could only submi a single ques ionnai e.
C owdsou cing ask: Ou ques ionnai e, implemen ed using he
Qual ics pla o m, ook abou i een minu es o comple e. On a e age,
Fig. 2. Collec ed da a pe sampled loca ion: NDVI alues and OSM ca ego ies wi hin adius dis ance, pe cei ed g eenness, and easons. Example in Pa c del Tu ´
o del
Pu xe , Ba celona.
R. Teeuwen e al.
Landscape and U ban Planning 245 (2024) 105009
6
we expec ed each loca ion o be a ed by i e people. We o mula ed ou
ques ions based on ela ed ques ionnai es used in en i onmen al heal h
esea ch (K uize e al., 2020; Zijlema e al., 2018), while keeping hem
simple and s aigh o wa d o c owdsou cing (Salesses e al., 2013;
Que cia e al., 2014; Milias e al., 2023; Candeia e al., 2017). Fig. 3
depic s an imp ession o he in e ace. Fi s , we in oduced he opic and
asked pa icipan s o p o ide some demog aphics. Second, o each
pa icipan , we andomly sampled i e loca ions om he same case-
s udy ci y. Fo each loca ion, we showed hem he pano amic GSV
image and ins uc ed hem o pan a ound o a leas 10 s. We hen
collec pa icipan s’ isual pe cep ions o g eenness by asking hem o
indica e o wha ex en hey ind he place ege a ed (on a 5-poin Like
scale: no a all (1) o e y (5)); and wha cha ac e is ics o he loca ion
mo i a ed hei choice (in open ex ). We included quali y checks con-
sis ing o a eCAPTCHA bo es and an a en ion check and collec ed he
numbe o panning clicks pa icipan s made. Finally, we asked pa ici-
pan s some mo e demog aphics and asked how clea hey ound he
c owd-sou cing asks.
3.4. Da a analysis
A e collec ing all necessa y da a (see Fig. 2), we could assess how
well NDVI and OSM cap u e people’s isual pe cep ions o g eenspaces.
Quan i a i e (s a is ical) analysis: Fi s , we il e ed ou pa ici-
pan s o isual pe cep ions ha did no mee ou quali y s anda ds (e.g.,
h ough bo de ec ion and an a en ion check, and checking i pa ici-
pan s panned a ound in he pano ama). We hen calcula ed desc ip i e
s a is ics based on he NDVI alue, OSM ca ego y, and pe cei ed ege-
a ion le el o each loca ion, and agg ega ed a ings in o a median alue
pe loca ion o u he analysis. To compa e pe cei ed ege a ion le els
o he bina y OSM da a ca ego ies (i.e., some hing ei he is agged as a
g eenspace, o no ), we also con e ed pe cei ed ege a ion le els in o
bina y alues, by de ining pe cei ed g eenspaces as loca ions wi h a me-
dian pe cei ed ege a ion le el o 4 ( ai ly) o 5 ( e y) ege a ed (Zij-
lema e al., 2018). We s a is ically es ed ou hypo heses and conduc ed
se e al explo a o y analyses. Table 1 summa izes ou hypo heses (H1-3)
and he non-pa ame ic me hods we used o es hem. Fi s , we es ed
o co ela ions be ween isual pe cep ions (on a 5-poin Like scale)
and NDVI da a using Spea man’s
ρ
(H1). Second, we compa ed he
pe cei ed g eenness dis ibu ions (on a 5-poin Like scale) o a ious
OSM ca ego ies using he Mann-Whi ney U and K uskal-Wallis es s
(H2.1); and calcula ed he pe cen age o OSM egula -size g eenspaces
ha a e pe cei ed as g eenspaces (in bina y alues) (H2.2). Thi d, we
implemen ed h ee algo i hms o selec g eenspaces om he di e en
da a sou ces and compa ed hem on how well hey cap u ed pe cep ions
o g eenspace using McNema ’s es . We also pe o med sensi i i y an-
alyses o iden i y how ou esul s change when we inc ease he bu e
zone adius om which we iden i y he median NDVI alue and he
p esence o OSM g eenspaces. Gi en ha we pe o med 8 di e en
signi icance es s (i.e., 1 o H1, 5 o H2, and 2 o H3), we applied a
Bon e oni co ec ion o ou signi icance h eshold o 0.05
8 =0.006. We
used he same cu o o explo a o y analyses.
