Testing of multispectral scanner data for prospecting of ferro-eclogite in the F0rdefjord area, western Norway

NGU-BULL436,2000-PAGE 19 3

Testing of multispectral scanner data for prospecting of ferro-eclogite in the F0rdefjord area, western Norway

BJ0RN A.FOLLE5TAD,AREKORNELlU55EN&NIGELJ.COOK

Foll estad,BA,Kornel iussen,A.&Cook, N.J.2000:Testingofmultispect ralscanner dataforprospecti ngofferro- eclogiteintheFordefjordarea,WesternNorway.Norgesgeologi skeundersekelseBulletin436,193-201.

The present investigati on aimed to determinewhet her remote sensing datacan be used in futureprospectingfor rutile-bearing eclogit e.Apreliminarystudywas carriedoutto establish if themet hodis suita ble for distingu ishing different key rock types atEnqebefje ll,western Norway,inwhich large volumesofrutile are hostedwithinferro- eclogite.Thest udywascarried outinthreeparts:laboratory measurementof reflect ivity curves on represent ative samples,fieldmeasurement ofsamplesusing aport able spectralradiomete randspect ralanalysisof data collected wit h the airbo rnehyperspect ral scanner. Both thelaborato ryand thefieldmeasurement sindicated that therock surfaces offerro-eclog itegenerallyhave alowerreflect ivitycompared wit h thesurfacesof bothleuco-eclogite and gneiss.Differ ent%reflectivity valuesin distinctpartsof thespect rumcanapparently beusedasatooltodiscriminate betweenareascharacte risedby gneiss,leuco-eclogite and ferro-eclogite.However,thepresenceof mosson the rock surfacewillchangethese characterist icssignificantlyand makeprecisecharacte risatio ndifficult. The airbo rne scanner testingwashamperedin this st udy bytheunavailability ofthosebandsthatprovedusefulfordistinguishing eclogite typesin thefield and laborat ory.However,they areuseful in that theyrevealameanstodiscrim inate between vegetatedand barerocksurfaces. Theapplicationofairborne measurement techniques andsuitable data processing would appear to be a promising meth od with potential for routine application in prospect ing.

Appli cation is, however, depende nt upon the availability of channelsfor the instrum entation being more ap prop riatethan inthis st udy. Also,the considerable influence ofveget at io nand moss onthedata wouldappear to be crit icaltosuccessfulimp lementat ionof themethod.

BjernA.Foll estad,AreKorn eliu ssen&Nige lJ.Cook,Geolog ica lSurvey of Nor wa y,N-7491 Trondheim,Nor wa y.

Introduction

In 1996, acovenant agreement for test ing the poten tia lof the multispec tra lscanner tech niq ue asa prospecting tool in an area ofSoqn&Fj ordane county was establishedbetween the Sogn&Fj ordanecounty administration (Department of RegionalPolicy)and the Geological Surveyof Norway(NGU).

The studieswere related to on-going geologicalinvestiga - tions in the Enqj ebefj ell area in the Ferdefjord district of weste rnNorway(Fig. 1).

TheEnqj ebefj ell eclogitedepositislocated onthe north- ernsideof Fe rdefj ordnearthe small com m unityof Vevring in Naustd alcom m une.The Fordefj ord area belon g s geologi- cally to the Western Gneiss region,sit uate d between the DevonianKvamshesten Basinto thesout hand the Devonian HasteinenBasin to thenort h.The Ferdefjordregion consists of a variety of amphibolitic,eclogit ic and gabbroic rocks, together wit h tonalitic, dioritic and granitic gneisses.The Enqjebe eclogite isa 2.5km-long, comp lexly deformed,lens- shaped body surrounded by alternating mafic and felsic countryrocks(Korneliussen&Foslie1985).Theprotolithto theeclogiteis believed to be aFe-andTi-rich gabbro ofPro- terozoi c age.Transform at ion intoeclogiteisrelated to Cale- doni anhigh-pressuremetamorp hi sm atca.400 Ma.Du ring this process, i1meni teinthe gabbroprotolit hwasreplacedby rutile.The eclogitebody is subdivided intoa leuco-eclogite and a ferro-eclogitevariety.Ferro-eclogite, whichisdistinctly enric hedin bothFe(>14%Fe203)and Ti(>3%Ti02;primaril y

N o km

l'

1

Fig.1.Locatio noftheFordefjord districtofwesternNorway and the area around Engebofj ellwheretheairborne multispectralscannerdata were collected(Figs.9 and 12).The eclogitearea is shaded.

as rutile), has been the target for rutile explorat ion.This investigation addressedwhetherremote sensing data can be usedin fut ureprospecting for rutile-bearing eclogit eof the Enqj ebefj elltype.

