Remo eID
Iden i ica ion o A lan ic wal us a haul ou si es in
G eenland using high- esolu ion sa elli e images
Ka l B ix Zingle sen, E a Ga de, Ki s y Langley and E a Mä zle
Technical Repo no. 111
G eenland Ins i u e o Na u al Resou ces
In collabo a ion wi h Asiaq G eenland Su ey
2
Ti le: Remo eID. Iden i ica ion o A lan ic Wal us a haul ou si es
in G eenland using high- esolu ion sa elli e images.
Au ho s: Ka l B ix Zingle sen, E a Ga de, Ki s y Langley, E a
Mä zle .
Publishe : G eenland Ins i u e o Na u al Resou ces
Final e sion: 11-01-2019
Da e o publica ion:
Financial suppo : Financial suppo ed by he Dancea (Danish Coo pe a ion
o En i onmen in he A c ic) en i onmen al suppo
scheme o he Chemicals Depa men a he En i onmen al
P o ec ion Agency, Minis y o En i onmen and Food o
Denma k. MST j.n . 112-00249.
Co e illus a ion Ka l B ix Zingle sen
P ojec : “Ko lægning og op ælling a pa edy i G ønland med
de alje ede sa elli billede ”
(Mapping and coun ing o mammals in G eenland wi h
de ailed sa elli e image y)
ISSN: 1397-3657
ISBN: 87-91214-89-0
EAN: 9788791214899
Web: h ps://na u .gl/
Ci a ion: Zingle sen KB, Ga de E, Langley K, & Mä zle E (2020)
Remo eID. Iden i ica ion o A lan ic Wal us a haul ou si es
in G eenland using high- esolu ion sa elli e images.
Technical epo no. 111, G eenland Ins i u e o Na u al
Resou ces, G eenland. ISBN
3
Con en s
Lis o igu es: ......................................................................................................... 5
Lis o ables: .......................................................................................................... 6
1. English summa y ................................................................................................. 7
2. G eenlandic summa y – Kalaalissu Naalisa neqa ne a ..................................... 8
3. Danish summa y – sammen a ning på dansk ..................................................... 9
4. In oduc ion ....................................................................................................... 10
5. Backg ound in o ma ion .................................................................................... 11
5.1. O e iew o a ailable e y high- esolu ion sa elli e image y ....................... 11
5.2. Recen s udies o emo e sensing me hods on pola mammals .................. 11
5.3. Wal us popula ions in G eenland ................................................................ 12
5.4. Wal us cha ac e is ics and beha iou ......................................................... 13
5.5. Pe iods and egions o in e es .................................................................... 15
6. Case s udy a eas selec ed ................................................................................ 16
6.1. Case s udy Sandøen ................................................................................... 16
6.2. Case s udy Lille Snenæs ............................................................................ 19
6.3. Case s udy Wols enholme Fjo d ................................................................. 20
7. The capabili ies o e y high- esolu ion imaging sa elli es ................................. 23
7.1. Reques and sea ch o da a ....................................................................... 24
8. P ocessing he da a be o e ex ac ion ............................................................... 27
8.1. P e-p ocessing he da a .............................................................................. 27
8.1.1. ans o ma ion and Pansha pening ....................................................... 27
8.1.2. Aligning he images o each si e ........................................................... 27
8.2. Fu he p ocessing o analy ical as e da a ................................................ 28
8.2.1. Land-Sea in e ac ion mask and p oximi y o sea analysis .................... 28
8.2.2. B igh ness and NIR a io....................................................................... 29
8.2.3. Mul i a ia e Al e a ion De ec ion ........................................................... 30
9. Objec based ea u e ex ac ion ........................................................................ 32
9.1. P inciple Componen Analysis and Independen Componen Analysis ....... 32
9.2. Fea u e ex ac ion a Sandøen si e ............................................................. 32
9.2.1. Me hods used o Sandøen case s udy ................................................ 33
9.2.2. PCA and ICA analyses o Sandøen case s udy .................................. 35
9.2.3. Resul s o Sandøen case s udy ........................................................... 35
4
9.2.4. Discussion conce ning he case s udy Sandøen .................................. 37
9.3. Fea u e ex ac ion a Lille Snenæs si e ....................................................... 37
9.3.1. Me hods ................................................................................................ 38
9.3.2. PCA and ICA analyses o Lille Snenæs .............................................. 39
9.3.3. Resul s and Discussion o Lille Snenæs.............................................. 39
9.4. Recommenda ions o objec -based analysis ................................................. 40
10. Sigh ings o wal us in he sa elli e images ...................................................... 42
10.1. Sigh ing a Sandøen si e .......................................................................... 42
10.2. Sigh ing a he Lille Snenæs si e .............................................................. 45
10.3. Sigh ing a he Wols enholme Fjo d ......................................................... 46
10.4. Recommenda ions om isual obse a ions ........................................... 49
11. Discussion ...................................................................................................... 49
12. Recommenda ions om he p ojec ................................................................ 51
13. Acknowledgemen s ........................................................................................ 52
Re e ences ............................................................................................................... 54
Appendix 1. .............................................................................................................. 57
5
Lis o igu es:
Figu e 1: O e iew map o G eenland including he selec ed haul ou si es o a eas
o wal us. ................................................................................................................... 7
Figu e 2: Wal us hauling ou on Sandøen island. Pho o: Fe nando Uga e ............. 13
Figu e 3: Sandøen Island cap u ed om opog aphical ae ial pho og aphy in 1985 16
Figu e 4: Sandøen Island cap u ed on Wo ldView-2 image o 2012. Cou esy o
Digi alGlobe Inc. ....................................................................................................... 17
Figu e 5: Sandøen Island cap u ed on Wo ldView-3 image o 2015, cou esy o
Digi alGlobe Inc. ....................................................................................................... 17
Figu e 6: Lille Snenæs case s udy a ea wi h he Do e Bug (bay) and H al osodden
loca ions ma ked. Sa elli e image: USGS Landsa 8 o 6/8-2015. ............................ 19
Figu e 7: No h Wa e Polynya open wa e and Wols enholme Fjo d. MODIS sa elli e
image om he NASA Wo ld iew se ice. ................................................................ 21
Figu e 8: P oximi y map o Sandøen Island. Dis ances om coas line..................... 29
Figu e 9: B igh ness analysis o Sandøen. ............................................................... 29
Figu e 10: NIR a io as e calcula ed om he nea -in a ed alues di ided by
b igh ness ................................................................................................................. 30
Figu e 11: Mul i a ia e Al e a ion De ec ion om wo images om Sandøen – om
2015-08-04 o 2015-08-06. The g een colo s ma k in ense change be ween he
images – a he beach conce ning ide wa es and o he g oup o wal us. ............. 31
Figu e 12: Sandøen, Pleiades-1B image (2015-08-06_141647_PHR1B1) o
Sandøen. Red box indica es a g oup o wal us. ....................................................... 33
Figu e 13: Sandøen, Wo ldView 2 scene (2012-08-16_144211-WV2). G een shading
a e a eas ou side o he egion o in e es de ined by dis ance om he sea. The ed
ou lines a e he objec s iden i ied wi h he p ocessing s eps gi en in Table 4. ......... 36
Figu e 14: Sandøen, Wo ld View 2 scene (2012-08-16_144211-WV2). PCA
componen 1. Red box indica es loca ion o he wal uses ........................................ 37
Figu e 15: Lille Snenæs, Pleiades scene (2015-08-20-1458000_PHR1A)............... 38
Figu e 16: Lille Snenæs a ea o in e es a e masking ou s eep slopes, snow, wa e
and 50 m inland om he coas . ............................................................................... 39
Figu e 17: Sa elli e image y om Sandøen island a he p e e ed haul ou si e. Wi h
ed ci cles a e ma ked he sigh ings o indi iduals o g oups o wal us hauling ou on
he beach a ea. ........................................................................................................ 42
6
Figu e 18: Obse ed indi idual wal us on 2015-06-28 om Wo ldView-3. To he le
a 30 cm pixel size esolu ion, and on he igh a 50 cm esolu ion. In bo h, he
wal us is appa en , howe e sha pe de ined o he le . .......................................... 43
Figu e 19: Obse a ions in he sa elli e images co e ing he Sandøen si e. Colo ing
acco ding o he da e o he image. Backg ound image o 2015-08-06 including a
la ge g oup o wal us. .............................................................................................. 44
Figu e 20: Obse a ions a he Lille Snenæs si e, scene by scene. ......................... 45
Figu e 21: Obse a ion om scene 2015-08-30. In he image i is di icul o exac ly
iden i y a wal us. I could be a s one o ano he objec . Howe e , i was no easily
appa en in he p e ious scenes. ............................................................................. 45
Figu e 22: Obse a ions om sa elli e image y in Wols enholme Fjo d. Poin colo s
ca ego ized acco ding o sa elli e image om which he obse a ion was done. ..... 47
Figu e 23: Posi ions o wal us a he da e 2018-06-06 eco ded wi h A gos
ansmi e s. Few posi ions a e o bes quali y (A gos quali y 3), and mo e han hal is
o poo posi ion quali y 0 (c. 500-1500 m accu acy). The ed poin s display he
obse a ions done om he 30 cm Wo ldView-3 image o h ee days la e a he da e
2015-06-09. .............................................................................................................. 48
Lis o ables:
Table 1: Gene al capabili ies o applied sa elli es .................................................... 24
Table 2: Sa elli es, o ma s in use, map p ojec ion and ini ial s eps o
pansha pening. ........................................................................................................ 25
Table 3: Lis o all images pu chased, hei si e loca ion, image name, sa elli e and
da e o acquision. ..................................................................................................... 26
Table 4: eCogni ion p ocessing s eps used o loca e wal uses in Sandøen scenes. 34
Table 5: PCA and ICA esul s o each Sandøen sa elli e scene. PH = Pleiades; WV=
Wo ldView. (b1-b8 a e he band numbe s o he o iginal scenes, whe e PH has 4
bands and WV has 8 bands) .................................................................................... 35
7
1. English summa y
G eenland Ins i u e o Na u al Resou ces and Asiaq G eenland Su ey ha e
analysed sa elli e images o e y high esolu ion o h ee case s udy a eas o
G eenland o obse e and de ec he p esence o A lan ic wal us. We chose his
species because indi iduals o g oups he eo es a pa icula a eas on land o ice
o hou s in calm and sunny wea he be ween pe iods o eeding.
