UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Detection of mountain pine beetle infestations using Landsat TM Tasseled Cap transformations Sharma, Rajeev 2001

Your browser doesn't seem to have a PDF viewer, please download the PDF to view this item.

Item Metadata

Download

Media
831-ubc_2001-0113.pdf [ 7.1MB ]
Metadata
JSON: 831-1.0074962.json
JSON-LD: 831-1.0074962-ld.json
RDF/XML (Pretty): 831-1.0074962-rdf.xml
RDF/JSON: 831-1.0074962-rdf.json
Turtle: 831-1.0074962-turtle.txt
N-Triples: 831-1.0074962-rdf-ntriples.txt
Original Record: 831-1.0074962-source.json
Full Text
831-1.0074962-fulltext.txt
Citation
831-1.0074962.ris

Full Text

DETECTION OF MOUNTAIN PINE BEETLE INFESTATIONS USING LANDSAT TM T A S S E L E D CAP TRANSFORMATIONS by RAJEEV SHARMA M.Sc , Meerut University, 1975 A THESIS SUBMITTED IN PARTIAL FULFILMENT OF THE REQUIREMENTS FOR THE D E G R E E OF MASTER OF SCIENCE in THE FACULTY OF GRADUATE STUDIES THE FACULTY OF F O R E S T R Y Department of Forest Resources Management We accept this thesis as conforming to the required standard THE UNIVERSITY OF BRITISH COLUMBIA November 2000 © Rajeev Sharma, 2000 In p resen t i ng this thesis in partial fu l f i lment of the requ i remen ts fo r an a d v a n c e d d e g r e e at the Univers i ty of Brit ish C o l u m b i a , I agree that t he Library shal l m a k e it f reely avai lable fo r re fe rence and s tudy. I fur ther agree that p e r m i s s i o n fo r ex tens ive c o p y i n g of this thesis fo r scholar ly p u r p o s e s may b e g ran ted by the h e a d o f m y depa r tmen t o r by his o r her representa t ives . It is u n d e r s t o o d that c o p y i n g o r pub l i ca t i on of this thesis for f inancia l ga in shal l no t b e a l l o w e d w i t h o u t m y wr i t t en p e r m i s s i o n . D e p a r t m e n t of fpY^'f /&<fcttA<&) pla^o^ny e^/— T h e Un ivers i ty of Brit ish C o l u m b i a V a n c o u v e r , C a n a d a Da te / V $ V - 3gn 2pOft D E - 6 (2/88) Detect ion of mounta in pine beet le infestat ions us ing Landsa t T M T a s s e l e d C a p Trans fo rmat ions Abstract Th is study invest igated the identif ication of p robab le mounta in pine beet le (Dendroctonus ponderosae Hopk.) a t tacked s i tes using T a s s e l e d C a p t ransformat ions namely , br ightness, g r e e n n e s s and we tness , der ived f rom Landsa t -7 digital da ta in parts of V a n d e r h o o f Fores t District, in P r i nce G e o r g e Fores t Reg ion , Brit ish Co lumb ia . Lodgepo le pine (Pinus contorta Dougl . ) const i tutes about 80 percent of the total forest vegetat ion in the study a rea . Abou t 85 percent of lodgepo le pine s tands are greater than 60 y e a r s of age , and hence suscept ib le to mounta in p ine beet le attack. Landsa t -7 E T M digital data (Bands ' ! , 2,3,4,5&7), acqu i red on Augus t 2, 1999 a n d S e p t e m b e r 12, 1999, w e r e the primary remote s e n s i n g data sou rce for the study. In addi t ion, T R I M m a p shee ts (1:20,000) der ived road and river vec tors , and 1:20,000 forest c o v e r m a p s and a beet le infestat ion c o v e r a g e m a p s of the a r e a (prepared b a s e d on aerial ske tch mapp ing and ground probes) w e r e col lateral da ta s o u r c e s . Methodo logy cons is ted of: i) p re -p rocess ing of satel l i te da ta (a tmospher ic and geomet r i c correct ions) , ii) computat ion of T a s s e l e d C a p coeff ic ients for the Landsa t -7 da ta , s i nce t hese w e r e not ava i lab le , iii) identif ication of mountain pine beet le a t tacked s tands , and iv) a c c u r a c y a s s e s s m e n t of a t tacked s tands . S o m e of the major observa t ions b a s e d on resul ts obta ined were : i) T a s s e l e d C a p ind ices for infestat ions of more than 30 at tacked t rees / site (< 0.09 ha in s ize) we re found to vary in a relatively narrow range; however for infestation s i tes with l ess than 30 tree / site the T a s s e l e d C a p ind ices had random and large d ispers ions ; ii) va lues of T a s s e l e d C a p ind ices for S e p t e m b e r w e r e found to be lower than those for A u g u s t for all the cove r types; iii) d i f ferences be tween m e a n br ightness, g r e e n n e s s , and w e t n e s s of healthy and at tacked s tands we re statist ical ly signif icant for Augus t ; iv) the identif ication accu racy for a t tacked lodgepo le pine s tands w e r e 38 .82 and 26 .17 percent for A u g u s t and Sep tember , respect ive ly ; v) a l inear relat ionship w a s obse rved be tween the number of a t tacked t rees at a site and identif ication a c c u r a c y for the A u g u s t da ta but not for S e p t e m b e r data ; vi) the poor identif ication a c c u r a c y w a s mainly d u e to the sub-p ixe l s i ze of infestat ions. ii Detection of mountain pine beetle infestations using Landsat T M Tasseled Cap Transformations Table of Contents Section Page Abstract ii List of Tables iv List of Figures v Acknowledgments vi 1.0 Introduction 1 1.1 Background 1 1.2 Conventional M P B Detection Methods 2 1.3 Possible Alternatives 5 1.4 Objectives 7 2.0 Tasseled Cap Transformations 8 2.1 The Concept 8 2.2 Tasseled Cap Coefficients 9 2.3 Characteristics of different cover types in Tasseled Cap feature space 11 3.0 Literature Review 13 3.1 Visual Image Interpretation Based Studies 13 3.2 Digital Image Analysis Based Studies 15 3.3 Recent Initiatives 18 4.0 Material and Methods 20 4.1 Study area 20 4.2 Data Used 22 4.3 Mountain Pine Beetle Infestations in the Study Area 23 4.4 Methodology 25 4.4.1 Pre-Processing of Satellite Data 25 4.4.2 Computation of Tasseled Cap Coefficients 26 4.4.3 Effect of Acquisition Dates, Topography, and Infestation Size 27 4.4.4 Identification of Attacked Stands 27 4.4.5 Accuracy Assessment 28 5.0 Results and Discussions 29 5.1 Tasseled Cap Transformations 29 5.2 Effect of Acquisition dates, Topography, and Infestation Size 31 5.3 Identifying Attacked Stands 39 5.4 Accuracy Assessment 40 6.0 Conc lus ions 44 7.0 Further Research Areas 45 References 46 Detect ion of mounta in pine beet le infestat ions us ing Landsa t T M T a s s e l e d C a p Trans fo rmat ions List of Tables Table Page 1. Mounta in p ine beet le life s t a g e s , s y m p t o m s , foliar co lor change and 3 control opt ions 2. T a s s e l e d C a p Coef f ic ients for V a r i o u s Landsa t Sate l l i tes 10 3. Spec t ra l Band-Wid th (micrometers) of Landsa t -5 T M and Landsa t -7 E T M + 11 4. Proport ion of red a t tacked t rees in different study a reas 14 5. Con fus ion matrix show ing beet le infested a rea (ha) identified by 15 v isua l interpretation 6. A c c u r a c y (%) ach ieved for identif ication of mounta in pine beet le infestat ions 16 7. T h e accu racy a s s e s s m e n t resul ts ob ta ined us ing spect ra l unmix ing p rocedu res 18 8. Relat ive proport ion (%) of different forest tree s p e c i e s in study a r e a 21 9. A g e - c l a s s distribution of l odgepo le pine s tands in study a rea 22 10. Detai ls of da ta used 22 11. S i z e character is t ics of beet le infestat ions in the total study a rea 2 3 12. A v e r a g e di f ference between healthy a n d a t tacked mounta in pine beet le s tands 39 (August 2, 1999) 13. A c c u r a c y es t imates b a s e d on 1 0 % random s a m p l e (August 2, 1999) 41 14. A c c u r a c y (%) b a s e d on number of a t tacked t rees/s i te (August 2, 1999) 41 15. A c c u r a c y es t imates b a s e d on 1 0 % random s a m p l e (Sep tember 12, 1999) 41 16. A c c u r a c y (%) b a s e d on number of a t tacked t rees/s i te (Sep tember 12, 1999) 41 iv Detect ion of mounta in pine beet le infestat ions us ing L a n d s a t T M T a s s e l e d C a p Trans fo rmat ions List of Figures Figure Page 1. Approx ima te locat ions of va r ious sur face c o v e r t ypes in T M T a s s e l e d C a p 11 feature s p a c e 2. Locat ion map of study a rea 20 3. A Landsa t -7 E T M (Sep tember 12, 1999) pseudoco lo r compos i t e 24 of study a rea . Locat ion of Tes t A r e a s is marked in red boxes . 4. Locat ion of different c o v e r types in br ightness - g r e e n n e s s (a), br ightness - w e t n e s s 30 (b) and w e t n e s s - g r e e n n e s s (c) feature s p a c e der ived f rom compu ted T a s s e l e d C a p coef f ic ients for Landsa t -7 Thema t i c M a p p e r Da ta . 5. C o m p a r i s o n of two date Br igh tness (a), g r e e n n e s s (b) and w e t n e s s (c) ind ices 32 for young lodgepole pine s tands (A: Augus t 2 1999, S : S e p t e m b e r 12, 1999, B R T : Br ightness, G R N : G r e e n n e s s , W E T : W e t n e s s ) . 6. C o m p a r i s o n of two date br ightness (a), g r e e n n e s s (b) and w e t n e s s (c) ind ices 33 for mature lodgepole pine s tands (A: A u g u s t 2 1999, S : S e p t e m b e r 12, 1999, B R T : Br ightness, G R N : G r e e n n e s s , W E T : W e t n e s s ) . 7. C o m p a r i s o n of two date br ightness (a), g r e e n n e s s (b) and w e t n e s s (c) ind ices 34 for sh rubs (A: Augus t 2 1999, S : S e p t e m b e r 12, 1999, B R T : Br ightness, G R N : G r e e n n e s s , W E T : W e t n e s s ) . 8. C o m p a r i s o n of two date br ightness (a), g r e e n n e s s (b) and w e t n e s s (c) ind ices 35 for roads (A: A u g u s t 2 1999, S : S e p t e m b e r 12, 1999, B R T : Br ightness, G R N : G r e e n n e s s , W E T : W e t n e s s ) . 9. C o m p a r i s o n of two date br ightness (a), g r e e n n e s s (b) and w e t n e s s (c) ind ices 36 land ings (A: A u g u s t 2 1999, S : S e p t e m b e r 12, 1999, B R T : Br ightness, G R N : G r e e n n e s s , W E T : W e t n e s s ) . 10. Br igh tness , g r e e n n e s s and w e t n e s s ind ices for mature lodgepo le pine s tands 37 on different aspec t c l a s s e s ( S E : Southeas t , F:Flat , NW:Nor thwes t aspec t ; A : Augus t , S : Sep tember , B R T : Br ightness, G R N : G r e e n n e s s , W E T : W e t n e s s ) . 11. Re la t ionsh ip be tween number of mounta in p ine beet le infested t rees and 38 T a s s e l e d C a p ind ices 12. P robab le mounta in pine beet le infestation m a p der ived f rom T a s s e l e d C a p 42 ind ices -Tes t a rea A 13. P robab le Mounta in pine beet le infestation m a p der ived f rom T a s s e l e d C a p 4 3 ind ices - Tes t a rea B 14. P robab le mounta in pine beet le infestation m a p der ived f rom T a s s e l e d C a p 4 3 ind ices - Tes t a rea C v Detect ion of mounta in pine beet le infestat ions us ing Landsa t T M T a s s e l e d C a p Trans fo rmat ions Acknowledgements I a m grateful to my superv isor , Dr. Pe te r A . Mur tha, Facul ty of Forest ry , Univers i ty of Brit ish Co lumb ia , for his gu idance and e n c o u r a g e m e n t throughout my entire g radua te p rogram. S ince re thanks for their interest, gu idance , and t ime are ex tended to the thes is commi t tee members , Dr. J o h n A . M c L e a n , Facu l ty of Forest ry , Univers i ty of Brit ish C o l u m b i a ; and Dr. Pe te r L. Marsha l l , Facul ty of Forest ry , Univers i ty of Brit ish C o l u m b i a . In addit ion, severa l people he lped in var ious w a y s dur ing the c o u r s e of this s tudy. J a n i c e Thurs ton, R P F , P la teau Fores t P roduc ts Ltd., Vanderhoo f , prov ided the Landsa t da ta and beet le infestation c o v e r a g e m a p s of the study a rea . V a l Fletcher, R P F , Fo res t Heal th Samp l i ng /Su rvey Specia l is t , Ministry of Fores ts , Brit ish C o l u m b i a , prov ided digital forest c o v e r m a p s of the study a rea and re levant publ icat ions. Jerry M a e d e l , G I S and R e m o t e S e n s i n g Co-coord ina to r , F I R M S Lab , Facul ty of Forestry, Universi ty of Brit ish C o l u m b i a , w a s a lways there to provide required support related to var ious remote sens ing and G I S sof tware. J a m i e Hea th , my fel low graduate student, gave m e a field tour of mounta in pine beet le infested lodgepo le pine s tands . I s incere ly thank all of them. T h e f inancial suppor t rece ived f rom Dona ld M c P h e e Scho la rsh ip is gratefully a c k n o w l e d g e d . Final thanks to my wife, T a r a , for her con t inuous suppor t and e n c o u r a g e m e n t throughout my graduate schoo l . v i Detect ion of mounta in pine beet le infestat ions us ing Landsa t T M T a s s e l e d C a p Trans fo rmat ions 1. Introduction 1.1 B a c k g r o u n d Mounta in pine beet le (Dendroctonus ponderosae Hopk. ) , hereafter referred to a s M P B , is one of the most se r ious pests of lodgepole pine (Pinus contorts Dougl . ) forests in Brit ish C o l u m b i a . Brit ish C o l u m b i a h a s 48 .79 mill ion ha of product ive forest land. Th is represen ts 5 1 . 4 8 % of total prov ince a rea . T h e current t imber harvest ing land b a s e is 23.14 mill ion ha . , 4 7 . 