i
Roo op-Place Sui abili y Analysis o U ban Ai Mobili y Hubs:
A GIS and Neu al Ne wo k App oach
Ca los Ja ie Delgado Gonzalez
ii
Roo op-Place Sui abili y Analysis o U ban
Ai Mobili y Hubs:
A GIS and Neu al Ne wo k App oach
Disse a ion supe ised by
Joel Dinis Bap is a Fe ei a da Sil a, PhD
NOVA In o ma ion Managemen School
Uni e sidade No a de Lisboa
Lisbon, Po ugal
and co-supe ised by
Robe o And é Pe ei a Hen iques, PhD
NOVA In o ma ion Managemen School
Uni e sidade No a de Lisboa
Lisbon, Po ugal
Ca los G anell Canu , PhD
GEOTEC
Uni e si a Jaume I
Cas ellón, Spain
Feb ua y 2020
iii
Decla a ion o O iginali y
I decla e ha he wo k desc ibed in his documen is my own and no om someone
else. All he assis ance I ha e ecei ed om o he people is duly acknowledged and all
he sou ces (published o no published) a e e e enced.
This wo k has no been p e iously e alua ed o submi ed o NOVA In o ma ion
Managemen School o elsewhe e.
Lisbon, Feb ua y 24, 2020
Ca los Ja ie Delgado Gonzalez
[ he signed o iginal has been a chi ed by he NOVA IMS se ices]
i
Acknowledgmen s
Fi s , I wan o exp ess my g a i ude o he h ee supe iso s, since wi hou a doub
hei con ibu ions we e c ucial o he imp o emen o his hesis. I wan o hank D .
Joel Sil a o his pa ien guidance, encou agemen , and oppo une suppo h oughou
he de elopmen o his esea ch. I wan o hank P o esso Robe o Hen iques because
his o ice was always open o any conce ns ha a ose in addi ion o his ad ice o
enhancing he documen . I would like o hank P o esso Ca los G anell also o his
pe inen eedback, ad ice, and mo i a ion o consolida e he documen in he bes
possible way.
Addi ionally, I also wan o hank P o esso s Ma co Painho and Sa a Ri ei o o also
gi ing me hei uncondi ional ad ice and suppo whene e I needed i . I wan o
hank my classma es, who also encou aged me o ge ahead in his esea ch, especially
I za, Maicol, Mihail, Mo i z and Vicen e.
I would like o s a e my since e g a i ude also o he E asmus Mundus Mas e
P og am o Science in Geospa ial Technologies headed by P o esso s Ma co Painho,
Ch is oph B ox, Ch is ian K ay, Joaquin Hue a, and Michael Gould. Ha ing ecei ed
an E asmus schola ship o be pa o his wonde ul p og am is one o he mos
ewa ding expe iences o my li e.
Finally, I wan o hank all my amily, who has always belie ed in me. Especially o
my pa en s Glo ia and Ca los, because despi e he housands o kilome e s ha
sepa a e us, hei cons an suppo and wo ds o encou agemen we e i al o gi e my
bes and hus achie e such desi ed goal.
“The applica ion o GIS is only limi ed
by he imagina ion o hose who use i ”
________________________
Jack Dange mond

i
Roo op-Place Sui abili y Analysis o U ban Ai
Mobili y Hubs:
A GIS and Neu al Ne wo k App oach
Abs ac
Nowadays, cons an o e popula ion and u ban expansion in ci ies wo ldwide ha e led
o se e al anspo - ela ed challenges. T a ic conges ion, long commu ing, pa king
di icul ies, au omobile dependence, high in as uc u e main enance cos s, poo public
anspo a ion, and loss o public space a e some o he p oblems ha a lic majo
me opoli an a eas. T ying o p o ide a solu ion o he u u e inne -ci y
anspo a ion, se e al companies ha e wo ked in ecen yea s o design ai c a
p o o ypes ha base hei echnology on cu en UAVs. The e o e, ehicles wi h
elec ical Ve ical Take-O and Landing (eVTOL) echnology a e apidly eme ging
so ha hey can be included in he U ban Ai Mobili y (UAM) sys em. Fo his o
become a eali y, space agencies, go e nmen s and academics a e gene a ing concep s
and ecommenda ions o be conside ed a sa e means o anspo a ion o ci izens.
Howe e , one o he mos ele an poin s o his u u e implemen a ion is he sui able
loca ion o he po en ial UAM hubs wi hin he me opoli an a eas. Since al hough
UAM ehicles can ake ad an age o in as uc u e such as oo s o buildings o clea
and land, se e al c i e ia mus be conside ed o ind he ideal loca ion.
As a solu ion, his hesis seeks o ca y ou an in eg al oo op-place sui abili y
analysis by in ol ing bo h he essen ial a iables o he u ban ecosys em and he
adequa e oo op su aces o UAM ope abili y. The s udy a ea selec ed o his
esea ch is Manha an (New Yo k, U.S), which is he mos densely popula ed
me opoli an a ea o one o he megaci ies in he wo ld. The applied me hodology has
an unsupe ised-da a-d i ing and GIS-based app oach, which is co e ed in h ee
sec ions. The i s pa is esponsible o analyzing he sui abili y o place when
e alua ing spa ial pa e ns gi en by he applica ion o Sel -O ganizing Maps on he
u ban ecosys em a iables a ached o he ci y census blocks. The second pa is based
on he de elopmen o an algo i hm in Py hon o bo h he e alua ion o he la ness
o he oo su aces and he de ini ion o he UAM pla o m ype sui able o i s
se lemen . The inal s age pe o ms a combined analysis o he sui abili y indexes
gene a ed o he de elopmen o UAM hubs. Resul s e lec ha 16% o he oo s in
he s udy a ea ha e high in eg al sui abili y o he de elopmen o UAM hubs, whe e
UAVs pla o ms and Ve is ops (small size pla o ms) a e he ypes ha can be he
mos se led in Manha an.
ii
The ep oducibili y sel -assessmen o his esea ch when conside ing Nüs e al. [45]
c i e ia (h ps://os .io/j97zp/) is: 2, 1, 2, 1, 1 (inpu da a, p ep ocessing, me hods,
compu a ional en i onmen , esul s). Gi Hub eposi o y code is a ailable in
h ps://gi hub.com/ca losjdelgadono aims/ oo op-
place_sui abili y_analysis_ o _U ban_Ai _Mobili y_hubs
iii
Keywo ds
A i icial Neu al Ne wo k
Ca chmen A ea
Census Block
Clus e
D i ing Dis ance
Elec ical Ve ical Take-o and Landing
Fla ness
Geog aphical SOM
K-means
Ligh De ec ion and Ranging
Machine Lea ning
Neu on
Pa allel P ocessing
Poin s-O -In e es
Py hon
Roo op
Sel -O ganizing Maps
Sui abili y Analysis
U-Ma ix
Unmanned Ae ial Vehicles
U ban Ai Mobili y
Ve ihub
Ve ipo
Ve is op
ix
Ac onyms
AHP
Analy ic Hie a chy P ocess
ANN
A i icial Neu al Ne wo k
ATC
Ai T a ic Con ol
BMU
Bes Ma ching Uni
EASA
Eu opean A ia ion Sa e y Agency
eVTOL
Elec ical Ve ical Take-o and Landing
FAA
Fede al A ia ion Adminis a ion
FTP
T ans e P o ocol Files
GDAL
Geospa ial Da a Abs ac ion Lib a y
GeoSOM
Geog aphical Sel -O ganizing Maps
GIS
Geog aphical In o ma ion Sys ems
LAS
LASe o ma o poin clouds
LIDAR
Ligh De ec ion and Ranging
MOD
Mobili y On-Demand
NASA
Na ional Ae onau ics and Space Adminis a ion
OSM
Open S ee Maps
PAV
Pe sonal Ai Vehicles
PM
Pa icula e Ma e
POI
Poin s-O -In e es
RANSAC
Random Sample Consensus
SOM
Sel -O ganizing Maps
SQL
S uc u ed Que y Language
UAM
U ban Ai Mobili y
UAS
Unmanned Ae ial Sys ems
UAV
Unmanned Ae ial Vehicles
1
1. In oduc ion
1.1 P oblem s a emen and mo i a ion
The g owing o e popula ion and u baniza ion in me opoli an ci ies in he wo ld ha e
gene a ed se e al anspo - ela ed challenges in he ield o u ban mobili y [49].
Fu he mo e, he p oli e a ing p i a e ca dependency and high le els o spa ial
agg ega ion o economic ac i i ies a e leading o signi ican a ic conges ions [16, 47].
Addi ionally, ecen Mobili y On-Demand (MOD) schemes such as ide-sha ing and
ca -sha ing as hose implemen ed by Ube and Ly , ha e con ibu ed o inc easing
le els o a ic densi y on some u ban a eas while ying o sol e limi ed pa king lo s
issues [16, 18]. Consequen ly, inne -ci y commu e ime has been meaning ully a ec ed,
whe e commu e s in ci ies like Chicago, New Yo k, and Los Angeles, on a e age, lose
97 hou s a yea due o a ic jams [29]. Se e al al e na i es a e p oposed o elie e
a ic conges ions such as non-mo o ized mobili y and p omo ing he use o public
anspo a ion. Ne e heless, hese op ions equi e new in as uc u es which a e
limi ed by he lack o space in densely-popula ed u ban a eas [19, 48]. In his way, all
he abo e al e na i es a e summa ized in solu ions ocused on a wo-dimensional
ma ix [9].
As a no el solu ion, Elec ical Ve ical Take-o and Landing echnology (eVTOL)
ai c a a e being designed and es ed o become a easible means o anspo o inne -
ci y commu ing [9]. The new app oach con e ges on a pa h ha con empla es he
h ee-dimensional space, which had p e iously been explo ed by he de elopmen o
Unmanned Ae ial Vehicles (UAV) [43, 44]. New elemen s a e hen added o he
concep o U ban Ai Mobili y (UAM), which seeks o op imize inne -ci y commu ing-
imes by a oiding exis ing adi ional impedances gene a ed on oad ne wo ks [48].
UAM scope con empla es no only he anspo o passenge s bu also eigh
modali ies o open he doo o mul iple pu poses [44]. Besides, he ea u es and
speci ica ions o he new eVTOL ehicles a e qui e p omising o hei sus ainabili y
in an u ban en i onmen . Concep s issued by he manu ac u e s and consul ancies
show no able di e ences be ween an eVTOL ai c a and a adi ional helicop e .
eVTOL ehicles will be ou imes quie e , wice sa e and en imes less expensi e o
build han adi ional helicop e s [48]. While se e al companies and in es o s wo k on
he de elopmen o new, e icien , and com o able p o o ypes o eVTOL ehicles,
leading space agencies a ound he wo ld gene a e egula o y policies o he use o
hese ehicles in UAM sys em [48]. Na ional Ae onau ics and Space Adminis a ion
(NASA), Eu opean A ia ion Sa e y Agency (EASA), and Fede al A ia ion
Adminis a ion (FAA) a e some o he en i ies ha a e cu en ly decla ing concep s
o he p ope implemen a ion o UAM [40, 55, 64]. P ocess ce i ica ion, ba e y
echnology, ehicle e iciency, pe o mance and eliabili y, ai a ic con ol, cos and

2
a o dabili y, sa e y, noise, emissions, pilo aining, and in as uc u e in ci ies
summa ize he challenges o o e come owa ds UAM implemen a ion [27].
Subsequen ly, in ecen yea s he in e es o academic and scien i ic communi ies in
di e en ields has a oused o con ibu e o he implemen a ion o UAM. Geospa ial
echnologies ha e been in ol ed in his con ibu ion, especially when c ea ing 3D
geo ences o he design o inne -ci y ligh ou e ne wo ks [25] and inding sui able
loca ions o UAM in as uc u es [11, 15]. The la e excels in he complexi y i
handles by in ol ing a syne gy o di e en u ban, social, economic and en i onmen al
concep s, which play an impo an ole in de ining po en ial UAM hubs loca ions.
No wi hs anding, he ac ha UAM concep s a e add essed o be applied in densely
popula ed u ban a eas o he imp o emen o anspo quali y leads o add an
e alua ion o he in as uc u e p esen in he ci y. The e o e, he lack o su icien
space in me opoli an a eas implies aking in o accoun he ypology in he UAM
pla o m designs and sizes o i s cons uc ion on he oo ops o buildings.
Al hough no much wo k has been de eloped on his subjec , some academics ha e
con ibu ed by applying and e alua ing geo-scien i ic app oaches in sea ch o his ideal
loca ion. A sui abili y analysis was gene a ed o he ci ies o Los Angeles (Uni ed
S a es) and Münich (Ge many) unde an expe knowledge app oach [15]. The e o e,
a e applying an Analy ic Hie a chy P ocess (AHP), he consensus o expe s allowed
us o e alua e he impo ance o he social-economic and en i onmen al a iables
in ol ed [15]. The esul ing map e lec s egions o sui abili y whe e UAM g ound
s uc u es can be loca ed. Howe e , an exac loca ion was no calcula ed by no
conside ing physical a ailabili y s uc u es o UAM pla o ms. In con as , ano he
app oach o he loca ion o sui able UAV si es o landing and clea ing was pe o med
when conside ing he physical s uc u e o he oo s. The s udy was based basically
on he classi ica ion oo op images gene a ed bo h om Ligh De ec ion and Ranging
(LIDAR) da a and using sa elli e images [11]. Machine-lea ning classi ica ion
echniques we e used o dis inguish di e en oo shapes in aining samples om
h ee ci ies ha we e manually labeled. Resul s o he classi ica ion allow sepa a ing
se en ypes o oo shapes whe e hose ha a e la a e he mos easible o UAV
landing and ake-o [11]. Despi e good esul s, his app oach does no include u ban
socio-economic and en i onmen al aspec s, and also i is no ex ending he scope o
e alua e he o al oo ops use acco ding o a ea speci ica ions o build di e en UAM
ypes.
In his o de o ideas, his esea ch seeks o es ablish an unsupe ised-da a-d i en and
GIS-based me hodology o ind ou po en ial loca ions o he de elopmen o UAM
hubs in Manha an, New Yo k Ci y (U.S). The p oposed me hodological p ocess aims
o inc ease he le el o de ail in he loca ion o po en ial UAM hubs by conside ing
bo h a place sui abili y analysis gi en by he u ban ecosys em a iables and oo op
sui abili y analysis when e alua ing physical cha ac e is ics o la ness and a ea
a ailable o delimi UAM pla o ms ypes.
3
1.2 Resea ch Ques ions
This hesis o mula ed he ollowing esea ch ques ions:
• Whe e a e he mos sui able places o de elop UAM hubs in he s udy a ea?
• Wha is he po en ial sui abili y o he oo ops in he s udy a ea o he
de elopmen o di e en UAM pla o ms ypes?
• Is i possible o inc ease he le el o de ail in he loca ion o sui able places o
de elop UAM hubs in he s udy a ea?
1.3 Objec i es
In o de o achie e he o mula ed esea ch ques ions, his hesis p oposed he
ollowing speci ic objec i es:
• Maximize he in o ma ion p o ided by inpu a iables o he place sui abili y
analysis when ex ac ing p oximi y ea u es om u ban Poin s o In e es
(POI) h ough ou ing algo i hms.
• Design and implemen a place sui abili y analysis when applying Sel -
O ganizing Maps o iden i y clus e s wi h simila u ban ecosys em pa e ns.
• De elop an algo i hm o es ima e he sui abili y o he oo s when e alua ing
he main exis ing la su aces and he po en ial UAM pla o m ype ha can
be se led on hem.
• Examine he esul s ob ained by ma ching bo h p oposed app oaches.
1.4 Assump ions
The ollowing assump ions we e assumed o he de elopmen o his hesis:
• Up un il now, a UAM sys em has no been o icially implemen ed in any ci y
in he wo ld. The e o e no UAM s uc u e has been ma e ialized a p esen .
• The easibili y o he de elopmen o UAM hubs is based on conside a ions
issued by space agencies, consul ancies, manu ac u e s and academics who a e
explo ing he u u e o UAM as an inne -ci y anspo sys em.