The s udy a eas a e di e se in ime
and en i onmen , co e ing sandy
beaches in Augus a Sandøen and
Lille Snenæs in No h Eas G eenland
and on mo ing loes o sea ice in la e
May and ea ly June a Wols enholme
Fjo d in No h Wes G eenland.
Simila ly, he esul s o he case
s udies a e a ied including
unques ionable obse a ions o wal us
a Sandøen and no a single ce ain
ecogni ion o wal us a Lille Snenæs
and Wols enholme Fjo d. The p ojec
has p o ided insigh in o he
capabili ies and limi a ions o e y
high esolu ion sa elli e images and
analysis he eo .
The analyses ha e included isual
inspec ion, applica ion o change
de ec ion using he Mul i a ia e
Al e a ion De ec ion algo i hm,
P incipal Componen Analysis and
Independen Componen Analysis.
We ha e p o en he echnology o be
e y success ul unde ideal
ci cums ances, bu also inadequa e
unde un i ing en i onmen al condi ions.
Figu e 1: O e iew map o G eenland including he
selec ed haul ou si es o a eas o wal us.
8
2. G eenlandic summa y – Kalaalissu Naalisa neqa ne a
Pinngo i ale i iup, Asiaq G eenland Su ey peqa igalugu sa elli ’ ikku
pi saaso ujussua mik e seqqa issumillu assilisa a o lugi Kalaalli Nunaanni
misissui igisa pingasu nalile soqqamme pai, aakkunani qanoq aa eqa igine soq
aku innia neqa poq. Aa i misissugassa u oqqa neqa simappu
ne illua sima ia lu ik sumii inni aalajange simasuni nunami sikumilu
eqqisseqqinnaa lu ik akunne passua ni seqinna issumi nala usaa ne anni
nalunaa so umina a ma a.
Misissuine i pi issani a a angiisinilu assigiinngi suni inge lanneqa a pu , Tunup
A annaa suani nunami allanngu saaliukkami eqqissisima i amilu inge lanneqa u
augus imi sissani sio aanna ni inge lanneqa a pu , assani Sandø-mi aammalu
Danma ksha nip eqqaani Lille Snenæs-imik aaneqa a umi, misissuineq alla ki aani
A ane sua mi Uummannap Kange luani immap sikuani inge laa umi maajip
naane anii junip aalla inne ani inge lanneqa poq. Misissuine i ine ne i aamma
assigiinngeqaa , Sandø-mi aa i nalunaa so neqa ne i nalo ninaa sumik
inge lanneqa sinnaasimappu , Lille Snenæs-imi A ane sua milu Uummannap
Kange luani aalajangii igiuminaassimappu . Taamaalilluni misiliilluni inge la sine up
aku ippaa sa elli - ikku e sa illuinna umik assiliilluni misissuine i qanoq
pe ia issagissaa igine su killeqa igine sullu.
Misissuine ne i assinik misissuine aqa pu , ma ema ikilu a o lugu naa so suine i
inge lanneqa a simappu assilisa allanngo a ne annik e sinne ule si sisa umik
pe iuseqa luni naa so so neqa a pu imminnullu a uumassu eqanngi sumik
pe iuseqa oqa a luni (Mul i a ia e Al e a ion De ec ion algo i hm, P incipal
Componen Analysis aamma Independen Componen Analysis).
A a angiisi eqqo uu illugi eknologi a o lugu nalunaa suineq naa so suusio ne lu
a o neqa sinnaasu uppe na sa pa pu , a a angiisinulli ulluu uunngi suni aamma
a o sinnaanngi soq uppe na sineqa luni.
9
3. Danish summa y – sammen a ning på dansk
G ønlands Na u ins i u og Asiaq G eenland Su ey ha analyse e sa elli billede a
mege høj opløsning o e s udieom åde i G ønland med de o mål a obse e e
og på ise ils ede æ elsen a h al os. Denne a ble alg , o di indi ide elle
g uppe he a h ile i bes em e om åde på land elle is i ime i olig og sol ig ej
mellem ou age ingspe iode .
Unde søgelsesom åde ne e o skellige i id og miljø, og dække sands ande i
augus ed Sandøen og Lille Snenæs i No døs g ønland og på ha is i be ægelse i
slu ningen a maj og begyndelsen a juni ed Wols enholme Fjo d i
No d es g ønland. Tils a ende e esul a e ne a cases udie ne o skellige, inklusi e
sik e obse a ione a h al os i indi ide og g uppe ed Sandøen, men de imod
ingen bes emmelse a h al osg uppe elle indi ide ed Lille Snenæs og
Wols enholme Fjo d. P ojek e ha gi e indsig i mulighede ne og beg ænsninge ne
ed sa elli billede a mege høj opløsning og analyse de a .
Analyse ne ha inklude e isuel inspek ion, an endelse a ænd ingsde ek ion ed
hjælp a mul i a ia al e a ionsde ek ionsalgo i me, ho edkomponen analyse og
ua hængig komponen analyse.
Vi ha be is , a eknologi og analyseme ode kan sikke an endes unde ideelle
oms ændighede , men også u ils ækkelig unde uegnede omgi else .
16
6. Case s udy a eas selec ed
6.1. Case s udy Sandøen
Sandøen (“Sand Island”) is an island in No h Eas G eenland a he posi ion N74°16'
W20° 9' in he en ance o Young Sound, on a sill o med by deposi s o g a el and
sand. The dynamical geological de elopmen s o he leached deposi s ha e
cons an ly changed he size and o m o he island documen ed in he sa elli e
images o he p ojec . In 2015, he island is o a cu le shape, c. 850-900 m long om
no h o sou h and 100 m and 500 m wide om eas o wes , o ally 18.5 hec a es.In
2012, he island was la ge owa ds sou h whe e 3 hec a es was los be ween 2012
and 2015, whe eas he eas e n pa gained 0.8 ha and he no he n ip gained 0.3 ha
in he same ime pe iod. Ae ial images om 1987 (Ko sgaa d e al. 2016) depic he
island as being 1,100 m long and o 27 ha. Since ha ime, a eas o he sou h and
eas ha e been los o he sea, in he wes e n pa , a ea has been gained. Gene ally,
he island is la and low wi h we lands and highe g ound in he sou h-eas e n a ea.
Figu e 3: Sandøen Island cap u ed om opog aphical ae ial pho og aphy in 1985
17
Figu e 4: Sandøen Island cap u ed on Wo ldView-2 image o 2012. Cou esy o Digi alGlobe Inc.
Figu e 5: Sandøen Island cap u ed on Wo ldView-3 image o 2015, cou esy o Digi alGlobe Inc.
Sandøen is one o wo emaining e es ial haul-ou si es o he A lan ic wal us in
No heas G eenland. Wal uses ha e been obse ed on land a o he si es in
No heas G eenland, bu Sandøen and Lille Snenæs a e he only wo emaining
haul ou si es known o be egula ly used by wal uses du ing summe and ea ly all.
Bo h si es a e mainly used by males, while emales a e dis ibu ed along he coas
a he no h (Bo n e al. 1995, 1997). P e ious e es ial haul ou si es in No heas
G eenland, Wes G eenland and No hwes G eenland ha e been abandoned due o
hun ing. In No heas G eenland hese includes wo si es no h o Sandøen, one in
Do e Bug and wo in he a ea close o Sco esby Sound. These wo la e si es a e
he mos sou he n haul ou si es known (Bo n e al. 1995).
18
In 1995, Bo n e al. es ima ed ha he haul ou si e a Sandøen was egula ly used
by a g oup o 20-50 male wal uses. Du ing Augus o 2002 and 2003, ID pho os o
27 and 37 wal uses hauling ou on Sandøen we e aken, espec i ely, and in 2002
and 2004 biopsies we e aken om 38 and 81 wal uses a Sandøen, espec i ely.
Maximum wal us day coun s we e 47 in 1991, 48 in 1994 and in 2004, 60 wal uses
we e obse ed o haul ou simul aneously on Sandøen (see e ences in Bo n and
Aqua one 2007). In 2009, i was es ima ed ha 1429 ( ange: 616 – 3316) wal uses
summe in Eas G eenland (Bo n e al. 2009).