Quali a i e analysis: To unde s and po en ial causes o di e ences
be ween isual pe cep ions and map da a, we conduc ed a e lexi e
hema ic analysis (Cla ke and B aun, 2013). Based on ou quan i a i e
indings, we iden i ied he places o which pe cep ions no ably de ia ed
om NDVI and OSM map da a and analyzed he spa ial cha ac e is ics o
hese places ha pa icipan s men ioned as easons. We used A las TI o
conduc ou hema ic analysis, employing induc i e coding and i e a i e
iden i ica ion o hemes.
4. Resul s
4.1. Desc ip i e s a is ics
This sec ion desc ibes he numbe o pe cep ions we collec ed, he
numbe o places we included in ou analysis, and he numbe o people
who pa icipa ed in ou s udy.
Be ween Ma ch and May 2023, 423 P oli ic pa icipan s, li ing in 21
di e en Eu opean coun ies, comple ed ou c owdsou cing ask. O
hese people, 409 passed ou quali y checks and we e he e o e included
in ou s udy. A majo i y ound he asks clea (100 %), panned he
pano amas a ound as eques ed (93 %), did no mo e away o adjacen
places (71 %) and we e no amilia wi h he places p esen ed (95 %).
Pa icipan gende s a y (49 % emale, 49 % male, and 2 % non-bina y,
hi d gende , o p e e o sel -desc ibe o no o say), as well as ages (12
% age 18–24, 19 % 25–34, 20 % 35–44, 20 % 45–54, 18 % 55–64, and
Fig. 3. Imp ession o c owdsou cing in e ace.
Table 1
O e iew o hypo heses and s a is ical me hods.
Hypo hesis Me hod
H1 The e is a s ong posi i e co ela ion be ween NDVI and pe cei ed g eenness. Spea man’s Rho
H2.1 Pe cei ed g eenness is highe o OSM egula -size g eenspaces han o pocke pa ks, play spaces, open public spaces,
and s ee s.
Mann–Whi ney U & K uskal-Wallis
H2.2 Pocke pa ks, play spaces, open public spaces, and s ee s can be pe cei ed as g eenspaces. Desc ip i e s a is ics (pe cen ages)
H3 Pe cei ed g eenspaces a e be e cap u ed in da a when selec ing hem based on a combina ion o OSM ca ego ies and
NDVI alues, opposed o only OSM ca ego ies, o only NDVI alues.
Desc ip i e s a is ics ( ue posi i es and
nega i es) & McNema ’s es
R. Teeuwen e al.
Landscape and U ban Planning 245 (2024) 105009
7
10 % 65 yea s o olde ). Mos pa icipan s a e ci y dwelle s (61 %).
When es ing di e ences in pe cep ions among pai s o demog aphic
g oups using he Mann-Whi ney U es , we did no ind any s a is ically
signi ican di e ences.
F om hese 409 pa icipan s, we ob ained a o al o 1956 pe cep ions
on g eenness, a e il e ing ou da a in case o echnical issues (e.g., he
pano ama did no load in ime) o in case he pa icipan did no in e ac
a all wi h he pano ama (i.e., made no clicks o pan he pano ama
a ound o zoom in).
Ou o 420 places, 413 ecei ed alid pe cep ions om ou pa ici-
pan s. On a e age, each place was a ed on g eenness by 5 people.
Table 2 shows desc ip i e s a is ics on he places, hei NDVI alues and
OSM ca ego ies, and associa ed pe cep ions pe ci y. People pe cei ed
180 o he places (44 %) as g een.
