

Ci a ion: Gonçal es, D.; Gonçal es,
G.; Pé ez-Al á ez, J.A.; And iolo, U.
On he 3D Recons uc ion o Coas al
S uc u es by Unmanned Ae ial
Sys ems wi h Onboa d Global
Na iga ion Sa elli e Sys em and
Real-Time Kinema ics and Te es ial
Lase Scanning. Remo e Sens. 2022,14,
1485. h ps://doi.o g/10.3390/
s14061485
Academic Edi o s: JoséJuan de
SanjoséBlasco, Ge mán Flo -Blanco
and Ramón Blanco Chao
Recei ed: 15 Feb ua y 2022
Accep ed: 17 Ma ch 2022
Published: 19 Ma ch 2022
Publishe ’s No e: MDPI s ays neu al
wi h ega d o ju isdic ional claims in
published maps and ins i u ional a il-
ia ions.
Copy igh : © 2022 by he au ho s.
Licensee MDPI, Basel, Swi ze land.
This a icle is an open access a icle
dis ibu ed unde he e ms and
condi ions o he C ea i e Commons
A ibu ion (CC BY) license (h ps://
c ea i ecommons.o g/licenses/by/
4.0/).
emo e sensing
Technical No e
On he 3D Recons uc ion o Coas al S uc u es by Unmanned
Ae ial Sys ems wi h Onboa d Global Na iga ion Sa elli e
Sys em and Real-Time Kinema ics and Te es ial
Lase Scanning
Diogo Gonçal es 1,2,* , Gil Gonçal es 1,3 , Juan An onio Pé ez-Al á ez 4and Umbe o And iolo 1
1INESC Coimb a, Depa men o Elec ical and Compu e Enginee ing, Polo 2, 3030-290 Coimb a, Po ugal;
[email p o ec ed] (G.G.); [email p o ec ed] (U.A.)
2Depa men o Ci il Enginee ing, Uni e si y o Coimb a, 3030-788 Coimb a, Po ugal
3Depa men o Ma hema ics, Uni e si y o Coimb a, 3001-501 Coimb a, Po ugal
4Depa amen o de Exp esión G á ica, Cen o Uni e si a io de Mé ida, Uni e sidad de Ex emadu a,
06800 Mé ida (Badajoz), Spain; [email p o ec ed]
*Co espondence: [email p o ec ed]
Abs ac :
A wide a ie y o ha d s uc u es p o ec coas al ac i i ies and communi ies om he ac ion
o ides and wa es wo ldwide. I is undamen al o moni o he in eg i y o coas al s uc u es, as
in e en ions and epai s may be needed in case o damages. This wo k compa es he e ec i eness
o an Unmanned Ae ial Sys em (UAS) and a Te es ial Lase Scanne (TLS) o ep oduce he 3D
geome y o a ocky g oin. The S uc u e- om-Mo ion (S M) pho og amme y echnique applied on
d one images gene a ed a 3D poin cloud and a Digi al Su ace Model (DSM) wi hou da a gaps. E en
hough he TLS e u ned a 3D poin cloud ou imes dense han he d one one, he TLS e u ned a
DSM which was no ep esen ing abou 16% o he g oin (da a gaps). This was due o he occlusions
encoun e ed by he low-lying scans de e mined by he displaced ocks composing he g oin. Gi en
also ha he su ey by UAS was abou eigh ime as e han he TLS, he SFM-MV applied on UAS
images was he mos sui able echnique o econs uc he ocky g oin. The UAS emo e sensing
echnique can be conside ed a alid al e na i e o moni o all ypes o coas al s uc u es, o imp o e
he inspec ion o likely damages, and o suppo coas al s uc u e managemen .
Keywo ds: d one; g oin; b eakwa e ; s uc u e om mo ion; 3D poin cloud
1. In oduc ion
In he coas al en i onmen , a wide a ie y o ha d s uc u es p o ec coas al ac i i ies
and communi ies om he ac ion o ides and wa es wo ldwide [
1
–
3
]. S uc u e ypes
ha e di e en con igu a ions and cons i u ion. Seawalls a e buil pa allel o he sho e o
de end he inland a eas and p e en sho eline e ea men [
3
,
4
]. De ached b eakwa e s
a e buil sho e pa allel o dissipa e wa e ene gy in he nea sho e [
3
], while je ies and
sho e-connec ed b eakwa e s play he unc ion o p o ec ing ha bou s and c ea ing a secu e
en i onmen o moo ing, ope a ing, and handling ships [
3
,
5
]. Finally, g oins a e buil
sho e-pe pendicula o mi iga e sho eline e osion and in e sec he upd i sedimen s [
1
,
6
].
E en hough di e en ma e ials can be used o build coas al s uc u es, mos o hem a e
cons i u ed by ocks and/o conc e e [1,7].