Remote sensing as a prospecting method

Rem ot e sensing technology applied to Eart hresource moni- toring and management hasbeen em ploye d on a globa l scale sincethe early 1970swhenthe first EarthResource Sat- elli te (ER5)was launched. Thissat ell it e, later renamed the

NGU-BULL436,2000-PAGE 194

Landsat Eart hresourc es satellite,was the first of a seriesof 6 satelli tesdesigned to provide a near-global coverage of the Earth 'ssurface on a regula r andpred ict able basis;Landsat 6 was,however, lost on launch. The return beam vidicon(RBV) which is a TVcamera-like instrument, and the multispectral scanner (MSS)were the principal sensors on Landsat 1,2and 3. These instrumentswere operating in the ranges 0.4-0.8 and 0.5- 1.1 IJm,respectively.An additional thermal band 7 (10.3 -12.6 IJm)was added on Landsat 3 and the 3RBV cam- eras were exchan gedfor 2 RBV cameras, offering improved resolution (see Richards & Jia 1999).TheThemat ic Mapper (TM), substituted fortheRBV cameras,permitted improved spectr al, spatial and radiometric scanner characteristics in Land sat 4 and 5.An addi ti onalband (7) in the2.08- 2.35IJm region was added to the six existing band s (1-6) at the requ estof thegeolog icalcom mun it y, duetothe im po rtance of the 2IJm reg ion in assessing hyd roth ermal alte rat ionon theEarth's surface.

Multispectral line scanners havebeen availab le for civil aircraft since the early1970s, and have cont ribute d to theut i- lisation , understanding and inte rp ret at io n of the scann er data obta ined through Eart h orbiti ngsystems.Theseaircraft scanners,e.g., theDaedalusAADS1240/1260mult ispectral line scanners, operate in the ultravio let,visible/reflective IR and the thermal part s of the spectrum, and permit data acquisition in 12 wavebands.Along with the AirborneThe- matic Mapper(ATM),these airborne missions were used,for example,for simulati onstudiesprior tothe launch of Landsat 4. At 12 km altitude, the ATM airborne scanner produces an equivalent area pixel size on the ground(30 m)as the Land- sat TM.The availabilityof new detector technologies madeit possible, throughoutthe 1980s, to develop aircraft scanners capable of recording in a large number of spectral bands (Richards&Jia 1999).For example,the Hyperspectral Digital ImageCol lection Exp eriment(HYDICE)carried out bythe US NavelResearchLabused 206 channels,and the Airborne Vis- ibleand Infrared Imagi ngSpectrometer(AVIRIS)flown by JPL used63 channels. Thesedevelop ment s in scanner technol- ogy, and in the subseque nt processing of gathered spectr al info rmat ionthrough datahandl in g,haveprov idedtheGER 63 Chan nel Dias (Digita l Airborn e Imaging Spectr omet er) used in this study. This inst rument is an airbo rne unit design ed for acq uisit ionof spectral inform ation (e.g.,envi- ronme nta lst udies, geolog ical mapp ing) .

The pot en tial of multi- andhyperspect ralscanner data for geologicalmappi ngand identi fication of mineralisedzones on the Eart h'ssurface has long been recognised and has been testedin many parts of the world(e.g.,(hang &Coll ins 1983,Lyon 1996,Hilkka etal. 1998).In Norway,however, only a few such studieshave been carried out. These have mostly dealtwith mapping of lineaments(e.g., Rindstad &Follestad 1982,Wester et al. 1990, Roberts & Karpuz 1995),surficial deposits,and naturaland anthropogenicpollution of soil and vegetation(e.g.,Belviken et al. 1977,R0d 1992).As abasis for evaluation of the remote sensing multispectral airborne scanner data, laboratoryand field testing of the reflectivityof mineralsin rock samples has been carried out.

BJ0RN A. FOLLESTA D,AREKORNELlUSSEN

s

M et hodol ogy and result s

On 18 July 1996,a flightover a Norwegiantest area in the F0r- defjord district was carried out using the Geophysical &Envi- ronmental Research Corporation's63-channel scanner in a 31-channeldigital mode(GER 63).The sitecovers the Enge- befjell area(Fig.1),including the elongate,eclogitised,gabb- roic intrusions containing enrichments in Ti and Fe(Kor- neliussen & Foslie 1985,Lutro & Ragnhildstveit 1996).The tests consisted of threepart s:

(1) Laboratory measurement of the reflect ivitycurves using the spectrometer at the Defence Research Laboratory (Forsvaret s forskn ings institutt,FFI);

(2) Field measurement using aportable spectral radiometer (GER3700);

(3)Spect ral data analysisof data collected withthe airborne mult ispectra lscann er.

Laboratorymeasurement of reflectivity

Spect ralmeasureme nt of rock samples(Table1)was carried outat theFFIlaborato ry using aPerkin-Elm er UV scanner.

Each sample was treated separately and scanned in the

Sample Rocktype Ti02 Fe203 Descript ionof the rock

no (%) (%) samples

R1 leuco-eclogite 0.5 10 light coloured rock. no moss

on surface

R2 leuco-eclogite 0.5 10 lightcoloured rock, no moss on surface

R3 leuco-eclogite 0.6 11.8 lightcoloured rock, no moss on surface

R4 leuco-eclogite 0.6 11.8 light colouredrock. no moss on surface

5R ferro-eclogite 3.2 16.9 dark colouredrock.no moss on surface

R6 ferro-eclogite 3.2 16.9 darkcolouredrock.no moss on surface

R7 ferro-eclog ite 3.4 18 dark colouredrock. nomoss onsurface

R8 ferro-eclogite 3.4 18 dark coloured rock, no moss onsurface

R9 ferro-eclogite 3 17 dark coloured rock.no moss on surface

R10 ferro-eclogite 3 17 dark coloured rock,some moss onsurface

Rll gneiss lightcolouredrock. no moss

on surface

R12 gneiss lightcolouredrock,no moss

on surface

Rl3 gneiss lightcolouredrock. some

moss on surface

R14 gneiss lightcoloured rock. no moss

on surface

R15 gneiss

R16 gneiss

Table1.Rock samples collected fromEnqebe in1996andanalysedfor contentsofTi0 2andFe203(givenaswt.%).