4.2. Quan i a i e analyses
4.2.1. H1: Pe cei ed g eenness in ela ion o NDVI alues
Hypo hesis and ou come: We hypo hesized o ind a s ong posi i e
co ela ion be ween NDVI and pe cei ed g eenness. Using Spea man’s
ρ
,
we es ed he co ela ion be ween how g een places a e pe cei ed o be,
and hei NDVI alues, and ound a s a is ically signi ican co ela ion o
mode a e s eng h (
ρ
: 0.459, p- alue <0.006), hus no suppo ing ou
hypo hesis.
Explo a o y analyses: No appa en NDVI alue h eshold ha di e -
en ia es g eenspaces om o he spaces could be iden i ied. When
compa ing co ela ions be ween ci ies, we ound ha co ela ion is
much weake in Ba celona (
ρ
: 0.269, p- alue <0.006) han in Ro e -
dam (
ρ
: 0.540, p- alue <0.006) and Go henbu g (
ρ
: 0.570, p- alue <
0.006) Fu he mo e, we did no ind s onge co ela ions when using
he maximum NDVI alue in a place’s p oximi y, as opposed o he
median (i.e.,
ρ
: 0.47, p- alue <0.006).
Sensi i i y analyses: To analyze how sensi i e ou co ela ion esul s
a e o he adius dis ance used o calcula e a place’s NDVI alue, we
ound ha by inc easing he adius dis ance o 25, 29, and 43 m, co -
ela ion s eng hs inc ease om 0.459 o 0.556, 0.585 and 0.600 (all p-
alue <0.006), while dec easing again o la ge dis ances. Fu he -
mo e, om 25 up o 100 m, he di e ences in co ela ions among case-
s udy ci ies la gely disappea ed.
4.2.2. H2: Pe cei ed g eenness in ela ion o OSM ca ego ies
Hypo hesis and ou come H2.1: We hypo hesized (H2.1) ha pe cei ed
g eenness is highe o OSM egula -size g eenspaces han o pocke -
size g eenspaces, play spaces, open public spaces, and s ee s. Using
he K uskal-Wallis es , we ound signi ican di e ence in pe cei ed
g eenness occu s be ween OSM ca ego ies (H: 107, p- alue <0.006).
Using a one- ailed Mann-Whi ney U es , wi h he al e na i e hypo hesis
ha egula -size g eenspaces a e pe cei ed mo e g een han o he s, we
ound ha egula -size g eenspaces (median pe cei ed g eenness: 4.0, n:
112) a e indeed pe cei ed g eene han: pocke -size g eenspaces (me-
dian: 4.0, U: 5878, n:86, p- alue <0.006); open public spaces (median:
2.0, U: 9723, n: 102, p- alue <0.006); and s ee s (median:3.0, U: 8741,
n: 107, p- alue <0.006); while no signi ican di e ence was ound wi h
play spaces (median: 4.0, n: 78); showing ha OSM egula -size g een-
spaces a e only pe cei ed g eene han pocke -size g eenspaces, open
public spaces, and s ee s.
Hypo hesis and ou come H2.2: We u he hypo hesized (H2.2) ha
also pocke -size g eenspaces, play spaces, open public spaces, and
s ee s can be pe cei ed as g eenspaces, as some li e a u e sugges s. We
conside ed a place o be pe cei ed as g eenspace when i was a ed on
median 4 ( ai ly) o 5 ( e y) ege a ed. We ound ha 70 % o all OSM
egula -size g eenspaces a e pe cei ed as g eenspaces. Fu he mo e, 47
% o pocke -size g eenspaces, 11 % o open public spaces, 36 % o play
spaces, and 23 % o s ee s (all excluding hose ha also lie wi hin di ec
p oximi y o a egula -size g eenspace) a e pe cei ed as g eenspaces.
Thus, we can con i m ha no only egula -size g eenspaces, bu also
pocke -size g eenspaces a e pe cei ed as g een mo e o en han 40 % o
imes (i.e., i g eenness a ings we e dis ibu ed equally o e ou 5-poin
scale, 2/5 o 40 % would be conside ed g een).