I is undamen al o moni o he in eg i y o coas al s uc u es and o de ec likely dam-
aged a eas, o suppo managemen in he p oposal o possible in e en ion and epai s. Fo
ins ance, he displacemen o s one blocks on ocky g oins, ubble mound sho e-connec ed
b eakwa e s and je ies may endange he s abili y o he s uc u es, lowe ing and e en
comp omising hei p o ec ion unc ions [
1
,
8
,
9
]. The adi ional moni o ing me hods o
ocky coas al s uc u es a e based on isual o pho og aphic inspec ions, howe e hese
Remo e Sens. 2022,14, 1485. h ps://doi.o g/10.3390/ s14061485 h ps://www.mdpi.com/jou nal/ emo esensing
Remo e Sens. 2022,14, 1485 2 o 15
needs o be pe o med by quali ied and ained pe sonnel, and a e echnically, logis ically
and empo ally limi ed [
10
–
12
]. Topog aphic su eys by o als s a ions and Global Na iga-
ion Sa elli e Sys em (GNSS) ha e also been used, ne e heless, despi e hei ad an ages in
e ms o spa ial accu acy, hese me hods equi e in ense human e o in he ield, and o en
do no p o ide a comple e su ey o he s uc u e [
12
]. Recen ly, emo e sensing echniques
ha e been applied o moni o coas al s uc u es. Ae ial and e es ial Ligh De ec ion and
Ranging (LiDAR) can p o ide a de ailed 3D econs uc ion o ocky coas al p o ec ions [
13
],
howe e he expensi e cos s and logis ical cons ain s o LiDAR deploymen emain he
bigges disad an ages o his me hod.
A aluable al e na i e o moni o ing coas al s uc u es is he use Unmanned Ae ial
Sys em (UAS). UAS can ope a e au onomously, p o ide high spa ial and empo al es-
olu ion wi h ela i ely low-cos , and a e iable ools o many ope a ional asks [
14
,
15
].
Gi en hei adap able and mul ipu pose p ope ies, UASs ha e imp o ed he coas al en i-
onmen al moni o ing o ad ance knowledge on beach–dune mo phodynamics [
16
–
23
],
coas al cli s [
24
–
26
], and ma ine pollu ion [
27
–
34
], among o he s. The 3D econs uc ion
wo k low consis s in applying he S uc u e- om-Mo ion (S M) pho og amme y ech-
niques o UAS image bulk, o ob ain a 3D dense poin cloud ep esen ing he a ge ed
a ea [
35
]. The use o G ound Con ol Poin s (GCPs) is needed o geo e e ence he 3D
dense poin cloud [
36
], howe e a Real-Time Kinema ic assis ed UAS (e.g., DJI Phan om 4
RTK; UAS-RTK) can accu a ely es ima e he came a posi ion, hus he use o GCPs can be
po en ially a oided [37].
The 3D econs uc ion o coas al s uc u es wi h S M pho og amme y may imp o e
he knowledge o s uc u al condi ion o a mou laye o assessing he s uc u e in eg i y
o e a ime in e al o a e ex eme e en s [
1
]. Acco ding o he Cons uc ion Indus y
Resea ch and In o ma ion Associa ion (CIRIA), he e a e ou le els o measu es o assess
he condi ion o a mou uni s [
1
]: (i) loca e uni s mo emen s by measu e wi h GNSS su ey;
(ii) geome ic su ey o desc ibe a mou laye wi h simila echnique o le el I; (iii) a mou
uni s posi ion and a eas whe e co e and unde laye a e exposed using o pho og aphic
me hods (pho og amme y, compa a i e pho og aphy); (i ) shape and size o a mou
uni s including a mou ac u es. E en hough di e en pa ame e s can cha ac e ize he
geome y o a mou laye , he mos impo an ones a e he packing densi y (numbe o
uni s pe a ea) and he a mou laye po osi y (p opo ion o oid pe olume) [
1
]. These
wo pa ame e s ha e a close ela ionship and mainly a ec he pe o mance o he coas al
s uc u e [38].
P e ious wo ks on he 3D econs uc ion o coas al s uc u es wi h UAS-based image y
a e limi ed o ew examples, which ocused on he accu acy e alua ion o S M pho og am-
me y p oduc [
39
–
41
]. The posi ional accu acy o Digi al Su ace Model (DSM) in espec
o GNSS ecei e was assessed by Hen iques e al. [
39
], who ob ained 10 cm and 8 cm o
ho izon al and e ical accu acies, espec i ely. González-Jo ge e al. [
40
] es ed ins ead he
pe o mance o 3D poin cloud o de ec ing displacemen s o a mou laye uni s. O e -
all, hese pionee ing and p elimina y wo ks showed ha UAS su eys can imp o e he
inspec ion o coas al s uc u es. Howe e , his echnique s ill needs o be be e e alua ed.
The aim o his pape is o assess he e ec i eness o UAS-RTK and TLS echniques o
ep oduce he 3D geome y o a ocky g oin. Two su eys we e conduc ed a Lei osa beach
on he No h A lan ic Po uguese coas . The L-shape g oin was digi ally econs uc ed
by 3D poin clouds and DSM. We e alua ed he econs uc ion o a L-shaped ocky g oin
by S M pho og amme y and TLS scans in e ms o mean su ace densi y and da a gaps.
O e all, his wo k e alua es he mos sui able echnique o moni o he a mou laye o
coas al s uc u es, such as g oins and b eakwa e s, o imp o e he inspec ion o likely
damages and suppo coas al managemen .
Remo e Sens. 2022,14, 1485 3 o 15
2. Me hods
2.1. S udy Si e
The s udy si e was he Lei osa beach, on he wa e-domina ed high ene ge ic No h
A lan ic Po uguese coas (Figu e 1a). In his coas al sec o , he meso idal egime has an
a e age ampli ude o 2.10 m, eaching a maximum ele a ion o 4 m du ing sp ing ides [
42
].