BJ0RNA.FOLLESTAD,AREKORNELlUSSEN&NIGELJ. COOK NGU-BULL436,2000-PAGE 195

- -meanRI+R2 70 60

60 - -meanR3+R4 50 - -mean

. f'

. f'

.U 40~ meanR5+R6 ^.U^~'-' 30 . - -R ll

C) !;:;

!;:;C) 30 - -mean R7+R8 0::^C) 20.

0:: 20 _~ RI2

~ - -meanR9+RI O 10

10

0 0

- -mean RII+RI 2 0 N r<)

"'"

0 a- 00 r- \C on 0 N r<)

"'"

~ ^{00r- r- 'Don ona-}

"'"

"'"

Wavelength (nm) Wavelength (nm)

- -meanR15+R16

Fig. 2.Reflectivity (in%)of themain rocksamples studied,comparedto a standardof BaS04'R1-R4:leuco-eclogite, RS-R8:ferro-eclogite,R11-

R16:gneiss. 70

60 - -mean

C 50 R15+R16

wave b an d rangeof 480-3129urn.Scanned datawer estand- :~

u

ardised again st Ba2S04.2H20reference valuesand stored in ^!;:;'-' 30 ASCII format.Results aresh o w n as a meanvalueof reflectivity 0::^'-' 20

(%) in thesp ect ral band from 500 nm(=O.5llm)to2400 nm. ^~ RIG

10 As seen in Fig. 2,%reflectivityvaluesareup to15%higherfor 0

0 a- 00 r- 'D on

"'"

samples of leuco-eclogite (m ean of R1 +R2,mean of R3+R4)

"'"

"'"

an d gnei ss(mean of R11+R12,mean of R13+R14, mean of ^N

R15+R16)compar edwithsam p les offerro-eclogite(m ean of Wavelength(nm) R5+R6,mean of R7+R8, mean of R9+R10).Reflecti vitydata

70

. f'

.::: t>

Il)

4=: 30

~

~

la o

O N "'" '-0 00 0

o ^'-0 N 00 "'"

"<t r- ^{"<t 00 N}

N

Wavel ength (nm)

- -meanRI+R2 - -RI

R2

80,.-- - - ---,

. f'

.~

~U 40

~~ 20

o

..

o 0"1 00 r - \ O V) ..q-

0 0 N("") ~V)

'o:::t r- 0 (""'j \0 0\ N N

Wavelength(nm)

- -mean R13+R14 - -R I3

Rl4

Fig.3. (top):Reflectivity(%)oftwo sub-samples(R1,R2)ofa sampleof leuco-eclogite,compared tothemean of themeasurement (Rl+R2).

(bottom):Reflectivity(%)of twofurthersub-samples ofleuco-eclogite {R3, R4}compared tothe measurement mean(R3+R4).

Wavelength(nm)

60

. £

. :::

t>

30

Il)

4=:

Il) 20

0:::

~ 10 0

0 N M

0 "<t 00 N

"<t r- 0 "<t

- -meanR3+R4 - -R3

R4

Fig.4.Reflectivit y(% )of sub-samplesofasample ofgneiss,compared to themean ofeachmeasurement.(top):Sub-samplesR11,R12 and mean R11+R12, (middle):Sub-samplesR15andR16and meanR15+R1 6,(bot- tom):Sub-samplesR13and R14 and meanR13+R14.

acrossthe wavele ngthband are presented in a series of spec- tra allowingmoredetailedinterpretation of the data(Ieu co- eclo g ite, Fig.3;gnei ss,Fig.4 and ferro-eclogite,Fig.5).Sam - ple numbers and contents of Ti02and Fe20 3are given in Table 1. Spectralpatternsare lithology-sp ec if ic, and sign if i- cant differ ences in Ti02and Fe20 3contents, for example among leuco-eclogites (Fig. 3),give curveswith sim il arpat - terns an d peaks (t he twosam ples of leuco-eclog it ehav eTi02 and Fe203valu esof 0.5,10 and 0.6,11.8 wt.%,resp ectiv el y).

Refl ectivityspectra for typical gneisses are sho wnin Fig.

4.Curves R11 and R12 aresub-samplesfrom thesame bul k sample and arenearly identical.However,thespectra ap p ar- ent ly differlittle from leuco-eclogites (co m pa rewit h Fig.3).

Gneiss sam p le R15and R16showthe sam e reflect iv it yvalu es

NGU-BULL436,2000 -PAGE 196

50

, £

./~

.

::

,..-.