Sensi i i y analyses: When we g adually inc eased he adius dis ance
which we use o de ine i a place lies in p oximi y o an OSM g eenspace,
we obse ed ha esul s o H2.1 emain a he s able up o 43 m. Ye o
H2.2, we obse ed ha pe cen ages decline: wi h a adius o 15 m, 70 %
o places loca ed nea OSM g eenspace a e indeed pe cei ed by people
as g een; while wi h 25, 29, and 43 m, he pe cen ages declined o 67 %,
66 %, and 63 %, espec i ely.
4.2.3. H3: Pe cei ed g eenness in ela ion o bo h NDVI alues and OSM
ca ego ies
Hypo hesis and ou come: We hypo hesized ha i pe cei ed g een-
spaces a e selec ed using a combina ion o OSM ca ego ies and NDVI
alues, hey a e be e eco ded in da a han when only OSM ca ego ies
o only NDVI alues a e used. To es ou hypo hesis, we implemen ed
h ee g eenspace selec ion algo i hms based on ou indings in H1 and
H2: 1) OSM-based, selec ing all loca ions nea OSM egula -size g een-
spaces; 2) NDVI-based, selec ing places wi h an NDVI alue la ge han
he median NDVI alue o all sampled loca ions in he same ci y; and 3)
combina ion-based, selec ing loca ions nea OSM egula -size g een-
spaces, and pocke -size g eenspaces and play spaces wi h an NDVI la ge
han he median. We compa ed hei esul s o he c owdsou ced pe -
cep ions o g eenspace. In 67.8 % o cases, he OSM-based algo i hm
co ec ly cap u ed pe cep ions o g eenspace, compa ed o 65.6 % o
he NDVI-based algo i hm, and 71.8 % o he combina ion-based al-
go i hm. McNema ’s one- ailed es e ealed ha he combina ion-
based algo i hm pe o med signi ican ly be e han he NDVI-based
algo i hm (n: 401, p- alue <0.006), while no signi ican di e ence
was ound wi h he OSM-based algo i hm.
Sensi i i y analyses: When we epea ed ou analysis wi h NDVI alues
and OSM ca ego ies wi hin la ge adius dis ances, we disco e ed ha
he pe cen ages o he OSM-based and combina ion-based algo i hms
g adually dec eased wi h dis ance, while hey inc eased o he NDVI-
based algo i hm, which is consis en wi h he indings in H1 and H2
(see Table 3). Rega dless o adius dis ance, he combina ion-based al-
go i hm ou pe o med he OSM-based algo i hm, while a a 43-me e
adius dis ance, he NDVI-based algo i hm achie ed he highes sco e
o 72.3 %. Using McNema ’s es , we obse ed he NDVI-based algo-
i hm ou pe o ms he OSM-based algo i hm signi ican ly (n: 412, p-
alue <0.006), while he di e ence wi h he combina ion-based algo-
i hm was no s a is ically signi ican .
Table 2
Desc ip i e s a is ics o places and associa ed pe cep ions pe case-s udy ci y.
case places NDVI [0–1] OSM ca ego y [%] pe cep ions
ci y n med. min max eg.-size g eensp. poc.-size g eensp. open space play space s ee n
Ba celona 139 0.139 0.017 0.379 30.9 % 20.1 % 31.7 % 20.1 % 28.1 % 647
Ro e dam 137 0.188 0.023 0.539 26.3 % 21.9 % 21.9 % 19.7 % 27.0 % 645
Go henbu g 137 0.140 0.019 0.492 24.1 % 20.4 % 20.4 % 16.8 % 22.6 % 664
To al 413 0.151 0.017 0.539 27.1 % 20.8 % 24.7 % 18.9 % 25.9 % 1956
R. Teeuwen e al.
Landscape and U ban Planning 245 (2024) 105009
8
4.3. Quali a i e analyses
The ollowing pa ag aphs p esen quali a i e indings ollowing up
on es ing hypo heses H1 and H2. Exempla y quo es deno ed as Q-i a e
included in he supplemen a y ma e ial, as is exempla y s ee -le el
image y.