A s udy wi h he pu pose o quan i ying he ophic ole o wal uses in Young Sound
equipped h ee male wal uses wi h sa elli e ansmi e s o eco d mo emen and
di ing ac i i y in Young Sound (Bo n and Aqua one 2007). The s udy was ca ied ou
du ing he open-wa e season in 1999 and 2001. One o he h ee wal uses was
acked bo h yea s. The s udy showed ha o e all he wal uses spen on a e age
abou 44% o hei ime in he Young Sound. When a sea inside Young Sound, he e
was a clea p e e ence o he shallow wa e a eas in he no he n pa s o he jo d
wes o Zackenbe g. Du ing he open wa e season, haul ou ime was on a e age
31,4% ( ange: 21 – 66%) and he wal uses spen on a e age abou 29,5% ( ange:
11 – 45%) o hei ime in he wa e in he Young Sound a ea. Remaining ime was
spen along he coas no h and sou h o Young Sound and o sho e in he
G eenland Sea.
When a sea, abou 32% o he ime he wal uses appea ed a dep hs om 0-6 m
and 11% o his ime hey we e a he su ace. The emaining ime was spen a
dep hs below 6 m.
The moni o ing p og amme BioBasis a Zackenbe g Basic o he G eenland
Ecosys em Moni o ing P og amme o Uni e si y o Aa hus egula ly obse e wal us,
du ing he mon h o July nea he Zackenbe g ield s a ion and he su oundings.
In 2010 GINR, conduc ed s udies o wal us on Sandøen including sa elli e ags and
no mal obse a ions wi hin he BioBasis p og amme we e no pe o med (Hansen, J.
e al 2011). Du ing he s udy’s co e imeline o 22 h July o 7 h Augus a daily
maximum coun was 23.2 indi iduals on a e age wi h he la ges numbe on 31s o
Augus (Bo n, E. e al 2011). The sa elli e ag ansmi e s displayed a els o
Sabine Island, Shannon, Hochs e e Fo land and Do e Bay, all no h o Sandøen, in
he pe iod Augus -Oc obe .
In he 23 d o July 2011, 12 wal uses we e obse ed by he Zackenbe g BioBasis
eam hauling ou , and on 12 h Sep embe one was seen on he su ounding d i ice.
In 2012 he eam achie ed one isi obse ing 12 indi iduals hauling ou , and in
2013, he eam obse ed 11 wal uses on haul ou on he island and addi ionally ou
in he wa e nea he beach. (Hansen e al 2012, Hansen e al 2013).
19
6.2. Case s udy Lille Snenæs
Lille Snenæs (“Li le Snow Headland”) is loca ed in Do e Bug (“Do e Bay”) also in
No h Eas G eenland, bu a he no h han Sandøen. The is hmus and nea by
coas s a e loca ed wes o Danma ksha n s a ion popula ed by he Si ius mili a y
pa ol. The pa ol membe s and o he isi o s egula ly obse e wal us a he si e,
which also holds a small cabin “Lille Snenæs hy en”. The coas line, Winge Kys , is
much ockie and a ied han a Sandøen, bu also inco po a es s ips o beaches o
g a el and sand.
Figu e 6: Lille Snenæs case s udy a ea wi h he Do e Bug (bay) and H al osodden loca ions ma ked. Sa elli e
image: USGS Landsa 8 o 6/8-2015.
Lille Snenæs is, as Sandøen, a e es ial haul-ou si e o wal us in No heas
G eenland. Many o he e es ial haul-ou si es p e iously used by wal us in he
a ea a e now abandoned, e.g. H al osodden (“wal us poin ”) in no hwes e n Do e
20
Bug , because o his o ical hun ing (Bo n e al. 1995). Lille Snenæs is loca ed
app oxima ely 10 km eas o H al osodden and in he summe o 1933, 50 wal uses
we e seen on he beach (see e e ences in Bo n e al. 1995). A la e s udy
sugges ed ha abou 50 males used he haul ou si e a Lille Snenæs (Bo n and
Knu sen 1990). The wal uses mig a e in o Do e Bay du ing he open wa e season
o eed in ensi ely on he mollusc’s banks and o es and moul on land a Lille
Snenæs be ween eeding excu sions (Bo n and Knu sen 1992). Du ing la e July un il
la e Sep embe /ea ly Oc obe , he sho e- as ice is usually absen in no heas e n
G eenland, and his is when he wal uses can exploi he insho e mollusc eeding
banks. The Lille Snenæs haul ou si e is exclusi ely used by males and i is he
same indi iduals ha occupy he beach du ing he open wa e season, and also in
subsequen yea s (Bo n and Knu sen 1992).
In he all, du ing la e Sep embe and ea ly Oc obe , he o ma ion o a dense co e
o land as ice o ces he wal uses o sho e in o he G eenland Sea. He e hey
win e in leads and c acks in he pack ice om 80 – 82°N o he coas o No heas
G eenland (Bo n and Knu sen 1992).
Bo n and Knu sen (1997) obse ed a g oup o abou 50 male wal uses a Lille
Snenæs du ing Augus and Sep embe 1989 and 1990. Eigh o he males we e
equipped wi h sa elli e-linked adio ansmi e s. The s udy showed ha he wal uses
ha use Lille Snenæs o eeding and es ing haul ou o app oxima ely 30% o hei
ime. When ice loes a e a ailable he wal uses spend abou 11% o his ime on ice,
while when ice is absen hey haul ou mo e on land (Bo n and Knu sen 1997).
Du ing he la e summe and ea ly all (Augus -Sep embe ) du a ion o haul-ou
pe iods o wal uses a Lille Snenæs a e aged 11 hou s on ice ( ange: 1 – 29 h) and
38 h on land ( ange: 13 – 64 h). Wal uses a Lille Snenæs haul ou less on land
when he wea he is we , windy and cold, which seem o be a gene al ai o
wal uses, ha p e e o haul ou in wa m and calm wea he . When o aging, wal uses
a e subme ged o abou 81% o he ime (Bo n and Knu sen 1997).
6.3. Case s udy Wols enholme Fjo d
The hi d si e, Wols enholme Fjo d, Ummannap Kange lua, is loca ed in No hwes
G eenland o Thule Ai Base, Pi u ik, a 76°35'N 69°11'W. The a ea o in e es is
loca ed a he en ance o he jo d be ween Mo iusaq hamle and Thule Ai Base,
Pi u ik, a ound he island Appa (Saunde s Ø).
The wal uses ound in he a ea a ound and in Wols enholme Fjo d belong o he
Ba in Bay s ock. These wal uses s ay app oxima ely 2/3 o he yea in No hwes
G eenland om Oc obe –June/July. In June/July when sea ice e ea s om he a ea
he wal uses c oss Smi h Sound and summe in shallow jo ds and inle s a he
Canadian side o Smi h Sound. In Canada, he mos equen ly used summe ing
g ounds a e in he eas e n pa o Ellesme e Island (Heide-Jø gensen e al., in
p ess).
21
In win e , when he Wols enholme Fjo d a ea is la gely co e ed by sea ice, sea bi ds
and ma ine mammals, including he wal uses, ely on he open-wa e polynya, he
No h Wa e Polynya. The No h Wa e Polynya ha is ound wes o Wols enholme
Fjo d is highly p oduc i e and is conside ed key habi a o o e win e ing animals.
The wal uses ha win e he e ha e access o shallow molluscs’ banks and a e
he e o e able o ind ood h oughou he win e mon hs (Heide-Jø gensen e al.
2016).
The e a e no e es ial haul-ou si es in Wols enholme Fjo d and wal uses he e haul
ou on he sea ice. Vibe (1950) men ioned a e es ial haul-ou a Ud le on he sou h
coas o Wols enholme Fjo d eas o Thule Ai Base bu his si e is now abandoned.
Vibe (1950) also desc ibed wal us o aging banks in No hwes G eenland. Fo aging
banks anged om he sou h-eas e n pa o Smi h Sound o Kane Basin cons i u ing
he no he n bounda y o wal us dis ibu ion. Fo aging g ounds in he Wols enholme
a ea was desc ibed o be mainly loca ed a ound Wols enholme Ø and Saunde s Ø
and no h o Kap Pa y.
Acco ding o Vibe (1950) males we e in su plus in he Wols enholme a ea, an
obse a ion ha was suppo ed by Bo n and K is ensen (1981) ha desc ibed ha
mainly males we e aken in he hun a ound Saunde s Ø. Heide-Jø gensen e al. (in
Figu e 7: No h Wa e Polynya open wa e and Wols enholme Fjo d. MODIS sa elli e image om he
NASA Wo ld iew se ice.
22
p ess) also ound ha emales wi h cal es we e gene ally a e in Wols enholme
Fjo d.
Bo n and K is ensen (1981) s udied he occu ence and hun ing o wal us in he
Thule a ea. They desc ibed ha when he sea ice is o ming in he jo ds du ing all
he wal uses a e o ced away om he a ea and he shallow eeding g ounds.
Howe e , hea y s o ms will o en b eak up he sea ice in he jo d and he wal uses
will hen mo e back in o he banks whe e hey s ay un il he ice one again is oo
hick. The wal uses ha e no way o main aining a b ea hing hole bu will ins ead y
o b eak he ice wi h i s head and shoulde s. This is doable un il he ice is
app oxima ely 5-10 cm hick a e which he wal uses a e o ced o lea e he a ea.