4 2 % of the product ive forest land. Lodgepo le pine forests m a k e up 3 5 % (17 mill ion ha) of Brit ish C o l u m b i a ' s forested l andscape and accoun t for 2 5 % (17.89 mill ion m 3 / year ) of the total t imber v o l u m e s harvested ( B . C . Ministry of Fo res t s 1998 and 2000) . T h e M P B is a natural part of lodgepo le pine e c o s y s t e m s and prefers lodgepo le pine t rees that are 80 yea rs of a g e or o lder and have large diameter. However , lodgepo le pine s tands 60 -80 y e a r s of age a re a l so suscept ib le but to a lesse r extent (Shore and Saf rany ik 1992). W h e n c l imat ic condi t ions (mild winters and w a r m s u m m e r s ) and food supp l ies (>25cm dbh lodgepo le pine trees) are favorable over a long per iod of t ime, the large e n d e m i c populat ion of beet les r e a c h e s ep idemic proport ions and a t tacks even young healthy t rees, thus greatly inf luencing forest landuse planning and harvest ing st rategies. M P B outbreaks a re not new to Brit ish C o l u m b i a . S i n c e the first recorded infestat ions in 1913 in the O k a n a g a n and Merr i t a r e a s , major infestat ions have occur red in Koo tenay Nat ional Pa rk and the Chi lcot in P la teau in the 1930s , on V a n c o u v e r Island during the 1940-50s , n e a r T a k l a and Bab ine lakes in the 1950s , and throughout m u c h of the southern interior, Chi lcot in P la teau and the S k e e n a and N a s s river a r e a s in the late 1970s and 1980s. T h e last major M P B outbreak w a s in 1982 in the Chi lcot in p la teau. B e t w e e n 1972 and 1998, the beet le kil led over 200 mill ion mature pine t rees in Brit ish C o l u m b i a (Unger 1993). Acco rd ing to forest industry es t imates , re leased on M a r c h 6, 2 0 0 0 (The D ispa tch 2000) , current M P B outbreaks are in ep idem ic proport ions in Brit ish C o l u m b i a . It is be l ieved that current infestat ions a re sp read over an a rea of approx imate ly 300 ,000 ha , in the wes t centra l part of British C o l u m b i a cons is t ing of L a k e s , Q u e s n e l , Mor i ce and V a n d e r h o o f Fo res t Distr icts. Approx imate ly 6 mill ion m 3 of t imber (equivalent to the total annua l a l lowab le cut of this a rea , va lued at $1 billion) w a s a l ready infested by M P B in the wes t central part of prov ince. T h e ep idemic nature of the infestat ions c a n be g a u g e d f rom the fact that the n u m b e r of app roved 1 Detect ion of mountain pine beet le infestat ions us ing L a n d s a t T M T a s s e l e d C a p Trans fo rmat ions logging s i tes rose to 48 ,000 from a typical annua l a v e r a g e of 6 ,000 s i tes. D u e to the a b s e n c e of a co ld -snap in the winter of yea r 1999, that might have b e e n effect ive in control l ing the beet le populat ion, the M P B ep idemic is expec ted to i nc rease aga in in the yea r 2 0 0 0 (The D ispa tch 2000) . Beet le infestat ions have a l so been reported in other parts of Brit ish C o l u m b i a a s we l l , such a s in Ar rows tone Prov inc ia l Pa rk ( M O E P 2000) , and Li l looet District ( B C , Ministry of Fo res t s 2000) . A s m u c h a s $312 .8 mill ion a l ready have been lost d u e to the 1998 wes t centra l ep idemic . A n addit ional $3 .9 billion of government revenue cou ld be lost ove r the next 10 yea rs , if t imber currently at r isk to M P B attack we re to be infested in the Ca r i boo , P r ince G e o r g e and Pr ince Ruper t forest regions (The D ispa tch 2000) . There fore , M P B m a n a g e m e n t is one of the high priority a reas for the Brit ish C o l u m b i a government . The beet le m a n a g e m e n t s y s t e m essent ia l ly h a s three c o m p o n e n t s : i) prevent ion, ii) detect ion and mapping and iii) control m e a s u r e s . However , sa lvag ing beet le-k i l led t rees and direct control methods a imed at reducing beet le populat ion in current ly infested t rees have been the tradit ional m a n a g e m e n t strategies. Prevent ive m e a s u r e s inc lude priorit izing s tands b a s e d on the potential for d a m a g e , facilitating planning of a c c e s s roads for t hese s tands , and initiating prevent ive managemen t t reatments. Cur rent beet le control s t ra tegies inc lude fell and burn, pest ic ide appl icat ions, s ing le tree sanitat ion harvest ing, partial cutt ing in smal l pa tches , and large c learcu ts . Large c learcu ts are used to reduce large beet le populat ions in severe ly infested a r e a s , and involve convent iona l logging pract ices (Macmi l lan et al., 1986). 1.2 Conventional M P B Detection Methods S u c c e s s f u l beet le infestat ions result in bright red t rees in the s u m m e r fol lowing at tack (Table 1). T h e s e can be identified either through aer ia l s u r v e y - b a s e d ske tch -mapp ing , wh ich a im at v isua l identif ication of red attack t rees, or g round b a s e d me thods (Forest Heal th S u r v e y s G u i d e b o o k 1995). Aer ia l surve i l lance, espec ia l l y of modera te to high-r isk s tands , he lps in detect ing the initial p h a s e s of beet le invasion and a l lows for the early implementat ion of effect ive control m e a s u r e s . However , aer ia l survey methods are ser ious ly h a m p e r e d by the fact that the current yea r a t tacks cannot be identified from the air. W h a t c a n be de tec ted is last yea r ' s attack, after the infested lodgepole pine fol iage has turned red in color. Howeve r , by this t ime the beet les have a l ready left the infected t rees and co lon ized other t rees. There fore , there is a lmos t a one yea r t ime g a p between s u c c e s s f u l beet le attack and its detect ion f rom aer ia l survey b a s e d methods . 2 o ro a ro CO CO CO 0 c (1) o i_ 3 o CO (/} g Q. O o o IZ ro a> ro o o L ~ ro is w" E o -*-< CL E >% U) (/)" 0 D) ro -*-» tn i m Q_ 0 ro H 0 E E CO I I ro >-sT1 ro l I 0)1 0 ro 0 >-Q . 0 CO 0 E E 3 co tn O) 0 "D ro cn o tn 0 . 5 1 o * " 1 ° "3 OQ 3 T3 ro 0" ro Q-0 cn ,_ ro 0 I o cog.g> d 0 l_ ro D_ 0~ ro £ ro "o —1 ro tn —* 3 0 T3 ro ro £ ro cs a. 0 3 ro — cn Li. 0 0 cn ro tn 0 .Si 0 0 = 0 ^ £ 5 1 1 Q. 0 D) C *L_ 0 CO ro cn w~ 0 3 °> |-5 cn c <" 1° 5 i 0 a . 0 £-CO 73 0 ro E £ 0 0 _ ro P E CD 0 0 2 m *-• ro ° i 2 E m 2? E 2 o o 0 W a. a. o O tn </) 0 o o 1 = o ro o T3 0 i_ ro tn 0 3 H: o T3 "O 0 C£ c: o ro T3 3 X 0 J= o a . > . 0 0 _Q T3 0 "O 0 0 Q . 0 k_ (/) ro 0 0 ro 0 •*—> ro o c w 0 XI 3 . C o co Detect ion of mounta in p ine beet le infestat ions us ing Landsa t T M T a s s e l e d C a p Trans fo rmat ions G r o u n d su rveys are conduc ted w h e n pocke ts of d isco lo red t rees first a p p e a r in a s tand to verify the c a u s a l agent and the status of the brood (Unger 1993). W h e n the beet le populat ion is at ep idemic proport ions, ground b a s e d su rveys , though most ef fect ive in detect ing current at tack, cannot meet the d e m a n d . T h e r e is not enough trained m a n p o w e r ava i lab le for s u c h su r veys and time and cos t requi rements a re very high (The D ispa tch 2000) . S h o r e and Saf rany ik (1992) deve loped a susceptibi l i ty and risk rating s y s t e m for M P B in lodgepole pine s tands . Th i s s y s t e m cons is t s of computa t ion of a Suscept ib i l i ty Index and a Beet le P r e s s u r e Index. T h e Suscept ib i l i ty Index is a m e a s u r e of the potential l oss of s tand basa l a r e a in the event of M P B infestat ion, and is a long-term rating. It is d e s i g n e d to rate the suscept ibi l i ty of the s tand a s a who le , not just the pine componen t of the s tand . T h e Beet le P r e s s u r e Index is a measu re of the magni tude of a M P B populat ion affect ing a s tand and is d y n a m i c in nature. W h e r e a s the Suscept ib i l i ty Index for a s tand c h a n g e s s lowly over a large per iod of t ime, the Beet le P r e s s u r e Index may c h a n g e ove r a short per iod of t ime. Th i s method is highly input intensive. T h e inputs required are: a) Suscept ibi l i ty Index: i) A v e r a g e basa l a rea / ha pine > 15cm dbh ; ii) A v e r a g e basa l a r e a / h a of all s p e c i e s > 7.5 c m dbh ; iii) A v e r a g e a g e of dominant and co-dominant s p e c i e s ; iv) S tand densi ty of all s p e c i e s > 7 .5cm dbh ; v) Lat i tude, longitude; and e levat ion; b) Beet le P r e s s u r e Index: i) N u m b e r of infested t rees inside the s tand ; ii) N u m b e r of infested t rees outs ide the s tand within 3 km; a n d iii) D i s tance f rom the s tand being rated to the neares t e d g e of the M P B infestation. Al l the inputs required are not part of convent iona l forest inventory da tabase , part icularly, the ave rage basa l a rea / ha of p ine >15cm dbh , wh ich is the most important var iab le. B e s i d e s , all the inputs required for the Beet le P r e s s u r e Index are i n - season inputs and wou ld have to be co l lec ted using expens i ve ground su rveys . S h o r e etal., (2000) carr ied out an a c c u r a c y a s s e s s m e n t of the predict ion capabi l i t ies of this s ys tem us ing 38 s tands in the C a r i b o o Fores t R e g i o n of Brit ish Co lumb ia . A l inear relat ionship ( R 2 = 0.67) w a s found be tween the percen tage of the total s tand 4 Detect ion of mounta in pine beet le infestat ions us ing L a n d s a t T M T a s s e l e d C a p T rans fo rmat ions basa l a rea killed by the M P B and the Suscept ib i l i ty Index. They sugges ted that the unexp la ined 33 percent variabil i ty cou ld have been c a u s e d by var iat ions in M P B populat ion leve ls be tween s tands , d i f ferences in host res is tance , and differential ef fects of many var iab les s u c h a s age , s tand densi ty, and locat ion. Lodgepo le pine forests are charac te r i zed by h o m o g e n e o u s e v e n - a g e d s tands . T h e s e s tands c a n be equa ted to large mono-c ropped agricul tural a reas in te rms of uniformity of spect ra l r e s p o n s e behavior ; hence , they are particularly sui table for detect ion / mapp ing f rom satel l i te b a s e d remote sens ing . Th is wou ld be helpful in c lass i fy ing lodgepole pine s tands b a s e d on a g e and thus identifying the extent of suscept ib le lodgepo le pine s tands . L o d g e p o l e pine o lder than 80 y e a r s of age are highly suscept ib le to M P B at tack and are cons ide red mature. B e c a u s e of the s u c c e s s f u l prevent ion of forest f ires during past 40 yea rs , the proport ion of mature t rees to immature t rees has i nc reased to 3:1 in the Pr ince Ruper t Fo res t R e g i o n and 3:2 in the P r i nce G e o r g e and Ca r i boo R e g i o n s (The D ispatch 2000) . Mature lodgepole pine s tands provide ideal b reed ing g rounds for M P B . Cons ide r ing the large extent of mature lodgepole pine in Brit ish C o l u m b i a and the l imitations of the convent iona l survey methods in t imely detect ion of M P B infestat ions in s u c h a vas t a rea , there is a need to explore and eva lua te al ternate me thods to detect and m a p M P B infestat ions. 1.3 Possible Alternatives Sate l l i te -based remote sens ing technology is one opt ion. Vegeta t ion health monitor ing, wh ich h a s direct re levance to forest d i s e a s e and d a m a g e detect ion, h a s been one of the most r e s e a r c h e d and extens ive ly appl ied appl icat ions of satel l i te data . App l ica t ion of satel l i te da ta in detect ion, mapp ing , and monitoring of M P B infestat ions is attractive b e c a u s e of the s tand charac ter is t ics , satell i te da ta capabi l i t ies, s e n s o r character is t ics , availabil i ty of desk top P C image ana l ys i s sof tware, and cost /benef i t aspec ts . B e c a u s e of their synopt ic and repetit ive da ta acquis i t ion capabi l i t ies, and the digital nature of the mult i -spectral da ta acqu i red , satel l i te b a s e d s e n s o r s hold promise to provide information on : i) the distribution of suscep t ib le lodgepo le pine, ii) the distribution pattern of exist ing infestat ions, iii) identif ication of poss ib le direct ion of beet le s p r e a d , and iv) identif ication of exist ing red at tack a reas . 