4.3.1. De ia ions be ween pe cei ed g eenness and NDVI alues ( ollowing
H1)
To unde s and de ia ions be ween pe cei ed g eenness and NDVI
alues, we explo ed why people ega d places g een, while NDVI alues a e
low, and ice e sa. We selec ed places o analysis based on ou quan-
i a i e esul s, using a 43-me e adius dis ance, i.e., whe e co ela ions
we e s onges .
Rega ding places ha do ha e a high su ounding NDVI alue bu a e
no deemed g een (n =6, see Table A1 and Fig. A1 in supplemen a y
ma e ial), we iden i ied ha hese a e ypically cha ac e ized by he
place being in be ween wo dis inc i e sides, esul ing in mixed opinions
among pa icipan s (Q-1). Speci ically, hese places a e cha ac e ized by
g eenness on one side, wi h g ass, ees, and occasionally o he ege-
a ion o na u al ea u es. Howe e , he o he side is gene ally domi-
na ed by buil -up elemen s (e.g., buildings, conc e e, and in as uc u e)
o , when he e is some ege a ion p esen (e.g., ees, g ass, g eene y, o
p i a e ga dens, some imes loca ed u he away o combined wi h
o he na u al ea u es), i emains oo li le o oo ba en (Q-2, Q-3). Less
o en, we obse ed ha g eene y may be p esen , bu is physically
inaccessible, o ins ance due o heigh di e ence (Q-4).
Rega ding places ha a e pe cei ed by people as g een bu ha e a
low NDVI alue (n =10, see Table A2 and Fig. A2 in supplemen a y
ma e ial), i s , we iden i ied ha ege a ion is o en p esen and a ying
in ype, bu only on a low le el (e.g., only g ass, o he g ound-co e ing
g eene y, o low-le el bushes), s ill young (e.g., iny ees), o sca e ed
a ound in small bi s (e.g., s and-alone ees, some ege a ion in e e y
ga den) (Q-5). We also obse ed ha ege a ion may be lush, bu only
on a limi ed a ea, in p i a e ga dens, o loca ed u he -on (Q-5). Sec-
ond, pa icipan s also men ioned o he na u al ea u es: i e on s, sand
o small iles on he g ound, wooden ences, o a seemingly good local
clima e (e.g., shaded, clean ai ) (Q-6, Q-7). Thi d, we obse ed spaces
a e cha ac e ized by a lack o ea u es, o example: dis an om a ic,
secluded, o quie (Q-8). We do no e, howe e , ha some pa icipan s
s ill cha ac e ized places domina ed by buil -up elemen s as “g een o an
u ban en i onmen ” (Q-10), in which cases he judgmen seemed
con ex ual a he han absolu e (Q-9). Las ly, men ions o a ac i eness
we e mo e p e alen among places pe cei ed as g een, han ice e sa
(Q-11).
4.3.2. De ia ions be ween pe cei ed g eenness and OSM ca ego ies
( ollowing H2)
Subsequen ly, we explo ed why people ega d places g een, while OSM
does no ag hem as such, and ice e sa. Again, we selec ed ou cases
based on quan i a i e esul s, now using he 15-me e adius dis ance a
which OSM pe o med bes .
As o places agged by OSM as g eenspaces, bu no pe cei ed as such
by people (n =12, see Table A3 and Fig. A3 in supplemen a y ma e ial),
we iden i ied wo main easons. Fi s , despi e being agged in OSM as
g een and seemingly equipped o use by people, e.g., wi h benches o an
ele a ed pedes ian walkway, some places we e no pe cei ed by people
as g een (Q-12). People s a e ege a ion is oo low, young (e.g., iny
ees), sca e ed, d y, cons ained, loca ed oo a away, o only on one
side, o he space is oo open and emp y (Q-13, Q-14, Q-15, Q-16). In
hese cases, he ege a ion was o e uled by buil -up s uc u es: majo
oads o amways, high building blocks, conc e e and o he pa ed
a eas, and associa ed sense o a bad local clima e (Q-17, Q-18, Q-19).