The dominan wa e egime comes om NW wi h a e age signi ican wa e heigh (Hs) o
2 m and pe iod om 7 s o 15 s [
43
]. In ense e osion occu ed in he las decades, mainly
caused by he li o al d i e en ion a he Mondego es ua y je y, and by he dec ease
in sedimen deposi ion om he Mondego i e [
44
,
45
]. An a e age sho eline e ea o
2 m/yea was egis e ed by compa ing sa elli e images om 1958 and 2010 [
46
], while he
dune c es e ac ed o abou 2 m sou he n he g oin o e 2018/2019 win e season [
16
]. An
L-shaped ocky g oin was buil o p o ec he u ban coas al agglome a ion o Lei osa. The
sho e-pe pendicula sec o o he g oin ex ends abou 200 m, wi h an a e age ele a ion o
6 m abo e he mean sea le el (MSL), while he sho e-pa allel sec o is sho e (150 m) and
sligh ly highe (7 m abo e MSL). Bo h sec o s a e composed by s one blocks o abou 2.5 m
wi h a shape ac o o 60–80% and a mass o abou 15 onnes. Since he main longsho e
anspo on his coas is o ien ed N-S [
42
,
44
,
45
], he g oin in e sec s sedimen upd i ,
de e mining accumula ion and sho eline ad ance no he n he s uc u e, in on o he
u banized a ea o Lei osa (Figu e 1b). The mean slope on he sho e-pe pendicula sec o
is 21
◦
(1:2.7 m) while on he sho e- pa allel wall is 30
◦
(1:1.75 m—Figu e 1j). An ini ial,
cu so y isual inspec ion e ealed signi ican displacemen s o s one blocks a he wo
heads o he g oin, which can de e mine u he disloca ion and he loss o he o iginal
shape and unc ion o he s uc u e (Figu e 1c– ).
Remo e Sens. 2022, 14, x FOR PEER REVIEW 3 o 16
2. Me hods
2.1. S udy Si e
The s udy si e was he Lei osa beach, on he wa e-domina ed high ene ge ic No h
A lan ic Po uguese coas (Figu e 1a). In his coas al sec o , he meso idal egime has an
a e age ampli ude o 2.10 m, eaching a maximum ele a ion o 4 m du ing sp ing ides
[42]. The dominan wa e egime comes om NW wi h a e age signi ican wa e heigh
(Hs) o 2 m and pe iod om 7 s o 15 s [43]. In ense e osion occu ed in he las decades,
mainly caused by he li o al d i e en ion a he Mondego es ua y je y, and by he de-
c ease in sedimen deposi ion om he Mondego i e [44,45]. An a e age sho eline e-
ea o 2 m/yea was egis e ed by compa ing sa elli e images om 1958 and 2010 [46],
while he dune c es e ac ed o abou 2 m sou he n he g oin o e 2018/2019 win e sea-
son [16]. An L-shaped ocky g oin was buil o p o ec he u ban coas al agglome a ion o
Lei osa. The sho e-pe pendicula sec o o he g oin ex ends abou 200 m, wi h an a e age
ele a ion o 6 m abo e he mean sea le el (MSL), while he sho e-pa allel sec o is sho e
(150 m) and sligh ly highe (7 m abo e MSL). Bo h sec o s a e composed by s one blocks
o abou 2.5 m wi h a shape ac o o 60–80% and a mass o abou 15 onnes. Since he main
longsho e anspo on his coas is o ien ed N-S [42,44,45], he g oin in e sec s sedimen
upd i , de e mining accumula ion and sho eline ad ance no he n he s uc u e, in on
o he u banized a ea o Lei osa (Figu e 1b). The mean slope on he sho e-pe pendicula
sec o is 21° (1:2.7 m) while on he sho e- pa allel wall is 30° (1:1.75 m—Figu e 1j). An
ini ial, cu so y isual inspec ion e ealed signi ican displacemen s o s one blocks a he
wo heads o he g oin, which can de e mine u he disloca ion and he loss o he o iginal
shape and unc ion o he s uc u e (Figu e 1c– ).
Figu e 1. S udy si e loca ion and mo phological cha ac e iza ion. (a) Loca ion o he s udy si e and
he Mondego es ua y in Po ugal; (b) Google sa elli e image o he g oin in he local coo dina e
sys em (ETRS89/PT-TM06). Red and blue ags indica e he wo heads o he g oin shown in (c,d).
G een lines indica e he wo p o iles shown in (j). (c,d) Pic u es aken by DJI Phan om 4RTK
Figu e 1.
S udy si e loca ion and mo phological cha ac e iza ion. (
a
) Loca ion o he s udy si e and
he Mondego es ua y in Po ugal; (
b
) Google sa elli e image o he g oin in he local coo dina e
sys em (ETRS89/PT-TM06). Red and blue ags indica e he wo heads o he g oin shown in (
c
,
d
).
G een lines indica e he wo p o iles shown in (
j
). (
c
,
d
) Pic u es aken by DJI Phan om 4RTK showing
Remo e Sens. 2022,14, 1485 4 o 15
he wo g oin heads (namely, sub-a ea 1 and 3, see Sec ion 2.4) (
e
,
) de ails o he wo g oin heads
shown in (
c
,
d
), espec i ely; (
g
–
i
) equipmen used in he ieldwo k campaign, namely he Te es ial
Lase Scanning (TLS) Leica ScanS a ion C10 (
g
), he Unmanned Ae ial Sys em (UAS) DJI Phan om
4RTK (
h
) and he GeoMax Zeni h 10 GNSS ecei e (
i
); (
j
) wo c oss sec ions o he s udied g oin
loca ed in (b). Da k g een iangles illus a e he slope o he be ms.