C>

V

c _{20 .}

C>

0:: R6

~ 10

0

0

0 ,.... 'D 0, N on 00

0 "T 00 N t"-

-

"T r-.

_{- - -}

-

Wavelength (nm)

50

. f'

.:: u

C>

r:

cC> ₂₀ V-

0:: R8

~ 10

0 ^>

- ^.

0

,....

0 "T QC N t"-

-

"T t"-

- - -

-

Wavelength (nm)

Fig.5.(top): Reflectivit y(%)oftwosub-samp les(R5, R6)of a sampleof ferro-eclog ite,compared tothe meanofthe measurement(R5+R6).(bot- tom):Reflecti vity(%)oftw ofurt her sub-sam plesof ferro-eclogit e(R7,R8) com pared to themeasurementmean(R7+R8).

in the visiblepart of thespect rumassam p le Rl l.In thecen- tralpart of theinfrared spectr um(900-1800nrn),how ever, reflectivity valuesaresomewhat hig herthan for the other gneisssamp les. Spectr a of two sam plesofgneiss,R13and R14,wit handwit houta surfacecovering ofmoss,areshown inFig.4(bottom). Compari son oft hesecurvesclearly demon-

BJ0RNA.FOLLESTAD,AREKORNELlUSSEN&NIGH J.COOK

strates that the reflectivity of sampleR13is affec ted by the presence ofmoss.Acharacteristicfeature of thespectrais a shar p gradient in the near-infrared region(900nm),charac- teristicfor chlorophyll.The R14sample displaysa less marked reflectivityincreaseinthe 900 nm band,butthis sub-sample has less moss on thesurface.These data indicate that moss on the samplesurfacewilldramaticallychange the%reflec- tivity andgive the reflecti vitycurve for chlorophyll(i.e.vege- tat ion).

Four sub-sam plesof ferro-eciogitefrom two sites,R5/R6, with Ti02 and Fe203conte ntsof 3.2;16.9%and 3.4;18.9%, anda second pair,R7/R8,weretested (Fig.5).Thesesam ples show rather similar reflecti vit y spectra.Reflectivitycurves for ferro-eclo git e arecharacterised by rather low reflectivity val- ues(ca. 20%)in thevisib lepartof the spectrum.Compared wit h the spect rafor both leuco-eciogite and gneiss,this is a 10% redu ct ion.The reduct io n of reflec ted energy is seen bot h in thevisibleandin the infraredparts ofthe spectrum.

Field measurementusing a portab lespectral radiometer (GER 3700)

TheGER Mark Vspect ro meter is a portable inst rument for measuring reflecte d electromag netic energy.Itoperates in the visibleand infrared parts of the spectrum,from 300 to 2500nm(or 0.3 to 25urn).The spectralresolution is 10 nm, witha sampli ngint ervalof2 nm.A BaS04 plate is usedas a sta ndar dfortheinst rument, and the BaS04values are given togetherwiththe spectral readi ngsat the measuredsite.This BaS04 stan dard allows for normalisat ion of site readings which,as will be explainedlater,isimpo rtant.The fieldstudy was intend edto demonstrate if different rock types in the areacould be differentia ted at outcrop aswell as in hand spec imen.Such applica tionsaredescribed by Kale& Rown (1980).

Theworkingspectralrangeofthe instrumen tGER Mark V covers the samewavelength region astheGER 63-channel

imaging spectrometer,operat ing in 31-channel mode, which was flown over the area in July 1996. Due to logist ical problems,fieldwork using theGERMark V spectrometerwas not carriedout until 11 August 1996.At thistime of the year, however,the variations in incoming electromag- netic energy are not considered to have anymaj or influenceon spectral measurements.In our case, the time

Fig. 6.3-di me nsional image of Enqebofjell show ing the majorrock unitsandlocations of sam plesused forreflectivity data.The cent ralpart of the image shows ferro-eclog- ite(violet),leuco-eclogit e(yellow), amphib- olite (lig htgreen)andgneiss(grey).The red line indicatesa distanceof1 km.The model was const ructedusingtheInt ergraph Vaxel Analyst prog ram, withgridresolution 5 x 5 m.

BJ0RN A.FOLLESTAD,AREKORNELlUSSEN &NIGELJ.COOK NGU-BULL 436,2000-PAGE 197

45.000 40.000

35.000 30.000 :~~ 25.000

~E 20.000

'"

?ft.

15.000 10.000 5.000

r-, QC QC -o M 0' M r-, 0'

~ ^{i2 QC QC ~ -r- N ;;}

e-, M 0' ~

° ^{-o ~ ~ N QC} ~

M M -r- -r ~ -o C ~ OC 0' 0' 0. 0, .,., -q

Wavelength (om)

- :-10,52 - NorS3 NorS4 - 1'10,55

- NorS6

- 1'0,5 7 -Nor58 - NorS9 - 59:'>lc3s 0

- No,510

- 1'10,5 11

Fig.7.(top): All readi ngs illustrated byasing lesamplefrom eachsam- ple site 52 -511, see Fig 6.The readingsforthe sam plesitesof thedifferentrockscannot beread- ily distinguis hedfromone anot h- er.(bo ttom): Spectr a from three samplessites,54 to 56,of ferro-ec- logite.