Second, again, we obse ed ha some places a e cha ac e ized by wo
dis inc sides: majo apa men buildings on one side, e sus a na u al
ock landscape wi h ege a ion on he o he ; conc e e and cons uc ions
wo ks, e sus a ca e ully designed g een-looking space; and a majo
oad, e sus an ex ensi e ege a ed a ea. These di e ences some imes
caused disag eemen among people, depending on wha a ac ed hei
a en ion he mos (Q-20 e sus Q-21).
Places ha a e no agged as g eenspaces o egula size in OSM, bu
s ill a e pe cei ed as g een by people, ou numbe ed all o he quali a i e
cases: 102 places. We iden i ied wo main hemes (see Table A4 and
Fig. A4 in supplemen a y ma e ial). Fi s , p esence and numbe o ees,
o he g eene y such as bushes, sh ubs, and smalle plan s, and o a lesse
ex en g ass played a majo ole, while people also men ioned a ia ion
in ege a ion, lowe s, and ields (Q-22, Q-23, Q-24). We also obse ed
ege a ion con igu a ion was explici ly o implici ly e e ed o (Q-25):
ege a ion on di e en heigh s (e.g., g ass ields, ee canopies, and
ege a ed walls) (Q26); and places ha a e spacious o ha e ege a ion
all a ound and a ex ending (Q-27). Second, we see again ha people
judged u ban g eenspace con ex ually a he han absolu ely: hey a e
g een “ o an u ban se ing” (Q-28). These included esiden ial neigh-
bo hoods wi h lo s o p i a e g eenspace and na u al buildings ma e ials
(Q-29); and egula o dense oad-side ege a ion (Q-30). Also, o he
quali ies we e associa ed wi h g eenness: wa e , shade and esh ai , and
quie ness, a ac i eness, and sa e y (Q31, Q-32, Q-33). Ye we do no e
ha he con lic be ween buil -up and g eenspace emained, wi h people
mo i a ing hei g eenness by he lack o p esence o buil -up elemen s,
such as buildings, a ic, conc e e, and pa king lo s (Q-34, Q-35, Q-36,
Q-37).
Table 4 summa izes he ou comes o ou hypo hesis es s and
explo a o y, sensi i i y, and hema ic analyses.
5. Discussion
5.1. In e p e a ion o esul s
Ou indings sugges ha NDVI and OSM da a cap u e how g een
people ind places o be a he well, ye signi ican disc epancies emain.
Fig. 4 shows exempla y loca ions whe e pe cep ions o g eenspace
de ia e om NDVI and OSM map da a.
We ound no e idence o a s ong co ela ion be ween how g een
places a e pe cei ed and he NDVI alues in hei immedia e icini y.
Howe e , we did disco e a signi ican mode a e co ela ion. Ou
indings o a signi ican co ela ion con as wi h hose o Ko hencz and
Blaschke (2017) and Leslie e al. (2010), who ound no signi ican co -
ela ion o NDVI alues wi h pa k isi o ’s pe cep ions o g eenspace, o
wi h people’s pe cep ions o hei home en i onmen . Wha dis in-
guished ou me hod om Leslie e al. (2010) is ha we collec ed da a o
Table 3
Quan i a i e esul s pe adius dis ance. Pe ow, highes sco es a e emphasized in bold.
adius dis ance [m]
15 25 29 43
H1 co ela ion pe cep ion & NDVI 0.459 0.556 0.585 0.600
H2.2 pe cen age OSM g eenspaces pe cei ed as such 69.6 % 66.9 % 66.4 % 62.9 %
H3 co ec ness OSM-based algo i hm 67.8 % 67.0 % 66.8 % 65.3 %
co ec ness NDVI-based algo i hm 65.6 % 69.7 % 70.8 % 72.3 %
co ec ness combina ion-based algo i hm 71.8 % 70.2 % 70.0 % 68.2 %
R. Teeuwen e al.
Landscape and U ban Planning 245 (2024) 105009
9
one single poin in place, a he han an en i e esiden ial neighbo hood,
and unlike Ko hencz and Blaschke (2017), we collec ed da a o a
b oade ange o public spaces, po en ially wi h a wide ange o NDVI
alues.