A ield wo k campaign was conduc ed wi h an Unmanned Ae ial Sys em (UAS) and a
Te es ial Lase Scanning (TLS) on 5 o No embe 2021 (Figu e 1g–i). A dual equency
GNSS ecei e was also used in Ne wo k Real Time Kinema ic (NRTK) using he Po uguese
s a ions ReNEP (h ps:// enep.dg e i o io.go .p , accessed on 6 Janua y 2022). o collec he
coo dina es o : (1) 13 con ol poin s o assess he S uc u e- om-Mo ion esul s (Sec ion 2.2);
and (2) 12 a ge s o geo e e ence and co- egis e he TLS scans (Sec ion 2.3).
2.2. Unmanned Ae ial Sys em Da a Acquisi ion and P ocessing
The UAS su ey was ca ied ou wi h a DJI Phan om 4 RTK (P4RTK), equipped wi h a
RGB came a (1
00
CMOS senso wi h 20 M e ec i e pixels, 8.8 mm ocal leng h and image
size o 5472
×
3648 pixels). The se up o d one ligh and came as pa ame e s was planned
in DJI GST RTK, choosing a linea ligh pa h (Figu e 2a). The ligh al i ude was se o
50 m, conside ing an image o e lap o 80% (bo h la e al and on al). A o al o 103 nadi al
images was collec ed wi h a mean G ound Sample Dis ance (GSD) o 1.36 cm/pixel which
is compu ed conside ing he ligh al i ude, senso dimensions and ocal leng h. The ligh
las ed abou 10 min. Al hough he P4RTK can p o ide accu a e geo e e enced da a [
16
,
47
],
we d ew 13 c osses on he blocks aces, e enly dis ibu ed along he g oin, o e alua e he
image posi ional accu acy in he pos -p ocessing phase. The measu emen s/collec ion o
hese poin s wi h he GNSS NRTK las ed abou one hou .
Remo e Sens. 2022, 14, x FOR PEER REVIEW 4 o 16
showing he wo g oin heads (namely, sub-a ea 1 and 3, see Sec ion 2.4) (e, ) de ails o he wo g oin
heads shown in (c) and (d), espec i ely; (g–i) equipmen used in he ieldwo k campaign, namely
he Te es ial Lase Scanning (TLS) Leica ScanS a ion C10 (g), he Unmanned Ae ial Sys em (UAS)
DJI Phan om 4RTK (h) and he GeoMax Zeni h 10 GNSS ecei e (i); (j) wo c oss sec ions o he
s udied g oin loca ed in (b). Da k g een iangles illus a e he slope o he be ms.
A ield wo k campaign was conduc ed wi h an Unmanned Ae ial Sys em (UAS) and
a Te es ial Lase Scanning (TLS) on 5 o No embe 2021 (Figu e 1g–i). A dual equency
GNSS ecei e was also used in Ne wo k Real Time Kinema ic (NRTK) using he Po u-
guese s a ions ReNEP (h ps:// enep.dg e i o io.go .p , accessed on 6 Janua y 2022). o
collec he coo dina es o : (1) 13 con ol poin s o assess he S uc u e- om-Mo ion esul s
(Sec ion 2.2); and (2) 12 a ge s o geo e e ence and co- egis e he TLS scans (Sec ion 2.3).
2.2. Unmanned Ae ial Sys em Da a Acquisi ion and P ocessing
The UAS su ey was ca ied ou wi h a DJI Phan om 4 RTK (P4RTK), equipped wi h
a RGB came a (1″ CMOS senso wi h 20M e ec i e pixels, 8.8 mm ocal leng h and image
size o 5472 × 3648 pixels). The se up o d one ligh and came as pa ame e s was planned
in DJI GST RTK, choosing a linea ligh pa h (Figu e 2a). The ligh al i ude was se o 50
m, conside ing an image o e lap o 80% (bo h la e al and on al). A o al o 103 nadi al
images was collec ed wi h a mean G ound Sample Dis ance (GSD) o 1.36 cm/pixel which
is compu ed conside ing he ligh al i ude, senso dimensions and ocal leng h. The ligh
las ed abou 10 min. Al hough he P4RTK can p o ide accu a e geo e e enced da a [16,47],
we d ew 13 c osses on he blocks aces, e enly dis ibu ed along he g oin, o e alua e he
image posi ional accu acy in he pos -p ocessing phase. The measu emen s/collec ion o
hese poin s wi h he GNSS NRTK las ed abou one hou .
Figu e 2. Unmanned Ae ial Sys em (UAS) da a acquisi ion and p ocessing. (a) ield da a acquisi ion
ep esen ed on UAS o hopho o. Red a ows show he UAS ligh pa h, whi e diamonds ep esen
he con ol poin s placemen o geo e e encing p ocess; (b) UAS da a p ocessing wo k low:
h ough s uc u e om mo ion (S M) and mul i- iew s e eo echnique (MVS), images and con ol
poin s we e p ocessed o gene a ing a 3D poin cloud and a Digi al Su ace Model (DSM).
Figu e 2.
Unmanned Ae ial Sys em (UAS) da a acquisi ion and p ocessing. (
a
) ield da a acquisi ion
ep esen ed on UAS o hopho o. Red a ows show he UAS ligh pa h, whi e diamonds ep esen
he con ol poin s placemen o geo e e encing p ocess; (
b
) UAS da a p ocessing wo k low: h ough
s uc u e om mo ion (S M) and mul i- iew s e eo echnique (MVS), images and con ol poin s we e
p ocessed o gene a ing a 3D poin cloud and a Digi al Su ace Model (DSM).