5 Detect ion of mountain p ine beet le infestat ions us ing Landsa t T M T a s s e l e d C a p Trans fo rmat ions T h e underly ing principle for use of remote s e n s i n g is that every su r face feature h a s a un ique spect ra l ref lectance s ignature, wh ich may c h a n g e in the spat ia l and tempora l doma ins . Spat ia l and tempora l variability in vegetat ion re f lec tance ar ise from severa l vegetat ion related propert ies including leaf a rea index, canopy structure, land cove r type, ' leaf opt ical propert ies, c a n o p y c rown cover , understory vegetat ion, and s t ress c a u s e d by a number of fac tors including d i s e a s e and pests . T h e s e var iat ions in leaf spect ra l r e s p o n s e are captured through vegetat ion ind ices (Vl 's) , wh ich are d imens ion less , radiometr ic m e a s u r e s usual ly involving l inear comb ina t ions and /o r ratios of spect ra l bands (L i l lesand and Kieffer 2000) . V l ' s may be c o m p u t e d f rom digital counts or sur face ref lectance, and require no addi t ional anci l lary information. Vege ta t i on ind ices a l s o min imize the effects of factors like the soi l background , i l lumination and v iew geomet ry on the canopy radiometr ic r esponse . Vege ta ted a r e a s general ly y ield high v a l u e s for V l ' s b e c a u s e of their relatively high infrared ref lectance and low vis ib le ref lectance. T h e highest index va lue is a s s u m e d to represent m a x i m u m vegetat ion g r e e n n e s s . V l ' s have b e e n related to seve ra l vegetat ion p h e n o m e n a s u c h a s s e a s o n a l vegetat ion dynamics , forest c l ea rance , leaf a r e a index measu remen ts , b i omass est imat ion, percent g round cove r est imat ion, photosynthet ical ly act ive radiation est imat ion, and s e a s o n a l and inter-annual var iat ions in the vegeta t ion. In turn, t hese vegetat ion attributes are used in var ious mode l s to study photosyn thes is , ca rbon budgets , water ba lance , vegetat ion health monitor ing, e c o s y s t e m productivity, land c o v e r c lass i f icat ion and ca rbon and b iogeochemica l c yc l es (Asrar ef al., 1984; G o w a r d and H u e m m r i c h 1992; Hue te 1988; Jus t i ce e r a / . , 1985; L i l lesand and K i e f f e r 2 0 0 0 ; Se l le rs e r a / . , 1994). V l ' s a re current ly used by var ious agenc ies for monitor ing vegetat ion to a d d r e s s food secur i ty , c rop product ion, and fire probabil i ty. V l ' s c a n be ca tegor ized into two groups (Baret and Guyo t 1991 ; Ma jo r e r a / . , 1990; Q i e r a / . , 1994): • Ratio-based vegetation indices: T h e s e are s imple ratios of b a n d s or band combina t ions . E x a m p l e s : Rat io Vegeta t ion Index (RVI) or S imp le Rat io ( S R ) , N o r m a l i z e d Di f ference Vegeta t ion Index (NDVI) , • Orthogonal vegetation indices: b a s e d on perpend icu lar d i s tance of vegetat ion or other features f rom the soi l line in spect ra l s p a c e . E x a m p l e s : Pe rpend icu la r Vegeta t ion Index -PVI (R ichardson and W e i g a n d 1977), G r e e n n e s s Vegeta t ion Index-GVI der ived f rom T a s s e l e d C a p Transformat ions (Kauth and T h o m a s 1976). 6 Detect ion of mountain pine beet le infestat ions us ing L a n d s a t T M T a s s e l e d C a p Trans fo rmat ions T h e T a s s e l e d C a p Transformat ion is one of the mos t widely used vegetat ion ind ices. Initially, this w a s p roposed for agriculture dominated s c e n e s , but it h a s been found subsequen t l y useful for mode l ing fire haza rds (Pat terson and Y o o l 1998), mapp ing natural g r a s s l a n d s (Lauver and Whis t le r 1993), monitoring vegeta ted a r e a s (B raga e r a / . , 1991), d iscr iminat ing con i ferous s t a n d s f rom d e c i d u o u s s tands (Crist e ra / . , 1986), est imat ing s tand densi ty (Crist e r a / . , 1986; H o r l e r a n d A h e m 1986), est imat ing canopy water content ( C o h e n 1991), est imat ing forest s u c c e s s i o n (Hal l e r a / . , 1991b), est imat ing the a g e and structure of forests ( C o h e n et al., 1995); monitoring forest regenerat ion (Pr ice and J a k u b a u s k a s 1998), identifying forest t ypes and c h a n g e detect ion in Brit ish C o l u m b i a ( S a c h s ef al., 1998), and c h a n g e detect ion in forest vegetat ion condit ion (Copp in e r a / . , 2000) . However , these t ransformat ions have not been used to identify M P B infestat ions. S u c c e s s f u l M P B infestat ions kill lodgepole pine t rees, wh i ch c a u s e distinct foliar co lor c h a n g e s in the pine s tands . G r a d u a l death of lodgepo le pine c a u s e s fol iar co lor to c h a n g e f rom green to yel low and then red. Th is is a c c o m p a n i e d by the g radua l dry ing of the fol iage. Both these mani festat ions are assoc ia ted with d i f ferences in spec t ra l r e s p o n s e f rom the tree canopy . There fore , red a t tacked lodgepole pine s tands shou ld have different re f lectance va lues c o m p a r e d to non-a t tacked s tands . T h e s e spect ra l c h a n g e s c a n be identif ied through the vegetat ion ind ices. 1.4 Objectives T h e overal l object ive of this study w a s to identify p robab le M P B at tacked lodgepo le pine s tands b a s e d on T a s s e l e d C a p t ransformat ions of Landsa t -7 E T M + data . T h e detai led object ives we re to: • C o m p u t e T a s s e l e d C a p coeff ic ients for L a n d s a t - 7 E n h a n c e d Thema t i c Mapper ; • S tudy c h a n g e s in T a s s e l e d C a p t ransforms: br ightness, g r e e n n e s s and we tness , c a u s e d by var iat ions in topography, M P B infestation s i z e , and satel l i te da ta acquis i t ion da tes ; • Identify M P B at tacked lodgepole pine s tands f rom healthy lodgepo le pine s tands using s ing le date and two date T a s s e l e d C a p T rans fo rms ; and • A s s e s s the identification accu racy of M P B infested lodgepo le pine s tands . Th i s thes is is o rgan ized into the fol lowing sec t ions : the concep t of T a s s e l e d C a p Trans format ions ; a l iterature rev iew on appl icat ion of remote s e n s i n g in M P B attack detect ion in Brit ish C o l u m b i a ; material and methods ; results and d i scuss ions ; and c o n c l u s i o n s including poss ib le further resea rch a reas . 7 Detect ion of mounta in p ine beet le infestat ions us ing Landsa t T M T a s s e l e d C a p Trans fo rmat ions 2.0 Tasseled Cap Transformations 2.1 The Concept VTs a re d i f ferences and ratios of spect ra l b a n d s that a re used for vegetat ion monitor ing. The re are other fo rms of l inear da ta t ransformat ions, wh ich h a v e a l so been used for vegetat ion monitor ing. T h e T a s s e l e d C a p t ransformat ion, d e v e l o p e d by Kauth and T h o m a s (1976) for Landsa t M S S data , is an examp le of s u c h a t ransformat ion. They o b s e r v e d that da ta in a multi-d imens iona l spect ra l s p a c e is not distr ibuted uniformly throughout. Instead, the data tend to be concent ra ted in certain reg ions of this s p a c e , giv ing r ise to a structure. T h e s e st ructures present in data f rom a part icular s e n s o r are directly re lated to the actua l phys ica l charac ter is t ics of the s c e n e c l a s s e s . H e n c e , extract ion of the information of t hese c l a s s e s will be best, if the data structure c a n be v iewed in its entirety and sepa ra ted f rom other da ta structures. In an agricultural a rea , this structure is of the form of a ' T a s s e l e d C a p ' . T h e T a s s e l e d C a p t ransformat ions cons is t of: i) identifying da ta st ructures for a part icular s e n s o r and appl icat ion; ii) chang ing the v iewing perspec t i ve (i.e. rotating the axes ) s u c h that those structures c a n be v iewed most directly; and iii) def in ing feature d i rect ions wh ich co r respond to spect ra l var iat ions in a part icular c l a s s (Crist and Kau th , 1986). Kauth and T h o m a s (1976) used four Landsa t M S S bands in l inear comb ina t ions to p roduce four ind ices ca l led Br igh tness (BR) , G r e e n n e s s (GN) , Y e l l o w n e s s (YN) , and N o n s u c h (NS) . T h e M S S da ta are rotated s u c h that the majority of information is conta ined in first two c o m p o n e n t s (i.e. br ightness and g reenness ) that are directly related to phys ica l s c e n e character is t ics . Br igh tness is a weighted s u m of all bands and is def ined in the direct ion of the principal var iat ion in soi l re f lectance. T h e s e c o n d componen t G r e e n n e s s is or thogonal to br ightness and is a contrast be tween the near- inf rared and vis ib le bands . It is strongly related to the amoun t of g reen vegetat ion present in the s c e n e . T h e s e two c o m p o n e n t s have proved useful for evaluat ing soi l and vegetat ion features in Landsa t data (Kauth e r a / . , 1979; T h o m p s o n and W e h m a n e n 1980). Cr is t and C i c o n e (1984) ex tended the T a s s e l e d C a p concep t to Landsa t T M data and found that the s ix bands of da ta effectively occupy three d i m e n s i o n s - br ightness, g r e e n n e s s and we tness , def ining p lanes of soi ls , vegetat ion, and a transit ion z o n e be tween them. T h e third feature W e t n e s s , is a contrast be tween shor t -wave infrared (SWIR) and v is ib le/near- in f rared (VNIR) da ta and is related to canopy and soi l moisture. 8 Detect ion of mounta in p ine beet le infestat ions us ing Landsa t T M T a s s e l e d C a p Trans fo rmat ions T a s s e l e d C a p t ransformat ions use the G r a m - S c h m i d t or thogonal izat ion p rocedure to genera te componen ts that a re or thogonal to e a c h other, mean ing that the information in these c o m p o n e n t s is uncorre lated, unl ike the information in or iginal bands . It a l so s e r v e s to reduce the dimensional i ty of the data . Ano the r techn ique, Pr inc ipa l C o m p o n e n t A n a l y s i s ( P C A ) , has often been used to unders tand data d imensional i ty and to genera te or thogonal componen t s . Wi th P C A the interpreter i m p o s e s no prior order or phys ica l interpretation on the pr incipal d i rect ions. T h e s e are determined by s u c c e s s i v e di rect ions of m a x i m u m var iat ion. In T a s s e l e d C a p Trans fo rmat ions , the ax is are a l igned in direct ions that h a v e phys ica l s ign i f icance. P C A is, thus, a stat ist ical procedure requir ing no a priori knowledge of features s u c h a s so i ls and vegetat ion ( J a c k s o n , 1983). T h e pr incipal c o m p o n e n t s genera ted for different s c e n e s cou ld be different depend ing on the variabil ity in the data for that s c e n e . In T a s s e l e d C a p Trans fo rmat ions , b e c a u s e the componen ts are def ined b a s e d on phys ica l character is t ics of relevant s c e n e c l a s s e s (soil and vegetat ion), they are app l icab le for different s c e n e s cover ing those s a m e c l a s s e s . Cr is t and Kauth (1986) pointed out that only the v iewing perspect ive c h a n g e s in these t ransformat ions; the data are fundamenta l ly the s a m e before and after its appl icat ion. 2.2 Tasseled Cap Coefficients T h e br ightness, g r e e n n e s s and w e t n e s s images are genera ted by mult iplying e a c h T M band , pixel by pixel , by a co r respond ing coeff icient. T h e coef f ic ients are unit vec to rs that indicate direct ion. T h e genera l form of equat ion is: T C ; = A * D N 1 + B i * D N 2 + C i * D N 3 + D i * D N 4 + E i * D N 5 + F i * D N 7 where: A i - Fi are band spec i f ic coef f ic ients for T a s s e l e d C a p c o m p o n e n t i, ( i = 1 represents br ightness, i = 2 represents g r e e n n e s s and i = 3 represents we tness ) ; and D N i - D N 7 a re Digital N u m b e r s in spect ra l bands 1 to 5 and 7 T a s s e l e d C a p coef f ic ients for Landsa t -4 and 5 are g iven in Tab le 2. Cr is t and Kauth (1986) sugges ted that a c h a n g e to the s e n s o r requi res reworking of coef f ic ients. Th is is n e c e s s a r y b e c a u s e of c h a n g e s in the s e n s o r character is t ics . In c a s e of E T M + , the spect ra l bandwidth w a s slightly different f rom Landsa t -5 (Table 3) and the radiometry w a s different ( S c i e n c e Wr i ters G u i d e to Landsa t -7 , 1999). A survey of the literature revea led that t hese coeff ic ients for Landsa t -7 E T M + were not ava i lab le or compu ted . There fore , for this study new coef f ic ients we re requi red. i 9 Detect ion of mounta in p ine beet le infestat ions us ing Landsa t T M T a s s e l e d C a p T rans fo rmat ions T a b l e 2. T a s s e l e d C a p Coef f ic ients for V a r i o u s Landsa t Sate l l i tes Landsa t M S S : (Source : Kauth and T h o m a s 1976) Fea tu re B a n d 4 B a n d 5 B a n d 6 B a n d 7 Br igh tness 0.433 0.632 0.586 0.264 G r e e n n e s s -0 .290 -0 .562 0.600 0.491 Y e l l o w n e s s -0 .829 0 .522 -0 .039 0.194 N o n s u c h 0.223 0.012 -0 .543 0.810 Landsa t -4 Themat i c M a p p e r (Sou rce : Cr is t , and C i c o n e 1984) Fea tu res B a n d l B a n d 2 B a n d 3 B a n d 4 B a n d 5 B a n d 7 Br igh tness 0 .33183 0.33121 0 .55177 0 .42514 0 .48087 0 .25252 G r e e n n e s s -0 .24717 -0 .16263 -0 .40639 0 .85468 0 .05493 -0 .11749 W e t n e s s 0 .13929 0 .22490 0 .40359 0 .25178 -0 .70133 -0 .45732 Landsa t -4 Themat i c M a p p e r (Source : Cr is t et al., 1986) Fea tu res B a n d l B a n d 2 B a n d 3 B a n d 4 B a n d 5 B a n d 7 Br igh tness 0 .3037 0 .2793 0 .4743 0 .5585 0 .5082 0 .1863 G r e e n n e s s -0 .2848 -0 .2435 -0 .5436 0 .7243 0 .0840 -0 .1800 W e t n e s s 0 .1509 0 .1973 0 .3279 0 .3406 -0 .7112 -0 .4572 H a z e 0 .8832 -0 .0819 -0 .4580 -0 .0032 -0 .0563 0 .0130 Fifth 0 .0573 -0 .0260 0 .0335 -0 .1943 0 .4766 -0 .8545 Six th 0 .1238 -0 .9038 0.4041 0 .0573 -0.0261 0 .0240 Landsa t -5 Themat i c M a p p e r (Source : Cr is t e r a / . , 1986) Fea tu res B a n d l B a n d 2 B a n d 3 B a n d 4 B a n d 5 B a n d 7 Br igh tness 0 .2909 0 .2493 0 .4806 0.5568 0 .4438 0 .1706 G r e e n n e s s -0 .2728 -0 .2174 -0 .5508 0.7221 0 .0733 -0 .1648 W e t n e s s 0.1446 0.1761 0.3322 0.3396 -0 .6210 -0 .4186 H a z e 0.8461 -0.0731 -0 .4640 -0 .0032 -0 .0492 0 .0119 Fifth 0 .0549 -0 .0232 0 .0339 -0 .1937 0 .4162 -0 .7823 Six th 0 .1186 -0 .8069 0 .4094 0.0571 -0 .0228 0 .0220 10 Detect ion of mounta in p ine beet le infestat ions us ing Landsa t T M T a s s e l e d C a p Trans fo rmat ions Tab le 3. Spec t ra l Bandwid th (micrometers) of Landsat -5 T M and Landsat -7 E T M + (Source : L i l lesand and Kieffer 2000) Spectral Band-Width Landsat-5 TM Landsat-7 ETM+ B a n d l 0.45-0.52^ 0.45-0.52n Band2 0.52-0.60^  0.53-0.61 |i Band3 0.63-0.69^ 0.63-0.69u Band4 0.76-0.90^  0.78-0.90n Band5 1.55-1 .75M 1.55-1.75u Band6 (Thermal) 10.4-12.5^ 10.4-12.5LI Band7 2.08-2.35n 2.09-2.35u Band8 (Panchromatic) N/A 0.52-0.90|a 2.3 Characteristics of different cover types Figure 1 i l lustrates genera l locat ions of s o m e important s c e n e c l a s s e s in the T M T a s s e l e d C a p feature s p a c e . A typical c rop ove r the growing cyc le represents spec t ra l d e v e l o p m e n t through e m e r g e n c e , green ing, c a n o p y c losure and s e n e s c e n c e . D e p e n d i n g on the c rop a n d its deve lopment , this trajectory c a n be different, but it will lie within the ' C r o p s a n d so i ls ' region in the feature s p a c e . T h e w e t n e s s d imens ion improves del ineat ion be tween deve lop ing vegetat ion and s e n e s c i n g vegetat ion. TM W*tn«*-@-Figure 1. Approx imate locat ions of var ious sur face c o v e r t ypes in T M T a s s e l e d C a p feature s p a c e ( C r i s t e r a / . , 1986). 