Bo n and K is ensen (1981) also desc ibed ha he hun o wal uses nea Saunde s
Ø in Feb ua y is occu ing when he sea ice is s ill hin. Hun e s loca e he wal uses
by hei exhales when a he su ace o b ea h and ha poon hem while s ill in he
wa e . In sp ing, he wal uses sunba he on he loa ing ice loes be ween hei
o aging ips and hey a e hen hun ed when on he ice o in he wa e .
23
7. The capabili ies o e y high- esolu ion imaging sa elli es
This p ojec in ol es he u iliza ion o Ea h obse a ion image y sa elli e da a as he
co e da a sou ce o me hods o iden i ying and coun ing animals. The applied
sa elli es a e o high and e y high esolu ion and ha e low, pola , sun-synch onous
o bi s a ound Ea h. The sa elli es can, when asked and p og ammed, change
cou se and angle sligh ly o obse a ion o speci ic a eas o he ea h a ce ain
pe iods.
Fo obse a ion o adia ion o ligh om he Ea h’s su ace, he sa elli es measu e a
numbe o op ic bands, each eco ding selec ed wa eleng hs o ligh including nea
in a ed and in a ed wa eleng hs, measu ed as he spec al esolu ion. S anda d
bands a e he panch oma ic band (PAN) eco ding g eyscale in high de ail, and he
mul ispec al (MS) bands o less high de ail eco ding wa eleng hs o ed, g een and
blue colou s (RGB) as well as nea -in a ed (NIR). O he wa eleng h bands migh
also be eco ded, such as Sho Wa e In a ed (SWIR), albei a a lowe esolu ion.
Each band is s o ed as a nume ic as e image ile di ided in o equally sized pixels
ep esen ing a pa o he obse a ion on he su ace o he Ea h and as such a
geome ic a ea. The size o each pixel on g ound is he spa ial esolu ion o G ound
Sample Dis ance, GSD, which a ies be ween he image y p oduc as p ima ily a
unc ion o he al i ude o he sa elli e and he capabili ies o he sa elli e’s lens and
image senso . The angle o obse a ion om sa elli e o g ound (o -nadi ) also a ec
he GSD, as he la ge he o -nadi alues he la ge he GSD alues. An o -nadi
angle below 20 deg ees is o en ecommended.
The mul ispec al bands o ed, g een and blue is mixed o a na u al colou RGB
image and esampled o a highe de ail (pan sha pening) using he panch oma ic
band. Fo he cu en analysis also he nea -in a ed band (NIR) o calcula e
b igh ness o he image and u he mo e he NIR- a io, which assis s he de ec ion o
pa icula objec s in he image such as wal us indi iduals and changes in alues
indica ing mo ing objec s.
High and Ve y High Resolu ion Sa elli es a e a ailable only comme cially, unlike
go e nmen al sa elli es such as he low o medium esolu ion MODIS, Landsa and
Sen inel sa elli e amilies, which a e ee o cha ge. The mos dominan dis ibu o s
a e he US company Digi alGlobe Inc. wi h Wo ldView-2 and -3, QuickBi d and
o he s, and he Eu opean company Ai bus De ence and Space.
24
Sa elli e/ GSD PAN RGB & NIR O he MS
Wo ldView-3 0.31 m 1.24 m Coas al, yellow, ed edge, nea -in a ed2
Wo ldView-2 0.46 m 1.84 m Coas al, yellow, ed edge, nea -in a ed2
QuickBi d 0.61 m 1.63 m None
Pléiades 1A/B 0.50 m 2.00 m None
Spo -6 1.50 m 6.00 m None
Table 1: Gene al capabili ies o applied sa elli es
7.1. Reques and sea ch o da a
In o de o ecei e his comme cial sa elli e image y da a, i is necessa y o ei he
eques new obse a ions ia asking, o a chi ed image y. Fo asking eques s, he
cus ome is equi ed o se a ime ame, an a ea o in e es geog aphically, a
maximum cloud co e , and a desi ed numbe o e isi s needed. The success o he
asking inqui y in ob aining obse a ion esul s will ely on possibili ies o e isi s by
he sa elli e, wea he in he a ea o in e es a he isi , and e en ual con lic ing o
mo e impo an eques s o obse a ion. Tasking o he highly capable Wo ldView-2
and Wo ldView-3 is popula wi hin he US go e nmen and esea ch o ganisa ions
and consequen ly, o eign inqui ies a e o en o lowe p io i y.
Conce ning a chi ed da a, a ime ame, maximum cloud co e , and a ea o in e es
is also equi ed om he cus ome in sea ch o a ailable da a. The ou come o he
inqui y is dependen on whe he obse a ions a e gene ally a ailable in he a ea as a
esul o p e ious asking, and popula i y o he a ea, e.g. a eas o high ac i i y such
as u ban cen es a e mo e likely o ha e good and epea ed co e age han
wilde ness o low human ac i i y. In he case o G eenland, some sa elli es like Spo -
6 and Pléiades 1 A/B only ha e e y ew image y scenes a ailable, while he
a chi es o Digi alGlobe (Wo ldView-2, -3 and Quickbi d) hold nume ous scenes
ac oss he coun y, and pa icula ly o he a eas o in e es o US esea ch p ojec s
such as o la ge ou le glacie s om he G eenland Ice Shee . Howe e , he numbe
o scenes pe yea a ies g ea ly, and o some a eas, he e can be mon hs o yea s
be ween he obse a ions.
The cos o equisi ion o asked image y is much highe han a chi ed image y,
pa icula ly o he Ve y High Resolu ion image y, so i possible a chi ed image y
should be eques ed due o he cos s. I was he aspi a ion o he p ojec ha da a
could be eques ed ia he EnhancedView ag eemen be ween he image y da a
p o ide Digi alGlobe Inc. and he Go e nmen o he Uni ed S a es’ o ganisa ion
Na ional Geospa ial-In elligence Agency, NGA. This ag eemen makes i possible o
US go e nmen al as well US esea ch o ganisa ions o eques and u ilize image y o
e y high esolu ion, and he A c ic-An a c ic image y da a is handled by Pola
Geospa ial Cen e a Uni e si y o Minneapolis, whom Asiaq, and o some ex en
GINR collabo a e wi h. Howe e , ou inqui y was no en i ely wi hin he scope o he
ag eemen , which should p ima ily suppo US esea che s wo king on g an s om
Na ional Science Founda ion, NSF, o he Na ional Space Agency, NASA, who
25
un o una ely a e no in ol ed in his p ojec di ec ly. This ac made i impossible o
Pola Geospa ial Cen e , PGC, o accommoda e he inqui y o he p ojec ully;
howe e , he cen e assis ed in asking o Digi alGlobe sa elli es. The p ojec was
able o ecei e esea ch and educa ional discoun h ough he cus ome accoun o
Uni e si y o Copenhagen. DHI GRAS A/S, a Danish consul ancy company, assis ed
as inqui y agen o he p ojec .
The o al quan i y o aw da a om he asking and a chi e eques s is 52 Gigaby es,
o which he majo i y co e s Wols enholme Fjo d, which is a geog aphically la ge
a ea han Sandøen and Lille Snenæs and consequen ly equi es mo e da a o
co e age.
Da a om he dis ibu o s a e o en au oma ically p e-p ocessed p oduc s. To some
ex en , he image y is co ec ed o dis o ions om he acquisi ion angle om he
sa elli e senso o he g ound le el and opog aphically o dis o ions o he e ain
mo phology which obscu es how he image display he g ound geome ically. O he
dis o ions a e usually adiance o ligh in he a mosphe e (TOA, Top o A mosphe e
Re lec ance) om he sunligh e lec ing in he clouds, o sunligh e lec ed om he
su ace o he Ea h (Su ace Re lec ance). Usually i is possible o depend on he
p e-p ocessed da a, bu in some cases, i is necessa y o pe o m calib a ions
cus om o he image. None o he images used o he p ojec needed special
calib a ions.
Da a om he dis ibu o s a e deli e ed in he gene ally used digi al o ma ,
GEOTIFF, including me ada a on he obse a ion and chain o p e-p ocessing, and
can hus be eadily applied o geog aphical in o ma ion sys ems. The p oduc a e
sligh ly di e en depending on hei amily o sa elli e, e.g. Digi alGlobe o Ai Bus.
Sa elli e Fo ma Me ada a Map p ojec ion O he
Wo ldView-3 GEOTIFF IMD Local UTM Tiled, no pansha pened
Wo ldView-2 GEOTIFF IMD Local UTM Tiled, no pansha pened
QuickBi d GEOTIFF IMD Local UTM Pansha pened
Pléiades 1A/B GEOTIFF DIM EPSG:4326 Pansha pened
Spo -6 GEOTIFF DIM Local UTM Pansha pened
Table 2: Sa elli es, o ma s in use, map p ojec ion and ini ial s eps o pansha pening.
32
9. Objec based ea u e ex ac ion
Wi hin emo e sensing classi ica ion calcula ions, you can ei he use pixel-based
ex ac ion o ea u es pixel by pixel, o objec based ea u e ex ac ion connec ing a
g oup o pixels in o egions sha ing simila p ope ies, e en ually ac oss di e en
image y bands and by sea ching o speci ic ypes o shapes.
The so wa e package, eCogni ion (h p://www.eCogni ion.com/), was chosen o help
localize wal uses on he sa elli e images. I is an objec -based image analysis
so wa e, meaning ha i ope a es in a mo e human like ashion by conside ing
g oups o pixels in con ex , as opposed o he adi ional pixel-based me hods. This
can be a bene icial app oach when ying o iden i y ea u es in an image ha do
ha e ce ain speci ic cha ac e is ics ega ding shape, size and colou .