• The ma e ial om which he oo s a e cons uc ed was no conside ed.
4
2. Li e a u e Re iew
2.1 U ban Ai Mobili y
U ban Ai Mobili y (UAM) is a “new” concep ha seeks o ga he all hose ai
passenge s and ca go-ca ying anspo sys ems wi hin me opoli an a eas by means
o bo h manned and unmanned ai c a [44]. Ne e heless, i is a e m ha has been
he e olu ion o se e al p ojec s and ideas de eloped be o e. Concep s such as Pe sonal
Ai Vehicles (PAV) and On-Demand Mobili y (ODM) we e men ioned by Na ional
Ae onau ics and Space Adminis a ion (NASA) in 2006 [15, 40, 41]. A ound 2007,
hanks o he ODM idea, he concep o ai - axis was al eady eme ging as a esponse
o he inc ease in a ic conges ion on he oads [27, 40]. NASA and Fede al A ia ion
Adminis a ion (FAA) e en began o o mula e easibili y s udies o he ai - axi
ma ke and ac o s ela ed o Ai T a ic Con ol (ATC) [40]. I s applicabili y has
depended on he echnological ad ances achie ed especially in he ield o Unmanned
Ae ial Sys ems (UAS). Today he eliabili y gi en by Unmanned Ae ial Vehicles
(UAVs) has been he eason o i s p oli e a ion in mul iple pu poses such as
hobbyis s, ca go-se ices, eme gency ca e, de ense, e c. Ne e heless, egula o y
conside a ions o bo h UAM and UAS a e being designed o imp o e he ATC [59].
Rega ding he passenge anspo in he UAM scheme, he e a e se e al academics,
indus ies, and go e nmen s ha a e s eng hening and p omo ing he sus ainabili y
o his inne -ci y and in a-ci y anspo sys em as a new al e na i e o mobili y [44].
Cu ing-edge echnological ad ances in he ield o elec onics and ae onau ics ha e
led o he de elopmen o designs and p o o ypes based on elec ical Ve ical Take-
O and Landing (eVTOL) echnology. Se e al p o o ypes o eVTOL ai c a a e being
designed and es ed in o de o gua an ee he ou key ealms o he e ical mobili y
ecosys em: ai c a sys ems, ce i ica ion and law, social accep ance and in as uc u e
[48]. Ehang, Volocop e , Ai bus Vahana, and Su eFly ha e been he pionee s
de eloping eVTOL ai c a p o o ypes since 2015 and es ing hem wi h success ul
ligh s in 2017 and 2018 [1, 2, 48]. Up un il he de elopmen o his esea ch, he mos
ecen es eco ded by Volocop e was a Helsinki ai po in Augus 2019 [10, 64].
Ano he impo an manu ac u e o eVTOL ai c a is he Ge man s a up Lilium.
They ha e designed a Je - ype ai - ehicle called LiliumJe which was i s es ed in
May 2019. I s design is also based o ul ill wi h EASA and FAA s anda ds [35, 46].
Amazon wi h P ime Ai [30] design and Boeing gian ca go d one a e explo ing and
es ing new designs o he deli e y o goods o di e en sizes. This adds o he many
o he uses ha UAVs ha e nowadays. In summa y, he e is e idence o a a ie y o
companies wo king ha d o de elop p o o ypes ha mee he equi emen s o a
e ical mobili y ecosys em.
5
2.1.1 Special Fac o s o U ban Ai Mobili y
2.1.1.1 Socio-economic ac o s
One o he ealm keys o he UAM is social accep ance owa ds he p ojec [28,
48]. The e o e he communi y mus unde s and he bene i s and p oblems ha can
cause he implemen a ion o a new mode o anspo in he ci y. UAM ies o
minimize his impac by conside ing ha his new mobili y model does no physically
eplace exis ing mobili y s uc u es [9]. In addi ion, some o he exis ing
in as uc u es will se e as a pi o o he ealiza ion o u u e inne -ci y mobili y [9,
27]. Thus, he loca ion o hospi als, me o s a ions, bus s ops, ade cen e s, g een
a eas and o he Poin s-o -in e es (POI) will wo k o es ablish he app op ia e
geospa ial en i onmen o u u e UAM in as uc u es. Popula ion mus also know
wha a e he mechanisms ha will gua an ee a sa e and eliable means o anspo
[48]. Fu he mo e, he popula ion needs o be a pa icipan in he de ini ion o ou es
and hubs. 3D geo ences ha e been simula ed by Hildemann e al. [25] e lec ing
possible condi ioned a eas o ai - axis ansi . The delimi a ion o a eas wi h high
popula ion densi y and job densi y a e ele an o ind he communi ies ha can
bene i om UAM [15]. In his way, no only po en ial use s o ci ies should be
conside ed in he s udies, bu also all ci izens who pa icipa e in he inne -ci y mobili y
in gene al [27, 48]. Sus ainabili y s udies o he UAM show a o able esul s especially
o he passenge ma ke [48] howe e , a o dabili y will play an impo an ole a he
beginning o he implemen a ion o UAM sys em. The e o e, spa ially iden i ying
a iables ela ed o income, job densi y, and mode o anspo used o ge job place
could suppo he loca ion o UAM in as uc u e [15, 48]. Addi ionally, i is ele an
o unde s and he cos o bo h mone a y and ime a el o commu e s using simila
anspo sys ems. One o he mos iable ma ke schemes is he ope a ion o ai axis
unde an on-demand app oach [43]. Ne e heless, he e migh be o he schemes which
in e es in es o s. Ai po shu le, ai -ambulance, police, company shu le and o ice-
o-o ice a els, ca go-deli e y a e some o he po en ial UAM ma ke s [43]. Ano he
in e es ing app oach was es ablished by Fadhil 2018 [15] by in ol ing o ice en al
p ices as an es ima ion o business ips. Po en ial demand o UAM could be loca ed
whe e en al p ices a e highe , leading o p obable se lemen s o UAM s uc u es [15].
2.1.1.2 En i onmen al ac o s
Noise pollu ion is one o he mos wo isome a iables when implemen ing eVTOL
ehicles [9, 27, 48, 56]. Al hough acco ding o he concep s, eVTOL ai c a gene a e
less noise han helicop e s, he e will be an inc ease in noise le els in ce ain pa s o
he ci y [48]. Ne e heless, manu ac u e s and expe s es ima e ha noise le els may
be simila o hose ha a hal -size uck can emi [27]. An eVTOL ai c a 90 me e s
om he g ound can emi abou 63dB measu ed a g ound le el [27, 48]. This could
be conside ed "accep able", bu he e a e o he implica ions ha should be s udied.
6
The e o e, his leads o ecognizing he ele ance o he noise a iable in geospa ial
analysis o UAM in as uc u e se lemen . An impo an idea o mi iga e noise was
designed by Ancli e al. [4] when ying o enhance pa hs o eVTOL ai c a when
conside ing a eas wi h high exis ing noise. In his way, highways and main oads o
he ci y could unc ion as noise abso p ion zones caused by he p opulsion o he
ehicles [4]. Consequen ly, a eas wi h high emissions o pa icula e ma e (PM) will
indica e he p esence o ho -spo s whe e he de elopmen o UAM sys ems by
implemen ing eVTOL echnology would imp o e he u ban ai quali y [43, 52]. O he
en i onmen al ac o s may be ela ed o he p esence o bi ds [56] and wea he
condi ions ha would limi he a ic o UAM ehicles.
2.1.2 U ban Ai Mobili y pla o ms
Keeping a success ul business model o UAM equi es an adequa e in as uc u e o
he ope a ion o eVTOL ehicles [9, 48]. No only landing and ake-o pla o ms
should be conside ed, bu ba e y cha ging in as uc u es, ATC o su eillance and
main enance a eas should be conside ed [9, 60]. Thus, he objec i e is o ind s a egic
places o he de elopmen o bo h g ound-based and oo op s uc u es o landing
and ake-o [9]. No wi hs anding, ad e se condi ions in highly popula ed ci ies can
make i di icul o analyze, especially due o he a ailabili y o space [60]. Se e al
concep s ha e been s a ed o classi y he ype o in as uc u e necessa y o he
de ini ion o he UAM in as uc u e ne wo k. Ube Ele a e [27] e e s o Ve ipo s
and Ve is ops as he main ypes o in as uc u e needed o eVTOL ope a ions.
Ve ipo s a e conside ed as a eas wi h mul iple landing and clea ance pla o ms o
eVTOL ehicles [27, 43, 48]. In addi ion, hey mus ha e enough space o es ablish
acili ies ha suppo main enance, echa ging, and s a [27, 48]. Ube Ele a e [27]
es ima es Ve ipo s would ha e a maximum capaci y o 12 eVTOL ai c a aking as
an example he cu en helipo s in New Yo k. On he o he hand, Ve is ops a e
conside ed as a pla o m wi h a single pad o he landing and clea ing o an eVTOL
[48]. Acco ding o Ube Ele a e [27] he ad an age o hese pla o ms is gi en by he
as emba ka ion and disemba ka ion o passenge s bu wi hou he p esence o
complex suppo acili ies. Fadhil [15] designed ano he ca ego y o a pla o m called
Ve ihub, which is conside ed as he bigges . These pla o ms could also ha e acili ies
o epai and main enance, in addi ion, hey would se e as pa king o eVTOL
ehicles [15]. Rega ding UAVs he e a e no ixed speci ica ions o he size o landing
pads o adi ional UAVs; howe e , po able landing pad designs ange om h ee
squa e me e s depending on he d one size.

7
2.2 Re iew o Place Selec ion Me hods o UAM pla o ms
Al hough se e al s udies and in es iga ions ela ed o si e selec ion analysis o place
sui abili y analysis ha e been de eloped, only a ew ha e been applied o ind ou he
loca ion o UAM pla o ms speci ically. Howe e , a bibliog aphic e iew has iden i ied
wo app oaches, he i s one gi en by he u ban en i onmen condi ions o he place
and he second one unde physical oo cha ac e is ics as po en ial landing and
clea ance spo s o UAM ehicles.
2.2.1 Place Selec ion App oach
Rega ding UAM g ound in as uc u e, Fadhil [15] de eloped his esea ch by
implemen ing a sui abili y analysis unde an Analy ic Hie a chy P ocess (AHP)
me hodology wi h a Delphi analysis. The e o e, his analysis in ol ed an expe
knowledge app oach o gene a e sui abili y maps o UAM in as uc u e in he ci ies
o Los Angeles and Münich [15]. E en hough he gene a ed maps allow o iden i y
egions wi h di e en le els o sui abili y, he exac loca ion whe e UAM pla o ms
can be es ablished is no de ined [15]. Simila s udies applied no speci ically o UAM
loca ion s a ions bu o nodes o anspo ne wo ks ha e also been included in his
e iew. Mul i-C i e ia and Single-C i e ion models we e analyzed by Vie a [63] in he
solu ion o Hub Logis ic P oblems. Wi hin he Mul i-C i e ia models analyzed by
Vie a [63], me hods such as AHP, Fuzzy se s, Weigh ed Sum, Topsis, Gene ic
algo i hms among o he s a e applied. Apa om expe knowledge app oach me hods
such as hose al eady men ioned, da a mining echniques ha e been applied o ind
pa e ns in egions o u ban a eas ha can la e be ca ego ized [39]. S a egic si e
loca ions ha e been ca ied ou using Sel -O ganizing Maps (Chap e 2.3) o ind co-
loca ion pa e ns when using oad ne wo ks on loca ion-based se ices app oach [66].
Spa ial accessibili y analysis is ano he impo an componen in si e loca ion.
The e o e, he e alua ion o a el dis ances be ween poin s o in e es di ec ly
in luences u ban planning [34]. Calcula ion o Euclidean dis ance, walking dis ance,
bicycle dis ance, and D i ing Dis ance a e some o he me hods ha Geog aphical
In o ma ion Sys ems (GIS) applies o de e mine cha ac e is ics o p oximi y and
op imal ou e [34]. D i ing Dis ance has been used o show dispa i ies and de iciencies
in he inne -ci y public anspo , which is a s a ing poin o u ban planning and
ci y design [34, 51]. Sma ci ies look o anspo nodes o ha e apid accessibili y
o ci izens by inc easing connec i i y in a el modes [8, 14, 38]. The design o
anspo nodes and u ban ou e planning ha e been de eloped by applying D i ing
Dis ance algo i hms di ec ly om da abases. Al hough some GIS applica ions embed
ou ing ools, da abase add-ons such as pgRou ing o Pos g eSQL da abases un
obus ou ing analysis when using Open Sou ce oad ne wo ks [17]. Thus, he quali y
o oad ne wo ks in luences he speed and accu acy o he esul s. Open S ee Maps
8
(OSM) and con ibu ing g oups such as Geo ab ik o e ee su icien da a om oad
ne wo ks wi h node a ibu es, edges, cos and o he impedances in he oad sys em.
D i ing Dis ance algo i hm om pgRou ing lib a y usually calcula es dis ances when
implemen ing he Dijks a algo i hm, which allows gene a ing a ealis ic simula ion o
he si ua ion [36]. Algo i hms ha e no only been used o calcula e op imal ou es bu
also o conside co e age a eas o ca chmen a eas based on dis ances [20].
2.2.2 Roo op Shape Iden i ica ion App oach
An in e es ing app oach was he one de eloped by Cas agno e al. [11] when
implemen ing a me hodology o he de ec ion o sui able oo s so ha UAV can land
in case o eme gency. Al hough his app oach was no p ope ly designed o eVTOL
ai c a , i akes ad an age o he use o high- esolu ion sa elli e images and 3D poin
clouds (LIDAR) o pe o m a oo op shape classi ica ion [11]. Cas agno e al. [11]
me hodology, akes oo op LIDAR da a which is ans o med in o RGB as e iles
by ipling he ele a ion in o ma ion. Subsequen ly, he as e s a e he inpu o a
Con olu ional Neu al Ne wo k and Random Fo es (Machine-Lea ning app oach)
algo i hms ha classi y he images acco ding o he shape ea u es. Howe e , he
labeling o aining samples is done manually [11]. Cas agno's esea ch leads o ha
image classi ica ion pe o med wi h LIDAR da a gene a es be e esul s han using
sa elli e images, bu a combina ion o he wo app oaches inc eases accu acy [11].
Despi e ob aining a good classi ica ion o he oo shape, sui abili y o UAV landing
pads is limi ed o inding la oo s. The e a e o he app oaches o iden i y oo planes,
ne e heless hese a e no aimed a assessing he la ness o landing and clea ance o
UAM pla o ms, bu b eaking down he planes ha make up he oo . Speci ic
algo i hms such as Region-G owing and Random Sample Consensus (RANSAC) ha e
been implemen ed by Albano [3] and Chen e al. [12]. Clus e ing echniques ha e also
been used o plane oo segmen a ion, he e o e k-means and ussy k-means ha e been
applied o oo econs uc ion [54]. C oss-line Elemen G ow h is also a echnique has
been p o ided by Wu e al. [65] o as and accu a e delinea ion o planes unde
ai bo ne LIDAR da a.