11 Detect ion of mounta in pine beet le infestat ions us ing Landsa t T M T a s s e l e d C a p Trans fo rmat ions In T a s s e l e d C a p feature s p a c e , forest vegetat ion o c c u p i e s the front of the c a p and is referred to a s "badge of t rees" (Crist ef al., 1986). It is mainly b e c a u s e of the i nc rease in s h a d o w s in a forest s tand a s c o m p a r e d to crop or g rass canop ies . A forest s tand con ta ins a higher percen tage of o p a q u e s tems , a s c o m p a r e d to crop or g r a s s canopy , thus inc reas ing the inc idence of d e e p s h a d o w s both on the lower level of the c a n o p y a n d on l e a v e s / n e e d l e s in the tree c rowns . W a t e r s h o w s min imum br ightness and g r e e n n e s s va lues , but m a x i m u m w e t n e s s va lues and is sepa rab le f rom other c l a s s e s in all d imens ions . 12 Detect ion of mounta in p ine beet le infestat ions us ing Landsa t T M T a s s e l e d C a p Trans fo rmat ions 3.0 Literature Review S e v e r a l s tud ies have been carr ied out in the past to detect bark beet le infestat ions in Brit ish C o l u m b i a us ing both aer ia l and satel l i te b a s e d remote sens ing data . Aer ia l da ta , mainly aer ia l co lor infra-red (CIR) photographs, have been used to identify s p r u c e beet le {Dendroctonus rufipennis kby.) attack (Banne r 1986; C h u r c h e r and M c L e a n 1984; Mur tha and C o z e n 1985; Mur tha 1985; Mur tha and Fourn ie r 1992), D o u g l a s fir beet le infestat ions (Hal l ef al., 1981 ; Hal l ef ai, 1983), M P B infestat ions ( H o b b s 1983 ; H o b b s and Mur tha 1984; Mur tha and Wiar t 1987; Mur tha and Wiar t 1989a andb) . K n e p p e c k and A h e r n (1988) a n a l y z e d A i rborne M E I S (Mult i -detector E lect ro-opt ica l Imaging Scanne r ) data , acqu i red at three different spat ia l resolut ions (1.4m, 3 .4m and 6m) , to a s s e s s the detectabil i ty of red attack t rees. They c o n c l u d e d that: i) with 6 m resolut ion, individual red attack t rees cou ld not be identified on normal co lor compos i te , ii) with 3 .4m resolut ion l ess red attack t rees we re identif ied c o m p a r e d to 2 3 c m aer ia l pho tographs at a 1:100,000 s c a l e , and iii) a 1.4m resolut ion image w a s found best for detect ion of red attack t rees. A number of s tud ies (e.g., A h e r n 1988; A h e r n and Arch iba ld 1986; Harr is etal., 1978; Mur tha ef al., 2000 ; R e n c z and Neme th 1985; S i ro is and A h e r n 1988; Tay lo r 1998) have been carr ied out using sate l l i te-based remote sens ing to detect M P B infestat ions. B a s e d on the ana l ys i s me thods emp loyed , the s tud ies c a n be ca tegor i zed into two c l a s s e s : i) those b a s e d on v isua l interpretation, and ii) those b a s e d on digital image ana l ys i s techn iques . 3.1 Visual Image Interpretation Based Studies At tempts to use satell i te remote s e n s i n g data for M P B detect ion in Brit ish C o l u m b i a date f rom 1978, w h e n Harr is etal., u sed Landsa t M S S da ta to detect M P B infestat ions. T h e resul ts we re not encourag ing , largely b e c a u s e of the scat tered nature of infestat ions and the c o a r s e spat ia l resolut ion (80m) of Landsa t M S S . Us ing s ingle date M S S data , an identif ication a c c u r a c y of only 2 5 % w a s ach ieved . A h e r n and Arch iba ld (1986), b a s e d on v isua l ana lys i s of fa lse co lor compos i t es (red, near infrared, and short w a v e infrared spect ra l bands) s h o w e d that gray a t tacked a reas were identif iable b e c a u s e of a dist inct c y a n co lor on fa lse co lor images . 13 Detect ion of mounta in p ine beet le infestat ions us ing Landsa t T M T a s s e l e d C a p Trans fo rmat ions S i ro is and A h e r n (1988) carr ied out a study to identify red attack lodgepo le pine s tands near Bab ine L a k e , Mor i ce Fores t District, Brit ish C o l u m b i a . Digital da ta f rom S P O T M L A (Mul t i -spectra l L inear Ar ray , resolut ion 20 m) and P L A (Panch romat i c L inear Ar ray , resolut ion 10 m) acqu i red on A u g u s t 11, 1986, we re used in the study. M P B infestat ions of f ive or more red t rees, d e m a r c a t e d on forest s tand m a p s at a 1:20,000 sca le , we re used a s ground truth. No rma l co lor aer ia l photographs, at a 1:10,000 sca le , acqu i red in A u g u s t 1986 we re used to verify the resul ts. Different band combina t ions of digitally e n h a n c e d S P O T data for three test a r e a s w e r e visual ly interpreted to est imate the lower limit of d a m a g e detec tab le f rom S P O T data . In all three a r e a s , af fected s tands cons is ted of lodgepo le pine s tands be tween 141-250 yea rs of age , c rown c losu res of 36-45 percent, and heights of 28 .5 to 37.4 m. However , the proport ion of red a t tacked t rees in e a c h test a rea w a s different (Table 4). Tab le 4. Propor t ion of red a t tacked t rees in different study a reas (Sirois and A h e r n 1988) Pa rame te rs A r e a 1 A r e a 2 A r e a 3 S i z e 2-3 ha 2 ha 0 .8ha Proport ion of red c rowns 8 0 - 9 0 % * red co lored c rown 20 % ( 35 red trees) 4 0 % (>50 trees) *red colorat ion of c rown w a s c a u s e d due to heat and s m o k e f rom s l ash burn ing; s u c h c rowns were a s s u m e d to have s imi lar spect ra l propert ies a s that c a u s e d by M P B T h e authors conc luded that it w a s not poss ib le to identify a r e a s of scat te red at tack a n d that the min imum red attack d a m a g e detec tab le with the S P O T satel l i te w a s approx imate ly 1 to 2 ha in s i ze with 80 to 100 percent of red c rowns . Th i s deg ree of mortality w a s found to be very high for control p rograms where the requi rement w a s to detect infestat ions of f ive or more t rees. H o w e v e r it w a s conc luded that S P O T data wou ld be usefu l for inventory update fol lowing an outbreak that c a u s e d ex tens ive mortality. Tay lo r (1998) c o m p a r e d the identif ication accu racy of M P B and Douglas- f i r beet le infestation detect ion obta ined us ing aer ia l ske tch mapp ing and v isua l interpretation of Landsa t normal co lor compos i te images . Ove rv iew aer ia l ske tch mapp ing , conduc ted f rom either he l icopters or aircraft, is the primary techn ique used to detect beet le a t tacks in Brit ish C o l u m b i a and rel ies upon v isua l identif ication of ye l low brown and /o r red tree c rowns . Th i s study w a s conduc ted in the Fort St. J a m e s Fores t District of the Pr ince G e o r g e Fores t R e g i o n . Lodgepo le pine w a s the pr imary forest type and be longed to the mature or overmature a g e g roups (>80 y e a r s of age) . Landsa t data acqu i red on July 22 , 1998 w e r e used in the study. Beet le infestation on three s a m p l e forest 14 Detect ion of mounta in pine beet le infestat ions using Landsa t T M T a s s e l e d C a p Trans fo rmat ions c o v e r m a p s we re del ineated using aer ia l ske tch mapp ing , conduc ted on A u g . 6 and 28 , 1998. Both supe rv i sed c lassi f icat ion of digital da ta , a s wel l a s v isua l interpretation of satel l i te data by two independent ski l led photointerpreters, w e r e at tempted to del ineate bark beet le infested a reas . T h e resul ts p resented in this paper are s u m m a r i z e d in T a b l e 5. Tab le 5. C o n f u s i o n matrix showing beet le infested a rea (ha) identif ied by v isua l interpretation (Taylor 1998) Interpreter #1 Aer ia l Ske t ch Mapp ing Landsa t Infested Heal thy Tota l Infested 167 5191 5358 Heal thy 1111 2 5 2 1 6 2 6 3 2 7 Total 1278 30407 3 1 6 8 5 Interpreter #2 Aer ia l Ske t ch Mapp ing Landsa t Infested Heal thy Total Infested 271 6 2 5 9 6 5 3 0 Heal thy 1007 2 4 1 4 8 2 5 1 5 5 Total 1278 30407 3 1 6 8 5 W h a t c a n be inferred from this is that the identif ication a c c u r a c y of beet le infested a r e a s b a s e d on v isua l interpretation of Landsa t satel l i te da ta w a s less than f ive percent in both c a s e s , wh ich is too low to be accep tab le . Bes i des , both er rors of om iss ion and c o m m i s s i o n we re very high. B e c a u s e of this, the authors conc luded that Landsa t imagery w a s not a su i tab le rep lacement for aer ia l ske tch mapp ing . 3.2 Digital Image Analysis Based Studies Dur ing a joint study between the C a n a d a C e n t e r for R e m o t e S e n s i n g ( C C R S ) and the Brit ish C o l u m b i a Ministry of Fores ts , R e n c z and N e m e t h (1985) eva lua ted the capabi l i t ies of Landsa t M S S and Themat i c M a p p e r Data (s imulated f rom ai rborne mult i -spectral s c a n n e r digital data) to detect M P B infestat ions in four test a r e a s (each 5 x 8 km in s ize) , near Tat la L a k e , Whi t ton L a k e , C lea rwa te r L a k e and Carpen te r L a k e in the C a r i b o o region of Brit ish C o l u m b i a . Landsa t M S S data we re acqu i red on two da tes (Sep tembe r 18, 1975 and A u g u s t 17, 1981). 15 Detect ion of mounta in p ine beet le infestat ions us ing Landsa t T M T a s s e l e d C a p Trans fo rmat ions A i rborne s c a n n e r da ta we re acqu i red on A u g u s t 21 , 1982. Norma l co lor and co lo r infrared photographs (1:30,000 sca le ) , a l so acqu i red on A u g u s t 2 1 , 1982, we re u s e d to identify red a t tacked t rees. S te reo a i rphotos (1:800 sca le ) acqu i red ove r e a c h test a r e a w e r e interpreted for c rown c losure and for tree counts of healthy, red at tack and gray at tack t rees to ca lcu la te levels of infestat ions. Infestation s i tes identified from the la rge-sca le photography w e r e ca tegor i zed a s gray at tack (>30 % gray), red attack high (>67 % red) and red at tack m e d i u m (>10 % red). Digital ana l ys i s w a s carr ied out for the fol lowing four da ta se ts / comb ina t ions us ing a supe rv i sed c lass i f icat ion: - S imu la ted Landsa t T M ( B a n d s 2345) Comb ina t i on 1: S ing le date Landsa t M S S ( B a n d s 4567) Comb ina t i on 2: T w o date Landsa t M S S der ived No rma l i zed Di f ference Vegeta t ion Index (NDVI) Comb ina t i on 3: 4 band compos i te of Landsa t M S S ( B a n d s 5 and 7 f rom two dates) Tra in ing s i tes we re se lec ted f rom infestation s i tes identif ied on the 1:30,000 s c a l e aer ia l photographs. Identification accu racy w a s ca lcu la ted for the three ca tegor ies (red high, red med ium and gray) for the S imu la ted Themat i c Mappe r . In the c a s e of the M S S , only two c l a s s e s (red a t tacked and gray at tacked) were a s s e s s e d . T h e resul ts ob ta ined are s u m m a r i z e d in Tab le 6. Tab le 6. A c c u r a c y (%) ach ieved for identif ication of M P B infestat ions ( R e n c z and N e m e t h 1985) Study a rea S imu la ted T M Landsa t M S S Comb ina t ion 1 Comb ina t i on 2 Comb ina t ion 3 1 2 3 1and2 3 1and2 3 1and2 3 Tat la L a k e 66 94 66 60 55 10 16 65 60 C learwa te r L a k e 25 48 60 50 50 0 32 0 60 Whit ton L a k e 80 80 30 4 5 32 Carpen to r L a k e 68 84 39 1=Red H igh , 2= R e d M e d i u m , 3 Gray Excep t at the C lea rwa te r L a k e site where the infestation w a s very sma l l in s i z e (few trees) and scat tered, infestation s i z e s in the remaining s i tes we re greater than 1.5 ha . P e r h a p s the scat tered infestation w a s the reason for the poor identif ication a c c u r a c y o b s e r v e d at the in C lea rwa te r L a k e 16 Detect ion of mounta in pine beet le infestat ions us ing L a n d s a t T M T a s s e l e d C a p Trans fo rmat ions site. A s e c o n d observat ion w a s that the 3 0 m S imu la ted T M data prov ided better identif ication a c c u r a c y than the 8 0 m M S S data, obv ious ly b e c a u s e of better d iscr iminat ion capabi l i ty. R e n c z and Neme th (1985) conc luded t ha t " . . . .spat ia l and spect ra l resolut ion of T h e m a t i c M a p p e r da ta will permit insect d a m a g e , speci f ical ly red at tack in lodgepo le pine, to be moni tored w h e r e a reas of outbreak e x c e e d 1.5 ha" . They