The image is subdi ided, o segmen ed, in o objec s o a ying sizes depending on
he p ope ies and con ex o neighbou ing pixels and depending on he minimum
objec size de ined by he use . Fu he classi ica ion and ea u e ex ac ion can hen
be ca ied ou on hese objec s.
In he ollowing sec ions we ou line he me hods and esul s o he image analyses
o wo egions in NW G eenland; Sandøen and Lille Snenæs. A discussion o he
esul s o each egion is gi en and inally ecommenda ions a e sugges ed based on
expe ience gained so a .
9.1. P inciple Componen Analysis and Independen Componen Analysis
In o de o assess he bene i s, speci ically simplici y and obus ness o he
p ocessing scheme; and o include as much po en ial in o ma ion as possible in he
analyses, P inciple Componen Analysis (PCA) (Comon, 1994) and Independen
Componen Analysis (ICA) (Hy ä inen and Oja 2000) we e de i ed o selec ed
scenes. Bo h o hese me hods a e commonly used o ex ac hidden signals om
mul i a ian da ase s, such as mul iband sa elli e scenes. I is assumed ha he
da ase is a linea mix u e o unknown a iables. All he a iance in he image is
analysed allowing a linea ans o ma ion o he da a on o new un-co ela ed axes.
The esul ing new se o componen s a e equal o he numbe o bands in he o iginal
image and a e as independen as possible. The g ea es amoun o a iance is
cap u ed in he i s componen , wi h p og essi ely less in successi e componen s.
The PCA is based on second-o de s a is ics; limi ed o de ining o hogonal axis; and
elies on Gaussian ea u es. The ICA me hod is based on highe -o de s a is ics and
assumes ha da a is non-gaussian. The o iginal image can be econs uc ed om
all he componen s, bu da a edundancy can be minimised by selec ing only ce ain
componen s.
9.2. Fea u e ex ac ion a Sandøen si e
The Sandøen scenes we e p ocessed i s since he e a e easily iden i iable wal us
in he images (Fig. 1). This is a homogeneous sand island, wi h li le ex a ma e ial o
c ea e noise in he images.
33
Figu e 12: Sandøen, Pleiades-1B image (2015-08-06_141647_PHR1B1) o Sandøen. Red box indica es a g oup
o wal us.
9.2.1. Me hods used o Sandøen case s udy
Each sa elli e scene (Table 3) was analysed as an indi idual p ojec in eCogni ion.
The inpu iles o each p ojec included he iles c ea ed in he p e ious p ocessing
s eps, i.e. 1) he clipped scene; 2) b igh ness image; 3) NIR- a io image; 4) all MAD
images ha used he selec ed scene; 5) he dis ance_ o_ocean as e . Sandøen is a
small enough island ha he clipped image (sec ion 8.1 P e-p ocessing he da a)
wi h he en i e island could be included in he eCogni ion analysis.
In o de o limi he sea ch a ea o egions mos likely con aining wal uses, a
segmen a ion was i s applied using he dis ance_ o_ocean as e wi h h esholds o
5 and 50. This c ea ed a zone ound he coas in which u he analysis was ca ied
ou . Focussing he a ea o in e es o a coas al band assumes ha he wal uses do
no mo e la ge dis ances inland, an assump ion ha is suppo ed by animals
iden i ied in he Sandøen scenes.
34
The p ocessing s eps used o each scene a e lis ed in Table 4.
1. Assign class based on dis ance o sea o c ea e zone o in e es nea coas
(mul i h eshold segmen a ion).
2. Mul i esolu ion segmen a ion using all a ailable da a (excep dis ance as e )
3. Assign class based on h esholding band 3
4. Assign class based on h esholding NIR a io
5. Assign class based on h esholding leng h/wid h a io o objec s
6. Assign class based on dis ance o objec s o class AOI ( emo es indi idual
spa ial ou lie s)
7. Assign class based on h esholding band 8
8. Expo AOI shape iles (be su e o assign class as me a da a
SCENE 1. Mul i-
h eshold
segmen a ion
2. Mul i-
esolu ion
segmen a ion
3. Assign
class;
mean b3
4.
Assign
class;
mean
NIR
a io
5. Assign
class;
leng h/wid h
6.
Assign
class;
Dis o
AOI
7.
Assign
class;
mean b8
8.
Expo
AOI
0804 PH <=5 <
unclassi i ied
<=50 <
no AOI on
dis _ o_sea
10 unclassi ied
< 440:
b3low
b3low >
0.24 :
NIRhigh
X
0806 PH <=5 <
unclassi i ied
<=50 <
no AOI on
dis _ o_sea
10 unclassi ied
< 500:
b3low
b3low >
0.27 :
NIRhigh
X
0819 PH <=5 <
unclassi i ied
<=50 <
no AOI on
dis _ o_sea
10 unclassi ied
< 474:
b3low
b3low >
0.21 :
NIRhigh
NIRhigh
< 100Pxl
: AOI
X
0828 PH <=5 <
unclassi i ied
<=50 <
no AOI on
dis _ o_sea
10 unclassi ied
< 330:
b3low
b3low >
0.21 :
NIRhigh
NIRhigh < 2
: AOI
X
0816 WV2 <=5 <
unclassi i ied
<=50 <
no AOI on
dis _ o_sea
10 unclassi ied
< 364:
b3low
b3low
>=146 :
AOI
X
0828 WV3 <=5 <
unclassi i ied
<=50 <
no AOI on
dis _ o_sea
10 unclassi ied
< 310:
b3low
b3low >
163 :
AOI
X
Table 4: eCogni ion p ocessing s eps used o loca e wal uses in Sandøen scenes.
35
9.2.2. PCA and ICA analyses o Sandøen case s udy
Be o e pe o ming he PCA and ICA, he images we e masked so ha only he
egion o in e es , i.e. he coas al band emained. This limi s he amoun o
unnecessa y in o ma ion in he image ha may educe he use ulness o he de i ed
componen s.
The analyses we e ca ied ou in he so wa e ENVI using he Tho anomaly
de ec ion ool. Fo each sa elli e scene, a isual inspec ion was made o iden i y he
componen s which ga e he bes isualisa ion o he wal uses. The componen and
associa ed band in he o iginal image, which mos in luences ha componen a e
gi en in Table 5. The PCA componen 1 was used o segmen a ion (Table 4, s ep 2)
in he eCogni ion p ojec .
Analysis PCA componen ICA componen
Componen 1 2 3 4 5 6 7 8 1 2 3 4 5 6 7 8
150804(PH) b3
150828(PH) b3
b1
b1
150819(PH) b3 b4
ack
150806(PH) b3
b2
edge
120816(WV2) b3
b4
150828(WV3) b3,
b5,
b7
b1,b5 b
7
b2,
b5,
b8
b5
Table 5: PCA and ICA esul s o each Sandøen sa elli e scene. PH = Pleiades; WV= Wo ldView. (b1-b8 a e he
band numbe s o he o iginal scenes, whe e PH has 4 bands and WV has 8 bands)
9.2.3. Resul s o Sandøen case s udy
Table 4 shows ha using he o iginal sa elli e scenes wi h s anda d de i ed p oduc s
o b igh ness, NIR a io and MAD, he i s h ee s eps we e highly e ec i e a
educing he numbe o po en ial objec s and we e common o all scenes. The ac
ha he h eshold used in s ep 3 is scene dependen limi s he abili y o au oma e he
p ocedu e.
P ocess 4 (o 7 o he Wo ldView scenes) was hen su icien o 4 o he 6 scenes
o selec he wal us objec s. Fo he emaining wo cases a spa ial o geome ic
condi ion was se o emo e emaining noise.
An example o he esul an classi ica ion is shown in Figu e 13, wi h all scenes gi en
in Appendix 1.
36
Figu e 13: Sandøen, Wo ldView 2 scene (2012-08-16_144211-WV2). G een shading a e a eas ou side o he
egion o in e es de ined by dis ance om he sea. The ed ou lines a e he objec s iden i ied wi h he p ocessing
s eps gi en in Table 4.
The PCA and ICA esul s a e gi en in Table 4. A isual check showed ha he PCA
was mo e e ec i e a cap u ing he wal uses (Figu e 14). Fo all scenes, componen
1 o he PCA was he bes , sugges ing ha his may also be ela i ely obus . This is
no su p ising since he i s componen con ains mos o he a iance in he da a. I
is in e es ing o no e ha i is band 3 ha mos in luences he PCA componen 1, and
also band 3 ha de e mines s ep 3 in Table 4. The ICA analyses on he o he hand
we e no so e ec i e a cap u ing he wal uses and we e less consis en in he
componen ha a ed bes .
Using he PCA componen 1 oge he wi h he o iginal clipped image o
segmen a ion in eCogni ion, good objec de ini ion was ob ained in ha he wal us
bodies we e cap u ed as objec s. Howe e , he alues o he PCA componen 1 a e
scene dependen , meaning ha a s anda d h eshold o iden i ying he wal us could
no be used.