2.3 Sel -O ganizing Maps: SOM and GeoSOM
Sel -O ganizing Maps (SOM) is one o he mos ecognized ypes o A i icial Neu al
Ne wo k (ANN) c ea ed by Kohonen which is designed o he ex ac ion and
isualiza ion o pa e ns employing an unsupe ised lea ning p ocess [5, 31]. The e o e,
SOM is conside ed as a Machine-Lea ning me hod ha p o ides solu ions o p oblems
ha a e modeled unde a da a-d i en app oach [33]. Analysis o pa e ns and
s uc u es a e achie ed a e SOM pe o ms a da a educ ion ask by p ojec ing high-
9
dimensional da a o a lowe space ha is usually wo-dimensional o h ee-dimensional
[22, 42]. The ac ha he ou pu space esul ing om da a educ ion p ese es
opological ela ionships makes SOM a mo e powe ul me hod compa ed o o he s i.e
k-means; in addi ion o allowing i o be applied o di e en pu poses such as mining
da a, da a isualiza ion, clus e ing, and classi ica ion [5, 6, 32]. The essen ial idea o
SOM is ha he ou pu space ha is usually a wo-dimensional g id composed o uni s
called neu ons, can ma ch he inpu pa e ns p esen ed in he inpu space ( aining
pa e ns) o he es ablishmen o he Bes Ma ching Uni (BMU). The BMU is he
esul o an i e a i e p ocess whe e andomly an inpu pa e n is p esen ed o each o
he SOM neu ons, hen dis ances a e calcula ed, and he closes one is assigned as
BMU [22]. Fo mula (1) explains he p ocess o ind he BMU, whe e x is he inpu
ec o , ‖ . ‖ i is gene ally Euclidean dis ance, mc co esponds o he SOM neu on and
mi o he BMU neu on.
‖𝑥𝑥 − 𝑚𝑚𝑐𝑐‖=𝑚𝑚𝑚𝑚𝑚𝑚𝑖𝑖{‖𝑥𝑥 − 𝑚𝑚𝑖𝑖‖} (1)
A e BMU is ound all SOM neu ons a e upda ed and mo ed close o aining
pa e ns in each i e a ion. The SOM upda e is gi en by he o mula (2), whe e α( )
is he lea ning a e in speci ic ime and ℎ𝑐𝑐𝑖𝑖(𝑡𝑡) is he neighbo hood unc ion a ound
he BMU uni c.
𝑚𝑚𝑖𝑖=𝑚𝑚𝑖𝑖(𝑡𝑡)+ 𝛼𝛼(𝑡𝑡)ℎ𝑐𝑐𝑖𝑖(𝑡𝑡)(𝑥𝑥 − 𝑚𝑚𝑖𝑖) (2)
The e a e se e al neighbo hood unc ions be ween hem: Bubble, Gaussian and
Cu gass [22]. These neighbo hood unc ions include ano he pa ame e called he
neighbo hood adius, which along o he lea ning a e dec eases in each SOM upda e.
The e o e, SOM is inc easingly adjus ing o aining pa e ns by na owing BMUs
and hei opological neighbo s [22]. The SOM aining p ocess is ca ied ou in wo
phases, un olding and ine- uning. Some o he SOM pa ame e s a e con igu ed
depending on he p oblem being e alua ed; howe e , Hen iques [22] explains some
special conside a ions be o e he algo i hm is execu ed. SOM quali y can be e alua ed
h ough wo e o s, quan iza ion e o and opog aphical e o [22, 32]. The i s one
is mo e o ien ed o measu e he adap i e capaci y o he neu al ne wo k, while he
second one assesses he opology p ese a ion o he SOM [22].
On he o he hand, GeoSOM is a a ia ion o he SOM algo i hm conside ing he
na u al geog aphical componen o aining pa e ns [22, 23]. In his way, he GeoSOM
concep a ec s he BMU sea ch o a ce ain aining pa e n by conside ing only
geog aphically close neu ons [7, 23]. The e o e, he en i e selec ion p ocess o he
winning uni selec ion is ca ied ou in wo s eps: De ini ion o he geog aphic
10
neighbo hood and inal sea ch when conside ing emaining mul idimensional
componen s [7, 24]. BMU sea ch is pe o med when he pa ame e “geog aphical
ole ance” k is se o ze o (k=0), which o ces he BMU o be ha uni ha is
geog aphically closes [24]. The ac ha k is equal o ze o desc ibes a di e en scena io
han when k inc eases i s alue since he sea ch adius is ampli ied, and when k eaches
he map size, a basic SOM hen execu ed [7, 24]. Thus, GeoSOM ou pu s will ha e
simila beha io o SOM aining bu only conside ing he spa ial coo dina es, which
leads o each neu on wo king as a “low-pass il e ” o o he non-geospa ial a iables
[23]. Se e al isualiza ions modes will hen allow e i ying ha he GeoSOM app oach
yields he c ea ion o clus e s composed o geog aphically con iguous a eas by o cing
uni s ha a e close in he ou pu space o be close in he inpu space [7, 24]. I is
ema kable o men ion ha GeoSOM algo i hm no only e alua es spa ial
homogenei y when conside ing spa ial au oco ela ion bu also he e ogeneous a eas
ha al hough geog aphically may be close, non-geog aphic a ibu es migh depic
low co ela ion [24]. Fu he mo e, GeoSOM clus e s can be used o di e en pu poses
in many hema ic ields, such as u ban and en i onmen al planning [24].
17
Figu e 6. LIDAR iles used in he oo op la ness assessmen
3.1.3 Da a p ep ocessing
P ep ocessing o he place sui abili y analysis consis ed o joining all he in o ma ion
coming om socio-economic and en i onmen al a iables o he census blocks since
his is he minimum uni o in o ma ion o his analysis. The e o e, all alphanume ic
in o ma ion om ables was joined o census block iden i ie s. Besides, ai - a ic and
oad- a ic noise in o ma ion was also associa ed wi h he census blocks by a e aging
noise alues h ough geop ocessing ools. Polygon- ype POIs we e ans o med o
poin geome y upon inding he cen oid; howe e , polygons wi h a signi ican ex en
such as pa ks we e ans o med when densi ying he polygon wi h andom-placed
poin s h ough GIS ools. Be o e he execu ion o he SOM, GeoSOM and k-means
algo i hms (Chap e 3.2.2), da a ans o ma ion was implemen ed due o he
a iabili y o measu emen uni s o each o he inpu da a in ol ed in he analysis.
Among he main ans o ma ion me hods, Min-Max and Z-sco e no maliza ions we e
es ed o he e alua ion o spa ial pa e ns om GeoSOM sui e so wa e.
Fu he mo e, e i ica ion o null alues and he po en ial p esence o ou lie s was
assessed. Howe e , due o he na u e o some a iables, especially hose gene a ed
a e he p oximi y ea u e ex ac ion p ocess (Chap e 3.2.1), null alues and possible

18
ou lie s we e p ese ed h oughou he analysis when conside ing hei in o ma ion
was ele an o pa e n iden i ica ion. Rega ding he sui abili y analysis o oo s,
LIDAR da a was downloaded h ough a T ans e P o ocol Files (FTP), and o ganized
in a olde o easy handling. LIDAR iles we e examined indi idually o e i y he
exis ence o he in o ma ion, and in some cases, i was necessa y o apply coo dina e
sys em ans o ma ion o p ope geoloca ion. Ou lie s e i ica ion o ele a ion alues
was also pe o med be o e he calcula ion o ele a ion s a is ics o each o he oo ops
(Chap e 3.3.1).
3.2 Place Sui abili y Analysis
3.2.1 P oximi y Fea u es Ex ac ion
D i ing Dis ance algo i hm om pgRou ing lib a y (Chap e 3.5) is calcula ed o
p oximi y ea u e ex ac ion, which allows expansion and maximiza ion o he
in o ma ion a ailable in he Manha an census blocks. The e o e, ec o geog aphic
laye s wi h dimension 0 (poin ype) o each o he POI ca ego ies a e s o ed in
Pos g eSQL da abase wi h Pos GIS ex ension (Chap e 3.5). A ou able Manha an
oad ne wo k is downloaded om Open S ee Maps (OSM) h ough Geo ab ick
eposi o y. Then by using osm2pg ou ing (Chap e 3.5), he oad ne wo k can be
s o ed in he Pos g eSQL da abase. This ne wo k is cha ac e ized by ha ing
in o ma ion abou he sou ce nodes, des ina ion nodes, edges, and cos . Census block
cen oids a e also s o ed in he Pos g eSQL da abase. Once he in o ma ion is
o ganized, he i s s ep is o calcula e he closes node o each elemen o he di e en
POI ypes and census blocks by using Euclidean dis ance (Figu e 7-a). In his way,
new ables o close s nodes a e gene a ed o each geog aphical laye .
Subsequen ly, D i ing Dis ance algo i hm is execu ed by assuming a maximum
dis ance o 2km om each closes node o he census blocks. The esul is a new able
wi h all possible nodes and edges ha a e wi hin 2km away o each census block
(Figu e 7-b). Since all he in o ma ion is p ope ly ela ed and e e enced in he
da abase, SQL que ies allow us o ob ain he numbe o POI elemen s ha a e
associa ed wi h he nodes wi hin 2km. Mo eo e , i allows knowing wha is he
sho es dis ance o each each o he POI ype nodes (Figu e 7-c). SQL ou ines lead
associa ing hese esul s as wo new columns pe POI ypes; which a e he numbe o
eachable POIs and minimum dis ance o each each POI ype. Thus, by using D i ing
Dis ance we a e adding eal p oximi y in o ma ion o he census blocks h ough he
POI geog aphic loca ion.
19
(a) (b) (c)
Figu e 7. Example o he p oximi y ea u es ex ac ion p ocess. (a) Selec ion o closes node
o each POI and census block. (b) Selec ion o all possible nodes wi hin 2km D i ing
Dis ance. (c) Selec ion o he sho es pa h o each POI ype om he census block node
3.2.2 Clus e ing algo i hms: SOM, GeoSOM, k-means
P io o SOM Execu ion, i is necessa y o apply a no maliza ion o he a iables due
o hey ha e di e en measu emen uni s. Min-Max no maliza ion echnic was
selec ed o his pu pose. The e o e, he new a ibu e alues o he census blocks
( aining pa e ns) will be be ween 0 and 1. Fo SOM aining algo i hm om
GeoSOM so wa e, i is necessa y o con igu e he inpu pa ame e s eques ed by he
ool. I is impo an o no e ha he quali y o he esul s is di ec ly ela ed o he
pa ame e s selec ed o SOM [22, 37, 62]. Al hough heo e ically, he e is no be e
con igu a ion o he ini ializa ion o SOM pa ame e s [22], special conside a ions om
he li e a u e e iew we e s udied o pa ame e selec ion. Subsequen ly, a sensi i i y
analysis a e unning he algo i hm se e al imes con ibu es o he de ec ion o he
minimum possible e o o bo h quan iza ion and opog aphical. Then his can helps
in he bes con igu a ion selec ion. Rega ding SOM s uc u e pa ame e s, mos
a emp s we e made using a hexagonal opology because i o e s a connec ion o i s
six neighbo s and gene a es smoo he maps [22]. Di e en inpu space sizes we e
examined conside ing ha he la ge i is, pa e ns and g ouping s uc u es a e clea e
o de ine [22, 58].
On he o he hand, wi hin he aining pa ame e s, di e en alues o he numbe o
i e a ions, lea ning a e, and neighbo hood adius we e es ed in each SOM a emp .
These a ia ions we e es ablished conside ing he wo phases o SOM aining,
un olding and ine- uning. The lea ning a e and neighbo hood adio alues a e usually
smalle in he uning phase han in he un olding. In con as , he numbe o i e a ions
in he un olding phase is usually less han in he uning phase. These conside a ions
a e aised because in he un olding phase all neu ons a e sca e ed h oughou he
inpu space, while in he uning p ocess, he neu ons a e loca ed in a eas whe e
aining pa e ns a e highly concen a ed [22]. O he aining pa ame e s such as
20
lea ning unc ion and neighbo hood unc ion we e selec ed acco ding o li e a y
e iew. The In e se unc ion was selec ed o he i s one by allowing g ea e ne wo k
mobili y in he aining phase and achie ing apid adjus men s [22], while Gauss
unc ion was selec ed o he second acco ding o showing a smoo he g anula i y in
he esul ing map [22]. Once he selec ed con igu a ion shows sa is ac o y esul s, he
ou pu space is analyzed by means o he U-Ma ix, which p o ides in o ma ion abou
he dis ance be ween neu ons [26, 62] (Figu e 8-a). The g ayscale pale e colo shows
high dis ances in da k ones, and low in he ligh ones. Hin s isualiza ion p o ides
in o ma ion abou he numbe o aining pa e ns associa ed wi h each neu on (Figu e
8-b).
(a) (b)
To simpli y he isualiza ion o he pa e ns in he ou pu space, a new SOM is ained
wi h he same size in X as he ini ial one bu wi h only one neu on in Y. Figu e 9
shows ha o he p e ious example, he esul ing U-Ma ix size is 5x5, so he new
U-Ma ix will ha e a size o 5x1. This new U-Ma ix is aken as a base-line o he
clus e p ojec ion pe neu on in he U-Ma ix 5x5, whe e pie-cha s depic s he
p opo ions o associa ed clus e s in each neu on (Figu e 9). Visualiza ion o he
pa e ns and s uc u es h ough Box-His o diag ams and componen planes allow
obse ing po en ial simila i ies and di e ences p esen in ce ain a iables o each o
he gene a ed clus e s. Because his hesis is aimed a aking he ini ial clus e s only
as a guide, new clus e s a e c ea ed om GeoSOM so wa e by iden i ying a eas and
clus e s o he U-Ma ix ha may end o e lec simila pa e ns.
Figu e 8. Example U-Ma ix 5x5 neu ons. (a) U-Ma ix
dis ances isualiza ion (b) U-Ma ix hin s isualiza ion
21
Figu e 9. S a egy o he isualiza ion o he ini ial clus e s h ough clus e p ojec ion pe
neu on om a U-Ma ix base-line
Rega ding he GeoSOM algo i hm, his is ained om GeoSOM sui e wi h he bes
con igu a ion o pa ame e s ound in SOM. Tha is because he only di e ence
be ween SOM and GeoSOM is ha he la e conside s he ela ionship o spa ial
con igui y o e ed by he na u e o he geog aphic da a. The me hodological p ocess
desc ibed is again execu ed o he esul s o GeoSOM, hus gene a ing a base-line U-
Ma ix ha will p ojec clus e pe neu on in he ini ially gene a ed U-Ma ix. Pa e n
e alua ion is pe o med o GeoSOM h ough Box-His o and componen planes.
Al hough SOM is a powe ul clus e ing echnique when inding in i s aining p ocess
he bes ma ching p o o ype in addi ion o upda ing opological neighbo s on he map
[5], he goal o unning k-means is o ha e a poin o compa ison wi h an unsupe ised
clus e ing algo i hm o he han SOM. The e o e, k-means is execu ed by using he
same ini ially no malized alues, besides o ake in o accoun he same numbe o
clus e s selec ed o SOM. Subsequen isual inspec ion o he dis ibu ion o he
spa ialized clus e s will allow he selec ion o one o hese algo i hms. The selec ed
algo i hm will be used o pe o m a clus e analysis o he non-s anda dized da a. In
his way, a desc ip ion o he main cha ac e is ics o each clus e can be ca ied ou
when e alua ing he collec i e beha io o he non-no malized a iables. The de ined
clus e s will be he main inpu o con inue wi h he classi ica ion o place sui abili y
le els.
22
3.2.3 Classi ica ion o Place Sui abili y Le els
While his hesis is aimed a de eloping a non-expe knowledge and unsupe ised
da a-d i en based me hodology, sui abili y le els will be calcula ed om he esul s
ob ained by he clus e ing algo i hm selec ed. Fu he mo e, he analysis ensu es
sui abili y le els a e es ablished wi hin census blocks ha ha e simila pa e ns.
Consequen ly, he li e a u e e iew allows an in e p e a ion o he impo ance o he
en i onmen al, socio-economic and p oximi y a iables in ol ed in he s udy. High
ai / oad a ic noise and PM alues can in luence he design o na iga ion ou es o
ai - axis [4, 9, 27, 48, 56], as well as de ining pilo a eas o imp o e ai quali y [27,
48]. Socio-economic a iables such as medium household incomes and g oss en wi h
high alues could indica e he loca ion o po en ial inhabi an s who can a o d ai - axi
se ices o business ips [15, 27, 48]. In addi ion, demog aphic a iables and wo ke s'
da a could con ibu e o inding a eas whe e UAM ehicles would acili a e no only
public anspo bu also ca go and deli e y ac i i ies [9, 28, 44]. P oximi y a iables
calcula ed by he D i ing Dis ance algo i hm p o ide de ailed in o ma ion abou he
numbe o accessible POIs o each o he census blocks. The e o e, high alues o
hese a iables would indica e possible egions nea nodes o ec ea ion, heal h,
anspo a ion, public sa e y, and ci izen ca e. In con as , low alues o a ibu es
ha desc ibe he minimum dis ance o each POIs show g ea e impo ance in
accessing he men ioned nodes.