When we in es iga ed he in luence o adius dis ances, we disco -
e ed ha he s onges co ela ion was 0.600 when using median NDVI
alues wi hin a 43-me e adius dis ance. We saw he g ea es change in
co ela ion s eng h wi h inc easing adius dis ance in Ba celona, ising
om 0.269 o 0.577, implying ha pe cep ions o places in Ba celona
a e based on g eene y loca ed u he away: One could hypo hesize ha
Ba celona’s public spaces a e mo e spacious o ha e mo e ma u e ees
ha can be seen om a dis ance, as opposed o g asslands o small
ege a ion ha is mo e e enly dis ibu ed in space.
We ound e idence o suppo ou hypo hesis ha OSM egula -size
g eenspaces a e pe cei ed signi ican ly g eene han pocke -size
g eenspaces, s ee s, and public open spaces, while OSM seems o use
open space ags almos exclusi ely o places whe e ege a ion is no
dominan . We also disco e ed ha nea ly hal o OSM pocke -size
g eenspaces a e pe cei ed as g een, adding o he body o e idence
ha pocke -size g eenspaces a e impo an o g een ci ies as well
(Wood e al., 2017; Labib e al., 2020; Pescha d e al., 2014). Su p is-
ingly, no signi ican di e ence in g eenness pe cep ion was ound be-
ween OSM egula -size g eenspaces and play spaces. We disco e ed
ha many play spaces a e unanimously pe cei ed by people as g een,
e en hough OSM does no p o ide any indica ion o he p esence o
g eene y. Fu he mo e, examples o g eene y ha a e unexpec edly
missing om OSM include o es s and g o es loca ed on he ou ski s o
ci ies ha a e no ep esen ed in OSM. O he de ia ions we e no due o
a lack o g eene y in OSM da a, bu a he o how we il e ed ou
g eenspaces. Tha is, we selec ed g eenspaces o signi ican size based on
a minimum size o 0.5 ha o adjacen g een land (Ambien e I alia, 2003).
Some g eenspaces, howe e , a e mapped in such g anula i y in OSM —
o example, e e y indi idual pa ch o g ass sepa a ed om o he s by
na ow oo pa hs — ha ou algo i hm il e ed hem ou .
We also obse ed ha some places in OSM a e labeled as g een bu
a e no pe cei ed as such. When we look a hese places in OSM, we see
ha hal o hem a e agged as pa ks. The e m pa k is explici ly included
in he WHO de ini ion o g eenspace ha we used (WHO Regional O ice
o Eu ope, 2017a), and many o he de ini ions o g eenspace in he
li e a u e (Taylo and Hochuli, 2017), and pa icipan s o en seemed o
ega d pa ks equi alen o g eenspaces. Acco ding o OSM, a pa k is “an
a ea o open space o ec ea ional use, usually designed and in semi-na u al
s a e wi h g assy a eas, ees and bushes” (OpenS ee Map, 2023). As his
de ini ion and ou indings sugges , OSM pa ks a e ypically bu no
always ege a ed.
We demons a ed ha combining OSM ca ego ies and NDVI alues
can help o be e selec pe cei ed g eenspaces om hese da a in many
cases, p o iding an answe o he ques ion aised by Liao e al. (2021)
whe he combining mul iple da ase s imp o es pe o mance. Su p is-
ingly, we obse ed ha he NDVI-based selec ion algo i hm ou -
pe o med all o he s a a 43-me e dis ance. Quali a i e esul s
sugges ed e ining hese algo i hms wi h in o ma ion on o he spa ial
cha ac e is ics om OSM has g ea po en ial, such as p oximi y o wa e
o p esence o g eene y in all di ec ions; p oximi y o a ic in a-
s uc u e o high- ise buildings; and p esence o p i a e ga dens. O he
quali ies, such as ege a ion a ie y, quie ness, a ac i eness, and
sa e y, may be mo e di icul o cap u e in la ge-scale da a, bu a e
s udied in ela ed wo k (Milias e al., 2023; Candeia e al., 2017; Salesses
e al., 2013; Que cia e al., 2014).