37
Figu e 14: Sandøen, Wo ld View 2 scene (2012-08-16_144211-WV2). PCA componen 1. Red box indica es
loca ion o he wal uses
9.2.4. Discussion conce ning he case s udy Sandøen
Sandøen is a ela i ely small and homogeneous a ea whe e wal uses a e easily
iden i iable in he scenes. The scenes we e well co- egis e ed meaning ha he MAD
analyses showed ac ual changes in he en i onmen al as opposed o shi s in he
scene loca ion. In essence, Sandøen is a pe ec example o de i ing a simple
p ocessing scheme. Full au oma ion o he p ocessing is howe e p oblema ic since
h esholds a e used in many o he key s eps and hese a e scene dependen .
In his case, he use o PCA and ICA did no p o ide a be e al e na i e p ocessing
op ion compa ed o he use o he o he de i ed p oduc s, i.e. MAD, b igh ness, and
NIR a io.
9.3. Fea u e ex ac ion a Lille Snenæs si e
Lille Snenæs is a much la ge a ea o in e es co e ing a coas line o app oxima ely
11 km (Figu e 15). I is a highly a ied coas line wi h snow pa ches, i e mou hs,
ocky ou c ops, boulde s, s eep cli s and sandy beaches. No wal uses ha e been
isually iden i ied on he a ailable sa elli e images.
38
Figu e 15: Lille Snenæs, Pleiades scene (2015-08-20-1458000_PHR1A)
9.3.1. Me hods
As o Sandøen, he inpu iles o he eCogni ion analysis included he iles c ea ed
in he p e ious p ocessing s eps (sec ion 8.1 P e-p ocessing he da a), i.e. 1) he
clipped scene; 2) b igh ness image; 3) NIR- a io image; 4) all MAD images ha used
he selec ed scene; 5) he dis ance_ o_ocean as e . Due o he size o he egion,
addi ional s eps had o be in oduced o limi he da a.
In addi ion o c ea ing a bu e a ound he coas , addi ional segmen a ion was ca ied
ou o classi y snow, wa e and s eep slopes in eCogni ion. These shapes we e
expo ed and used o clip he image u he (Figu e 16: Lille Snenæs a ea o in e es
a e masking ou s eep slopes, snow, wa e and 50 m inland om he coas .).
Focussing he a ea o in e es o a coas al band and emo ing ce ain g ound ypes,
assumes, as o Sandøen ha he wal uses do no mo e la ge dis ances inland, bu
also ha hey a e no o be ound on snow pa ches, no on s eep slopes.
39
The p ocessing s eps applied o Sandøen we e applied o Lille Snenæs as a s a ing
poin and h esholds a ied o expe imen wi h he de i ed ea u es.
9.3.2. PCA and ICA analyses o Lille Snenæs
The PCA and ICA analysis we e again pe o med in he so wa e ENVI using he
Tho anomaly de ec ion ool.
9.3.3. Resul s and Discussion o Lille Snenæs
Lille Snenæs p esen s many challenges no p esen in he Sandøen si e, no leas
he size o he egion and he a iabili y o he physical en i onmen . In o de o ocus
Figu e 16: Lille Snenæs a ea o in e es a e masking ou s eep slopes, snow, wa e and 50 m inland om he coas .
40
o egions o in e es we applied a se o ules ega ding whe e we expec ed o ind
wal us. By clipping away egions o e y s eep opog aphy, wa e o snow he sea ch
a ea and a iabili y o wi hin he sea ch a ea dec eased. In e ms o compu a ional
ime, his made he da a much easie o wo k wi h.
The scenes om his a ea we e ex emely di icul o co- egis e (see sec ion 8.1.2
Aligning he images o each si e) wi h he esul ha he e is a shi be ween hem.
As a esul o his he MAD analyses a e likely no e y use ul as hey will show shi s
in he scenes, in addi ion o any en i onmen al changes. A seconda y p oblem
associa ed wi h his is ha he land/ocean mask used o clip he images is less
accu a e.
In he absence o a known a ge ea u e, i.e. an iden i ied wal us, i is e y di icul o
de elop o assess any p ocessing scheme. The p ocessing s eps documen ed in
Table 1 equi e h esholds, which a e scene dependen , bu also en i onmen
dependen . The PCA and ICA esul s a e scene dependen . An objec in di e en
scenes will no show up in he same PCA axis due o dependence on he empo al
beha iou o o he pixels in he scene.
The wal uses a Lille Snenæs haul-ou o app oxima ely 30% o hei ime and when
ice loes a e a ailable abou 11% o his ime is on ice. The wal uses haul-ou less
when he wea he is we , windy and cold compa ed o sunny and wa m. The
p obabili y o wal uses hauling ou on he Lille Snenæs beach o ice loes in nea
icini y o he beach is hus be ween abou 20-30%. Howe e , i he wea he was
cold and windy he day he images we e aken he p obabili y a e e en less (only
sa elli e images om days wi h clea skies a e used). Wal uses hauling ou on ha
speci ic beach a Lille Snenæs should in heo y be easily ecognizable as is he case
wi h wal uses hauling ou a Sandøen. Howe e , we ecommend ha sa elli e
images a e ob ained and analysed o each day when wea he condi ions a e
a ou able om July h ough Augus o inc ease he p obabili y o wal uses being a
he beach a ha exac ime when he image a e aken.
9.4. Recommenda ions o objec -based analysis
Based on he expe ience gained so a , he ollowing ecommenda ions can be
made;
1) Good co- egis a ion o he sa elli e images educes p ocessing ime by
allowing a single, accu a e land mask o be used. I is also essen ial o educe
noise in he MAD analyses.
2) Images should be a mosphe ic co ec ed o elimina e a mosphe ic and
illumina ion e ec s. This is pa icula ly impo an o scenes acqui ed so a
no h whe e hese e ec s a e la ge .
41
3) The accu acy o he land mask is impo an i used in he i s s ep o de ine
he a ea o in e es , since i is e y possible o cu o a eas o po en ially
in e es ing coas line wi h a s anda d pa ame e .
4) The g ea e he esolu ion o he da a, he mo e pixels ha will co e he
wal us ea u e. The 0.5 m esolu ion images used in he cu en s udy a e
p obably an uppe limi o an animal he size o a wal us.
5) PCA and ICA analyses gi e an al e na i e ep esen a ion o he sa elli e da a.
The ad an age wi h he de i ed componen s is ha hey educe edundancy
in he da a so ha no all o he componen s need o be used in u he
analysis. Selec ed componen s can be e ec i e addi ional in o ma ion o
include in he eCogni ion segmen a ion s ep o help de ine he objec s o
in e es . Fu he use o he componen s in classi ica ion based on mean
h esholds will be scene dependen .
6) Wi hou a posi i e iden i ica ion o a wal us in a gi en a ea o in e es i is e y
di icul o assess how ans e able he p ocessing s eps de eloped o he
Sandøen egion a e. I is highly likely ha di e en en i onmen al se ings will
equi e sligh ly di e en p ocessing s eps. In iew o his he ollowing s eps
a e conside ed p uden when assessing a new a ea;
a. Use agged wal uses o iden i y egions o wal us ac i i y.
b. Wi hin hose egions iden i y a eas o mos likely wal us isi s based on
opog aphy and geology (e.g. gen ly sloping sand banks).
c. Acqui e images and c op o a ea o in e es . Use MAD analyses and
isual check o spo wal uses. I co- egis a ion is poo , isually check
only. Once a posi i e iden i ica ion has been made, a p ocessing ee
can be de eloped.
48
coo dina es, bu he Dopple E ec posi ioning om A gos ins umen s on boa d
sa elli es wi h a lowe accu acy. The ad an age is, howe e , a lowe powe
consump ion and hus longe ange. The esul s o he sa elli e agging o wal us in
he egion is published in Ga de e al 2018.
The eco dings om he animals co e a wide a ea and h oughou mo e days han
he sa elli e images acqui ed. The sa elli e image y o he pe iod in June 2015
consis o a SPOT-6 image o 150 cm esolu ion and a Wo ldView-3 image o 30 cm
esolu ion (Figu e 23). The 1x1 km squa es in he c. 900 sq. km. g id suppo ed he
sys ema ic obse a ion ac oss he la ge a ea, howe e di e en be ween he scenes
acco ding o hei co e age, e.g. co e age o he SPOT-6 image is e y much la ge
han he Wo ldView-3 image. Tes ing demons a ed ha dis inguishing any objec s in
he SPOT-6 150 cm esolu ion images was highly unce ain, so u he obse a ions
we e disca ded using SPOT-6 as a sou ce.
Figu e 23: Posi ions o wal us a he da e 2018-06-06 eco ded wi h A gos ansmi e s. Few posi ions a e o bes
quali y (A gos quali y 3), and mo e han hal is o poo posi ion quali y 0 (c. 500-1500 m accu acy). The ed poin s
display he obse a ions done om he 30 cm Wo ldView-3 image o h ee days la e a he da e 2015-06-09.
49
10.4. Recommenda ions om isual obse a ions
The isual inspec ion o he sa elli e images has he same limi a ions as he
au oma ed and analysing me hods conce ning he accu acy o he co-posi ioning o
he images o de ec ion o mo ing o ixed objec s, e.g. a s one o a wal us; he
image esolu ion, e.g. 0.5 m o 1.5 m; and he en i onmen al condi ions on he
g ound du ing he image acquisi ion, e.g. clouds, haze, ocky o sandy su ace
geology and solid o ac ionalized sea ice.