When conside ing hese s a emen s, his hesis p oposes hen o es ablish pe clus e
a s a is ical Jenks dis ibu ion in each o he a iables. Jenks o “na u al b eaks”
dis ibu ion allows es ablishing classi ica ions by op imizing he di e ences be ween
hem [13]. The e o e, Jenks dis ibu ion is pe o med o es ablish h ee di e en class
anges pe a iable in each clus e . By bea ing in mind he impo ance o he desc ibed
a iables, a sco e is assigned o each o he classes in each a iable. This sco e is buil
in a ange om 1 o 3. Whe e 3 will be assigned o classes o g ea e impo ance, 2
medium impo ance, and 1 low impo ance. Figu e 10 allows us o obse e he
p ocessing scheme abou he gene a ion o Jenks classi ica ions gi en pe a iable in
each clus e . Subsequen ly, he summa ion o he sco es o he a iables pe clus e is
pe o med in o de o ha e o alized alues o all sco es in each clus e . Then hese
o alized sco es a e classi ied again by implemen ing Jenks dis ibu ion o keep he
same class anges p e iously es ablished. Thus, he inal sui abili y ca ego ies a e
es ablished in each o he clus e s.

23
Figu e 10. Scheme o he ca ego iza ion o Place Sui abili y Le els o each o he n clus e s
calcula ed
3.3 Roo op-Fla ness Assessmen
This sec ion o he hesis is aimed a explaining he me hodology applied o he
e alua ion o oo - la ness, which is de eloped in wo phases. The i s one seeks o
es ima e he basic oo ele a ion s a is ics. The second s age is esponsible o ca ying
ou he la ness e alua ion o hen de e mine sui abili y le els o build UAM
pla o ms.
3.3.1 Roo op Ele a ion S a is ics Calcula ion
The i s s ep in he calcula ion o oo ele a ion s a is ics is o c op he iles o LIDAR
3D cloud poin s wi h he oo ops oo p in ec o laye . The algo i hm designed in
Py hon akes ad an age o he exis ing unc ions and me hods in he A cPy lib a y
o he LIDAR managemen da a. In his way, he algo i hm pe o ms a ou ine o
ex ac he 3D cloud o poin s in each oo o Manha an. Due o he possibili y ha
a oo can geog aphically sha e in o ma ion om mo e han one LIDAR ile, he
algo i hm conside s hese si ua ions by ex ac ing he 3D poin s by sec ions and hen
annexing hem. The e o e, he comple eness o he in o ma ion pe oo is gua an eed.
Taking as e e ence he wo k done by Cas agno e al. [11] in he oo LIDAR da a
p e-p ocessing, elimina ion o po en ial ou lie s was conside ed in he algo i hm. These
ou lie s can gene a e noise in he analysis when including opog aphic su ace poin s.
24
Once he new LIDAR gene a ed iles a e clean, oo ele a ion da a is s o ed as pandas
da a ames o accele a e da a p ocessing. Thus, he algo i hm calcula es main
s a is ics pe oo such as mean ele a ion, maximum ele a ion, and s anda d de ia ion
o he heigh s. This las s a is ic is o g ea impo ance since i wo ks as a p oxy in
he iden i ica ion o almos comple ely la oo s by ha ing alues e y close o ze o.
Then a geo- isualiza ion o he s a is ics desc ibed yields o examine he spa ial
dis ibu ion and hei pa e ns.
3.3.2 Roo op Fla Su aces Ex ac ion
Once he c opped LIDAR iles a e ob ained, he i s s age o he algo i hm is
esponsible o gene a ing a as e ile o he ele a ions o each oo . A cPy LAS
me hods allow he as e iza ion o he 3D poin s whe e each pixel will ake he
co esponding ele a ion alue. Pixels a e gene a ed wi h a size o 30 cm o inc ease
he le el o image de ail. This s ep is based on he me hodology applied by Cas agno
e al. [11] in he gene a ion o LIDAR images o oo shapes iden i ica ion. Ha ing
hese images o each o he oo s, he nex s ep is o pe o m a segmen a ion p ocess
o iden i y su aces wi h simila heigh . This p ocess is ca ied ou by simula ing
GDAL polygoniza ion unc ion, whe e neighbo ing pixels ha sha e he same alue
a e g ouped as a polygon. Ne e heless, his p ocess is coded unde A cPy unc ions,
whe e su aces wi h he same heigh alue a e gene a ed and s o ed as compac
polygons. These polygons allow us o es ima e he a ea o con inuous su aces wi h
equal heigh based on he g ouping o pixels a high esolu ion o each oo (Figu e
11).
Figu e 11. Main s eps o oo op la su aces ex ac ion
Al hough o he me hods o p ecision segmen a ion and oo econs uc ion ha e been
p oposed when wo king wi h LIDAR poin s di ec ly [3, 12], LIDAR image
segmen a ion simpli ies he calcula ions necessa y o his s udy by implemen ing GIS
in eg a ed ools. The algo i hm o de s hese a eas o selec he i e la ges and hen
calcula e he pe cen age hey ep esen wi hin he en i e oo a ea (Figu e 12). These
25
a eas a e hen classi ied acco ding o he size necessa y o he cons uc ion o di e en
UAM pla o ms. Sizes o eVTOL ai c a p oposed by Fadhil, Seeley, and Ube
Ele a e [15, 27, 53] we e aken as a e e ence o he de ini ion o size anges o each
o he possible UAM hubs ype. Howe e , a new ca ego y was p oposed o he landing
and clea ance o UAVs in o de o maximize he po en ial o oo op use (Table 3).
Po en ial UAM Hubs
Type
A ea Range
(m
2
)
Ve ihubs > 3400
Ve ipo s 1000 - 3400
Ve is ops 160 - 1000
UAV ehicles 3 - 160
Table 3. A ea anges p oposed o he ca ego iza ion o UAM pla o ms
3.3.3 Roo op Sui abili y Ca ego iza ion
Because hese polygons can ha e di e en o ms, he designed algo i hm e alua es he
compac ness o each o hem by means o Elonga ion me hod o Leng h-Wid h
p oposed by Ha is [21] and desc ibed by he o mula (3). WMBB e e s o he wid h
o he minimum bounding-box and LMBB o he leng h o he minimum bounding-box.
To calcula e he a io, he maximum measu e o he polygon mus be in he
denomina o .
𝐿𝐿𝐿𝐿 =𝑊𝑊𝑀𝑀𝑀𝑀𝑀𝑀
𝐿𝐿𝑀𝑀𝑀𝑀𝑀𝑀 (3)
Ra io alues close o 1 indica e g ea e compac ness han hose close o 0. Then, he
algo i hm sa es his alue pe polygon in each oo . As a second e alua ion me ic o
he de ec ion o polygons wi h ec angula shapes and inclined o ien a ion, he
algo i hm also includes he pe cen age o a ea co e ed by he polygon wi hin he
bounding-box egion. Me ic elonga ion is hen ca ego ized in o qua iles o he
de ini ion o sui abili y anges.
26
Figu e 12. Example o a ibu es calcula ed by he algo i hm in each o he main oo
su aces
3.4 Combined Ca ego iza ion o he Sui abili y Indexes o
UAM Pla o ms
The las s age in he p oposed me hodology is he gene a ion o he inal sui abili y
o he loca ion o UAM pla o ms. The e o e he p ocess consis s o combining he
place sui abili y calcula ed om he de ini ion o clus e s and oo op sui abili y gi en
by he la ness assessmen . Consequen ly, he spa ial o e lapping o bo h geog aphical
laye s (census blocks and oo s) will allow es ablishing amalgama e sui abili y
ca ego ies. This combined sui abili y leads o bo h spa ially and s a is ically o e i y
he iden i ica ion o o al sui abili y le els in each o he po en ial UAM pla o m
ypes which can be de eloped in he di e en clus e s. Thus, an es ima e o he
numbe o oo ops wi h high in eg al sui abili y o UAM pla o ms can be ound, in
addi ion o geog aphically knowing whe e hey lie in each o he clus e s. In he same
way, he ou pu s le knowing which oo s de ini ely do no show in eg al sui abili y
o he de elopmen o he UAM hubs. Fu he mo e, joined sui abili y le els can also
be examined by ela ing hem o he ea u es gi en by he non-s anda dized u ban
ecosys em a iables in each o he clus e s. The o e lapping p ocess is made
conside ing ha no ambigui ies a e gene a ed in he combina ion o he ca ego ies by
knowing in ad ance ha en i ies in he oo op building oo p in laye used in he
analysis can only be inside a single census block.
S1 S2 S3 S4 S5
A ea (m2) 382 166 141 133 17
% To al Roo op 32 13.9 11.8 11.2 1.4
Ele a ion (m) 61 47 61 69 64
Elonga ion
0.7
0.5
0.4
0.7
0.5
% wi hin Bbox
0.5 0.2 0.4 0.6 0.3
33
Figu e 16. SOM Box-His o o he 10 base-line clus e s
A new Box-His o is gene a ed o e i y he a e age beha io o he no malized
a iables in he 8 clus e s aised (Figu e 18). This new Box-His o e lec s ce ain
simila i ies wi h he baseline. Howe e , i associa es mo e pa e ns ha end o be he
same in ewe clus e s. In addi ion, he clus e s ob ained e lec a se o di e en
beha io s ha cha ac e ize he uniqueness o he pa e ns p esen in hem. To
complemen he desc ip ion o he gene a ed clus e s, hey we e isualized on a map
o obse e hei spa ial dis ibu ion (Figu e 19). The e o e, he o e iew o he Map
wi h he SOM clus e s allows us o obse e he exis ence o clus e s ha end o
ep esen compac and non-dispe sed g oups (Clus e s 2, 4, 5, 6, and 8). On he
con a y, Clus e s 1, 3, and 7 show g ea e dispe sion h oughou he s udy a ea,
especially Clus e 1. Ne e heless, i is impo an o no e ha al hough Clus e 8 is
mos ly ep esen ed by a compac g ouping o blocks, nea he Down own (Sou h o
he s udy a ea) shows a small g ouping o census blocks wi h he same cha ac e is ics.
Figu e 17. U-Ma ix gene a ed o he new 8 SOM clus e s

34
Figu e 18. Box-His o o he no malized a iables o each o he 8 SOM clus e s
Figu e 19. Manha an census blocks clus e ed by SOM algo i hm
4.1.3 GeoSOM Clus e s
A e unning SOM, GeoSOM is execu ed using he bes inpu pa ame e s
con igu a ion. In he same way, a new GeoSOM base-line wi h a U-Ma ix size 10x1
is gene a ed in o de o p ojec a clus e pe neu on o hen ob ain 10 base-line clus e s
(Figu e 20). P opo ions o he aining pa e ns in each clus e a e analyzed oge he
wi h he Box-His o (Figu e 21) and he componen planes (Figu e A- 3, Figu e A- 4 -
Annex 1). The simila i y in he a e age beha io o no malized a iables and dis ances
35
be ween neu ons indica es ha some baseline clus e s may be showing e y simila
pa e ns. Thus, om GeoSOM a new design o he clus e s is ca ied ou and he s eps
pe o med in SOM a e epea ed. Eigh new clus e s a e gene a ed on he U-Ma ix
(Figu e 22) whe e he a e age a iabili y o each o he clus e s is displayed in Box-
His o (Figu e 23). Simila o SOM, some o he clus e s can con e ge o simila i ies in
some a iables; howe e , in o he s, hey can show o ally di e en .
(a) (b)
Figu e 20. (a) GeoSOM U-Ma ix 10 x 15 ob ained wi h bes pa ame e con igu a ion, (b)
Re lec ed aining pa e n p opo ions om he 10 base-line clus e s
Figu e 21. GeoSOM Box-His o o he 10 base-line clus e s
Figu e 22. U-Ma ix gene a ed o he new 8 GeoSOM clus e s
36
Figu e 23. Box-His o o he no malized a iables o each o he 8 SOM clus e s
As a nex s ep, a isual analysis o he beha io o clus e s gene a ed wi h GeoSOM
is ca ied ou . Clea ly and as expec ed by he na u e o GeoSOM, he spa ial adjacency
as a opological ela ionship plays an impo an ole in de ining clus e s. Mos clus e s
show a clea endency o o m a single compac g oup, wi h ce ain excep ions. Jus
a ew census blocks om Clus e s 1, 2, 3, 4 and 5 a e ligh ly a om hei
co esponding main g oup (Figu e 24).
Figu e 24. Manha an census blocks clus e ed by GeoSOM algo i hm
37
4.1.4 Visual Compa ison o Clus e Resul s
Once he esul s ha e been ob ained om SOM and GeoSOM, his hesis seeks o
e alua e he quali y o he in o ma ion acqui ed in he a iables by applying a
di e en g ouping algo i hm. The e o e k-means, as men ioned in he me hodology,
was he algo i hm selec ed o gene a e clus e s based solely on he mul i-dimensional
dis ances gi en by he selec ed a iables. k-means was execu ed o ob ain he same
numbe o clus e s as in he o he algo i hms. I is in e es ing o no e ha al hough
he e a e some dispe sions o census blocks o some o he g oups o med, mos o
hem a e concen a ed in la ge blocks. Be o e add essing some o he main di e ences
ound be ween he algo i hms, i is impo an o men ion ha colo s and numbe s
we e nominally se o ease o in e p e a ion (Figu e 25). The e o e, i is no possible
o asse ha SOM Clus e 1 is equal o k-means Clus e 1 e en when hey can be in
a simila geog aphical posi ion. Wi h his in mind, we can obse e ha he p esen
clus e pa e ns gene a ed using SOM and k-means show a e y simila spa ial
dis ibu ion. G ea e di e ences a e p esen ed in concen a ions gi en by Clus e s 2
and 7. Conce ning GeoSOM we can con i m ha i is he g ouping me hod ha
gene a es mo e compac clus e s when conside ing con igui y in aining pa e ns.
Ne e heless, se e al o he clus e s gene a ed wi h GeoSOM show a spa ial
dis ibu ion sligh ly simila o hose ob ained in SOM and k-means, especially in he
Down own o Manha an. Thus, he ac ha he esul s o SOM and k-means e lec
simila pa e ns in hei clus e s is e idence ha he a iables included in he analysis
a e in o ma i e enough o he con ex o he p oblem.
Figu e 25. Map compa ison om clus e s gene a ed using SOM, GeoSOM, and k-means
38
To con inue wi h he me hodology p oposed in his in es iga ion, i is necessa y o
selec one o he p oposed me hods. GeoSOM, being he only one o he me hods which
adds he geog aphic con igui y o aining pa e ns as a g ouping ac o , ies o
gua an ees ha spa ially he g ouped census blocks keep g ea e compac ness.
Fu he mo e, GeoSOM esul s a e selec ed since hey depic mo e in sync wi h he
de ini ion o he i s law o geog aphy p oposed by Toble , "e e y hing is ela ed o
e e y hing else, bu nea hings a e mo e ela ed han dis an hings" [57]. Howe e ,
i is impo an o include in he analysis ha al hough GeoSOM esul s show mo e
compac homogenei y o aining pa e ns when conside ing spa ial a iables,
he e ogenei y o non-geospa ial a iables can occu [24]. Subsequen ly, a deepe
analysis o he clus e s when e e sing he no maliza ion o he a iables will allow
gene a ing mo e exhaus i e s a emen s.
4.1.5 GeoSOM Clus e s Desc ip ion
This sec ion aims o analyze and b eak down he main pa e ns associa ed wi h each
o he GeoSOM clus e s when conside ing he non-no malized alues o each a iable.