further conc luded that the s i ze of the outbreak is the major inf luencing factor in success fu l insect d a m a g e detect ion. It w a s stated that infestat ions must be larger than 3.0 ha for detect ion to be rel iable. Mur tha ef al., (2000) used spect ra l unmix ing techn iques to der ive M P B attack probabil i ty us ing Landsa t -5 T M digital data acqu i red on A u g u s t 23 , 1998. Th i s study w a s carr ied out in a part of the V a n d e r h o o f Fores t District in the Pr ince G e o r g e Fores t R e g i o n . Th i s a r e a is charac te r i zed by smal l and scat tered M P B infestat ions that a re sub-p ixe l in s i z e (i.e. sma l le r than the spat ia l resolut ion of Landsa t T M ) . In a broad s e n s e , spect ra l unmix ing techn iques tend to identify the relative proport ion of cove r types contr ibut ing to the compos i t e re f lectance f rom a pixel . T h e methodo logy cons is ted of: i) se lect ing forest po lygons wh ich (a) con ta ined more than 50 percent under lodgepo le pine and (b) whe re lodgepo le pine s tands w e r e more than 60 y e a r s of age ; ii) e l iminat ing po lygons whe re satell i te da ta w a s under c loud / s h a d o w s ; iii) stratifying the remain ing po lygons into three ca tegor ies b a s e d on lodgepo le pine s tand age : 60 -145 , 146-170 and >170 years ; iv) running a spect ra l unmix ing algor i thm on the s ix b a n d s (B, 1,2,3,4,5 and 7) of Landsa t data and generat ing images showing the pe rcen tage of e a c h pixel cons is t ing of an e n d m e m b e r ; v) generat ing po lygon level a v e r a g e s for e a c h e n d m e m b e r f rom pixels under that po lygon ; vi) ass ign ing a compos i te at tack va lue to e a c h po lygon b a s e d on red and g reen at tack fract ions; and vii) c lass i fy ing forest po lygons into 10 probabil i ty c l a s s e s of at tack b a s e d on at tack fract ion. Fo r this purpose , attack fraction w a s conver ted to a pe rcen tage and d iv ided into 10 c l a s s e s . T h e higher the va lue of attack fract ion, the higher the probabil i ty of that s tand have been a t tacked by M P B and v i ce -ve rsa . F ie ld verif ication of results through plant s t ress detect ion g l a s s e s ass i s ted v isua l identif ication of red a t tacked t rees through an aer ia l survey in part of the study a rea . Th i s w a s fo l lowed by final accu racy a s s e s s m e n t . T h e authors p resented a c o m p a r i s o n of percent attack probabil i ty of po lygons in all 10 probabil i ty c l a s s e s with p r e s e n c e or a b s e n c e of red a t tacked t rees a s identif ied from the aer ia l survey. T h e overal l identif ication a c c u r a c y at the po lygon level w a s 79 .30 percent (Table 7). 17 Detect ion of mounta in pine beet le infestat ions us ing Landsa t T M T a s s e l e d C a p Trans fo rmat ions Tab le 7. T h e accu racy a s s e s s m e n t resul ts obta ined us ing spect ra l unmix ing p rocedures (Murtha et al., 2000) L A N D S A T R e f e r e n c e data (aerial survey) Tota l N o n -d a m a g e d D a m a g e d Fract ion %* 0-31 32-41 42 -49 50-57 58 -65 66 -76 77-88 8 9 -100 0-31 59 20 12 6 3 1 3 104 32-41 2 2 42 -49 13 13 50-57 14 14 58-65 20 20 66-76 14 14 77-88 24 24 89-100 26 26 TOTAL 59 22 2 5 20 20 17 25 29 217 * At tack fraction probabil i ty percent Overa l l accu racy = (59+2+13+14+20+14+24+26) =172 / 217 = 7 9 . 2 6 % Al though this app roach d o e s not indicate the number of red at tack t rees, the relative s i ze of infestation within a po lygon and the relat ive locat ion of red a t tacked t rees within a po lygon , it d o e s indicate a reas whe re efforts shou ld be d i rected for cost ly and t ime c o n s u m i n g ground su r veys to plan for control m e a s u r e s . A s c o m p a r e d to convent iona l hazard rating s y s t e m s (e.g. S h o r e and Saf rany ik 1992; S h o r e at el., 2000) , this app roach is faster and more cos t effect ive. A total a r e a of 346 ,300 ha, cove red by 19 Brit ish C o l u m b i a M o F m a p shee ts , cou ld be a n a l y z e d us ing only a part of Landsa t s c e n e within one month. Ano the r advan tage of this method is that an entire s tand can be charac te r i zed a s d a m a g e d or d a m a g e d . Th i s is much c l ose r to the forest inventory app roach than is c lass i f icat ion of indiv idual p ixels, wh ich charac te r i ze the c l a s s of a part icular pixel but not that of a forest s tand . B e s i d e s , forest po lygons cou ld be further c lass i f ied accord ing to the d a m a g e sever i ty. 3.3 Recent Initiatives SELES: Spatially Explicit Landscape Event Simulator project A joint team of exper ts from the P r i nce Ruper t Fo res t R e g i o n , Mor i ce and L a k e s Fores t Distr icts, Brit ish C o l u m b i a Pa rks , a c a d e m i c institutions and the private sector , led by M o F planning 18 Detect ion of mounta in pine beet le infestat ions us ing L a n d s a t T M T a s s e l e d C a p T rans fo rmat ions s y s t e m s biologist Don M o r g a n , is deve lop ing me thods for model ing the genera l behav io r of the bark beet les at a l a n d s c a p e level . Th i s mode l is expec ted to help forest m a n a g e r s to look at beet le movemen t a c r o s s a l a n d s c a p e over t ime, and predict bark beet le patterns. O n e of the important c o m p o n e n t s of this project is the u s e of satel l i te imagery, acqu i red ove r a t ime per iod of 1993-1999 , to track infestat ions over t ime and study beet le infestation s p r e a d pattern in the L a k e s Fo res t District. Future expec ta t ions for use of satel l i te imagery are "to pick out detai led infestation information s o that cost ly overv iew fl ights and ground probing will not be n e c e s s a r y " . Resu l t s f rom this study are yet not ava i lab le ( B C M o F 1999). Obse rva t i ons f rom the satel l i te da ta b a s e d s tud ies on M P B detect ion desc r i bed earl ier, c a n be s u m m a r i z e d a s fol lows: Infestations larger than 1-2 ha c a n be de l inea ted with reasonab le a c c u r a c y us ing v isua l ana lys is techn iques. H o w e v e r infestat ions of this s i ze are very large f rom the point of v iew of control measu res . S tud ies carr ied out s o far have largely rel ied upon v isua l interpretation techn iques . Theoret ica l ly , infestat ions of a s smal l a s 0.09 ha s i ze shou ld be poss ib le to detect e v e n with data of 30 -m spat ia l resolut ion satell i te da ta . T h e majority of the s tud ies have at tempted direct detect ion by employ ing c lass i f icat ion a lgor i thms operab le at the pixel level , w h e r e a s in many c a s e s infestation s i z e is subp ixe l in nature. Spec t ra l unmixing p rocedures , wh ich opera te at subp ixe l level , have s h o w n by far the mos t promis ing result in identifying M P B infestat ions at the s tand level . A l though M P B detect ion us ing satell i te data a re gain ing momen tum, it has to be suppor ted by more s tud ies involving ana lys i s of different spat ia l resolut ion data , appl icat ion of va r ious digital ana lys i s techn iques , and us ing al ternate concep ts . 19 Detec t ion of mounta in p ine beet le infestat ions us ing Landsa t T M T a s s e l e d C a p Trans fo rmat ions 4.0 Material and Methods T h e c h o i c e of the s tudy a r e a w a s largely de te rm ined by the avai labi l i ty of satel l i te da ta , aer ia l da ta and relevant g round truth informat ion, acqu i red dur ing a separa te study on detect ion of M P B infestat ions us ing spec t ra l unmix ing techn iques (Mur tha e r a / . , 2000) . 4.1 S tudy area T h e study a r e a c o v e r s parts of the L a k e s Fo res t District (Pr ince Ruper t Fo res t Reg ion ) , border ing T w e e d s m u i r Prov inc ia l P a r k in west and V a n d e r h o o f Fores t District (Pr ince G e o r g e Fores t Reg ion ) in the east in Brit ish C o l u m b i a (F igure 2). F igure 2. Loca t ion m a p of s tudy a rea 20 Detect ion of mounta in p ine beet le infestat ions using L a n d s a t T M T a s s e l e d C a p Trans fo rmat ions It l ies be tween longi tudes 124° 30 ' and 125° 30 ' W a n d lat i tudes 53° 00 ' and 53° 40 ' N, and is cove red by N T S M a p shee t N u m b e r 9 3 F at a 1:250,000 sca le . K n e w s t u b b L a k e , Na ta lkuz Lake , Tate lkuz Lake , T s a c h a L a k e , Johnny L a k e , M o o s e L a k e and C a p o s e L a k e are s o m e of the major water bod ies in this a rea . Mos t of the a r e a is i naccess ib le . Logg ing roads a c c e s s only the eastern part, whi le the wes te rn part towards T w e e d s m u i r P a r k h a s no roads . T h e major tree s p e c i e s in this a rea is lodgepole pine wh ich const i tu tes about 80 percent of the total forest cove r in the a rea (Table 8). T h e remain ing 20 percent cons is t of b a l s a m poplar (Populus balsamifera), a s p e n (Populus tremuloides), a lp ine fir (Abies lasiocarpa), b lack sp ruce (Picea mariana), and interior sp ruce (Picea engelmannii). T h e a g e of l odgepo le pine s tands var ies from less than 20 yea rs to a high of over 300 yea rs (Table 9). Abou t 66 percent of the lodgepole pine s tands are more than 81 yea rs of a g e and a re suscep t ib le to M P B attack (Shore and Saf rany ik 1992). Tab le 8. Re la t ive proport ion (%) of different forest tree s p e c i e s in study a r e a T r e e s No. of po lygons % A r e a of po lygons % B a l s a m poplar 2 0.01 18 0.01 A s p e n 313 2.64 3742 1.58 A lp ine fir 579 4.88 15677 6.59 Lodgepo le pine 8489 71 .53 186551 78 .35 S p r u c e 1196 10.08 17236 7.24 B lack sp ruce 264 2.22 2 0 3 0 0.85 Enge lmann s p r u c e 9 0.09 199 0.08 Whi te S p r u c e 1015 8.55 12617 5.30 Total 11867 2 3 8 0 7 0 Data sou rce : M O F forest cove r m a p s 0 9 3 F 1 3 , 14, 15, 16, 23 , 24 , 25 , 26 , 33 , 34, 35 , 36, 43 , 44 , 45 , 46 , 53,5 4, 55 , 56 (1:20,000 sca le ) 21 Detect ion of mounta in p ine beet le infestat ions us ing Landsa t T M T a s s e l e d C a p Trans fo rmat ions Tab le 9. A g e - c l a s s distribution of lodgepo le pine s tands in study a rea A g e group A r e a (ha) Pe rcen t <20 5085 2.75 21 -40 2406 1.29 41 -60 19288 10.34 61-80 39346 21 .09 81 -100 8662 4 .64 101-120 13613 7.29 121-140 3 4 6 7 5 18.59 141-160 35849 19.22 161-180 16095 8.62 181-200 8567 4 .59 201 -220 2075 1.11 221 -240 6 6 3 0.36 241 -260 37 0.02 261 -280 136 0.07 281 -300 25 0.01 301-320 29 0.01 186551 100.00 Data S o u r c e : M O F forest cove r m a p s 0 9 3 F 1 3 , 14, 15, 16, 23 , 24, 25 , 26 , 33 , 34, 35 , 36, 43 , 44 , 45 , 46 , 53,5 4, 55 , 56 (1:20,000 sca le ) 4.2 Data Used Landsa t -7 E T M digital da ta ( B a n d s l , 2, 3, 4, 5 and 7), acqu i red on A u g u s t 2, 1999 a n d S e p t e m b e r 12, 1999, we re used in this study. Deta i ls of other da ta used are g iven in Tab le 10. T R I M map shee ts (1:20,000 sca le ) w e r e used to der ive road and river vec to rs used for geometr ic rectif ication of satell i te data . Digital forest c o v e r m a p s we re used to identify lodgepo le pine s tands. Tab le 10. Detai ls of data used Da ta S o u r c e R e m o t e S e n s i n g Data : Landsa t -7 E n h a n c e d Themat i c M a p p e r digital da ta acqu i red on Augus t ,2 , 1999 (path 50, row 23) and S e p t e m b e r 12, 1999 (path 49 , row 23) P la teau Fores t P roduc ts Ltd., Vanderhoo f , B C M P B Infestation m a p T R I M data F I R M S L a b / M o F Fores t C o v e r M a p B C M o F 22 Detect ion of mountain pine beet le infestat ions us ing L a n d s a t T M T a s s e l e d C a p Trans fo rmat ions 4.3 Mountain Pine Beetle Infestations in the Study Area T h e number of infested t rees at a site has a great bear ing on its identif ication us ing satel l i te data , generat ion of training s i tes, and a c c u r a c y a s s e s s m e n t . There fo re , this information w a s co l lec ted from a beet le infestation m a p of the study a r e a , p repared by P la teau Fores t P roduc ts Ltd., Vanderhoo f , B C . T h e map ind icates s i tes of both red a t tacked t rees (beet le at tack of 1998) and the current attack of 1999. Information on red a t tacked t rees w a s co l lec ted f rom interpretation of normal co lor aer ia l photographs, by a G P S survey carr ied out in the s e c o n d quarter of A u g u s t 1999, and by field verif ication f rom m i d - S e p t e m b e r to the end of Oc tober , 1999. Information on g reen attack t rees w a s co l lec ted b a s e d on a wa lk th rough (m id -Sep tember to end of October ) and a beet le probe (using a 100m grid) in se lec ted a r e a s a long the Van t ine and Ma lapu t roads (mid-S e p t e m b e r to end of N o v e m b e r 1999). T h e beet le infestation map w a s in Microstat ion file format with no d a t a b a s e a t tached to it. There fore , a 10 percent s a m p l e (530 g round observa t ion points) w a s randomly se lec ted f rom the map and a da tabase showing number of red, and red p lus g reen attack t rees for e a c h site w a s genera ted (Table 11). T h e s i ze of the beet le infestat ion w a s very smal l and w a s scat tered in nature. Approx imate ly 87 percent of the infested s i tes had less than 10 red a t tacked t rees. S i n c e g reen at tacked (GA) t rees are a l so p resent a long with red a t tacked (RA) t rees at most of the infestation s i tes, at a 30-meter spat ia l resolut ion both R A and G A t rees wou ld be recorded by the satell i te b a s e d sensor . There fore , both red a t tacked and g reen at tacked t rees together we re ca l led "a t tacked trees". S i n c e the s i ze of a rea cove red by a t tacked t rees is still subp ixe l in s i ze , p ixels represent ing these s i tes we re ca l led probab le s i tes of M P B attack. Tab le 11: S i z e character is t ics of beet le infestat ions in the total study a rea . ( B a s e d on a 10 percent random s a m p l e f rom beet le infestat ion map, p repared by P la teau Fores t Produc ts , Ltd.) N u m b e r of M P B R e d at tack* A t tacked t rees ( R e d + Cur rent infested t rees per at tack)** site No . of s i tes % No. of s i tes % <10 461 86 .98 393 75 .15 11-20 57 10.75 77 14.53 21-30 8 1.51 27 5.10 31-50 3 0.57 17 3.22 >50 1 - 16 3.02 Total 530 100.00 530 100.00 2 3 Detect ion of mounta in p ine beet le infestat ions us ing Landsa t T M T a s s e l e d C a p T rans fo rmat ions It w a s p roposed that three test s i tes wou ld be identif ied within the s tudy a r e a . M P B infestat ion s i tes in one test a rea wou ld be used to cal ibrate the T a s s e l e d C a p vegeta t ion ind ices and the other two a reas wou ld be used for a c c u r a c y a s s e s s m e n t (F igure 3). Tes t a r e a s , A and B we re comple te ly cove red by the beet le infestat ion m a p provided by P la teau Fores t P r o d u c t s Ltd., but Tes t A r e a C w a s only partly c o v e r e d . A da tabase on M P B s i tes w a s p repared for all the three test a reas b a s e d on comp le te enumera t ion . F igure 3. A Landsa t -7 E T M (Sep tembe r 12, 1999) pseudoco lo r compos i t e of s tudy a rea . Loca t ion of Tes t A r e a s is m a r k e d in red boxes . 