The mos a ou able condi ions o wal us de ec ion on land a e a ligh and e en
sandy beach, which should be isible du ing he haul ou season o he wal us,
usually in Augus . Fo he jo d a ea, he sea ice should be solid and smoo h, so i
will appea mos ly as a whi e backg ound emphasizing he da ke animals, bu
ne e heless wi h a eas o open wa e and channels o he wal us o swim and eed
be ween haul ou on he ice.
The 0.5 m pansha pened image esolu ion o Pleiades-1A/B, Wo ldView-2 and
Wo ldView-3 is su icien o obse a ion o wal us g oups o indi iduals. Mo e han a
single image o each a ea is needed o de e mine mo ing om ixed objec s. I is no
possible o de ec any animals om coa se esolu ion han 0.5 m pixels.
Fo la ge a eas, i is encou aged o lay ou a 1x1 km g id o suppo a s uc u ed and
me iculous inspec ion and ma k each g id cell as inspec ed be o e mo ing o he
nex .
11. Discussion
Some deg ee o isual inspec ion o he high- esolu ion sa elli e image y is
manda o y o de ec ce ain objec s be o e any supe ised, unsupe ised
classi ica ion o objec based calcula ion is s a ed. This is simply o ge an idea o
wha p ope ies o he objec can be ound h ough he analysis done ia he
classi ica ions: he leng h and wid h, shape o shadows, and po en ial pixel alues in
isual colou s ( ed, g een, blue) and he o humans non- isual colou s o nea
in a ed (NIR) and any o he bands a ailable o possible de ec ion.
Ob iously, he choice o season and e en speci ic weeks is o g ea es impo ance
and de e mines he absence o p esence o animals in he a ea. Fo he Sandøen
case s udy he usual haul ou season is well known, as he g oup isi ing he island
is moni o ed annually. This was no he case o Lille Snenæs, which is less o en
moni o ed. Fo he sea ice a ea in he mou h o Wols enholme Fjo d, he sa elli e
obse a ion o 2015 coincided in da e mo e o less he sa elli e acking o animals in
he a ea. Howe e , he acking coo dina es didn’ allow o exac posi ional ma ch
be ween a acking poin and a pixel in he image, and he en i onmen al condi ions
in he images didn’ allow o con iden iden i ica ion o animals. The obse a ion da a
and p ocedu es ca ied ou o Wols enholme Fjo d we e solid conside ing he
possibili ies wi h bo h GPS(?) acks and a ailable image y, howe e ailed due o
50
accu acy in posi ion o he sa elli e images as well as he animal ags, and mo eo e
hei iming be ween images and ag da a and he en i onmen al ci cums ances
conce ning loa ing sea ice.
A likely eason o missing obse a ions is he absence o wal us on ice when hey
swim o eed in wa e . I he ice is hin du ing au umn and win e , wal us s ay in he
wa e , his is also ue du ing mo e un a ou able wea he condi ions. The sa elli e
pho os a e only pu chased om days wi h a clea sky, as he op ical sa elli es used
a e no able o obse e he g ound h ough clouds o s ong haze. In sp ing, he
wal us haul-ou on d i ing ice loes, which a e no s a iona y making he ime ame
o obse a ion limi ed.
The ci cums ances a ound he p esence and absence o wal us, he en i onmen al
condi ions and sa elli e image capabili ies s ongly a ec he au oma ed analysis as
well. Conce ning change de ec ion h ough Mul i a ia e Al e a ion De ec ion (MAD) a
i m co-posi ion be ween he images is ul ima ely equi ed no o de ec alse
changes: we could use he MAD on Sandøen wi h good posi ional co espondence
bu no on Lille Snenæs wi h poo posi ioning. The MAD analysis da a is applied o
he objec -based ea u e ex ac ion, which weigh i s esul s s ongly. Mo e simple
analysis o b igh ness, nea in a ed and dis ance o he coas is no inco po a ing
changes and can hus be applied independen o he MAD. Howe e , we chose no
o pe o m an objec -based ea u e ex ac ion o P incipal Componen Analysis
(PCA) o Independen Componen Analysis (ICA) o he Lille Snenæs, because o
he absence o he MAD analysis.
The en i onmen ally s able and geog aphically con ined Sandøen island is he
pe ec and success ul example o bo h isual inspec ion and au oma ed analysis
p ocessing. Howe e , ou s udy p oo ha such posi i e ac o s a e no a all places
a ailable. Un o una ely, he en i onmen al condi ions du ing he acquisi ions o he
sa elli e images o he Lille Snenæs and Wols enholme Fjo d a eas did no allow o
con iden de ec ions using any au oma ed o isual me hods.
51
12. Recommenda ions om he p ojec
In his s udy, we ha e done a li e a u e e iew conce ning wal us ecology and
beha iou , in es iga ed he gene al en i onmen al ci cums ances o hei haul ou
a eas on he obse a ion da es selec ed, conside ed he abili ies o he a ailable
op ical sa elli e senso s and p epa ed he downloaded images o u he analysis.
The emo e sensing analyses ha e inco po a ed da a om MAD change de ec ion,
b igh ness and NIR a io calcula ions done on he images, as well as calcula ions o
p oximi y o coas . In he analysis, we ha e applied a P incipal Componen Analysis
and Independen Componen Analysis o he image da a and used he image bands
and calcula ions o pe o m objec -based ea u e ex ac ion. Mo e simply, we ha e
also isual inspec ion o e e y image o possible o ac ual wal uses. In
Wols enholme Fjo d his was assis ed by ac ual GPS acking o indi iduals.
In his s udy, we ha e done a li e a u e e iew conce ning wal us ecology and
beha iou , in es iga ed he gene al en i onmen al ci cums ances o hei haul ou
a eas on he obse a ion da es selec ed, conside ed he abili ies o he a ailable
op ical sa elli e senso s and p epa ed he downloaded images o u he analysis.
The emo e sensing analyses ha e inco po a ed da a om MAD change de ec ion,
b igh ness and NIR a io calcula ions done on he images, as well as calcula ions o
p oximi y o coas . In he analysis, we ha e applied a P incipal Componen Analysis
and Independen Componen Analysis o he image da a and used he image bands
and calcula ions o pe o m objec -based ea u e ex ac ion. Mo e simply, we ha e
also isual inspec ion o e e y image o possible o ac ual wal uses. In
Wols enholme Fjo d his was assis ed by ac ual GPS acking o indi iduals.
Based on ou expe iences om abo e, we ecommend he ollowing o ca ying ou
u u e obse a ions and de ec ions o wal us o simila species om sa elli e
image y:
1) Tho ough s udy o li e a u e, expe iences and acking da a a ailable o
decide on ele an and con ined geog aphical a ea and he ime o he yea .
2) A p ecise coas line ec o map will assis he co ec ou line o a eas o
in e es used in o de ing o image y om he p o ide s, and also o co-loca e
he images and p oduce coas line p oximi y maps.
3) A de ailed and well posi ioned digi al e ain model can g ea ly assis in
iden i ying gen ly sloping o la a eas along he coas , which can possibly be
used as a haul ou si es by he wal uses, as opposed o ough and s eep
coas lines.
4) Accu a e co- egis a ion o sa elli e images is necessa y o pe o m good MAD
analyses and p oduce co ec p oximi y o coas line masks as well as he
isual de e mina ion o mo ing objec s om s a iona y objec s.
5) The 0.5 m esolu ion is su icien o he de ec ion o wal us by isual
inspec ion and in au oma ed analysis. Less de ailed esolu ion is no iable.
Highe esolu ion, 0.3 m, al hough desi able, is no manda o y.
52
6) Mul i a ia e De ec ion Analysis poin ing ou changes be ween wo images
pe o ms well o highligh ing wal us on land. I does no wo k i co-loca ion is
poo o he en i onmen mo ing.
7) Visual iden i ica ion o a wal us o a posi i e Mul i a ia e De ec ion Analysis
esul mus be used o he design o a emo e sensing analysis p ocessing
ee.
8) PCA and ICA analyses can de ec componen s in he images, which can be
e ec i ely used in he objec -based segmen a ion and ea u e ex ac ion.
9) A posi i e isual iden i ica ion o a wal us in an image is equi ed o assess
he esul s o he emo e sensing analyses.
10) Sa elli e agging o wal us indi iduals is op imal o iden i ying he ime o hei
p esence in he a ea. Howe e , o di ec ly compa e ack posi ions wi h pixels
in he images, he ags need o be o he mo e p ecise GPS posi ioning
sys em a he han he low p ecision Dopple E ec sys em.
13. Acknowledgemen s
The p ojec eam pa icula ly hanks he En i onmen al P o ec ion Agency o he
Minis y o En i onmen and Food o Denma k, bu also G eenland Ins i u e o
Na u al Resou ces and Asiaq G eenland Su ey o unding ou wo k.
The p ojec epo au ho s would like o hank ou p ojec eam membe s and
collabo a o s:
E a Mä zle , Head o Remo e Sensing g oup a Asiaq G eenland Su ey, o he
s ong p ojec managemen and aining a angemen s, as well as emo e sensing
da a p epa a ion and discussions on wal us mo emen beha iou s and iming o
image acquisi ion.
Mikkel Lydholm Rasmussen, Remo e Sensing Specialis o DHI GRAS A/S o his
se ice du ing image selec ions and pu chases, aining o he p ojec g oup on VHR
sa elli e image da a managemen and objec -based ea u e ex ac ion in eCogni ion,
and emo e sensing discussions du ing he p ojec .