Remo e Sens. 2022,14, 1485 5 o 15
The UAS image y pos -p ocessing phase was pe o med wi h Agiso Me ashape so -
wa e. The applica ion o S M pho og amme y ollowed he wo k low shown in Figu e 2b.
Fi s ly, images we e o ien ed calcula ing he ex e nal and in e nal came a pa ame e s
h ough bundle block adjus men . A se o 3D poin s ( ie poin s) was ob ained ep esen ing
he scale in a ian ea u es iden i ied in o e lapping images. Secondly, since he e inemen
o ie poin s imp o es he accu acy o econs uc ion [
26
,
48
], we emo ed poin s wi h a
ep ojec ion e o g ea e han 0.5. A e wa ds, he bundle block adjus men was epea ed
using he e ined ie poin s. Thi dly, he geo e e encing p ocess was pe o med by selec ing
only 3 g ound con ol poin s om he se o 13 su eys poin s. The emaining 10 poin s
we e used as check poin s. All 13 con ol poin s we e manually picked in he co esponding
images. Fou hly, 3D poin cloud was gene a ed h ough dense ma ching by compu e i-
sion algo i hms based on mul i-s e eoscopy (Figu e 2b). Finally, a e g idding he 3D poin
cloud wi h egula cells (i.e., squa es 5
×
5 cm), we gene a ed he DSM in CloudCompa e.
Each cell g id was calcula ed by he mean heigh o all poin s wi hin he cell. The emp y
cells we e linea ly in e pola ed using he nea es cells.
2.3. Te es ial Lase Scanning Da a Acquisi ion and P ocessing
The TLS su ey was pe o med wi h a Leica ScanS a ion C10, scan ange o 300 m.
The posi ional accu acy o a single TLS s a ion is 6 mm o a ange o 50 m. A o al o
se en scans we e dis ibu ed along Lei osa g oin, wi h mos o s a ions chosen on he wo
g oin heads (Figu e 3a). A se ies o 12 ci cula a ge s we e dis ibu ed in he TLS scan
anges o di e en s a ion, o s ich he scan da a in he pos -p ocessing phase. Ta ge s
we e placed among he e ain in o de o be isible a leas om wo consecu i e s a ions.
Be o e s a ing he scanning p ocedu e, a ge s we e isually picked in he TLS sc een,
and measu ed wi h he GNSS NRTK ecei e . When mo ing o he ollowing s a ion, we
de ec ed he same p e ious a ge s o building he scan ne wo k. Du ing he TLS su ey,
he ide ose in luencing he candida e zones o he scans posi ion. In he end, he TLS
su ey las ed abou 8 h.
Remo e Sens. 2022, 14, x FOR PEER REVIEW 6 o 16
Figu e 3. Te es ial Lase Scanning (TLS) da a acquisi ion and p ocessing. (a) Field da a acquisi ion
on UAS o hopho o. G een iangles show he di e en s a ions in which he TLS was placed o
scanning ( om 1 o 7), blue-whi e ci cles ep esen he a ge s ( om 1 o 12) placed o s i ching he
scans acqui ed om di e en s a ions. (b) TLS da a p ocessing wo k low: di e en scans a e co-
egis e ed and joined using a ge s, o inally gene a ing a 3D poin cloud and a Digi al Su ace
Model (DSM).
The mean su ace densi y aims a cha ac e izing he mean poin s concen a ion o e
he su ace, iden i ying zones wi hou poin s (da a gaps). The 3D poin clouds we e p o-
jec ed along Z di ec ion and con e ed in o a egula g id. The mean su ace densi y (𝜌)
was es ima ed by coun ing he poin s inside each g id cell and hen calcula ing he mean
o he o e all cells densi y:
ρ = 1
N∑ci
N
i=1
(1)
whe e 𝑁 is he numbe o g id cells in he egion o in e es (𝐶) and 𝑐𝑖 is he poin coun -
ing o local densi y in g id cell 𝑖.
The da a gaps indica e he numbe s o emp y g id cells in he egion o in e es using
𝛿𝑔= #𝐷𝑔×𝐴𝑐
(2)
whe e #𝐷𝑔 is he ca dinali y o se 𝐷𝑔={𝑐𝑖:𝑐𝑖=0, o 𝑖 =1,…,𝑁} o med by he emp y
g id cells and 𝐴𝑐 is he g id a ea.
2.5. Compa a i e Analysis and S a is ical Measu es
To assess he accu acy and e ec i eness o UAS-RTK and TLS p oduc s o mapping
g oins, we implemen ed a wo-s age me hodology (Figu e 4): (i) 3D app oach o as-
sessing he posi ional accu acy o 3D poin cloud, and (ii) 2.5 D app oach o assessing he
DSM e ical accu acy.
Figu e 3.
Te es ial Lase Scanning (TLS) da a acquisi ion and p ocessing. (
a
) Field da a acquisi ion
on UAS o hopho o. G een iangles show he di e en s a ions in which he TLS was placed o

Remo e Sens. 2022,14, 1485 6 o 15
scanning ( om 1 o 7), blue-whi e ci cles ep esen he a ge s ( om 1 o 12) placed o s i ching
he scans acqui ed om di e en s a ions. (
b
) TLS da a p ocessing wo k low: di e en scans a e
co- egis e ed and joined using a ge s, o inally gene a ing a 3D poin cloud and a Digi al Su ace
Model (DSM).