Since he analysis in ol es se e al a iables, s a is ical g aphs we e g ouped acco ding
o socio-economic and en i onmen al a iables (Figu e 26, Figu e 27, Figu e 28, Figu e
29), he numbe o associa ed POIs (Figu e 30), and minimum dis ance o each each
POI (Figu e 31).
• Clus e 1:
I is composed o blocks loca ed in sou he n Manha an, speci ically in he inancial
dis ic , in addi ion o he h ee main islands (Go e no s, Libe y and Ellis) along
wi h some po s on he side o B ooklyn. I is cha ac e ized by being he clus e wi h
he leas numbe o POIs; howe e , i can ha e quick access o almos all ypes o
POIs excep o embassies. Despi e i s size, i concen a es high incomes and g oss
en ; his p obably because i akes pa in he ci y's inancial dis ic . On he o he
hand, i is he clus e wi h ewe wo ke s. Deducing ha i is p obably a sec o whe e
he e a e qui e a ew en ep eneu s. Conce ning ai quali y, i is one o he clus e s
wi h he highes concen a ions o PM, al hough he noise le els a e no as high
compa ed o o he clus e s. In summa y, his clus e may e lec he po en ial o he
de elopmen o UAM pla o ms whe e eVTOL ehicles a e ela ed o comme cial and
business ac i i ies. Al hough i could also be used o he ai - axis se ices by co e ing
he ou is si es ha a e in he men ioned islands.

39
• Clus e 2:
Loca ed a he Uppe Manha an and bo de ing he B onx, i is cha ac e ized by
being he second mos ex ensi e. I is he second g oup wi h he lowes incomes.
Compa ed o he o he clus e s, i has one o he highes noise le els, bu one o he
lowes in PM concen a ions. Possibly ela ed when being he clus e wi h he la ges
numbe o g een a eas (pa ks and g a eya ds). Howe e , i has a conside able numbe
o pa king lo s, bus s a ions and heal h cen e s. Taking his in o conside a ion, he
clus e does no e lec a high in e ac ion o he a iables o ind ou a popula ion
ha will a o d ai - axi se ices, excep o e y speci ic ac i i ies such as access o
he ew ou is si es o heal h cen e s.
• Clus e 3:
Loca ed in he sou h o Manha an and e y close o he inancial cen e , i s ands
ou because i is one wi h he lowes popula ion densi y, al hough i has a la ge
pe cen age o wo ke s. I is also cha ac e ized by showing e y low le els o income
and g oss en . I is a clus e whe e on a e age, he me o s a ions a e qui e dis an .
Howe e , as anspo nodes, i has se e al bus s a ions and pa king lo s. Wi h hese
desc ip ions, i can be deduced ha his a ea does no o e a ac i e ea u es o
massi ely build UAM pla o ms. These could be dedica ed o he speci ic use o heal h
cen e s and go e nmen en i ies p esen in his g oup.
• Clus e 4:
Loca ed in Mid own Manha an. I has access o mos POIs. In addi ion, he e is a
signi ican pe cen age o wo ke s; household incomes and g oss en a es a e
conside ably high. On he o he hand, i depic s he highes PM le el among all he
clus e s, which is consis en wi h i s high noise le el. Wi hou a doub , i is a clus e
ha o e s a p io i y scena io o he cons uc ion o UAM pla o ms o imp o e ai
quali y. Addi ionally, i shows ha i is a clus e wi h a mix u e o se ices ha can
be o e ed, ai - axis, ai -ambulances, ai -secu i y, among o he s.
• Clus e 5:
The g ouping is loca ed in Down own Manha an. One o he mos ema kable
cha ac e is ics is ha al hough i has a e y low popula ion densi y, i shows he
highes le els o bo h incomes and g oss en . Low p esence o heal h cen e s and poo
access o hospi als p edomina e. Also, i is a clus e ha has a mode a e numbe o
ou is si es and a ai p esence o anspo nodes. Clea ly, his clus e b ings oge he
he necessa y p ope ies o dis inguish he popula ion ha can mos a o d UAM
se ices.
40
• Clus e 6:
I has he la ges ex ension and is loca ed in Mid own Manha an, co e ing he en i e
egion nea Cen al Pa k. I highligh s due o b ings oge he all embassies, in addi ion
o ha ing quick access o go e nmen and secu i y en i ies. I also s ands ou o
concen a es he majo i y o schools and uni e si ies. As expec ed, easy access o g een
a eas. Howe e , i has high le els o noise and PM. I is also a clus e whe e incomes
and g oss en a e mode a ely high. I agg ega es mo e anspo nodes, al hough no
many pa king lo s. Conside ing hese desc ip ions, i is a clus e ha can p o ide
pla o ms o se e al ypes o UAM se ices. Ne e heless, by collec ing a la ge numbe
o go e nmen en i ies, some a eas could be used exclusi ely o diploma ic anspo
and secu i y con ol.
• Clus e 7:
Loca ed in he Uppe eas side o Manha an, i g oups blocks e y close o he Eas
Ri e coas and in ol es he Randalls and Roose el Islands. I ga he s he la ges
numbe o inhabi an s; howe e , hey ha e mode a ely low incomes and g oss en s.
I also shows o be he g oup wi h he lowes le els o noise. Besides, i has he la ges
numbe o hospi al cen e s, bu no e y easy access o me o s a ions. This clus e
may no be a p io i y o he de elopmen o se ices such as ai - axis, bu he
de elopmen o ai -ambulances.
• Clus e 8:
Loca ed a he no h o he Cen al Pa k in Uppe Manha an. I has he highes
noise le els and mode a e PM le els. Al hough i is a g oup wi h low-incomes le els,
i has a conside able pe cen age o wo ke s. I g oups he la ges numbe o pa king
lo s and has easy access o bus s a ions. I also shows ha ing conside able numbe o
heal h cen e s, schools, and uni e si ies. The e o e, al hough his does no show oo
many a ac i e ea u es o build se e al UAM pla o ms, i shows a o able
en i onmen al condi ions o he ansi o eVTOL ehicles.
(a) (b)
Figu e 26. Ba -cha s non-no malized a iables (a) popula ion densi y (b) popula ion
41
(a) (b)
Figu e 27. Ba -cha s non-no malized a iables (a) Pe cen age o wo ke s ha commu e
mo e han 60 minu es one-way, (b) A ea km2
(a) (b)
Figu e 28. Ba -cha s non-no malized a iables (a) Road- a ic and ai - a ic noise, (b) Ai -
quali y
(a) (b)
Figu e 29. Ba -cha s non-no malized a iables (a) Median household incomes, (b) Medium
G oss Ren
42
Figu e 30. Compa a i e ba -cha s o he numbe o POI p esen in each GeoSOM clus e
49
(a) (b) (c)
Figu e 36. Roo 1 (a) Lida image wi h main su ace, (b) sa elli e image, (c) 3D scene
(a) (b) (c)
Figu e 37. Roo 4 (a) Lida image wi h main su ace, (b) Sa elli e image, (c) 3D scene
(a) (b) (c)
Figu e 38. Roo 7 (a) Lida image wi h main su ace, (b) Sa elli e image, (c) 3D scene
(a) (b) (c)
Figu e 39. Roo 2 (a) Lida image wi h main su ace, (b) Sa elli e image, (c) 3D scene

50
(a) (b) (c)
Once hese a ibu es ha e been calcula ed o each o he 45.515 oo s e alua ed, he
algo i hm akes he a ea o he main su ace as a e e ence o he es ima ion o he
ype o pla o m o be buil . The li e a u e e iew allowed us o ga he enough
in o ma ion o es ablish es ima es o he a ea necessa y o he cons uc ion o UAM
pla o ms. The e o e, his hesis has aken as a e e ence o he pla o ms ypology
gi en by [15, 27, 53] o eVTOL ai c a and has added a special ca ego y o UAV.
The map in Figu e 41 allows us o app ecia e he dis ibu ion o he pla o m ypes
associa ed wi h he main sui able a ea o each o he oo s. The e o e, only conside ing
he main su ace o oo s, 33.070 (72.7%) oo s would wo k o he cons uc ion UAV
pla o ms, 11.631 (25.6%) oo ops o ins alla ion o Ve is ops, 703 (1.5%) oo ops
o Ve ipo s and 57 (0.1%) oo s o he la ge Ve ihubs. Only 54 oo s we e
disquali ied as hey did no ha e a minimum use ul a ea o UAM pla o ms. As a
complemen , Figu e 42 allows us o isualize he dis ibu ion o he pla o m ypes
a ailable o each o he su ace le els. The e o e we can deduce ha as he a ea in
he su ace le els dec eases, he numbe o a ailable pla o ms also dec eases,
especially he la ge ones such as Ve ihubs and Ve ipo s.
Figu e 40. Roo 5 (a) Lida image wi h main su ace, (b) Sa elli e image, (c) 3D Scene
51
Figu e 41. Map dis ibu ion o UAM Pla o ms ypology using he main su ace le el
Figu e 42. Wa le-cha o UAM ypology dis ibu ion a di e en su ace le els
UAV
Ve is op
Ve ipo *
Ve ihub*
No Use ul
Each squa e ep esen s 100 oo ops.
*Ve ipo s in su aces 2 and 3 depic
less han 100 oo s.
*Ve ihubs in su aces 3,4 and 5 depic
less han 100 oo s.
52
An addi ional and deepe analysis leads o he e alua ion o he oo op la -su aces
o al po en ial o he de elopmen o UAM pla o ms. Thus, when conside ing he
i e main su aces, Table 10 shows ha 19.993 (44%) oo s could gene a e 99.965 UAM
spo s. Thus, o all o Manha an, 45.461 oo s could ha e he po en ial o gene a e
167.262 places o landing and clea ing o ehicles in he UAM sys em.
Numbe o la Su aces
use ul o UAM
Pla o ms
Numbe o
Roo ops % To al Po en ial Hubs
Loca ions
1 3294 7.2 3294
2 8443 18.6 16886
3 7807 17.2 23421
4 5924 13.0 23696
5 19993 44.0 99965
To al 45461 100 167262
Table 10. To al po en iali y o he oo ops by coun ing he usable su aces a ailable
4.2.3 Roo op Sui abili y Analysis
The esul s o he las s ep o his me hodology was he calcula ion o he anges o
sui abili y le els o he main su ace le el. Due o 85% o he main su aces o all
Manha an oo s occupy mo e han 50% o he a ea in hei co esponding bounding-
boxes (ancilla y ac o ), sui abili y was es ima ed only conside ing he elonga ion
(compac ness ac o ). Subsequen ly, he alues gi en by he elonga ion ac o we e
ca ego ized in o qua iles o each o he pla o m ypes selec ed in his in es iga ion.
The maps Figu e 43, Figu e 44, Figu e 45 and Figu e 46 allow us o obse e he spa ial
dis ibu ion and inal sui abili y o each o he pla o m ca ego ies acco ding o he
main su ace. Figu e 47 shows ha independen o he numbe o oo s a ailable o
each ypology, High and Medium/High sui abili y indexes a e ai ly dis ibu ed in
each UAM pla o m ype. Ne e heless, Medium/High and Medium/Low indexes
show a g ea e numbe o oo s o each o he UAM pla o m ypologies (Table 11).
In gene al e ms o all Manha an, 16.5% (7.512) o he oo s e lec a High sui abili y,
36.5% (16.608) Medium/High, 44.6% (20.286) Medium/Low and he emaining 2.3%
(1.055) a Low sui abili y. Conce ning he UAM ype, he UAV pla o ms and
Ve is ops ha e he highes numbe o oo s a ailable in all sui abili y indexes.
53
Figu e 43. Sui abili y Index map o UAV pla o ms
Figu e 44. Sui abili y index map o Ve is ops
54
Figu e 45. Sui abili y index map o Ve ipo s
Figu e 46. Sui abili y index map o Ve ihubs

55
Figu e 47. Dis ibu ion o he sui abili y index le els o each UAM pla o m ype
Numbe o oo ops pe Sui abili y Index
UAM Hub
Type
High Medium/High Medium/Low Low To al
% pe UAM
Type
UAV 4357 11411 16637 665 33070 72.7
Ve is op 2982 4930 3392 327 11631 25.6
Ve ipo 163 248 237 55 703 1.5
Ve ihub 10 19 20 8 57 0.1
To al 7512 16608 20286 1055 45461 100
% pe Index 16.5 36.5 44.6 2.3 100
Table 11. Summa y able showing he dis ibu ion o oo ops in each oo sui abili y index
4.3 Final sui abili y ca ego iza ion o UAM Pla o ms
This sec ion o he in es iga ion shows he esul s ob ained a e he combina ion o
place-sui abili y le els o he census blocks ob ained om GeoSOM clus e s and
oo op-sui abili y le els de ined a e unning he oo op- la ness assessmen . The
ac ha he oo s o he buildings only belong o a census block gua an ees ha each
oo will ha e a unique classi ica ion o o al sui abili y. The e o e, when o e lapping
maps does no gene a e unce ain ies in he esul s. Howe e , 76 oo s we e no
o e lapped wi h he laye o Manha an census blocks, as mos o hem a e in a eas
belonging o Cen al Pa k and some islands whe e he e is no co e age by he laye
o census blocks. So, hese we e no conside ed in he inal s a is ics.
56
Figu e 48. Combined sui abili y aking he c i e ia o place and la ness o he main su ace
on he oo s
Figu e 48 depic s he spa ial dis ibu ion o he 12 combined le els o sui abili y, whe e
he i s sui abili y ca ego y e e s o place ea u es and he second one o he oo op
la ness. Roo s wi h High/High and High/Medium-high combina ions a e
geog aphically dis ibu ed h oughou he s udy a ea bu end o be mo e inland
Manha an. Howe e , nea hese a eas, he e a e also oo s wi h low sui abili y o
su aces o landing and clea ance o UAM ehicles bu wi h high place sui abili y
57
ep esen ed by he High/Low le el. On he o he hand, Medium/High and
Medium/Medium-high ca ego ies a e also sca e ed all o e he s udy a ea bu show
concen a ions bo h nea he sho eline and in cen al a eas o he bo ough. These
a eas a e o g ea ele ance since hey can be aken as a second p io i y o he u ban
planning o UAM hubs. I also s ands ou ha some oo s in he no he n, Down own,
and especially in Mid own Manha an ha e sui able su aces o he se lemen o
UAM pla o ms, bu u ban ecosys em condi ions do no show an app op ia e place.
Figu e 49. To al o oo ops dis ibu ed o each combined sui abili y class
Roo op Sui abili y Indexes
High Medium-
high
Medium-
low Low To al % Place
Sui .
Place
Sui abili y
Indexes
High 2541 5864 7080 311 15796 34.8
Medium 2752 5937 6877 385 15951 35.1
Low 2204 4784 6302 348 13638 30.1
To al 7497 16585 20259 1044 45385 100
% Roo Sui . 16.5 36.5 44.6 2.3 100
Table 12. Dis ibu ion o he oo ops in he inal ca ego iza ion o sui abili y o UAM
pla o ms
I is also in e es ing o obse e how he numbe o oo s is dis ibu ed e y simila ly
in he di e en le els o in eg al sui abili y, pa icula ly o he medium sui abili y
ca ego ies (Figu e 49). Fu he mo e, he e a e abou 2.500 (5.6%) oo s wi h High
sui abili y o bo h c i e ia and less han 350 (0.8%) wi h Low sui abili y (Table 12).
58
The ca ego y wi h he mos oo s is he one gi en by High place sui abili y and
Medium/Low oo sui abili y wi h 6.877 (15.6%) oo ops (Table 12). Acco ding o he
pla o m ype, UAVs and Ve is ops show a g ea e numbe o app op ia e oo s wi h
a combined High/High sui abili y, bu Ve ihubs does no show any oo classi ied as
High/High (Table B- 1– Annex 2, Figu e 50).