45.000. - - - ,

40.000+-- - - - - - - - - - - - - - -- - - - - - - -- - - ---j

15.000 10.000

- NorS4

-~orS5

- NorS6

5.000

~ :;;^{QC QCM >Co- M~ 0'}° ^{M>C ~ 0'}~ °~ :;;; ^{;:; 0r-, QC QCQC >Co-} _~^{-r ~ c- -rN NQC ~or; ;; MQC}

.,.

M "T -r ^or; C >C ~ c- o- 0

_-

-

^- -

^-

:;;

Wavelength (om)

of theday when themeasurem ent s were taken (from 08.40 to 15.33),togetherwith the air humidity,constitutegreater sources fordeviati on.No correct io nswere madefor potential chang esin measurem ent dueto dayand month, Field sites are shown in Fig.6 and tabu lated in Tabl e 2.Spectr a for leuco-e clo gite(52, 53),ferro-eclo gite(54,55,56),amph ib ol ite (57,58) and gneiss (59)are showninFig.7a.Minor differenc es betw een the different rock types are noted,yet variat io n betw eenrocktypesof thesame type isoftengreater,allow- inglit tlediscrimination.

As anexam p le,reading s of massiveferro -eclogit e were made at sam ple localit ies54,55 and 56 (Fig. 7).Ti02and Fe20 3conte nts(in wt.%)of ferro-eclogiteatthethreesites were 4.27,14.28;4.03,13.95; and4.36,12.93,respect ively.All three localities had some surface moss.The three curv es show some con sid erabl e variatio n, read ing svaryin g both wit h theextent of mosscover,butalso significantlywit h the time of daywhen themeasurem entswere taken.Theinfor- mationbase availab le fromthefieldmeasurem ent s indicates that the moss cover consti tutes the greatest influence on reflectivity and strongly impactson different iati on ofrock typ es.The rock surface offerro-eclogit e appears to have a somew hat high er ref lecti vity in theregio n 1180 - 1948nm, whereas in the wavelength 2000-2497 nm, the reflectivity of thislithology isgenerallylow er.Thedistinct and charact eris-

ticfeaturesin the 1180-1948 nm band cannot,how ever, be used for differenti ati on of eclogite types using airbo rne measurement s since the availab le chann elsfor theGER 68 syste m donotcover the 1180 nm - 1948 nm regionofthe spectr um(seeTable 3).

Airborne spectralscanning

The GER63-channel imagi ng system is a mul tispe ctra lair- born e spectro met er. Measurem ent s are carri ed out in the ultraviolet,visible,infrared and therm alpart sof the spect ra.

The maximum num ber of chann elsforuse in the scannersys- temis 63.In thetests reported here,31chann elshave been in operat ion modeand one chann el(32) is used forgyro;the channels used are liste d in Tabl e 3.Thethermal channels (chan nels29,30and31)might herebeusedfordiscrimina- tion,aswill beshow n later.

The data have been subject to minimal preliminary proce ssing.Basicdata processing consists of gyro,baselin e corr ection and panor amiccorrections.Gyro correction is car- ried outtoremove the effects of the aircraft motionbytrans- lati ng each scanlinehori zontall yin space by anamountpro- port ion al to the number of gyro counts . The gyroscope encodes theamount ofaircraft roll in the last chann el. Inthe image, stra ight features wit hi n the image,suchasroad s, are st raight, but the edges of the image may have irregular

NGU-BULL436,2000 -PAGE198 BJ0RN A.FOLLESTAD,AREKORNELlUSSEN

s

Locality Rock %Ti02 %Fe203 UTM-East UTM-North Spect ralmeasurement,Engebofj ell,11.aug.1996