24 Detect ion of mountain p ine beet le infestat ions us ing Landsa t T M T a s s e l e d C a p Trans fo rmat ions 4.4 Methodology T h e ana lys i s w a s carr ied out us ing a combina t ion of PC I E A S I P A C E image p rocess ing sof tware and E S R I A R C A / I E W geograph ic information s y s t e m sof tware. T h e major s teps in the ana lys i s inc luded: • P re -p rocess ing of satel l i te da ta • Computa t ion of T a s s e l e d C a p coeff ic ients for Landsa t -7 • Identification of M P B a t tacked s tands • A c c u r a c y a s s e s s m e n t of a t tacked tree identif ication 4.4.1 Pre-Processing of Satellite Data Geo-Registration of Data Set S u b - s c e n e s of the study a rea , ext racted f rom Landsa t da ta of both acquis i t ion da tes , the forest cove r map, beet le infestation cove rage , and T R I M m a p s of the study a r e a we re co- reg is te red , and t ransformed to N A D 8 3 da tum and Un iversa l T r a n s v e r s e Merca to r ( U T M ) project ion, us ing a nearest ne ighbor resampl ing algor i thm (root m e a n squa re error <0.5 pixel). T R I M m a p s we re used a s re fe rences for registrat ion of other da ta se ts . T h e G C P W O R K S modu le of E A S I / P A C E w a s used to perform the geo-regis t rat ion. Atmospheric Correction In order to get truly or nearly representat ive ref lectance f rom sate l l i te -based mult i -spectral da ta for var ious cove r feature types, an a tmospher i c correct ion is necessa ry . T h e re f lec tance of a ground feature recorded by E T M onboa rd the Landsa t -7 satell i te is contro l led by a combina t ion of: i) so lar i r radiance; ii) s e n s o r parameters (gain and offset); iii) re f lec tance f rom sur face features imaged ; iv) i l lumination and v iewing geomet ry of the s c e n e ; v) absorpt ion and scat ter ing by the a tmosphere plus external a tmospher i c contr ibut ions to the incoming spect ra l re f lec tance f rom an object ( J e n s e n 1996). T h o u g h many more compl ica ted and a d v a n c e d mode l s a re ava i lab le for a tmospher ic correct ion (Hal l etal., 1991a ; Mo ran etal., 1992; R ichter 1997), this study w a s carr ied out us ing the dark object subtract ion techn ique deve loped by C h a v e z (1988). Th i s techn ique is b a s e d on the p rem ises that: i) a tmospher i c effects add a uni form offset to an image; and ii) all or nearly all near infra-red energy incident on c lear d e e p water is a b s o r b e d . T h o u g h 25 Detect ion of mounta in pine beet le infestat ions us ing Landsa t T M T a s s e l e d C a p Trans fo rmat ions this techn ique is one of the o lder o n e s , it is still widely used for a tmosphe r i c correct ion b e c a u s e of its simplicity and its requirement for little information beyond the image itself (Pr ice ef al., 1997). T h e study a rea has a large number of both sma l l and large water bod ies . T e n training s i tes for c lear water, in the d e e p port ions of the large water bod ies , we re se lec ted us ing band 4 (near infrared spect ra l region, where ideally ref lectance shou ld be ze ro or near zero) for e a c h of the two Landsa t images . T h e range of digital numbers (DN) in e a c h band , for e a c h of t hese 10 s i tes, were recorded and used for a tmospher i c correct ion. 4.4.2 Computation of Tasseled Cap Coefficients T h e number of d imens ions (n) ava i lab le in spect ra l s p a c e is the n u m b e r of spect ra l b a n d s avai lab le from a senso r . In this c a s e , there are s ix poss ib le d imens ions f rom six spect ra l bands (bands 1-5 and 7) f rom Landsa t -7 E T M . T h e number of spect ra l ind ices (m) that may be ca lcu la ted is a l so equa l to the number of bands (n). In this c a s e there are s ix poss ib le ind ices namely: br ightness, g r e e n n e s s , w e t n e s s , haze , fifth, and sixth. Of ten , only the first three ind ices are of interest. In this study, coeff ic ients we re compu ted for t hese three ind ices only. T h e procedure cons is ted of identifying training s i tes for var ious c o v e r types, computa t ion of T a s s e l e d C a p coeff ic ients, and finally evaluat ion of br ightness, g r e e n n e s s and w e t n e s s der ived from the compu ted coeff ic ients. Fo r deve lopmen t of the three ind ices, four da ta points we re required for dry soi l , wet soi l , healthy vegetat ion, and s e n e s c e n t vegeta t ion. Dry soi l and wet soi l are required for deve lopment of the br ightness index. Dry soi l , wet so i l , and healthy vegetat ion are required for the g r e e n n e s s index. Al l the points are required for the w e t n e s s index. Identifying training s i tes for healthy and s e n e s c e n t vegetat ion and dry and moist soi l w a s d o n e us ing the Imageworks modu le of the E A S I / P A C E image p rocess ing sof tware. No rma l co lor aer ia l photographs of the study a rea were used to se lec t training s i tes. Howeve r , us ing s ing le date da ta it w a s not poss ib le to get the s a m e vegetat ion at the s e n e s c e n t s tage. B e s i d e s , it is a l so difficult to get s e n e s c e n t vegetat ion in a predominant ly forested l andscape . There fo re , s p a r s e sc rub vegetat ion w a s taken a s s e n e s c e n t vegetat ion. C lus te rs of p ixe ls be long ing to dry so i l , wet soi l , vegetat ion and s e n e s c e n t vegetat ion were se lec ted f rom the A u g u s t 2, 1999 Landsa t E T M + data. A v e r a g e d digital numbers (DN) obta ined for t hese s i tes we re u s e d to c o m p u t e the coef f ic ients of the different ind ices using the procedure g iven by J a c k s o n (1983). 26 Detect ion of mounta in pine beet le infestat ions us ing Landsa t T M T a s s e l e d C a p Trans fo rmat ions Br ightness, g r e e n n e s s and w e t n e s s images genera ted us ing these coeff ic ients w e r e eva lua ted by compar ing the relative distribution of va r ious c o v e r types in the br ightness - g r e e n n e s s , g r e e n n e s s - w e t n e s s and w e t n e s s - br ightness feature s p a c e aga ins t the distr ibution pattern obse rved for L a n d s a t - 5 T M (Figure 1). 4.4.3 Effect of topography, infestation size and acquisition dates In order to eva luate the effect of topography, an a s p e c t m a p w a s prepared us ing T R I M 20-meter contour m a p s in A R C / V I E W . T a s s e l e d C a p ind ices for a total of 90 observat ion points (Op) , w e r e randomly se lec ted from mature lodgepo le pine s tands on flat terrain (Op=38), nor thwest a s p e c t s (Op=22), and southeas t a s p e c t s (Op=30). T h e ind ices for both da tes we re ext racted f rom the respect ive t ransformed images . T h e effect of infestat ion s i z e on br ightness, g r e e n n e s s and w e t n e s s ind ices w a s eva luated by extract ing t hese ind ices for randomly se lec ted infestat ion s i tes (Table 11), vary ing in number of a t tacked t rees at e a c h site. T o study the effect of acquis i t ion da tes on br ightness, g r e e n n e s s and w e t n e s s , ind ices va l ues for both da tes (August 2, 1999 and S e p t e m b e r 12, 1999) were extracted for randomly se lec ted observa t ion points for young lodgepole pine (<20 yea rs in age , Op=37) ; mature lodgepo le pine s tands (>60 y e a r s of age) ; sh rubs (Op=27); gravel roads (Op=46); and landing s i tes within cu tb locks (Op=27). S i tes identif ied f rom the beet le infestation c o v e r a g e we re u s e d to ass i gn br ightness, g r e e n n e s s and w e t n e s s indexes to at tacked s tands for both the da tes of images . B e c a u s e of the sub-p ixe l s i z e of infestat ions and their scat tered distr ibut ion, it w a s not poss ib le to use supe rv i sed c lass i f icat ion. K - M e a n cluster ing techn iques w e r e used in this study ins tead. Th i s iterative p rocess , some t imes cal led "c luster bust ing" ( J e n s e n 1996; P r i ce e r a / . , 1997) w a s repeated to separa te probable M P B infestat ions f rom other forest and nonforest c l a s s e s . Con fus ion matr ices, a lso referred to a s con t ingency tab les or c lass i f icat ion error mat r ices prov ide a m e a n s of evaluat ing the themat ic a c c u r a c y of a c lass i f ied image by compar ing the c l a s s 4.4.4 Identification of Attacked Stands 4.4.5 Accuracy Assessment 27 Detect ion of mounta in pine beet le infestat ions us ing L a n d s a t T M T a s s e l e d C a p Trans fo rmat ions a s s i g n e d to a group of test p ixels to the actual g round information at those s i tes (Conga l ton 1991 ; L i l lesand and Kieffer 2000) , and were u s e d in this study for a c c u r a c y a s s e s s m e n t . T h e s e mat r ices a lso help by identifying wh ich ca tegor ies a re be ing con fused , either by e r roneous ly being exc luded f rom one c l a s s (error of omiss ion) or inc luded in ano ther (error of commiss ion ) . In a con fus ion matrix, d iagona l e lements represent the identif ication accu racy , co lumn totals represent errors of omiss ion and row totals indicate error of c o m m i s s i o n s . Initially, it w a s p roposed to a s s e s s identif ication a c c u r a c y in two test a reas . However , comp le te enumera t ion of the beet le infestation map cover ing Tes t A r e a s A a n d B, revea led that there we re only 15 and 32 s i tes, respect ive ly , wh ich had more than 30 a t tacked t rees. T h i s number w a s thought to be inadequate for accu racy a s s e s s m e n t . Instead a c c u r a c y a s s e s s m e n t w a s per formed on the total s tudy a rea . A total of 568 observat ion points (340 for a t tacked s i tes, 228 for healthy forest) we re used for a c c u r a c y a s s e s s m e n t . A confus ion matrix w a s genera ted us ing the beet le infestat ion m a p s a s re ference data and probable M P B infested a r e a s a s identif ied f rom satell i te da ta a s image data . A m o n g the var iab les p roduced f rom the con fus ion matrix were identif ication a c c u r a c y of M P B detect ion, producer 's accu racy or error of o m i s s i o n , user a c c u r a c y or error of c o m m i s s i o n , and overal l accu racy . Es t imates of identif ication a c c u r a c y we re a l so m a d e for s i tes c lass i f ied b a s e d on the number of infested t rees. 28 Detect ion of mountain p ine beet le infestat ions us ing L a n d s a t T M T a s s e l e d C a p Trans fo rmat ions 5.0 Results and Discussion 5.1 Tasseled Cap Transformations Derivation of Coefficients Original ly T a s s e l e d C a p coeff ic ients we re deve loped us ing spect ra l s ignatures f rom agricul tural c rops . In the present c a s e , forest cove r w a s c h o s e n a s s i tes for healthy vegeta t ion, first, b e c a u s e the present study primarily w a s b a s e d on spect ra l character is t ics of forest cover , and s e c o n d , b e c a u s e there were no agricultural c rops in this a rea . In agricul tural a r e a s , s e n e s c e n t vegetat ion is represented by c rops at maturity, w h e n the c rops a re dry and charac te r i zed by a ye l low/go lden ye l low color. F inding simi lar cove r types in forested a r e a s is not poss ib le . There fo re , sc rub vegetat ion w a s se lec ted a s si tes for s e n e s c e n t vegeta t ion. Tra in ing s i tes for wet soi l w e r e se lec ted f rom bogs . S igna tu res f rom land ings within cu tb locks w e r e u s e d for dry so i l . A l l the se lec ted s i tes were c h e c k e d for purity of spect ra l s ignatures and then the m e a n s ignature va l ues for e a c h site w a s der ived. T h e s e m e a n s ignature va l ues w e r e used in computa t ion of coef f ic ients fol lowing the procedure desc r ibed by J a c k s o n (1983). T h e coeff ic ients der ived were : Br ightness: (0 .1099*B1) + (0 .1557*B2) + (0 .3023*B3) + (0 .2931*B4) + (0 .7420*B5) +( 0 .4855*B7) G r e e n n e s s : ( -0.0865*B1) + ( - 0 .0585*B2 ) + ( -0 .2291*B3) + (0 .9375*B4) + (-0.1135*B5) + ( -0 .2115*B7) W e t n e s s : (-0.207CTB1) + ( -0.5812*B2) + ( -0 .6092*B3) + ( -0 .1515*B4) + (0 .4741*B5) + ( -0 .0205*B7) Evaluation of Coefficients T h e reliability of these coeff ic ients w a s tested by plotting the va l ues for different c o v e r types in g reenness -b r igh tness (Figure 4a) , we tness -b r igh tness (F igure 4b) and g r e e n n e s s - w e t n e s s (Figure 4c) two d imens iona l feature s p a c e . F r o m these f igures, the fol lowing observa t ions c a n be made . Ba re soi l has the highest br ightness va lue , g r e e n n e s s va l ues for water and so i ls are lowest, and sh rubs have high g r e e n n e s s va lues . M a n - m a d e features (gravel roads) s h o w e d the lowest w e t n e s s va lue . T h e s e observa t ions c o m p a r e favorably with those of Cr is t and C i c o n e (1986) for var ious cove r types us ing Landsa t -5 da ta (F igure 1). 