Figu e 50. P opo ions o combined sui abili y le els pe UAM hub ypes
Rega ding a clus e -o ien ed analysis, combined sui abili y a ios show ha he
High/High ca ego y is dis ibu ed h oughou he GeoSOM clus e s. Clus e 1 is he
one ha ing he highes p opo ion o his in eg al sui abili y le el (Figu e 51). Howe e ,
when being la ge , Clus e 5 and 6 show a g ea e numbe o oo s wi h his ca ego y
(Table B- 2– Annex 2). Ca ego ies wi h medium sui abili y o p io i iza ion in he
cons uc ion o UAM hubs (Medium/High and Medium/Medium-High) show a ai
dis ibu ion ac oss all clus e s, bu Clus e 8 shows lowe p opo ions o hese
ca ego ies (Figu e 51). Besides, Clus e s 6 and 8 a e he ones ha concen a e oo s
wi h low sui abili y o bo h place and oo -su ace o he de elopmen o UAM
pla o ms (Table B- 2– Annex 2). As a complemen a y analysis, amalgama ed da a
also allows us o obse e ha he ype o pla o ms is ai ly dis ibu ed o each o he
GeoSOM clus e s (Figu e 52). None heless, he only ones ha can de elop Ve ihubs
a e Clus e s 2, 4 and 7. Finally, i is also possible o connec hese esul s wi h he
disc imina ed desc ip ions o each clus e gene a ed in Chap e 4.1.5. Thus, po en ial
UAM se ices desc ibed can be ca ied ou by e idencing a ailable oo su aces o
di e en ypes o UAM hubs.
65
Swi ze land) 18, 11. Re ie ed Sep embe 3, 2019 om DOI:
h ps://doi.o g/10.3390/s18113960.
[12] CHEN, D., ZHANG, L., LI, J., and LIU, R. 2012. U ban building oo
segmen a ion om ai bo ne lida poin clouds. In e na ional Jou nal o Remo e
Sensing 33, 20, 6497–6515. Re ie ed No embe 28, 2019 om DOI:
h ps://doi.o g/10.1080/01431161.2012.690083.
[13] CHEN, J., YANG, S., LI, H., ZHANG, B., and LV, J. 2013. Resea ch on
Geog aphical En i onmen Uni Di ision Based on he Me hod o Na u al
B eaks (Jenks). In . A ch. Pho og amm. Remo e Sens. Spa ial In . Sci. XL-
4/W3, 47–50. Re ie ed Decembe 12, 2019 om DOI:
h ps://doi.o g/10.5194/isp sa chi es-XL-4-W3-47-2013.
[14] CZIOSKA, P., MATTFELD, D., and SESTER, M. 2017. GIS-based
iden i ica ion and assessmen o sui able mee ing poin loca ions o ide-sha ing.
T anspo a ion Resea ch P ocedia 22, 314–324. Re ie ed Oc obe 10, 2019
om DOI: h ps://doi.o g/10.1016/j. p o.2017.03.038.
[15] FADHIL, D. 2018. A GIS-Based Analysis o Selec ing G ound In as uc u e
Loca ions o U ban Ai Mobili y. Mas e ’s Thesis, Munich, 2018. Mas e .
Technical Uni e si y o Munich.
[16] FIEDLER, D., CAP, M., and CERTICKY, M. 2017. Impac o mobili y-on-
demand on a ic conges ion: Simula ion-based s udy. In IEEE ITSC 2017. 20 h
In e na ional Con e ence on In elligen T anspo a ion Sys ems : Mielpa que
Yokohama in Yokohama, Kanagawa, Japan, Oc obe 16-19, 2017. IEEE,
Pisca away, NJ, 1–6. Re ie ed Sep embe 15, 2019 om DOI:
h ps://doi.o g/10.1109/ITSC.2017.8317830.
[17] FIRDHOUS, M., BASNAYAKE, D., KODITHUWAKKU, K., HATTHALLA,
N., CHARLIN, N., and BANDARA, P. 2010. Rou e Ad ising in a Dynamic
En i onmen – A High-Tech App oach. Inno a ions in Compu ing Sciences and
So wa e Enginee ing.
[18] FOX, A. 2019. Ube and Ly may be making San F ancisco’s a ic wo se.
Science. Re ie ed Oc obe 18, 2019 om DOI:
h ps://doi.o g/10.1126/science.aax9717.
[19] FU, M., ROTHFELD, R., and ANTONIOU, C. 2019. Explo ing P e e ences o
T anspo a ion Modes in an U ban Ai Mobili y En i onmen : Munich Case
S udy. T anspo a ion Resea ch Reco d 7, 036119811984385. Re ie ed
Sep embe 27, 2019 om DOI: h ps://doi.o g/10.1177/0361198119843858.
[20] GIANQUINTIERI, L., CAIANI, E. G., BRAMBILLA, P., PAGLIOSA, A.,
VILLA, G. F., and BROVELLI, M. A. 2019. Open Da a in Heal h-Geoma ics:
Mapping and E alua ing Publicly Accessible De ib illa o s. In . A ch.
Pho og amm. Remo e Sens. Spa ial In . Sci. XLII-4/W14, 63–70. Re ie ed
No embe 10 , 2019 om DOI: h ps://doi.o g/10.5194/isp s-a chi es-XLII-4-W14-
63-2019.

66
[21] HARRIS, C. 1964. A scien i ic me hod o dis ic ing. Beha io al science 9, 3,
219–225. Re ie ed Decembe 8, 2019 om DOI:
h ps://doi.o g/10.1002/bs.3830090303.
[22] HENRIQUES, R. 2010. A i icial In elligence in Geospa ial Analysis:
applica ions o Sel -O ganizing Maps in he con ex o Geog aphic In o ma ion
Science. PhD Thesis. Ins i u o Supe io de Es a ís ica de Lisboa, Uni e sidade
No a de Lisboa, Lisboa.
[23] HENRIQUES, R., BACAO, F., and LOBO, V. 2010. Spa ial Clus e ing wi h
SOM and GeoSOM: Case S udy o Lisbon's Me opoli an A ea. In
GEOP ocessing 2010. P oceedings, he Second In e na ional Con e ence on
Ad anced Geog aphic In o ma ion Sys ems, Applica ions, and Se ices : 10-16
Feb ua y, 2010, S . Maa en, Ne he lands An illes. IEEE Compu e Socie y, Los
Alami os, Cali ., 148–152. Re ie ed No embe 29, 2019 om DOI:
h ps://doi.o g/10.1109/GEOP ocessing.2010.30.
[24] HENRIQUES, R., BACAO, F., and LOBO, V. 2012. Explo a o y geospa ial
da a analysis using he GeoSOM sui e. Compu e s, En i onmen and U ban
Sys ems 36, 3, 218–232. Re ie ed No embe 24, 2019 om DOI:
h ps://doi.o g/10.1016/j.compen u bsys.2011.11.003.
[25] HILDEMANN, M. and DELGADO, C. 2019. An adap able and scalable leas
cos ne wo k o ai - axis in u ban a eas S udy a ea: Manha an, New Yo k. In:
Ky iakidis, P., Hadjimi sis, D., Ska la os, D. & Mansou ian, A. (Eds.), 2019.
Accep ed Sho Pape s and Pos e s om he 22nd AGILE Con e ence on Geo-
in o ma ion Science. Cyp us Uni e si y o Technology 17-20 June 2019,
Limassol, Cyp us. ISBN 978-90-816960-9-8.
[26] HIMBERG, J. 2000. A SOM based clus e isualiza ion and i s applica ion o
alse colo ing. In In e na ional join con e ence on neu al ne wo ks. IEEE, 587-
592 ol.3. Re ie ed Decembe 28, 2019 om DOI:
h ps://doi.o g/10.1109/IJCNN.2000.861379.
[27] HOLDEN, J. and GOEL, N. 2018. Ube Ele a e. Fas -Fo wa ding o a Fu u e
o On-Demand U ban Ai T anspo a ion. Ube . Re ie ed Sep embe 27, 2019
om h ps://www.ube .com/ele a e.pd .
[28] HOLMES, B., PARKER, R., STANLEY, D., MCHUGH, P., GARROW, L.,
and MASSON, P. 2017. NASA S a egic F amewo k o On-Demand Ai
Mobili y. A Repo o NASA Headqua e s Ae onau ics Resea ch Mission
Di ec o a e (2017).
[29] INRIX. 2019. Global T a ic Sco eca d. INRIX Resea ch (2019). Re ie ed
Sep embe 27, 2019 om h ps://in ix.com/sco eca d/.
[30] JUNG, S. and HYUNSU, K. 2017. Analysis o Amazon P ime Ai UAV Deli e y
Se ice. Jou nal o Knowledge In o ma ion Technology and Sys ems 12, 2, 253–
266. Re ie ed Oc obe 15, 2019 om DOI:
h ps://doi.o g/10.34163/jki s.2017.12.2.005.
[31] KASKI, S., NIKKILÄ, J., KOHONEN, T., Ed. 1998. Me hods o in e p e ing
a sel -o ganized map in da a analysis.
67
[32] KOHONEN, T. 1982. Sel -o ganized o ma ion o opologically co ec ea u e
maps. Biological Cybe ne ics 43, 1, 59–69. Re ie ed No embe 27, 2019 om
DOI: h ps://doi.o g/10.1007/BF00337288.
[33] KOHONEN, T. 2013. Essen ials o he sel -o ganizing map. Neu al Ne wo ks
37, 52–65. Re ie ed No embe 16 , 2019 om DOI:
h ps://doi.o g/10.1016/j.neune .2012.09.018.
[34] LI, L., DU, Q., REN, F., and MA, X. 2019. Assessing Spa ial Accessibili y o
Hie a chical U ban Pa ks by Mul i-Types o T a el Dis ance in Shenzhen,
China. In e na ional jou nal o en i onmen al esea ch and public heal h 16, 6.
Re ie ed Oc obe 15, 2019 om h ps://doi.o g/10.3390/ije ph16061038.
[35] LILIUM AVIATION. 2019. Mission: Ai c a o e e yone (2019). Re ie ed
om h ps://lilium.com/ou -belie s.
[36] LIU, R., LIU, H., KWAK, D., XIANG, Y., BORCEA, C., NATH, B., and
IFTODE, L. 2016. Balanced a ic ou ing: Design, implemen a ion, and
e alua ion. Ad Hoc Ne wo ks 37, 14–28. Re ie ed Sep embe 29, 2019 om
DOI: h ps://doi.o g/10.1016/j.adhoc.2015.09.001.
[37] LOURENCO, J. 2009. Visualiza ion o Clus e s in Geo- e e enced Da a Using
Th ee-dimen ional Sel -O ganizing Maps. Mas e Thesis. Uni e sidade No a de
Lisboa, Lisboa.
[38] MENEGUETTE, R., DE GRANDE, R., and LOUREIRO, A. 2018. In elligen
T anspo Sys em in Sma Ci ies: Aspec s and Challenges o Vehicula
Ne wo ks and Cloud. Sp inge In e na ional Publishing.
[39] MOHAPATRA, S. and BHUYAN, P. 2012. Sel O ganizing Map o A i icial
Neu al Ne wo k o De ining Le el o Se ice C i e ia o U ban S ee s. IJTTE
2, 3, 236–252. Re ie ed No embe 21, 2019 om DOI:
h ps://doi.o g/10.7708/ij e.2012.2(3).06.
[40] MOORE, M. 2006. The Thi d Wa e o Ae onau ics: On-Demand Mobili y. In
NASA Pe sonal Ai T anspo a ion Technologies. SAE Technical Pape Se ies.
SAE In e na ional400 Commonweal h D i e, Wa endale, PA, Uni ed S a es.
Re ie ed Sep embe 22, 2019 om DOI: h ps://doi.o g/10.4271/2006-01-2429.
[41] MOORE, M., Ed. 2006. NASA Pe sonal Ai T anspo a ion Technologies. SAE
Technical Pape Se ies. SAE In e na ional400 Commonweal h D i e,
Wa endale, PA, Uni ed S a es. Re ie ed Sep embe 22, 2019 om DOI:
h ps://doi.o g/10.4271/2006-01-2413.
[42] MOOSAVI, V. 2017. Con ex ual mapping: Visualiza ion o high-dimensional
spa ial pa e ns in a single geo-map. Compu e s, En i onmen and U ban
Sys ems 61, 1–12. Re ie ed No embe 14, 2019 om DOI:
h ps://doi.o g/10.1016/j.compen u bsys.2016.08.005.
[43] NATIONAL AERONAUTICS AND SPACE ADMINISTRATION. 2018.
NASA UAM Ma ke S udy – Final Repo Re ie ed Sep embe 30, 2019 om
h ps://n s.nasa.go /a chi e/nasa/casi.n s.nasa.go /20190001472.pd .
[44] NATIONAL AERONAUTICS AND SPACE ADMINISTRATION. 2018.
U ban Ai Mobili y Ai space In eg a ion Concep s and Conside a ions. (2018).
68
Re ie ed July 19, 2019 om
h ps://n s.nasa.go /a chi e/nasa/casi.n s.nasa.go /20180005218.pd .
[45] NÜST, D., GRANELL, C., SILERYTE, R., HOFER, B., and OSTERMANN,
F. 2019. Rep oducibili y o GIScience MSc Theses - Sel -assessmen expe imen .
OSF Re ie ed Feb ua y 10, 2020 om DOI: h ps://doi.o g/10.17605/os .io/j97zp.
[46] O'HEAR, S. 2019. Lilium un eils i e-sea e ai axi p o o ype a e a success ul
maiden ligh o i s la es je (2019). Re ie ed Sep embe 27, 2019 om
h ps:// echc unch.com/2019/05/15/lilium-5-sea e /.
[47] ONYENEKE, C. 2018. Modeling he E ec s o T a ic Conges ion on Economic
Ac i i ies - Acciden s, Fa ali ies and Casual ies. BSI 3, 2, 7. Re ie ed
Sep embe 29, 2019 om DOI: h ps://doi.o g/10.11648/j.bsi.20180302.11.
[48] PORSCHE CONSULTING. 2018. The Fu u e o Ve ical Mobili y. Sizing he
ma ke o passenge , inspec ion, and goods se ices un il 2035 (2018).
Re ie ed Sep embe 27, 2019 om h ps://www.po sche-
consul ing.com/en/p ess/insigh s/de ail/s udy- he- u u e-o - e ical-mobili y/.
[49] RODRIGUE, J-P., COMTOIS, C., and SLACK, B. 2017. The geog aphy o
anspo sys ems (4 h edi ion). Rou ledge Taylo & F ancis G oup, London,
New Yo k ISBN: 978-1-31-561815-9.
[50] RURAL AGRICULTURAL PLANNING UNIT. 2018. UPRA Geop ocessing
op imized ools. Gi hub Online Reposi o y (2018). Re ie ed om
h ps://gi hub.com/Up aAnalisis.
[51] SALONEN, M. and TOIVONEN, T. 2013. Modelling a el ime in u ban
ne wo ks: compa able measu es o p i a e ca and public anspo . Jou nal o
T anspo Geog aphy 31, 143–153. Re ie ed Augus 21, 2019 om DOI:
h ps://doi.o g/10.1016/j.j angeo.2013.06.011.
[52] S SCHNELL, J., NAIK, V., HOROWITZ, L., PAULOT, F., GINOUX, P.,
ZHAO, M., and HORTON, D. 2019. Ai quali y impac s om he elec i ica ion
o ligh -du y passenge ehicles in he Uni ed S a es. A mosphe ic En i onmen
208, 95–102. Re ie ed Sep embe 21, 2019 DOI:
h ps://doi.o g/10.1016/j.a mosen .2019.04.003.
[53] SEELEY, B. 2017. Regional Sky T ansi IV: Pocke Ai pa k Design
Cons ain s. In 17 h AIAA A ia ion Technology, In eg a ion, and Ope a ions
Con e ence. Ame ican Ins i u e o Ae onau ics and As onau ics, Res on,
Vi ginia, 2017. Re ie ed Oc obe 11, 2019 DOI: h ps://doi.o g/10.2514/6.2017-
3440.
[54] SHAN, J. and SAMPATH, A. 2009. Building Ex ac ion om LiDAR Poin
Clouds Based on Clus e ing Techniques. In Topog aphic lase anging and
scanning. P inciples and p ocessing, Jie Shan and Cha les K. To h, Eds. CRC,
Boca Ra on, 421–444. Re ie ed Sep embe 20, 2019 DOI:
h ps://doi.o g/10.1201/9781420051438.ch15.