Measurement Level[ml 40dB

Time Filename Detector

51 Leuco-eclogite 1.23 11.74 310589 6823337 0 0,6 OFF 09.33 51MO.5IG

51 Leuco-eclogite 1.23 11.74 310589 6823337 1 0,6 OFF 09.35 51M1.51G

51 Leuco-eclogite 1.23 11.74 310589 6823337 2 0,6 ON 09.39 51M2.5IG

51 Leuco-eclogite 1.23 11.74 310589 68233 37 4 0,6 OFF 09.49 51M4.5IG

51 Leuco-eclogite 1.23 11.74 310589 682333 7 5 0,6 ON 09.56 51M5.5IG

52 Leuco-eclogite 1.17 12.14 310587 6823307 0 0,6 OFF 10.15 52MO.5IG

52 Leuco-eclogite 1.17 12.14 310587 6823307 1 0,6 OFF 10.20 52M1.51G

53 Leuco-eclogite 1.16 10.66 310490 6823273 0 0,6 OFF 10.39 53MO.5IG

53 Leuco-eclogite 1.16 10.66 310490 68232 73 1 0,6 OFF 10.44 53M1.5IG

54 Ferro-eclogite 4.27 14.28 310350 6823191 0 0,6 OFF 11.01 54MO.5IG

54 Ferro-eclogite 4.27 14.28 310350 68231 91 1 0,8 OFF 11.05 54M1.51G

55 Ferro-eclogite 4.03 13.95 310350 6823184 0 0,8 OFF 11.26 55MO.5IG

55 Ferro-eclogite 4.03 13.95 310350 68231 84 1 0,8 OFF 11.31 55M1.51G

56 Ferro-eclogite 4.36 12.93 310348 6823 187 0 0,7 OFF 11.14 56MO.5IG

57 Amphibolite 310575 6823470 0 0,6 OFF 11.55 57MO.5IG

57 Amphibolite 310575 6823470 1 0,6 OFF 11.57 57M1.51G

57 Amphibolite 310575 6823470 2 0,6 OFF 12.00 57M2.5IG

58 Amphibolite 310578 6823472 0 0,8 OFF 12.06 58MO.5IG

58 Amphibolite 310578 6823472 1 0,8 OFF 12.08 58M1.51G

58 Amphibolite 310578 6823472 2 0,8 OFF 12.12 58M2.SIG

58 Amphibolite 310578 6823472 3 0,8 OFF 12.15 58M3.5IG

59 gneiss 310695 6823340 0 0,6 OFF 08.39 59MO.5IG

59 gneiss 310695 6823340 1 0,6 OFF 08.45 59M1.51G

59 gneiss 310695 6823340 2 0,6 OFF 08.56 59M2.5IG

59 gneiss 310695 6823340 3 0,6 OFF 09.00 59M3.5IG

59 gneiss 310695 6823340 5 0,6 OFF 09.02 59M5.5IG

59 gneiss 310695 6823340 6 0,6 OFF 09.05 59M6.5IG

59 gneiss 310695 6823340 7 0,6 ON 09.08 59M7.5IG

59 gneiss 310695 6823340 8 0,6 ON 09.11 59M8.5IG

Table2.Samplelocaliti esfor fieldinvestigations,11^thAugust1996.

boun d ariesdueto aircraft roll. Eachscanlineis translatedby an integr alnumber of pixels to the righ t orleft;thereisnore- sampling.The spect ralcontent of thedataisnotaffected by gyro correctio n.Baseline correct ion ut iliseson-board refer- ence panelsto norm alise theeffec tsof detect or reference volta ges. Whentargets wit hext remeradiancedifferencesare imaged,the GER scanner utilises atime-varying reference voltage toincreasethe effect ivedyn am ic range.Theimageof thebaselin est riprecordsthedigi talcountsofaconstan t- reflectan cepanel.Inbaselinecorrect io n,thereference panel images are usedtonormalisethe datasothatthe consta nt- reflect ancepanelshavea consta ntdigita l count in theimage. Eachlineandeachchannelare norm alised individually.The 32baseline pixelsare then removedfrom the dat a set.

Panoramic correctioncorrects thegeomet ric dist ortions

caused bythe differe ntviewing geometry in different parts ofeachscanline. Pixelsarere-map pedint otheir correct posi- tion using nearest-neighbou r re-samp ling where possible, and linearinterpolation wherenew pixelsmustbe created. The spectralcontent forthe nearest-neigh bourpixelsis not affected by thiscorrection.Thedata setsare received as 8 mm (Exabyte)tape.Itshould, however,be noted that the chosenchannels(Table 3)donot favour detectionand differ- ent iat ionofferro-eclo gitein the1000 nm - 1948nm of the spect rum,as thereisonlyone channel(14)wit hi n thisrange.

ENVI software (Environment for Visualizing Images ver- sion 2.7) isused forimag eprocessing and classification of multispect ral analysis based onthe aircraft remote sensing data.Theimage processing system (ENVI)uses a generalised rast er dataform atstored asa binary st reamofbytesinband-

BJ0RNA.FOLLESTAD,AREKORNELlUSSEN& NIGELJ.COOK NGU-BULL436, 2000 -PAGE 199

sequentia l (BSQ) format. The basic function starts with an open imagefile wherethe different bands can bepresent ed aloneas a greyscaleimage,orcombined,as anRGB-colour image (Fig.8). This gives usthe possibilities of combining channels2(blue),5 (green)and 9(red) and thuscreatin g a falsecolour image fortheareascover bythe flight line 1(Fig.

1). Thiscolo ur combination shows that the area isheavily vegetated asthe reflecte dinfraredfrom channe l9dominates the image. Blue and greencolou rs are chosen for channels2 and 5wit h lowreflectivitydueto absorpti onbychlorophyll.

The blue/green colours thus show areasof bare rock;the fjord would havenochloroph yll in this context.From Fig. 8, it is onlyinthe elevated partsof the area thata bare rock-sur- face,withlittleor no vegetation ,is present.

Fig.9.Barren rock surface(p1J.nearly ba rren surface(p3) and an area dom inat edby vegeta tion(p2) are shown in a spectral plot(seeFig.12).