Fe nando Uga e, Head o Depa men o Bi ds and Mammals, G eenland Ins i u e o
Na u al Resou ces, o his assis ance in he ini ial discussions o wal us di ing and
haul ou beha iou be o e o de ing o sa elli e image y.
Mads Pe e Heide-Jø gensen, P o esso o Depa men o Bi ds and Mammals,
G eenland Ins i u e o Na u al Resou ces, o his assis ance du ing he phase o
applica ion o unding and suppo conce ning wal us beha iou du ing w i ing o he
epo .
53
Rikke Guldbo g Hansen, Scien is o Depa men o Bi ds and Mammals, G eenland
Ins i u e o Na u al Resou ces, o he suppo o da a om sa elli e acking o wal us
in Wols enholme Fjo d and Smi h Sound.
Michelle LaRue and Se h S aple on, Uni e si y o Minneso a, o hei assis ance and
discussions du ing he ini ial phase o he p ojec .
54
Re e ences
Ande sen LW, Bo n EW, S ewa REA, Die z R, Doidge DW and Lan hie C (2014):
A gene ic compa ison o Wes G eenland and Ba in Island (Canada) wal uses:
managemen implica ions. NAMMCO Sci. Publ. 9:33–52.
Acqua one, M. (2004): Body composi ion, ield me abolic a e and eeding ecology o
wal us (Odobenus osma us) in no heas G eenland. PhD hesis. Na ional
En i onmen al Resea ch Ins i u e 2004.
Bo n, E. W. (2005): The wal us in G eenland. Ilinniusio ik Unde isnings o lag,
Nuuk. pp. 1-80.
Bo n, E. W. and K is ensen, T. (1981): H al ossen i Thule. Na u ens e den, no. 4:
132-143.
Bo n, E.W. and Knu sen, L. O. (1990): Sa elli e acking and beha iou al
obse a ions o A lan ic wal us (Odobenus osma us osma us) in NE G eenland in
1989. Teknisk appo – G ønlands Hjemmes y e, Miljø- og Na u o al ningen. N .
20 – Ok obe 1990. 68 pp.
Bo n, E. W. and Knu sen, L. Ø. (1992): Sa elli e-linked adio acking o A lan ic
wal uses (Odobenus osma us osma us) in No heas e n G eenland, 1989-1991. Z.
Säuge ie kd. 57(5):275–287.
Bo n, E. W., Gje z, I., Ree es, R. R. (1995) Popula ion assessmen o A lan ic
wal us. No sk Pola ins . Meddelelse no.138:1–100.
Bo n, E. W. and Knu sen, L. Ø. (1997) Haul-ou and di ing ac i i y o male A lan ic
wal uses (Odobenus osma us osma us) in NE G eenland. J. Zool. 243(2):381–396.
Bo n, E.W. and Acqua one, M. (2007): An es ima ion o wal us (Odobenus
osma us) p eda ion on bi al es in he Young Sound a ea (NE G eenland). In:
Rysgaa d, S. & Glud, R. N. (Eds.), Ca bon cycling in A c ic ma ine ecosys ems:
Case s udy Young Sound. Medd . G ønland, Bioscience 58: 176-191.
Bo n, E.W., Boe mann, D.M., Heide-Jø gensen, M.P., Die z, R., Wi ing. L., Kyhn,
L., Rige , F.F., Laid e, K. and Uga e, F. (2009): Abundance o A lan ic Wal us
(Odobenus osma us osma us) in Eas G eenland. NAMMCO SC/17/WWG/07.
Bo n, EW, Ande sen, SB., Joensen, S., Knu sen, LØ, Co yell-Ma in, M., Ande sen,
LLW., (2011): 6.13 Wal us s udies wi h a no e on bi ds on Sandøen. In: Jensen, L.M.
and Rasch, M. (eds.) 2011. Zackenbe g Ecological Resea ch Ope a ions, 16 h
Annual Repo , 2010. Aa hus Uni e si y, DCE – Danish Cen e o En i onmen and
Ene gy. 114 pp.
Comon, P. (1994): Independen componen analysis—a new concep ? Signal
P ocessing, 36: 287–314.
55
F e well, PT, S aniland, IJ., Fo cada, F.. (2014): Whales om space: Coun ing
sou he n igh whales by sa elli e. Plos One, Feb ua y 2014, 9 (2), e88655.
Ga de, E., Jung-Madsen, S., Di le sen, S., Hansen, R.G., Zingle sen, K.B., and
Heide-Jø gensen, M.P. (2018): Di ing beha io o he A lan ic wal us in high A c ic
G eenland and Canada. J. Expe imen al Ma ine Biol. & Ecology 500 (2018): 89-99.
Hansen, LH., Hansen, J., Smedegaa d Nielsen, Humaidan, J., Ch is o e sen, KS.,
Schmid , NM., (2014): The BioBasis p og amme, In: Zackenbe g Ecological
Resea ch Ope a ions, 19 h Annual Repo 2013, DCE Uni e si y o Aa hus.
Hansen, L.H., Hansen, J., Ch is o e sen, KS., Lund, M., Smedegaa d Nielsen, P.,
Ch is ensen, MU., Schmid , NM., (2012): Zackenbe g Basis. The BioBasis
p og amme. In: Zackenbe g Ecological Resea ch Ope a ions, 18 h Annual Repo
2012, DCE Uni e si y o Aa hus.
Hansen, J., Hansen, LH., Ch is o e sen, KS, Albe , KR, Scholle Sko gaa d, M.,
Bay, C., K is ensen, DK, Be g, TB, Lund, M., Boulange -Lapoin e, N., Læ kedal
Sø ensen, P., Ch is ensen, MU., Schmid , NM, (2011): 4. Zackenbe g Basic. The
BioBasis p og amme. In: Jensen, L.M. and Rasch, M. (eds.) 2011. Zackenbe g
Ecological Resea ch Ope a ions, 16 h Annual Repo , 2010. Aa hus Uni e si y, DCE
– Danish Cen e o En i onmen and Ene gy. 114 pp.
Heide-Jø gensen, M.P., Sinding, M.S., Nielsen, N.H., Rosing-As id, A., Hansen,
R.G. (2016). La ge numbe s o ma ine mammals win e in he No h Wa e polynya.
Pola Biol. 39 (9): 1605–1614.
Heide-Jø gensen, M.P., J. Flo a, A. O. Ande sen, R. E. A. S ewa , N. H. Nielsen
and R. G. Hansen (2017). Wal us mo emen s in Smi h Sound: a Canada–G eenland
sha ed s ock. A c ic 70 (3):308.
Hy ä inen, A., Oja, E. (2000). Independen componen analysis: algo i hms and
applica ions, Neu al Ne wo ks, Vol. 13, Issues 4–5, June 2000, pg. 411-430, ISSN
0893-6080, h p://dx.doi.o g/10.1016/S0893-6080(00)00026-5.
Je e son, T., Webbe , M., Pi man, R. (2015). Ma ine Mammals o he Wo ld, 2nd
Edi ion. 2015. A Comp ehensi e Guide o Thei Iden i ica ion, he wal us, s. 440.
Ko sgaa d, N J., Nu h, C., Khan, SA., Kjeldsen, KK., Bjø k, AA., Schomacke , A.,
Kjæ , KH. (2016). Digi al ele a ion model and o hopho og aphs o G eenland based
on ae ial pho og aphs om 1978-1987 (G150 AERODEM) (NCEI Accession
0145405). NOAA Na ional Cen e s o En i onmen al In o ma ion. Da ase .
LaRue, M., S aple on, S. (2018): Es ima ing he abundance o pola bea s on
W angel Island du ing la e summe using high- esolu ion sa elli e image y: a pilo
s udy. Pola Biology.
56
Le e mann, N., Gala ius, A., Ehlmé, G., Rysgaa d, S. and Bo n, E.W. (2003).
Feeding beha iou o ee- anging wal uses wi h no es on appa en dex ali y on
lippe use. BMC Ecology 3:9.
McMahon, CR., Howe, H., an den Ho , J., Alde man, R., B olsma, H., Hindell,
MA. (2014). Sa elli es, he all-seeing eyes in he sky: coun ing elephan seals om
space. Plos One, Ma ch 2014, 9 (3), e 92,613 h
Nielsen, A., & Con adsen, K. (1997). Mul i a ia e al e a ion de ec ion (MAD) in
mul ispec al, bi- empo al image da a: A new app oach o change de ec ion s udies.
S aple on, S., LaRue, M., Lecom e, N., A kinson, S., Ga shelis, D., Po e , C.
A wood, T. (2014). Pola bea s om space: Assessing sa elli e image y as a ool wo
ack a c ic wildli e. PLoS ONE, July 2014, 9 (7), e101513.
Vibe, C. 1950. The ma ine mammals and he ma ine auna in he Thule Dis ic
(no hwes G eenland) wi h obse a ions on ice condi ions in 1939-41. Medd.
G ønland, Bioscience 150 (6): 1–115.
Wiig, Ø., Bo n E. W. and S ewa , R. E. A. 2014. Managemen o A lan ic wal us
(Odobenus osma us osma us) in he a c ic A lan ic. NAMMCO Sci. Publ. 9: 315–
342.
57
Appendix 1.
Resul s o Table 4 p ocessing s eps o each scene. Red ou lines a e he po en ial
wal us objec s.
Scene: 2012-08-16_144211-WV2