Ou quali a i e indings indica ed ha people seem o judge he
g eenness o a place con ex ually a he han absolu ely. Speci ically,
people s a ed, o example, ha “ o an u ban se ing, mo e ees han I
would ha e expec ed” (Q-28), sugges ing ha people ha e di e en ex-
pec a ions o g eenspaces wi hin ci ies opposed o ou side o hem.
Fu he mo e, hey s a ed “conside ing i ’s in he middle o a man made
squa e i seems qui e g een” (Q-9), indica ing ha wi hin he cons ain s
o he ype o unc ion o a gi en u ban space (e.g., a c oss oads o a
majo oad), people some imes simply conside ed a place as g een as can
be.
5.2. Implica ions o esea ch and p ac ice
Acco ding o ou indings, NDVI maps a e only mode a ely associ-
a ed wi h how g een people pe cei e places o be. Fo op imum esul s,
pe cep ion da a should only be exchanged o NDVI maps while keeping
his limi a ion in mind, ideally u ilizing median NDVI alues wi hin 43
m. When using OSM da a, simila limi a ions a ise, bu we sugges using
a sho adius dis ance o 15 m ins ead.
Ou esul s u he sugges ha inco po a ing NDVI da a in o OSM-
based analyses p oduces mo e accu a e esul s. In he case in o mal
and small-scale g eenspaces a e o in e es , NDVI alues may help o
il e ou hose pa ks ha a e no pe cei ed g een, o o iden i y pocke
pa ks and play spaces ha a e o en pe cei ed as g een.
Ou quali a i e indings sugges ha when iden i ying loca ions o
g eenspace in e en ions, u ban planne s could conside p io i izing
g eenspaces ha appea in la ge-scale da a bu a e no pe cei ed as
such, e.g., whe e buil -up ea u es a e oo dominan .
Ou indings could also se e as guidance when aiming o make ci ies
“jus g een enough”: g eenspace s a egies ha limi ad e se e ec s o
in e en ions o make neighbo hoods heal hie and mo e a ac i e,
Table 4
Summa y o quan i a i e and quali a i e indings.
H1 Pe cei ed g eenness in ela ion o NDVI
hypo hesis es No e idence o a s ong co ela ion wi h median NDVI.
explo a o y
analysis
Signi ican mode a e co ela ion ins ead, weak co ela ion o Ba celona, while mode a e o Ro e dam and Go henbu g, and mode a e bu less s ong
co ela ion wi h maximum NDVI.
sensi i i y analysis NDVI wi hin 43 m adius dis ance yields s onges co ela ion.
quali a i e
analysis
Places pe cei ed no -g een, bu wi h high NDVI: ha e wo dis inc i e sides; o he g eenspace is physically inaccessible. Places pe cei ed g een, bu wi h
low NDVI: ha e a ying ege a ion; o he na u al ea u es nea by; absen buil -up ea u es; a e a ed in con ex .
H2 Pe cei ed g eenness in ela ion o OSM
hypo hesis es s Regula -size g eenspaces a e pe cei ed as g eene han pocke -size g eenspace, open public spaces, and s ee s, bu no g eene han play spaces. Pocke -
size g eenspaces a e o en imes pe cei ed as g eenspaces.
sensi i i y analysis OSM wi hin 15 m adius dis ance yields bes ou comes.
quali a i e
analysis
Places pe cei ed no -g een, bu in OSM: a e s ill equipped o people; ha e dominan buil -up ea u es; o wo dis inc i e sides. Places pe cei ed g een, bu
no in OSM: ha e la ge amoun and good con igu a ion o ege a ion; a e a ed in con ex ; ha e o he na u al and so ea u es.
H3 Pe cei ed g eenness in ela ion o NDVI and OSM
hypo hesis es Algo i hm combining OSM and NDVI da a yields be e esul s han NDVI-based algo i hm, bu no signi ican di e ence wi h OSM-based algo i hm.
sensi i i y analysis Combina ion- and OSM-based algo i hm pe o m bes wi h 15 m adius dis ance, while NDVI-based algo i hm wi h 43 m adius dis ance sco es highes
o e all.
R. Teeuwen e al.