In the immediate vicinity of Engje bofje ll (Fig. 1), such areas only form a smallproportionof thetotal area, basedon a RGB-composi teof channels 5,8 and 13 constructed from data from flightline 4, shown in Fig.9.Thevariance in the spectra lreflectivityonall31 channels from testsitesvisit ed in the field represent ative of nearly barren rock surface (p l), barrenrock surface(p2)andanarea domin ated by vegeta- tion (p3)areshow n(Fig. 1O).ltis clearly shown that the heav- ilyvegetatedsite(p2) has a maximumof reflection in channel 10,equalto 823 nm.The barren rock surfaces on the ot her hand(p 1 andp2) show maximum valuesin channel 4,equal to521 nm.In the areas from1500 to 2400nm,the curves of reflection s arequite simi lar. Inthethermalchannels(chan- nels 29,30 and31),how ever,there seems to be a higher value of thermalradiation for barren rock samples. This might be explainedasa result of thetrappingof radiantenergy (green- Fig.8.Testflightline1.TheEnge-

be test area is show nas an RGB- colour image. Channel 13: red , Channel6:green, Channel 4:blue.

Enqebeneset is marked wit h E.

Channel Band(nm)

1 410.00

2 432.00

3 483.6

4 521.3

5 565.8

6 617.7

7 665.8

8 716.6

9 768.5

10 823.7

11 883.90

12 941.9

13 989.7

14 1048.3

15 2039.3

16 2084.00

17 2112.5

18 2145.2

19 2170.00

20 2208.00

21 2237.7

22 2270.7

23 2296.2

24 2333.3

25 2358.00

26 2388.6

27 2411.7

28 2446.9

29 9468.4

30 10206.5

31 10983.1

32 Gyro

Table3.Wavelength calibra- tionfor GER63-Channe lSen- sorin 31-Chan ne loperation.

NGU-BULL436.2000 -PAGE200 BJ0RNA.FOLLESTAD,AREKORNELlUSSEN& NIGELJ.COOK

Band5:g31.960718 'ol'\vay-Gordcfjord4.11f05.final

Band umb er

Fig.10.RGB-co m po sit e of channels S.8 and 13.scan ner flight 4(Fig. 1), withlocationmarksofspectralplotsshown inFig s 7-9.

The minimum noise fraction (MNF) transformation can be usedif a quick classification is required,as in our case.This transformation is used to determine the in herent spectral dimensionalityofimage data,to segregate noisein thedata, and to reduce the computationa l requirements for subse- quent processing(Boardman & Kruse1994).

In our work,the channelinformation for barren rock or nearly barren rock surfaces(markedwith p1,p3;Fig. 10)and vegetation (p2; Fig.10)is used as separationcriteria.Expres- sion of these properties can be presentedin terms of a two- dimensionaldiagram plotting datafrom channelSagainst thethermal channel 30(Fig.11).

Inthe algorithms on whichthis diagram is based,differ- ent transfo rmat ions arecarried out.The first transformation is based onanesti mated noise covariancematrix,which de- correlatesand re-scalesthenoise in thedata.This resultsin transfo rmed data in which thenoisehas unit variance andno band -t o-band correlat ion.The second transfo rmation is a standard principalcomponent stransformationofthenoise- white ned data.Thus,the data spacecanbe dividedinto one part assiste d with large eigen-values and coherent eigen images,anda complementary part withnear-unity eigen -val- ues andnoise-dominate dimages.Classify ingthe data set by marking the area of interest by a't rial and error procedure' allowsa best fit to be made.In Fig. 11,the selected data- point s areshow n(in red)and superimposedonto the colour image of the areainFig.12.It can be seenthat thisclassifica- tion easily picksout the non-vegetated areas.These areas, whichinclud e roads aswell as bare rock surfaces,are readily seen. Acom parison of the areaswit h gneissandleuco-eclog-

25 30

x:226 y:203 x:462y:275 x:420 y:315

20

15 10

# 1 Spectra l Pro file

5 4500 -

4000 -

3500 -

3000 -

C)

:::

-;

5727

<;:;t:

on

e

.,:

..

'E

4791

=

-:::l

(;

0

t

z

00 r- -o0 0';

'"

0...

-:::l

::0

a 1985

3120 4826 6533 8239 9846

house effect) under a cover of leaves.Thermal radiat io n might,aswill be discussed later. be one channelofinforma- tionwhich permitsa classificationof surface type.

A classification ofthe regionofint erest (ROI) wasmad e on the test flight4 data.The classificat ion menuinENVIsoft- ware givesaccess to supervisedand unsupervisedclassifica- tion.Utilit iesare also provided for end-mem ber collect io n, classifying previousrule images, calculating class statistics, ete.Supervised techniquesusethe end member collection utility to import training class spectra; ENVI's integrated region-of-intere st(ROI) selectionutilit ies areused to interac- tively defin e training classes.The selection rou tinesallow ext ract io nof training stat isti csfro m polygo ns,vectors,ete.

Fig11.Scatterplotofthe min im um noise fraction(M NF) transfo rm at io n usedtodet ermin ethe inhe ren t dim ensio nality ofchannels Sand 30.

Fig.12.The region of interest wit h the scatter plotfrom Fig .11supe rim- posed ont o it.

BJ0RNA.FOLLESTAD,ARE KORNELlUSSEN&NIGEL J.COO K

ite shows,however,thatthereisnosignificantseparat ion of the different rocktypes.