The esolu ion o scans was di ided in o he ollowing ca ego ies: (i) medium es-
olu ion, co esponding o 10 cm dis ance be ween poin s a 100 m; (ii) high esolu ion,
co esponding o 5 cm dis ance be ween poin s a 100 m.
The TLS pos -p ocessing me hodology was pe o med wi h Leica Cyclone so wa e
(Figu e 3b). The main s eps consis ed in: (i) impo ing he scans; (ii) eading a ge s coo di-
na es; (iii) co- egis a ion o he scans. The scans we e s i ched using he co esponden
picked a ge s o p oduce a single 3D poin cloud. The geo e e encing p ocess was also
included, p ojec ing he 3D poin cloud in o he Po uguese e e ence sys em. Using he
same g id adop ed o p oduce he UAS DSM (Sec ion 2.2), we also gene a ed he DSM om
TLS 3D poin cloud.
2.4. Th ee-Dimensional Poin Clouds and Digi al Su ace Model
We e alua ed and compa ed he 3D poin clouds o bo h UAS-RTK and TLS using wo
quali y pa ame e s, namely he mean su ace densi y and da a gaps. We compu ed he wo
pa ame e s o he en i e g oin s uc u e, and o h ee sub-a eas. Sub-a eas we e chosen
on he g oin heads and in he cen al g oin a ea (Figu e 4). The head on seaside is he mo e
a ec ed zone by wa es, and al eady shows damages and block displacemen s. The sou h
head shows also block displacemen s and co e s he dune idge, which is unde se e e
e osion [
16
]. Finally, he cen al a ea ep esen s a la wall o p o ec ing he u ban a ea (see
also Figu e 1b– ).
Remo e Sens. 2022, 14, x FOR PEER REVIEW 7 o 16
Fo he 3D app oach, he dis ance be ween poin s o TLS and UAS was calcula ed.
Fo his pu pose, in CloudCompa e so wa e, wo algo i hms we e adop ed: (i) Cloud- o-
Cloud (C2C) dis ance and (ii) Mul iscale Model- o-Model Cloud Compa ison (M3C2). The
C2C compu es di ec ly he dis ance be ween wo 3D poin clouds based a local model
i ing. This local model is calcula ed o each poin o he e e ence da a (TLS 3D poin
cloud) and is composed by: (1) he k nea es poin s o (2) he poin s inside a sphe e wi h
adius o he a ge da a (UAS 3D poin cloud). The M3C2 measu e compu es he dis-
ances be ween wo 3D poin clouds h ough he no mal di ec ion o a local su ace. This
local su ace is calcula ed by sea ching o poin s in a adius D and hen a cylinde is
compu ed along he no mal di ec ion ha in e sec s he a ge da a [49]. The e o e, he
e o o UAS poin cloud (Δ𝐷) was ob ained by:
Δ𝐷 =[Δ𝑑1⋯ Δ𝑑𝑁]
(3)
whe e Δ𝑑𝑖 is he dis ance be ween he poin 𝑖 o UAS and he co esponden in TLS poin
cloud, and 𝑁 is numbe o UAS 3D poin cloud. . He eina e , when Δ𝐷 is men ioned, i
includes dis ances calcula ed by C2C and M3C2.
Fo he 2.5D app oach, a di ec e ical di e ence was calcula ed using DSM o Di -
e ence (DoD). The o e all e o s we e ob ained as:
Δ𝑍 =[Δ𝑧1⋯ Δ𝑧𝑀]
(4)
whe e Δ𝑧𝑗=𝑧𝑗𝑇𝐿𝑆 −𝑧𝑗𝑈𝐴𝑆 is he e ical di e ence o j- h DoD cell.
Finally, he s a is ical analysis ocused a cha ac e izing he e o s p o ided by he
3D and 2.5D app oaches (Figu e 4). We i s ly i ed he his og ams o cha ac e ize he
dis ibu ion o e o s. Then, a obus s a is ical analysis sui able o non-no mal dis ibu-
ions was adop ed [50], in o de o compa e he esul s wi h he adi ional s a is ical
measu es. Fo adi ional measu es, he mean, he S anda d De ia ion (S d) and he Roo
Mean Squa e E o (RMSE) we e calcula ed. Fo obus measu es o non-no mal e o dis-
ibu ions, he median and No malized Median Absolu e De ia ion (NMAD) we e com-
pu ed. In addi ion, he obus RMSE (RRMSE) was also compu ed based on mean and
s anda d de ia ion [51].
Figu e 4. Da a p ocessing wo k low o he compa a i e analysis. Th ee sub-a eas (le box, ed
polygons) we e de ined o e he whole g oin s uc u e (le box, g een polygon). The dis ance be-
ween 3D poin clouds and Digi al Su ace Model we e compu ed using 3D and 2.5D app oach
based on poin - o-poin dis ance and e ical di e ences (cen al boxes). Finally, he e o s we e
assessed h ough s a is ical measu es compu a ion ( igh box).
Figu e 4.
Da a p ocessing wo k low o he compa a i e analysis. Th ee sub-a eas (
le
box, ed
polygons) we e de ined o e he whole g oin s uc u e (
le
box, g een polygon). The dis ance
be ween 3D poin clouds and Digi al Su ace Model we e compu ed using 3D and 2.5D app oach
based on poin - o-poin dis ance and e ical di e ences (
cen al
boxes). Finally, he e o s we e
assessed h ough s a is ical measu es compu a ion ( igh box).