[55] SUASNEWS. 2017. NASA Emb aces U ban Ai Mobili y (UAM), Calls o
Ma ke S udy (2017). Re ie ed Sep embe 26, 2019 om
69
h ps://www.suasnews.com/2017/11/nasa-emb aces-u ban-ai -mobili y-uam-calls-
ma ke -s udy/.
[56] THIPPHAVONG, D., APAZA, R., BARMORE, B., BATTISTE, V., BURIAN,
B., DAO, Q., FEARY, M., GO, S., GOODRICH, K., HOMOLA, J., IDRIS, H.,
KOPARDEKAR, P., LACHTER, J., NEOGI, N., NG, H., OSEGUERA-LOHR,
R., PATTERSON, M., and VERMA, S. 2018. U ban Ai Mobili y Ai space
In eg a ion Concep s and Conside a ions. In 2018 A ia ion Technology,
In eg a ion, and Ope a ions Con e ence, June 25-29, 2018, A lan a, Geo gia.
AIAA AVIATION Fo um. Ame ican Ins i u e o Ae onau ics and As onau ics,
[Res on (VA)], 2018. Re ie ed Oc obe 21, 2019 DOI:
h ps://doi.o g/10.2514/6.2018-3676.
[57] TOBLER, W. 1970. A Compu e Mo ie Simula ing U ban G ow h in he
De oi Region. Economic Geog aphy 46, 234. Re ie ed No embe 21, 2019
DOI: h ps://doi.o g/10.2307/143141.
[58] ULTSCH, A. and SIEMON, H., Eds. 1990. Kohonen's sel o ganizing ea u e
maps o explo a o y da a analysis. Kluwe Academic p ess.
[59] VASCIK, P., BALAKRISHNAN, H., and HANSMAN, R., Eds. 2018.
Assessmen o Ai T a ic Con ol o U ban Ai Mobili y and Unmanned
Sys ems.
[60] VASCIK, P. and HANSMAN, R. 2018. Scaling Cons ain s o U ban Ai
Mobili y Ope a ions: Ai T a ic Con ol, G ound In as uc u e, and Noise. In
2018 A ia ion Technology, In eg a ion, and Ope a ions Con e ence, June 25-
29, 2018, A lan a, Geo gia. AIAA AVIATION Fo um. Ame ican Ins i u e o
Ae onau ics and As onau ics, [Res on (VA)], 413. Re ie ed Sep embe 19,
2019 DOI: h ps://doi.o g/10.2514/6.2018-3849.
[61] VENUGOPAL, V. and KANNAN, S. 2013. Accele a ing eal- ime LiDAR da a
p ocessing using GPUs. In 2013 IEEE 56 h In e na ional Midwes Symposium
on Ci cui s and Sys ems (MWSCAS). IEEE, Pisca away, 1168–1171. Re ie ed
Decembe 5, 2019 DOI: h ps://doi.o g/10.1109/MWSCAS.2013.6674861.
[62] VESANTO, J. 1999. SOM-based da a isualiza ion me hods. In elligen Da a
Analysis 3, 2, 111–126. Re ie ed No embe 17, 2019 DOI:
h ps://doi.o g/10.1016/S1088-467X(99)00013-X.
[63] VIEIRA, C. and LUNA, M. 2016. Models and Me hods o Logis ics Hub
Loca ion: A Re iew Towa ds T anspo a ion Ne wo ks Design. Pesqui. Ope .
36, 2, 375–397. Re ie ed Sep embe 12, 2019 DOI: h ps://doi.o g/10.1590/0101-
7438.2016.036.02.0375.
[64] VOLOCOPTER. 2019. Pionee ing he U ban Ai Taxi Re olu ion. Whi e
Pape (2019). Re ie ed Sep embe 27, 2019 om
h ps://p ess. olocop e .com/images/pd /Volocop e -Whi ePape -1-0.pd .
[65] WU, T., HU, X., and YE, L. 2016. Fas and Accu a e Plane Segmen a ion o
Ai bo ne LiDAR Poin Cloud Using C oss-Line Elemen s. Remo e Sensing 8, 5,
383. Re ie ed Oc obe 9, 2019 DOI: h ps://doi.o g/10.3390/ s8050383.
70
[66] YU, W. 2016. Spa ial co-loca ion pa e n mining o loca ion-based se ices in
oad ne wo ks. Expe Sys ems wi h Applica ions 46, 324–335. Re ie ed
No embe 29, 2019 DOI: h ps://doi.o g/10.1016/j.eswa.2015.10.010.

71
Annex 1
Id
A em p
X
Y
La ice
No maliza ion
Type
I e a ions
(Rough)
Radio
(Rough)
Alpha
(Rough)
I e a ions
(Fine une)
Radio
(Fine une)
Alpha
(Fine une)
Q
E o
T E o
1
10
1
Hex
Z_Sco e
10000
4
0.3
20000
2
0.1
4.62871
0.03248
2
5
5
Hex
Z_Sco e
2500
5
0.3
5000
3
0.1
4.3898
0.014787
3
5
5
Rec
Z_Sco e
2500
5
0.3
5000
3
0.1
4.36213
0.028783
4
10
5
Hex
Z_Sco e
2500
5
0.3
5000
3
0.1
4.12333
0.02403
5
10
5
Hex
Z_Sco e
8000
8
0.4
16000
4
0.2
4.0094
0.025614
6
5
10
Hex
Z_Sco e
3000
5
0.3
6000
3
0.1
3.99226
0.017428
7
10
10
Hex
Z_Sco e
6000
5
0.3
12000
3
0.1
3.61997
0.024229
8
15
10
Hex
Z_Sco e
10000
5
0.3
20000
3
0.1
3.36671
0.019277
9
10
15
Hex
Z_Sco e
10000
7
0.4
20000
4
0.2
3.36263
0.016899
10
15
10
Hex
Z_Sco e
10000
10
0.5
20000
5
0.2
3.32511
0.016372
11
10
1
Hex
Min-Max
10000
4
0.3
20000
2
0.1
0.91351
0.02403
12
5
5
Hex
Min-Max
2500
5
0.3
5000
3
0.1
0.861302
0.004489
13
5
5
Rec
Min-Max
2500
5
0.3
5000
3
0.1
0.846135
0.022181
14
10
5
Hex
Min-Max
2500
6
0.3
5000
3
0.1
0.760353
0.015316
15
5
10
Rec
Min-Max
3000
5
0.3
6000
3
0.1
0.757865
0.007394
16
10
5
Hex
Min-Max
8000
8
0.4
16000
4
0.2
0.757237
0.011883
17
10
10
Rec
Min-Max
6000
5
0.3
12000
3
0.1
0.664587
0.00977
18
10
15
Hex
Min-Max
10000
7
0.4
20000
4
0.2
0.61566
0.015844
19
15
10
Rec
Min-Max
10000
10
0.5
20000
5
0.2
0.605454
0.016108
20
15
10
Hex
Min-Max
10000
5
0.3
20000
3
0.1
0.600095
0.015844
Table A- 1. Di e en pa ame e se ings used o he sensi i i y analysis o he SOM algo i hm execu ion
72
Figu e A- 1. Componen Planes om SOM algo i hm pa 1
73
Figu e A- 2. Componen Planes om SOM algo i hm pa 2
74
Figu e A- 3. Componen Planes om GeoSOM algo i hm pa 1
81
else:
chain_subp ocess += "%s = subp ocess.Popen(aCommands[%s],
s din=None,s dou =subp ocess.PIPE,shell=T ue)"%("ch"+s (j+1), j)
chain_wai += "as dou , as de =
%s.communica e()"%("ch"+s (j+1))
p in (chain_subp ocess)
p in (chain_wai )
chain_subp ocess = compile(chain_subp ocess, '<s ing>', 'exec')
exec(chain_subp ocess)
chain_wai = compile(chain_wai , '<s ing>', 'exec')
exec(chain_wai )
p in ("Ending pa allel p ocessing...")
p in (da e ime.da e ime.now())
p in ("Pa allel p ocess execu ed in " + s ((( ime.clock() -
ini ial_ ))/60))
i __name__ == '__main__':
main()
#---------------------------------------------------------------------------
----
# Name: Ini ial_Roo op_S a is ics_Calcula ion.py
# Pu pose: Calcula ion o he main s a is ics pe oo in he s udy a ea
#
# Au ho : Ca los Ja ie Delgado
#
# C ea ed: 14/10/2019
# Copy igh :
# Licence:
#---------------------------------------------------------------------------
----
#Impo ing lib a ies implemen ed
impo a cpy
impo pandas as pd
impo os
impo da e ime
impo sys
a cpy.en .o e w i eOu pu = T ue
#Main unc ion in he calcula ion p ocess
de main(i e a ion_p):
p in ("S a ing P ocess %s"%(i e a ion_p))
p in (da e ime.da e ime.now())
#Main pa hs o sa ing in o ma ion
olde LAS iles =
'D: Geo_Tech_Mas e Thesis_Resea ch Lida _Da a USGS_NYC2014'
shpRoo ops =
'D: Geo_Tech_Mas e Thesis_Resea ch P ocessing Inpu s Sui abili y_Analysis_
Inpu s.gdb buildings'
d a GDB =
'D: Geo_Tech_Mas e Thesis_Resea ch P ocessing es _lida _ esul s d a .gdb
'
olde LIDARC opped =
"D: Geo_Tech_Mas e Thesis_Resea ch P ocessing Lida _C opped_%s"%(i e a ion
_p)

82
olde BK_lasC opped =
'D: Geo_Tech_Mas e Thesis_Resea ch P ocessing LAS_C opped_%s'%(i e a ion_p
)
gdbFil e ed3dpoin s =
"D: Geo_Tech_Mas e Thesis_Resea ch P ocessing es _lida _ esul s Fil e ed_
3dpoin s_%s.gdb"%(i e a ion_p)
gdbSingle3dpoin s = "in_memo y"
gdbLida Vec o = "in_memo y"
gdbOu lie s = "in_memo y"
c ea e_Folde ( olde LIDARC opped)
c ea e_Folde ( olde BK_lasC opped)
c ea e_gdb
#Func ion o build he LIDAR as Fea u e Class only is execu ed in he
i s i e a ion
i i e a ion_p == 0:
g idL = buildLIDARg id( olde LAS iles, d a GDB)
#g idL = os.pa h.join(d a GDB, "lida _g id")
#Calling he unc ion ha build he oo op indexes and hen he
s a is cs
indexingRoo ops(shpRoo ops, g idL, d a GDB, olde LAS iles,
olde LIDARC opped,
olde BK_lasC opped, gdbOu lie s, gdbLida Vec o ,
gdbSingle3dpoin s, gdbFil e ed3dpoin s, i e a ion_p)
p in ("Ending P ocess 1")
p in (da e ime.da e ime.now())
de c ea e_Folde (pa h_c ea e):
os.mkdi (pa h_c ea e)
p in ("Folde C ea ed")
de c ea e_gdb(pa h_comple e):
a cpy.C ea eFileGDB_managemen (os.pa h.di name(pa h_comple e),
os.pa h.basename(pa h_comple e))
p in ("GDB C ea ed")
#Func ion ha build he LIDAR g id o managemen
de buildLIDARg id( oldLAS iles, pa hGDB):
lis LAS iles = lis LIDAR( oldLAS iles)
p in (len(lis LAS iles))
lag1 = 0
o l in lis LAS iles:
p in ( lag1)
ileLAS = os.pa h.join( oldLAS iles, l )
s LAS = a cpy.Desc ibe( ileLAS).spa ialRe e ence
i lag1 == 0:
a cpy.C ea eFea u eclass_managemen (pa hGDB, "lida _g id",
"POLYGON", spa ial_ e e ence=s LAS)
pa hLAS c = os.pa h.join(pa hGDB, "lida _g id")
a cpy.AddField_managemen (pa hLAS c, "al _id", "TEXT")
nameLAS ile = os.pa h.spli ex (os.pa h.basename( ileLAS))[0]
#Ge ing he en elope o each lida ile
desc = a cpy.Desc ibe( ileLAS)
xmin = desc.ex en .XMin
ymin = desc.ex en .YMin
xmax = desc.ex en .XMax
ymax = desc.ex en .YMax
83
geoEn elope = a cpy.A ay([a cpy.Poin (xmin, ymin),
a cpy.Poin (xmax, ymin),
a cpy.Poin (xmax, ymax),
a cpy.Poin (xmin, ymax)
])
polygonEne elope = a cpy.Polygon(geoEn elope)
cu so = a cpy.da.Inse Cu so (pa hLAS c,['al _id', 'SHAPE@'])
cu so .inse Row([nameLAS ile, polygonEne elope])
lag1 += 1
e u n pa hLAS c
#Func ion ha lis he .LAS iles
de lis LIDAR(pa hDi ec o y):
e u n [k o k in os.lis di (pa hDi ec o y) i k.endswi h('.las')]
#Co e unc ion o oo op indexing and la e call o s a is ics
de indexingRoo ops( oo Buildings, g idF, pa hGDB, olde O iLIDAR,
olde C op_LIDAR, olde bklas, old_Ou lie s,
gdbV_lida , gdb_Single3dP, gdb_Fil e ed3dP, i e a ion):
ou pu In e g id = os.pa h.join(pa hGDB, "in e _buildings_g id")
ou pu SumTableg id = os.pa h.join(pa hGDB, "summa yTable_building_g id")
ou pu _ able_ e i y = os.pa h.join( "in_memo y",
" able_ e i y_duplica ed_" + i e a ion)
ou pu _ abled_duplica es = os.pa h.join( "in_memo y",
" able_duplica es_" + i e a ion)
a cpy.analysis.In e sec ([g idF, oo Buildings], ou pu In e g id, "ALL",
None, "INPUT")
#C ea ing he combined able o he buildings and he co esponding ile
o he g id
a cpy.analysis.S a is ics(ou pu In e g id, ou pu SumTableg id, "al _id
COUNT", "FID_buildings;FID_lida _g id;al _id")
a cpy.analysis.S a is ics(ou pu SumTableg id, ou pu _ able_ e i y,
"FID_buildings COUNT", "FID_buildings")
a cpy.analysis.TableSelec (ou pu _ able_ e i y,
ou pu _ abled_duplica es, "FREQUENCY > 1")
#C ea ing pandas d only o oo ops ha sha e mo e han one LIDAR
ile
pdID_Duplica es = cToPandasDF(ou pu _ abled_duplica es,
["FID_buildings", "FREQUENCY"])
l_buildings = " l_buildings_%s"%(i e a ion)
a cpy.managemen .MakeFea u eLaye ( oo Buildings, l_buildings)
coun = 0
pdDF_S a is icspe Building = ""
lag_duplica ed = 0
a ay_pa hs_duplica ed = []
lp = 0
#Cu so o i e a e each o he oo ops and ex ac he lida
in o ma ion
wi h a cpy.da.Sea chCu so (ou pu SumTableg id, ["FID_buildings",
"FID_lida _g id", "al _id", "OBJECTID"]) as cu so :
p in ("P ocessing ex ac ion...")
o ow in cu so :
84
p in (coun )
p in ( ow[0])
a cpy.managemen .Selec Laye ByA ibu e( l_buildings,
"NEW_SELECTION", "OBJECTID = " + s ( ow[0]), None)
lasds = os.pa h.join( olde C op_LIDAR, "C_" + s ( ow[0]) + "_"
+ s ( ow[2]) + ".lasd")
a cpy.ddd.Ex ac Las(os.pa h.join( olde O iLIDAR,
ow[2])+".las", olde bklas, "DEFAULT",
l_buildings, "PROCESS_EXTENT", "_R" +
s ( ow[0]) , "MAINTAIN_VLR", "REARRANGE_POINTS",
"NO_COMPUTE_STATS", lasds)
ou lie 3dP_building = indOu lie s( old_Ou lie s, lasds, "O_" +
s ( ow[0]) + "_" + s ( ow[2]))
mul ipoin _las = lasToVec o Poin (os.pa h.join( olde bklas,
s ( ow[2])+ "_R" + s ( ow[0]) + ".las"),
gdbV_lida , "V_" + s ( ow[0]) +
"_" + s ( ow[2]))
single3dp_wo_oulie s = dele ingOu lie s(mul ipoin _las,
ou lie 3dP_building, gdb_Single3dP,
gdb_Fil e ed3dP, "F_" + s ( ow[0]) +
"_" + s ( ow[2]))
#Big s ep
alida ion_dup = ow[0] in se (pdID_Duplica es.FID_buildings)
i alida ion_dup is T ue:
equency =
pdID_Duplica es.loc[pdID_Duplica es['FID_buildings'] == ow[0],
'FREQUENCY'].iloc[0]
lag_duplica ed += 1
i lag_duplica ed < equency:
a ay_pa hs_duplica ed.append(single3dp_wo_oulie s)
else:
p in ("Analizing duplica es...")
a ay_pa hs_duplica ed.append(single3dp_wo_oulie s)
a cpy.managemen .Me ge(a ay_pa hs_duplica ed,
os.pa h.join(gdb_Fil e ed3dP,
"CM_" + s ( ow[0]) + "_" +
s ( ow[2])))
lag_duplica ed = 0
pdDF_Ele a ion = cToPandasDF(single3dp_wo_oulie s,
["POINT_Z"])
i lp == 0 and coun > 0:
pdDF_S a is icspe Building =
c ea ingS a is ics(pdDF_Ele a ion, pdDF_S a is icspe Building, 0, ow[0])
lp = 1
else:
pdDF_S a is icspe Building =
c ea ingS a is ics(pdDF_Ele a ion, pdDF_S a is icspe Building, coun ,
ow[0])
p in ("S a is ics...")
else:
#S a is ics a e calcula ed om pandas d o speed up he
p ocessing ime
pdDF_Ele a ion = cToPandasDF(single3dp_wo_oulie s,
["POINT_Z"])
pdDF_S a is icspe Building =
c ea ingS a is ics(pdDF_Ele a ion, pdDF_S a is icspe Building, coun ,
ow[0])
p in ("S a is ics...")
coun += 1
85
p in ("Expo ing o excel...")