Conclusions

Thelabor ator ytestsindicatethattherocksurfaces of ferro- elcog ite will generally have a low er reflectivity com pared wit h thesurface of both leuco-eclogite and gneiss.However, thepresence of mosson the rock surfacewillchange these characterist ics significant ly.The data collect ed in the field em p hasisethefactthatvegetat io ncoverand measurem ent conditio nsimpac t heavil yon theresult s.How ever,thefind- ingsdosuggest thatdiff erent%refl ecti vit yvalues indistinct partsof thespect rumcou ld possiblybeused asatoolto dis- crim inate betwe en areas characterised by gneiss, leuco- eclogite and ferro- eclog it e,if ot her factor sinfluencing the dataarecarefullyconside red.However,theunavailabil ity of the 1180-1948 nm band for airborne measurem ent makes app licat ion of a part of these difference sof conside rab le interest, yet of little practical benefit,when ap plying the meth odsasaprospecting toolusingGER63data.Other sen- sors suchasHYMAPdo cover these wavelengt hs and have greatpot enti al.

The airbornescanner data and interpretation thereof are useful in thattheyreveal ameansto discrimin atebetween areas of vegetation and barerock surface.However,thearea istoo heavily vegetated for closer separat ion based on a scanner resolutionof 10 mx10 m. Theairborne techniques are less successf ul in distinguishing different rock types,at leastleuco-from ferro-ecloqite,Adoption of more suita b le channels,covering that part of the spect ra in which differ- ences betw eenrock typeshavebeenidentifiedin both labo- ratory and field st udies,wo uld givemore usefulinformation.

Small-scale variati on sinvegetat ion cover and mossdensity are, however, a concern and may preclud e more routine applicationofsuch techniques in prospecting .

NGU-BULL 436 ,2000-PAGE201

Acknowledgements

We wish to thankBennyMoeller Soerensen(GER-INTRADANA/S)andthe lateProf.BrianSturtfor valuableco-ope ratio n;andDr. Stuart Marsh fo r hiscriticalread ingofanearlierversio nofthemanu script.

References

Boardman,J.& Kruse, F.1993:Knowled ge-based geologicmapp in gwit h imag in g spect rome ters:RemoteSensing Reviews(special issue on NASAInnovat iveResearchProgr am (IRP)results),8,3-28.

Bolviken,B.,Hon ey,F,Levine,R. Lyon,R.&Prelat, A.1977:Detectionof nat uralheavy-m et al-p oison edareasbyLan d sat-1digitaldata. Jour- nal ofGeochem icalExplora tion8,457-471.

Cha ng,S.& Collins,W.1983: Confir mationofthe Airbo rneBiogeophysi- calMineral Exp lo rationTech niq ueUsing Lab orato ryMeth ods.Eco- nomi cGeology78,723-736.

Hilkka,A.,Ruohomaki,T.Rain o,H.&Laitinen,J.1998:Applicatio nof AISA airborne imagin gspectro me ter tothe discriminati o nof Quat ernary soiltyp esin theMantsala-Ohkola areain Fin land.Geologica lSurvey ofFi nla nd Rep ortRS/1998/4.

Kale,A.B. &Rown,L.e.1980:Evaluat ionof multispe ctralmid dleinfrared aircraft imagesfor lit hologicmapp ingin the eastTint ic Mountains, Utah.Geology8, 234-239.

Korneliussen,A..&Fo slie,G.1985:Ruti le-bearing eclo gitesinthe Sunnf- jordregionofWestern Norw ay.Norges geologiskeunde rsekelseBulle- tin402,65-71.

Lutro,O.&Ragnhildstveit,J.1996:Geologicalmapof theFo rdefj ordarea, bedrock map, scale 1:50,000. (unp u blished). Norges geologiske undersek else.

Lyon , R. 1996: Spect ral Pro perties ofMinerals- Significa nt in Mineral Explor ati on and EnvironmentalStudies. Eleventh Them at ic Confer- enceGeologicRemote Sensing,LasVegas,Nevada,USA.

Rich ards,J.A. &Jia, X.1999:Remo teSensingDigitalImage Ana lysis,3,dedi- tion.Springer-Verlag Berlin HeidelbergNew York.

Rindsta d,R.&Foll est ad,BA 1982:Dig ital met hodsfo rlineamentanaly- sis.FirstThem atic Conference:Rem ot e SensingofAridand Semi- Arid Lands,Cairo,Egypt.Environm ental Res.Ins.Mich.AnnArbor, Michigan,United States,1982,955-961.

Roberts,D. &Karpu lz,M.R. 1995:St ructuralfeatures of theRyb achiand Sred niPeninsulas,Northw estRussia,asinterp ret edfrom Land sat- TM imagery.Norgesgeolog iskeundersekelse SpecialPublicatio n7, 145-150.

Rod ,J.K 1992:Perspe kt iviskvisualisering av et5POT-bilde sam regis trert med geolog isk tem a.Norges geolog iskeundersekelse Report92.256.

Wester,K.,Lunden,B.&Bax,G.1990:Analyt ically processedLandsatTM imagesfor visua lgeological interpret ation in northern Scand inav ian Caledonides.ISPRS JournalofPho tog am me tryandRemoteSensing 45, 443 -459.