The mean su ace densi y aims a cha ac e izing he mean poin s concen a ion o e
he su ace, iden i ying zones wi hou poin s (da a gaps). The 3D poin clouds we e
p ojec ed along Z di ec ion and con e ed in o a egula g id. The mean su ace densi y (
ρ
)
Remo e Sens. 2022,14, 1485 7 o 15
was es ima ed by coun ing he poin s inside each g id cell and hen calcula ing he mean o
he o e all cells densi y:
ρ=1
N
N
∑
i=1
ci(1)
whe e
N
is he numbe o g id cells in he egion o in e es (
C
) and
ci
is he poin coun ing
o local densi y in g id cell i.
The da a gaps indica e he numbe s o emp y g id cells in he egion o in e es using
δg=#Dg×Ac(2)
whe e #
Dg
is he ca dinali y o se
Dg={ci:ci=0, o i=1, . . . , N}
o med by he emp y
g id cells and Acis he g id a ea.
2.5. Compa a i e Analysis and S a is ical Measu es
To assess he accu acy and e ec i eness o UAS-RTK and TLS p oduc s o mapping
g oins, we implemen ed a wo-s age me hodology (Figu e 4): (i) 3D app oach o assessing
he posi ional accu acy o 3D poin cloud, and (ii) 2.5 D app oach o assessing he DSM
e ical accu acy.
Fo he 3D app oach, he dis ance be ween poin s o TLS and UAS was calcula ed. Fo
his pu pose, in CloudCompa e so wa e, wo algo i hms we e adop ed: (i) Cloud- o-Cloud
(C2C) dis ance and (ii) Mul iscale Model- o-Model Cloud Compa ison (M3C2). The C2C
compu es di ec ly he dis ance be ween wo 3D poin clouds based a local model i ing.
This local model is calcula ed o each poin o he e e ence da a (TLS 3D poin cloud) and
is composed by: (1) he k nea es poin s o (2) he poin s inside a sphe e wi h adius o
he a ge da a (UAS 3D poin cloud). The M3C2 measu e compu es he dis ances be ween
wo 3D poin clouds h ough he no mal di ec ion o a local su ace. This local su ace is
calcula ed by sea ching o poin s in a adius D and hen a cylinde is compu ed along he
no mal di ec ion ha in e sec s he a ge da a [
49
]. The e o e, he e o o UAS poin cloud
(∆D)was ob ained by:
∆D=∆d1· · · ∆dN(3)
whe e
∆di
is he dis ance be ween he poin
i
o UAS and he co esponden in TLS poin
cloud, and
N
is numbe o UAS 3D poin cloud. He eina e , when
∆D
is men ioned, i
includes dis ances calcula ed by C2C and M3C2.
Fo he 2.5D app oach, a di ec e ical di e ence was calcula ed using DSM o
Di e ence (DoD). The o e all e o s we e ob ained as:
∆Z=∆z1· · · ∆zM(4)
whe e ∆zj=zTLS
j−zUAS
jis he e ical di e ence o j- h DoD cell.
Finally, he s a is ical analysis ocused a cha ac e izing he e o s p o ided by he 3D
and 2.5D app oaches (Figu e 4). We i s ly i ed he his og ams o cha ac e ize he dis ibu-
ion o e o s. Then, a obus s a is ical analysis sui able o non-no mal dis ibu ions was
adop ed [
50
], in o de o compa e he esul s wi h he adi ional s a is ical measu es. Fo
adi ional measu es, he mean, he S anda d De ia ion (S d) and he Roo Mean Squa e
E o (RMSE) we e calcula ed. Fo obus measu es o non-no mal e o dis ibu ions, he
median and No malized Median Absolu e De ia ion (NMAD) we e compu ed. In addi ion,
he obus RMSE (RRMSE) was also compu ed based on mean and s anda d de ia ion [
51
].
3. Resul s
3.1. 3D Poin Clouds and Digi al Su ace Model
The Agiso Me ashape p ocessing wo k low las ed abou 1.5 h and e u ned a 3D
poin cloud wi h a geo e e encing RMSE on check poin s equal o 3.2 cm. The 3D poin
cloud ep esen ed he en i e g oin wi h abou 11 million poin s (Table 1), wi h a mean
Remo e Sens. 2022,14, 1485 8 o 15
su ace densi y o 1.7
×
10
3
poin s/m
2
and wi hou da a gaps. A e using he 12 a ge s
o link he TLS scans in Cyclone ha las ed abou 1 h, he RMSE o TLS s i ching was o
1.5 cm. The inal 3D poin cloud o TLS co e ed he g oin wi h abou 41 million poin s
(Table 1). E en hough he numbe o poin s and mean su ace densi y o TLS poin cloud
was abou h ee imes he UAS one, he TLS da a gap was abou 16% o he en i e g oin
a ea (6630 m2).
Table 1.
Cha ac e iza ion o he 3D poin clouds ob ained by he Unmanned Ae ial Sys em and
Te es ial Lase Scanning da a. Numbe o poin s, mean su ace densi y and da a gaps a e indica ed
o he whole g oin s uc u e and he h ee suba eas ( e e o Figu e 4 o sub-a eas iden i ica ion).