#S a is ics a e sa ed as excel ile o secu i y
pdDF_S a is icspe Building. o_excel(os.pa h.join( olde C op_LIDAR,
"Gene al_S a is ics_Pe _Building_%s.xlsx"%(i e a ion)),
index = T ue, heade =T ue)
#Func ion o ind ou lie s and ele a ion alues g ea e han 1
de indOu lie s( Ou lie s, lasda ase , nameLASd):
p in ("Finding ou lie s...")
a cpy.ddd.Loca eOu lie s(lasda ase , os.pa h.join( Ou lie s, nameLASd),
"APPLY_HARD_LIMIT", 1, 600,
"NO_APPLY_COMPARISON_FILTER", 0, 150, 0.5, 2500)
e u n (os.pa h.join( Ou lie s, nameLASd))
#Func ion o con e he LIDAR poin s in o ec o ea u e class
de lasToVec o Poin (lasC, gdbV, nameLAS_C opVec o ):
p in ("LAS c opped o ec o ...")
desc = a cpy.Desc ibe(lasC)
s _desc = desc.spa ialRe e ence
a cpy.ddd.LASToMul ipoin (lasC, os.pa h.join(gdbV, nameLAS_C opVec o ),
0.3, None, "ANY_RETURNS",
"CLASSIFICATION Class", s _desc, "las", 1,
"NO_RECURSION")
e u n (os.pa h.join(gdbV, nameLAS_C opVec o ))
#Func ion o dele e ou lie s om ec o ea u e class and la e s a is ics
calcula ion
de dele ingOu lie s( o al_3dpoin s, ou lie s3dpoin s, gdb_d a _single_las,
gdb_ il e ed, nameLASdel):
p in ("Dele ing ou lie s...")
a cpy.managemen .Mul ipa ToSinglepa ( o al_3dpoin s,
os.pa h.join(gdb_d a _single_las, "S_" + nameLASdel))
single3dp_laye = "single3dp_laye _1"
a cpy.managemen .MakeFea u eLaye (os.pa h.join(gdb_d a _single_las,
"S_" + nameLASdel), single3dp_laye )
a cpy.managemen .Selec Laye ByLoca ion(single3dp_laye , "INTERSECT",
ou lie s3dpoin s, None, "NEW_SELECTION", "INVERT")
ou pu _dele ed = os.pa h.join(gdb_ il e ed, nameLASdel)
a cpy.managemen .CopyFea u es(single3dp_laye , ou pu _dele ed)
a cpy.managemen .AddXY(ou pu _dele ed)
e u n(ou pu _dele ed)
#Func ion ha con e s Fea u e Class able o pandas d
de cToPandasDF( cobj, aA ibu es):
e u n (pd.Da aF ame( a cpy.da.Fea u eClassToNumPyA ay(in_ able =
cobj, ield_names = aA ibu es,
skip_nulls = False, null_ alue = -99999)))
#Func ion ha calcula es main s a is ics om pandas d
de c ea ingS a is ics( es _pandasd , ou pu _d _s s s, lag_d ,
id_building_m):
new_d = pd.Da aF ame( es _pandasd ['POINT_Z'].desc ibe())
new_d ["id_build"] = id_building_m
new_d ["s a s"] = new_d .index
new_d .se _index("id_build")
pi o _d = new_d .pi o (index = "id_build", columns = "s a s", alues =
"POINT_Z")
i lag_d == 0:
ou pu _d _s s s = pi o _d
else:
ou pu _d _s s s =
ou pu _d _s s s.append(pi o _d .loc[id_building_m], igno e_index=False)
86
e u n ou pu _d _s s s
#Cons uc o ha eci e he id o he p ocess acco ding o pa allel
p ocessing py hon unc ion using subp ocess
i __name__ == '__main__':
num_i e a ion_p ocess = s (sys.a g [1])
p in (num_i e a ion_p ocess)
main(num_i e a ion_p ocess)

87
Annex 5
Py hon code o oo op la ness assessmen .
#---------------------------------------------------------------------------
----
# Name: Pa allel_P ocessing_Launche _S age2.py
# Pu pose: Launching pa allel p ocesses o each o he codes in ol ed
in he
# oo op sui abili y analysis
#
# Au ho : Ca los Ja ie Delgado
# Subp ocess s uc u e om UPRA Gi Hub
# C ea ed: 22/10/2019
# Copy igh :
# Licence:
#---------------------------------------------------------------------------
----
impo os
impo sys
impo subp ocess
impo ime
impo da e ime
de main():
p in ("Launching pa allel p ocess...")
ini ial_ = ime.clock()
p in (da e ime.da e ime.now())
#Ge ing pa h o py hon shell applica ion
pydi = sys.exec_p e ix
pyexe = os.pa h.join(pydi , "py hon.exe")
p in (pyexe)
#De ining numbe o pa allel p ocess
num_pa allel_p ocesses = 5
#De ining he py hon iles ha will be execu ed hen using pa allel
p ocessing
sc ip _1 =
"D: Geo_Tech_Mas e Thesis_Resea ch Code Roo op_Fla ness_Elonga ion_Assess
men .py"
chain_subp ocess = ""
chain_wai = ""
aCommands = []
o i in ange(num_pa allel_p ocesses):
aCommands.append( "s a py hon %s %s"%(sc ip _1, i+1))
p in (aCommands)
#De ining he chain o pa ame e s necessa y o subp ocess execu ion
o j in ange(num_pa allel_p ocesses):
i j+1 < num_pa allel_p ocesses:
chain_subp ocess += "%s = subp ocess.Popen(aCommands[%s],
s din=None,s dou =subp ocess.PIPE,shell=T ue);"%("ch"+s (j+1), j)
chain_wai += "as dou , as de =
%s.communica e();"%("ch"+s (j+1))
88
else:
chain_subp ocess += "%s = subp ocess.Popen(aCommands[%s],
s din=None,s dou =subp ocess.PIPE,shell=T ue)"%("ch"+s (j+1), j)
chain_wai += "as dou , as de =
%s.communica e()"%("ch"+s (j+1))
p in (chain_subp ocess)
p in (chain_wai )
chain_subp ocess = compile(chain_subp ocess, '<s ing>', 'exec')
exec(chain_subp ocess)
chain_wai = compile(chain_wai , '<s ing>', 'exec')
exec(chain_wai )
p in ("Ending pa allel p ocessing...")
p in (da e ime.da e ime.now())
p in ("Pa allel p ocess execu ed in " + s ((( ime.clock() -
ini ial_ ))/60))
i __name__ == '__main__':
main()
#---------------------------------------------------------------------------
----
# Name: Roo op_Fla ness_Elonga ion_Assessmen .py
# Pu pose: Es ima e he la ness polygons o each oo op and calcula e
# he sui abili y by conside ing he compac ness/elonga ion o
# he polygons.
#
# Au ho : Ca los Ja ie Delgado
#
# C ea ed: 18/10/2019
# Copy igh :
# Licence:
#---------------------------------------------------------------------------
----
#Impo ing main lib a ies implemen ed in he algo i hm
impo a cpy
impo pandas as pd
impo os
impo da e ime
impo ime
impo sys
impo os
impo sys
om os impo lis di
om os.pa h impo is ile, isdi , join
impo glob
om ime impo sleep
a cpy.en .o e w i eOu pu = T ue
a cpy.en .ou pu Coo dina eSys em = a cpy.Spa ialRe e ence(102387)
#Main unc ion o assess he la ness o he oo ops
de main(indica o ):
p in ("S a ing Algo i hm...")
p in (s (indica o ))
p in (da e ime.da e ime.now())
p in (a cpy.CheckEx ension("3D"))
p in (a cpy.CheckEx ension("Spa ial"))
89
a cpy.CheckOu Ex ension("3D")
a cpy.CheckOu Ex ension("Spa ial")
ini ial_ = ime.clock()
pa h_p =
"D: Geo_Tech_Mas e Thesis_Resea ch P ocessing Resul s_P1 Union_3 LAS_C opp
ed_%s"%(indica o )
esul _ aw_images =
"D: Geo_Tech_Mas e Thesis_Resea ch P ocessing Resul _ aw_lida _images aw_
lida _images_%s"%(indica o )
ec o ized_ oo =
"D: Geo_Tech_Mas e Thesis_Resea ch P ocessing Roo _Vec o ized _ oo ops_%
s.gdb"%(indica o )
a cpy.en .wo kspace = ec o ized_ oo
expo _space =
"D: Geo_Tech_Mas e Thesis_Resea ch P ocessing Resul s_P2_De ailed"
pa h_ olde _image_ i =
"D: Geo_Tech_Mas e Thesis_Resea ch P ocessing Images_Ti _Masked IM10_%s"%
(indica o )
pa h_ oo s_manh =
'D: Geo_Tech_Mas e Thesis_Resea ch P ocessing Inpu s Sui abili y_Analysis_
Inpu s.gdb buildings'
sc a ch_gdb =
"D: Geo_Tech_Mas e Thesis_Resea ch P ocessing Roo _Vec o ized sc a ch_%s.g
db"%(indica o )
#C ea ion o a ea u e laye
oo _manh_ a ge = " oo _manh_ a ge "
a cpy.managemen .MakeFea u eLaye (pa h_ oo s_manh, oo _manh_ a ge )
c ea e_Folde (pa h_p)
c ea e_Folde ( esul _ aw_images)
c ea e_Folde (pa h_ olde _image_ i )
c ea e_gdb( ec o ized_ oo )
c ea e_gdb(sc a ch_gdb)
#Ge all he .las iles al eady c opped
ilez = lis LIDARC opped(pa h_p)
inal_pi o = ""
lag = 0
o las in ilez:
y:
p in ("P ocessing oo : " + s ( lag))
jname = las. eplace(".las","")
id oo = (jname.spli ("_")[2]). eplace("R","")
#Lida o Ras e
esul ing_image = os.pa h.join( esul _ aw_images, "I_" + id oo )
esul ing_image_masked = os.pa h.join( esul _ aw_images, "M_" +
id oo )
esul ing_poly_ oo = os.pa h.join( ec o ized_ oo , "V_" +
id oo )
a cpy.con e sion.LasDa ase ToRas e (os.pa h.join(pa h_p, las),
esul ing_image, "ELEVATION", "BINNING AVERAGE LINEAR", "INT", "CELLSIZE",
0.3, 1)
#Mask he new as e wi h he oo op oo p in
a cpy.managemen .Selec Laye ByA ibu e( oo _manh_ a ge ,
"NEW_SELECTION", "C = " + id oo , None)
ou _ as e = a cpy.sa.Ex ac ByMask( esul ing_image,
oo _manh_ a ge )
ou _ as e .sa e( esul ing_image_masked)
90
#Ras e o polygon
a cpy.con e sion.Ras e ToPolygon( esul ing_image_masked,
esul ing_poly_ oo , "NO_SIMPLIFY", "VALUE", "SINGLE_OUTER_PART", None)
pdDFRoo = cToPandasDF( esul ing_poly_ oo ,
["Id","g idcode","Shape_A ea", "Shape_Leng h"])
sumTo alA ea = pdDFRoo ["Shape_A ea"].sum()
#Ini ial compac ness calcula ion
pdso ed["Pe cen age"] = (pdso ed["Shape_A ea"] /
sumTo alA ea)*100
pdso ed["Compac ness"] = (4*np.pi)*(pdso ed["Shape_A ea"] /
np.powe (pdso ed["Shape_Leng h"],2))
#C ea ing he s uc u e o he esul ing able
i pdso ed.shape[0] < 5:
aO de = []
aPe cen = []
aEle a ion_A g = []
aCompac = []
aShape_F = []
aShape_A = []
o i in ange(1,pdso ed.shape[0]+1):
aO de .append("A%s"%(i))
aPe cen.append("P%s"%(i))
aEle a ion_A g.append("E%s"%(i))
aCompac .append("Compac _%s"%(i))
aShape_F.append("Shape_%s"%(i))
aShape_A.append("Pe _E ec_%s"%(i))
pdso ed["O de "] = aO de
pdso ed["Pe cen"] = aPe cen
pdso ed["Ele a ion_A g"] = aEle a ion_A g
pdso ed["Compac "] = aCompac
pdso ed["Shape_Fac o "] = aShape_F
pdso ed["Shape_A ea_E ec "] = aShape_A
else:
pdso ed["O de "] = ["A1","A2","A3","A4","A5"]
pdso ed["Pe cen"] = ["P1","P2","P3","P4","P5"]
pdso ed["Ele a ion_A g"] = ["E1","E2","E3","E4","E5"]
pdso ed["Compac "] = ["Compac _1", "Compac _2",
"Compac _3", "Compac _4", "Compac _5"]
pdso ed["Shape_Fac o "] = ["Shape_1", "Shape_2", "Shape_3",
"Shape_4", "Shape_5"]
pdso ed["Shape_A ea_E ec "] = ["Pe _E ec_1", "Pe _E ec_2",
"Pe _E ec_3", "Pe _E ec_4", "Pe _E ec_5"]
esul ing_poly_ oo _laye = " esul ing_poly_ oo _laye "
a cpy.managemen .MakeFea u eLaye ( esul ing_poly_ oo ,
esul ing_poly_ oo _laye )
aShapeRela ions = []
aShapePe cen ageBbox = []
#Asssesing he shape by gene a ing he en elope o each polygon
o ow in ange(0, pdso ed.shape[0]):
id_ alue = pdso ed.iloc[ ow]["Id"]
a ea_ alue = pdso ed.iloc[ ow]["Shape_A ea"]
a cpy.managemen .Selec Laye ByA ibu e( esul ing_poly_ oo _laye ,
"NEW_SELECTION", "Id = " + s (id_ alue), None)
ou _bbox = "in_memo y minbbx_%s_%s"%(id oo , ow)