29 Detect ion of mounta in p ine beet le infestat ions us ing Landsa t T M T a s s e l e d C a p Trans fo rmat ions in 2 100 • • • 100 150 • ?nn I • S H R U B S I LP-YOUNG I LP-M ATURE | I SOIL I WATER I • ROADS B R I G H T N E S S (a) 70 60 50 Wl 40 • • • • | • SI-RUBS I LP-YOUNG I LP-MATURE | I SOIL I WATER I • ROADS 100 150 B R I G H T N E S S (b) u X 50 20 30 40 50 * 1 | • SHRUBS i LP-YOUNG • LP-MATURE j I SOIL • WATER •ROADS (c) F igure 4. Locat ion of different cove r types in br ightness - g r e e n n e s s (a), b r igh tness - w e t n e s s (b) and w e t n e s s - g r e e n n e s s (c) feature s p a c e der i ved f rom compu ted T a s s e l e d C a p coef f ic ients for Landsa t -7 Thema t i c M a p p e r Da ta . ( L P - Y O U N G : young lodgepo le pine s tands ; L P - M A T U R E : mature lodgepo le p ine s tands) 30 Detect ion of mounta in pine beet le infestat ions us ing Landsa t T M T a s s e l e d C a p Trans fo rmat ions Br igh tness and g r e e n n e s s for a mature lodgepo le pine s tand are lower than for young lodgepo le pine. Th i s is due to the inc reased s h a d o w effect and structure of l eaves . C o m p a r e d to young plantat ions, o lder forest s tands contain a higher pe rcen tage of s tems , wh ich resul ts in i nc reased inc idence of d e e p s h a d o w s both on the lower layers of the canopy and on the l eaves in the tree c rowns (Crist e r a / . , 1986). 5.2 Effect of Topography, Acquisition Dates and Infestation Size Comparison of Two Date Tasseled Cap Indices Br ightness, g r e e n n e s s and w e t n e s s ind ices va l ues for both da tes w e r e ext racted for randomly se lec ted observat ion points (Op) for young lodgepo le pine (<20 y e a r s in age , Op=37, F igure 5), mature lodgepo le pine s tands (>60 yea rs of age , Op=70, F igure 6), s h r u b s (Op=27, F igure 7), gravel roads (Op=46, F igure 8), and landing s i tes within cu tb locks (Op=27, F igure 9). T h e r e w a s a d i f ference of 42 d a y s between the two imaging da tes . Dur ing this per iod it w a s expec ted that no signif icant c h a n g e wou ld have taken p lace in many c o v e r c l a s s e s . However , what w a s surpr is ing w a s that even landing s i tes had consistent ly lower br ightness va lues in S e p t e m b e r c o m p a r e d to A u g u s t for all the 27 s i tes and the magni tude of this d i f ference w a s a lmos t constant . Th is happened most probably b e c a u s e of lower incident energy in S e p t e m b e r than Augus t . In the c a s e of young lodgepole pine plantat ions, both br ightness a s wel l a s g r e e n n e s s we re much lower in Sep tember . Th is might have been due to the effect of i nc reased s h a d o w s in addit ion to lower incident energy. Ano the r interesting observat ion is that w e t n e s s rema ined relatively unaf fected, w h e r e a s br ightness and g r e e n n e s s s h o w e d relatively large variabil ity. Th i s con fo rms to observa t ions reported by C o h e n and S p i e s (1992). They reported that br ightness and g r e e n n e s s images captured the majority of the spect ra l var iat ions assoc ia ted with forest condi t ions, but we re strongly inf luenced by topograph ic var iat ions result ing in a large d y n a m i c range. W e t n e s s , on the other hand , w a s nearly insensi t ive to topograph ic var ia t ions and h e n c e had a smal l dynam ic range. T h e consistent ly low m e a n va lues for all the three ind ices for va r ious cove r types for the S e p t e m b e r da ta may be b e c a u s e of the c o m b i n e d inf luence of low incident energy , lower sun e levat ion ang le , and the resultant longer s h a d o w s in S e p t e m b e r a s c o m p a r e d to Augus t . T h e s e observa t ions indicate that the A u g u s t da ta w a s better for d iscr iminat ing var ious cove r types than the S e p t e m b e r data . 31 Detect ion of mounta in pine beet le infestat ions us ing L a n d s a t T M T a s s e l e d C a p Trans fo rmat ions YOUNG LODGEPOLE PINE STANDS 140 120 100 80 60 40 20 0 — • — A - B R T —M—S-BRT OBSERVATION POINTS (a) YOUNG LODGEPOLE PINE STANDS 120 100 8 80 UJ I 60 LU £ 40 CD 20 —m S-GRN m c o c o i ^ - < - i r > O T c o i - ~ i - i - CM CM CN co co OBSERVATION POINTS (b) YOUNG LODGEPOLE PINE STANDS 70 5 20 10 0 r - m o c o r - T - m c n c o h -T - CM CM CM co co OBSERVATION POINTS (C) F igure 5. C o m p a r i s o n of two date Br igh tness (a), g r e e n n e s s (b) and w e t n e s s (c) ind ices for young lodgepo le pine s tands (A: Augus t 2 1999, S : S e p t e m b e r 12, 1999, B R T : Br igh tness , G R N : G r e e n n e s s , W E T : We tness ) . 32 Detect ion of mounta in p ine beet le infestat ions using Landsa t T M T a s s e l e d C a p Trans fo rmat ions MATURE LODGEPOLE PINE OBSERVATION POINTS (a) MATURE LODGEPOLE PINE 00 CO LU o - A - G R N - S - G R N OBSERVATION POINTS (b) MATURE LODGEPOLE PINE 70 65 co o n co 60 LU 55 50 45 -A -WET -S-WET CD T-T- CM CD CM CD « -n n * t m OBSERVATION POINTS CD CD CD CD (C) Figure 6. C o m p a r i s o n of two-date br ightness (a), g r e e n n e s s (b) and w e t n e s s (c) ind ices for mature lodgepole pine s tands (A: Augus t 2 1999, S : S e p t e m b e r 12, 1999, B R T : Br igh tness , GRIM: G r e e n n e s s , W E T : W e t n e s s ) . 33 Detect ion of mounta in p ine beet le infestat ions us ing Landsa t T M T a s s e l e d C a p T rans fo rmat ions SHRUBS Figure 7. C o m p a r i s o n of two date br ightness (a), g r e e n n e s s (b) and w e t n e s s (c) ind ices for sh rubs (A: Augus t 2 1999, S : S e p t e m b e r 12, 1999, B R T : Br igh tness , G R N : G r e e n n e s s , W E T : W e t n e s s ) . 34 Detect ion of mountain p ine beet le infestat ions us ing Landsa t T M T a s s e l e d C a p T rans fo rmat ions ROADS Figure 8. C o m p a r i s o n of two date br ightness (a), g r e e n n e s s (b) and w e t n e s s (c) i nd ices for roads (A: A u g u s t 2 1999, S : S e p t e m b e r 12, 1999, B R T : Br ightness, G R N : G r e e n n e s s , W E T : W e t n e s s ) . 35 Detect ion of mounta in p ine beet le infestat ions using Landsa t T M T a s s e l e d C a p T rans fo rmat ions LANDINGS oo oo CD 20 10 0 -10 -20 o oo" - •co _ /SV/<NW^O W-JHi<-|l CM CM • . A -GRN - S - G R N OBSERVATION POINTS (b) LANDINGS 80 cn 60 CO I 40 | 20 0 » A-WET —m S-WET t^- O OO CD OBSERVATION POINTS O) CN T - CM CM (C) Figure 9. C o m p a r i s o n of two date br ightness (a), g r e e n n e s s (b) and w e t n e s s (c) ind ices for landings (A: Augus t 2 1999, S : S e p t e m b e r 12, 1999, B R T : Br igh tness , G R N : G r e e n n e s s , W E T : W e t n e s s ) . 36 Detect ion of mounta in p ine beet le infestat ions us ing Landsa t T M T a s s e l e d C a p Trans fo rmat ions Effect Of Topography on Tasseled Cap Indices Topog raphy a l so p lays an important role in inf luencing spec t ra l s igna tu res f rom land su r faces . In order to eva luate the effect of topography, an aspec t map w a s p repared us ing T R I M 20-mete r contour m a p s in A R C / V I E W . T a s s e l e d C a p ind ices for 90 observa t ion points w e r e se lec ted f rom lodgepo le pine s tands on flat terrain (Op=38), northwest a s p e c t s (Op=22), and sou theas t a s p e c t s (Op=30). T h e va lues for T a s s e l e d C a p ind ices for both da tes w e r e ex t rac ted f rom the respec t ive index images . T h e range of T a s s e l e d C a p ind ices obta ined ind icated that b r igh tness and g r e e n n e s s of lodgepole pine s tands on southeas t s l opes w a s near ly twice t hose on nor thwest aspec ts , whi le those f rom s tands on flat terrain fell in be tween (F igure 10). A s reported earl ier, we tness w a s not found to be sens i t ive to var iat ions in topography. It w a s near ly the s a m e for all lodgepole pine s tands i r respect ive of a s p e c t c l a s s . 100 A-BRT S-BRT A-GRN S-GRN A-WET S-WET Figure 10. Br ightness, g r e e n n e s s and w e t n e s s ind ices for mature lodgepo le pine s tands on different aspec t c l a s s e s ( S E : Southeas t , F: Flat, N W : Nor thwest aspec t ; A : Augus t , S : Sep tembe r , B R T : Br ightness, G R N : G r e e n n e s s , W E T : W e t n e s s ) . Effect of Infestation Size on Tasseled Cap Indices Figure 11 s h o w s the relat ionship be tween br ightness, g r e e n n e s s and w e t n e s s and number of a t tacked t rees per si te for the A u g u s t and S e p t e m b e r images . F o r s i tes conta in ing l e s s than 30 at tacked t rees, there w a s a large d ispers ion in va lues . After this th resho ld , all three ind ices var ied within a relatively narrow range. 37 Detect ion of mounta in p ine beet le infestat ions using Landsa t T M T a s s e l e d C a p T rans fo rmat ions F igure 11. Re la t ionsh ip be tween number of M P B infested t rees and T a s s e l e d C a p ind ices There are many r easons that may accoun t for such a pattern. In vegeta ted a r e a s , pixel level ref lectance is in f luenced by the relat ive proport ions of vegetat ion types, e x p o s e d soi l if any, the ref lectance interaction be tween soi l and vegetat ion, and s h a d o w s , all modi f ied by a tmosphe re (R ichardson and W e i g a n d 1990). O n e of the major p rob lems in extract ing vegeta t ion information from satell i te b a s e d s e n s o r s is that the spat ia l resolut ion of s e n s o r s is genera l ly larger than the 38 Detect ion of mountain p ine beet le infestat ions us ing Landsa t T M T a s s e l e d C a p T rans fo rmat ions vegetat ion objects, a s is the c a s e in this study a rea . Apparen t ly , less than 30 infested lodgepo le pine t rees per pixel are not large e n o u g h to in f luence ref lectance. Th is observat ion w a s signi f icant f rom the point of v iew of the detect ion of M P B infestat ions in the study a rea by us ing a pixel level a lgor i thm. A s s h o w n in Tab le 1 1 , 9 5 percent of all the infestation s i tes had l ess than 30 a t tacked t rees/s i te and hence , had a very low c h a n c e s of gett ing de tec ted b a s e d on pixel level a lgor i thms. Th i s indicated that use of subp ixe l level a lgor i thms, s u c h a s spect ra l unmix ing techn iques , wou ld be more appropr ia te than pixel level a lgor i thms (Murtha ef al., 2000) . 5.3 Identifying Attacked Stands In order to test the separabi l i ty of a t tacked s tands f rom healthy lodgepole pine s tands b a s e d on the T a s s e l e d C a p ind ices, a t-test w a s per formed on a se t of randomly p icked 22 observa t ion s i tes wh ich had >30 at tacked t rees /si te a n d 22 observa t ion points for non-a t tacked mature lodgepole pine s tands . T h e resul ts we re eva lua ted at 40 d e g r e e s of f reedom. T h e d i f fe rences in the br ightness, g r e e n n e s s and w e t n e s s v a l u e s of healthy and at tacked s tands w e r e s igni f icant at the .05 s ign i f icance level (Table 12). Howeve r , t hese observa t ions are b a s e d on a very sma l l samp le s i ze . Tab le12 : A v e r a g e Di f ference be tween Heal thy a n d A t tacked M P B S tands (August 2 , 1999) Br igh tness G r e e n n e s s W e t n e s s Healthy A t tacked Heal thy A t tacked Heal thy A t tacked M e a n 57.41 62.71 32 .53 29.91 57.34 59.71 S D 6.83 8.42 3.91 3.70 1.51 1.57 t-value 2 .2929 2 .2829 5 .1032 t 0.025, 40 - 2.021 Classification Based on Tasseled Cap Indices C h a n g e detect ion techn iques we re in tended to be used for identifying a t tacked s t a n d s f rom healthy ones . Th is w a s due to the expec ted c h a n g e in T a s s e l e d C a p ind ices f rom A u g u s t to S e p t e m b e r b e c a u s e of further dis integrat ion of ch lorophyl l , need le structure, and mois ture s t ress c a u s e d by disrupt ion of water and nutrient up take due to growth of blue stain fungus (Ceratocystis montia, or Ophiostoma setosum *) in the xy lem and ph loem t issue. *Wh ich of t hese two fungi mainly grow in the lodgepo le pine after the success fu l M P B attack, is still an a rea of act ive research (Breui l 2 0 0 0 , pe rsona l communica t ion) . 