Pa ame e s Unmanned Ae ial Sys em Te es ial Lase Scanne
Whole s uc u e
(6630 m2)
Numbe o poin s 11.3 ×10641.3 ×106
Mean su ace densi y (poin s/m2) 1.7 ×1036.2 ×103
Da a gaps (%) 0 16.8
Sub-a ea 1
(560 m2)
Numbe o poin s 1.0 ×10615.2 ×106
Mean su ace densi y (poin s/m2) 1.8 ×10327.2 ×103
Da a gaps (%) 0 1.4
Sub-a ea 2
(530 m2)
Numbe o poin s 0.9 ×1060.7 ×106
Mean su ace densi y (poin s/m2) 1.6 ×1031.3 ×103
Da a gaps (%) 0 0
Sub-a ea 3
(600 m2)
Numbe o poin s 1.2 ×10621.8 ×106
Mean su ace densi y (poin s/m2) 2.0 ×10336.4 ×103
Da a gaps (%) 0 0.8
O e all, he numbe o poin s and mean su ace densi y among he h ee sub-a eas
we e much a iable on TLS da a, while did no change signi ican ly o UAS. In ac , he
mean su ace densi y o TLS was highe on sub-a eas 1 and 3, due o he mo e nume ous
scans and s a ions in hese zones. Finally, o he sub-a eas 1–3, which a e he mos
in e es ing a eas o be moni o ed on he g oin s uc u e, TLS su ey e u ned an a e age
1.1% o su ace da a gaps. On he o he hand, despi e he lowe means su ace densi y, no
da a gaps we e ound on UAS su ey. Figu e 5depic s he 3D poin clouds and Digi al
Su ace Models (DSM) ob ained in he da a p ocessing s ep. Al hough he da a gaps we e
p esen on TLS 3D poin s, he in e pola ion o he emp y cells (5
×
5 cm) allowed o mi iga e
his e ec on DSM.
3.2. Compa a i e Analysis and S a is ical Measu es o 3D Poin Clouds and DSM
Fo analysing he spa ial dis ibu ion o e o s, Figu e 6shows he esul s o 3D and
2.5D app oaches o he sub-a eas. O e all, mo e han 50% o e o s we e concen a ed
be ween 0 and 10 cm. The accumula ed g een ba s p esen ed highe alues in sub a eas
1 and 3. In ac , he p esence o da a gaps was p opo ional o he inc ease o e o s. The
oughness o hese sub-a eas may ha e penalized he e o s o M3C2 algo i hm whe e i
is no possible o compu e he dis ance in se e al poin s due o w ong calcula ion o he
no mal di ec ion o local su aces (see Sec ion 2.4 and Figu e 4). The e o s o sub-a ea
2 egis e ed lowe a iabili y among he di e en algo i hms. These esul s we e due o he
egula oughness o he g oin su ace. O e all, he non-no mali y o he e o s dis ibu ion
was co obo a ed wi h a cou se analysis o he his og am shapes. The main eason o his
assump ion is he igh hea y ails o he his og ams.
Remo e Sens. 2022,14, 1485 9 o 15
Remo e Sens. 2022, 14, x FOR PEER REVIEW 9 o 16
Figu e 5. P oduc s ob ained by Unmanned Ae ial Sys em (a) and Te es ial Lase Scanning (b). In
de ails, 3D poin clouds (uppe ow) and Digi al Su ace Models (DSM, lowe ow). Red numbe ed
polygons ep esen he h ee chosen sub-a eas (1–3, please e e o Sec ion 2.4 o explana ion). The
axis labels e e o he p ojec ed Po uguese e e ence sys ems (ETRS89/PT-TM06).
Fo an in-dep h assessmen , he accu acy measu es we e abula ed highligh ing he
main di e ences be ween 3D and 2.5D app oaches in sub-a eas (Table 2). In gene al, he
con en ional measu es o e es ima ed he e o s be ween UAS and TLS in compa ison
wi h he obus measu es (in he wo s case, di e ences we e abou 14 cm). The sub-a ea
2 p esen ed he lowe a iabili y bo h in e ms o S d and NMAD, while sub-a ea 1
showed high dispe sion in ela ion o he mean alue. O e all, he RMSE egis e ed alues
in a ange o 30 cm due o he a iabili y o e o s exp essed in e m o S d. In ac , he
o mula ion o RMSE conside s he squa e o e o s ha magni ies he la ges alues
mainly obse ed in sub-a eas 1 and 3. Howe e , i can be obse ed ha he RRMSE p e-
sen ed a lowe a iabili y concen a ing alues in a ange o 10 cm. Al hough he
Figu e 5.
P oduc s ob ained by Unmanned Ae ial Sys em (
a
) and Te es ial Lase Scanning (
b
). In
de ails, 3D poin clouds (uppe ow) and Digi al Su ace Models (DSM, lowe ow). Red numbe ed
polygons ep esen he h ee chosen sub-a eas (1–3, please e e o Sec ion 2.4 o explana ion). The
axis labels e e o he p ojec ed Po uguese e e ence sys ems (ETRS89/PT-TM06).
Fo an in-dep h assessmen , he accu acy measu es we e abula ed highligh ing he
main di e ences be ween 3D and 2.5D app oaches in sub-a eas (Table 2). In gene al, he
con en ional measu es o e es ima ed he e o s be ween UAS and TLS in compa ison
wi h he obus measu es (in he wo s case, di e ences we e abou 14 cm). The sub-
a ea 2 p esen ed he lowe a iabili y bo h in e ms o S d and NMAD, while sub-a ea
1 showed high dispe sion in ela ion o he mean alue. O e all, he RMSE egis e ed
alues in a ange o 30 cm due o he a iabili y o e o s exp essed in e m o S d. In
ac , he o mula ion o RMSE conside s he squa e o e o s ha magni ies he la ges
alues mainly obse ed in sub-a eas 1 and 3. Howe e , i can be obse ed ha he RRMSE