39 Detect ion of mounta in p ine beet le infestat ions us ing Landsa t T M T a s s e l e d C a p Trans fo rmat ions However , d i f ferences we re obse rved not only in a t tacked s tands , but in all other c o v e r c l a s s e s a s wel l . S u c h uniform c h a n g e s a c r o s s all the cove r types pe rhaps c a n be exp la ined by the lower sun ang le and the lower incident radiation in S e p t e m b e r c o m p a r e d to Augus t . The re wou ld a l so be more s h a d o w s within forest s tands due to the lower sun e levat ion ang le , wh i ch might m a s k the spect ra l c h a n g e s c a u s e d due to deter iorat ion in lodgepo le pine s tand heal th. E v e n in the event of no c h a n g e , the healthy s tands wou ld have lower re f lec tance in S e p t e m b e r than A u g u s t and be con fused with a t tacked s tands . Th is p rob lem is further comp l i ca ted b e c a u s e of the extremely smal l s i ze of the infestat ions. Pr ior to any c h a n g e detect ion, it is imperat ive that images be regis tered with e a c h other. Wh i le mult i-date image registration w a s within accep tab le a c c u r a c y of 0.5 p ixe ls (15 meters) s u c h a shift a c r o s s two date images with smal l infestation s i z e s cou ld itself c rea te a large d e g r e e of per-pixel spect ra l ambiguity. Moreover , it is sugges ted that the mult i - temporal da ta for c h a n g e detect ion shou ld have been acqu i red during nearly s a m e per iod of t ime s o a s to avo id the inf luence of ex t raneous factors on the re f lec tance of features / c o v e r t ypes of interest (Copp in and B a u e r 1996). There fore , c h a n g e detect ion ana lys i s w a s not pu rsued further. S i n c e d i f ferences be tween the br ightness, g r e e n n e s s , and w e t n e s s of healthy and a t tacked lodgepole pine s tands we re statistically signif icant, an unsuperv i sed c lass i f icat ion w a s per formed on all the three tasse led c a p ind ices, for both da tes separate ly . Mult ip le i terations w e r e run: a) to separa te vegetat ion f rom non-vegetat ion features; b) sepa ra te lodgepo le pine s tands f rom other vegetat ion; and c) c lass i fy lodgepole pine s tands into different spect ra l s u b c l a s s e s . C lus te rs belonging to a t tacked s tands we re identified b a s e d on information on the locat ion of a t tacked s tands identified from the beet le infestat ions c o v e r a g e map . 5.4 Accuracy Assessment A s s e s s m e n t of identif ication accu racy for a t tacked s tands we re m a d e for both the da tes , us ing the procedure g iven by L i l lesand and Kieffer (2000). P r o d u c e r s a c c u r a c y inc ludes the ef fects of error of om iss ion , w h e r e a s , use rs accu racy takes into accoun t the error of c o m m i s s i o n . Overa l l identif ication a c c u r a c y of a t tacked s tands w a s found to be only 3 8 . 8 2 % and 2 6 . 1 7 % (Tab les 13 and Tab le 15) for the A u g u s t and S e p t e m b e r images , respect ive ly . A l inear relat ionship w a s obse rved be tween the number of a t tacked t rees and identif ication a c c u r a c y for A u g u s t (Table 14). However , the s a m e relat ionship w a s not o b s e r v e d for the S e p t e m b e r image (Table 16). T h e low identif ication accu racy obse rved for the majority of the re ference g round points largely w a s d u e to 40 Detect ion of mounta in pine beet le infestat ions us ing L a n d s a t T M T a s s e l e d C a p Trans fo rmat ions Tab le 13. A c c u r a c y es t imates b a s e d on a 1 0 % random s a m p l e (August 2, 1999) R e f e r e n c e da ta * A t tacked Heal thy Total Satel l i te A t tacked 132 60 192 Healthy 208 168 376 Total 340 228 568 * G r o u n d data prov ided by P la teau Fores t Ltd., V a n d e r h o o f Producer's Accuracy: User's Accuracy At tacked = 132 /340 = 3 8 . 8 2 % A t tacked = 132 /192 = 6 8 . 7 5 % Heal thy = 168/228 = 7 3 . 6 8 % Heal thy = 168/376 = 4 4 . 6 8 % Overa l l a c c u r a c y = (132+168)/568 = 5 2 . 8 2 % Tab le 14. A c c u r a c y (%) b a s e d on the number of a t tacked t rees/s i te (August 2, 1999) R e f e r e n c e da ta * N u m b e r of a t tacked t rees per site Satel l i te <10 11-20 21 -30 31-80 >80 T O T A L 31.34 34 .55 53 .85 53 .85 72 38.82 (63/201) (22/64) (7/13) (14/26) (26/36) (132/340) * G r o u n d data provided by P la teau Fores t Ltd., V a n d e r h o o f Tab le 15. A c c u r a c y es t imates b a s e d on a 1 0 % random s a m p l e (Sep tembe r 12, 1999) R e f e r e n c e da ta * Satel l i te A t tacked Heal thy Total A t tacked 89 103 192 Heal thy 251 125 376 Total 340 228 568 * G r o u n d data prov ided by P la teau Fores t Ltd., V a n d e r h o o f Producer's Accuracy: User's Accuracy At tacked = 89 /340 = 2 6 . 1 7 % A t tacked = 8 9 / 1 9 2 = 4 6 . 3 5 % Heal thy = 125/228 = 5 4 . 8 2 % Heal thy = 125 /376 = 3 3 . 2 4 % Overa l l accu racy = (89+125)/568 = 3 7 . 6 7 % Tab le 16. A c c u r a c y (%) b a s e d on the number of a t tacked t rees/s i te (Sep tember 12, 1999) Satel l i te R e f e r e n c e da ta * N u m b e r of a t tacked t rees per site <10 11-20 21 -30 31-80 >80 T O T A L 21.89 31.25 38 .46 30 .76 33 .33 26 .17 (44/201) (20/64) 5/13 (8/26) (12/36) 89 /340 * G r o u n d data prov ided by P la teau Forest Ltd., V a n d e r h o o f 41 Detect ion of mounta in p ine beet le infestat ions us ing Landsa t T M T a s s e l e d C a p Trans fo rmat ions the ex t remely smal l and sca t te red nature of the M P B infestat ions in the s tudy a r e a . B e c a u s e of this, p ixel b a s e d a c c u r a c y a s s e s s m e n t p rocedures were not adequa te . Instead, a larger unit, s u c h a s s tand , shou ld be taken a s a base for a c c u r a c y a s s e s s m e n t . Th is is suppor ted by the o b s e r v e d co r respondence of the distr ibution of a t tacked s i tes (identif ied f rom beet le infestat ion coverage) and probab le a t tacked a reas identif ied from the three T a s s e l e d C a p ind i ces (F igure12 -14). TEST AREA A PROBABLE MOUNTAIN PINE BEETLE AFFECTED LODGEPOLE PINE STAND IDENTIFIED BASED ON TASSELED CAP TRANSFORMATIONS 9 PROBABLE MPB ATTACKED STANDS | WATER/CLOUD SHADOW + GROUND BASED MPB ATTACK SITES -ROAD DATA S O U R C E : LANDSAT-7 ETM DIGITAL DATA ACQUIRED ON S E P T E M B E R 12.1999 Figure 12. P robab le M P B infestat ion m a p der ived from T a s s e l e d C a p ind ices - Tes t a rea A . 42 Detect ion of mounta in p ine beet le infestat ions us ing Landsa t T M T a s s e l e d C a p T rans fo rmat ions TEST AREA B PROBABLE MOUNTAIN PINE BEETLE AFFECTED LODGEPOLE PINE STAND IDENTIFIED BASED ON TASSELED CAP TRANSFORMATIONS DATA SOURCE . LANDSAT-7 ETM DIGITAL DATA ACQUIRED ON SEPTEMBER 12,1989 Figure 13. P r o b a b l e M P B infestat ion m a p der ived from T a s s e l e d C a p ind ices - Tes t a r e a B. TE8T AREA C PROBABLE MOUNTAIN PINE BEETLE AFFECTED LODGEPOLE PINE STAND IDENTIFIED BASED ON TASSELED CAP TRANSFORMATIONS S PROBABLE MPB ATTACKED STANDS | WATER/CLOUD SHADOW - ROAD + GROUND BASED MPB ATTACK SITES DATA SOURCE: LANDSAT-7 ETM DIGITAL DATA ACQUIRED ON SEPTEMBER 12,1999 F igu re 14. P r o b a b l e M P B infestat ion m a p der ived from T a s s e l e d C a p ind ices - Tes t a rea C . 43 Detect ion of mounta in p ine beet le infestat ions us ing Landsa t T M T a s s e l e d C a p Trans fo rmat ions 6.0 Conclusions T h e fol lowing conc lus ions c a n be drawn f rom this s tudy: • T h e location of different cove r types in b r i gh tness -g reenness , g r e e n n e s s - w e t n e s s , and wetness-b r igh tness feature s p a c e b a s e d on Landsa t -7 T a s s e l e d C a p coef f ic ients c o m p a r e d wel l with the respect ive pattern o b s e r v e d for s u c h c o v e r types for Landat -5 . T h i s s u g g e s t s that compu ted T a s s e l e d C a p coef f ic ients cou ld be u s e d for generat ion of br ightness, g r e e n n e s s and w e t n e s s ind ices for the in tended appl icat ions. • T h e va lues of T a s s e l e d C a p ind ices for S e p t e m b e r we re found to be lower than those for A u g u s t for all the cove r types. T h e main reason for this w a s thought to be the lower incident so la r energy in Sep tember . • T a s s e l e d C a p ind ices for infestat ions of more than about 30 at tacked t rees per site we re found to vary in a relatively narrow range. Howeve r , for infestation s i tes with l ess than 30 af fected t rees per site the T a s s e l e d C a p ind ices had random and large d ispers ions . • T a s s e l e d C a p ind ices for mature lodgepo le pine s tands we re strongly in f luenced by topograph ic var iat ions. • D i f ferences between m e a n br ightness, g r e e n n e s s , and w e t n e s s of M P B at tacked s tands and healthy lodgepole pine s tands we re statist ical ly s igni f icant for Augus t . • It w a s poss ib le to separa te p robab le suscep t ib le lodgepo le pine s tands us ing c luster ing techn iques operated on a combina t ion of all three T a s s e l e d C a p ind ices. • A c c u r a c y a s s e s s m e n t s m a d e at the pixel level g a v e unsat isfactory resul ts. T h e identif ication accu racy for at tacked lodgepo le pine s tands w a s 38 .82 and 26 .17 percent for A u g u s t and Sep tember , respect ively. • A l inear relat ionship w a s o b s e r v e d be tween the number of a t tacked t rees at a site and identif ication accu racy for the A u g u s t da ta but not for the S e p t e m b e r da ta . • T h e poor identif ication a c c u r a c y w a s mainly due to the sub-p ixe l s i ze of infestat ions. There fore , pixel b a s e d c o m p a r i s o n for a c c u r a c y a s s e s s m e n t d o e s not s e e m to be an adequa te procedure for a c c u r a c y a s s e s s m e n t . • T h e distribution pattern of a t tacked s tands , a s identif ied f rom ground b a s e d beet le p robes , c o m p a r e s wel l with the spat ia l pattern of suscep t i b le lodgepole pine s tands identif ied b a s e d on T a s s e l e d C a p indices. There fore , a c c u r a c y a s s e s s m e n t at the s tand level ins tead at pixel level , appea rs to be the appropr ia te level in the c a s e of smal l infestation s i z e s . 44 Detect ion of mounta in pine beet le infestat ions using Landsa t T M T a s s e l e d C a p Trans fo rmat ions 7.0 Further Research Areas In order to eva luate the eff ic iency of t asse led c a p t ransformat ions in identif ication of M P B infestat ions, s imi lar s tud ies may be conduc ted in test a r e a s w h e r e the M P B infestat ions are at least of a Landsa t T M pixel s i z e (i.e. 30 x 30 meters) in spat ia l extent. O n e s u c h a r e a cou ld be near T w e e d s m u i r Nat ional Pa rk whe re la rge-sca le inc idence of M P B infestat ions h a s been reported. Topograph ic var iat ions strongly inf luence the spect ra l re f lec tance f rom the s a m e object. Therefore, an evaluat ion of the ava i lab le p rocedures for normal iz ing the effect of t hese var iat ions on spect ra l ref lectance, shou ld a l so be done . Ano ther study cou ld be the appl icat ion of mult i -spectral (4-meter spat ia l resolut ion) da ta f rom the I K O N O S satell i te to identify M P B infestat ions. Digital da ta f rom s u c h high spat ia l resolut ions could be effect ive in detect ion of smal l -scat tered infestat ions, a s w a s the c a s e in the present study a rea . It wou ld be des i rab le to use a sub-p ixe l algori thm, s u c h a s spec t ra l unmix ing p rocedures , to detect and identify M P B infestat ions wh ich are sub-p ixe l in spat ia l extent. Howeve r , it wou ld be worthwhi le to eva luate the eff ic iency of spect ra l unmix ing p rocedu res in detect ing the M P B infestat ions of vary ing s i z e s , conta ined within a pixel . Ano ther important aspec t , wh ich n e e d s to be systemat ica l ly exp lo red , is the identif ication of the s tage w h e n the c h a n g e s in the spect ra l re f lectance behav io r of a success fu l l y a t tacked tree begin to take p lace. A n a l y s i s of mult i - temporal satell i te data of op t imum spat ia l resolut ion may provide s o m e a n s w e r s on this. 4 5 Detect ion of mounta in pine beet le infestat ions us ing Landsa t T M T a s s e l e d C a p T rans fo rmat ions References A h e r n , F . J , 1988. T h e ef fects of bark beet le s t ress on the foliar spect ra l re f lec tance of lodgepo le pine. International Journa l of R e m o t e S e n s i n g , 9 (9): 1451 -1468 . A h e r n , F . J . , and Arch iba ld , P .D . , 1986. Thema t i c M a p p e r information about C a n a d i a n Fo res ts : early results f rom a c r o s s the country, P roc . of the 10th C a n a d i a n S y m p o s i u m on R e m o t e S e n s i n g , pp: 683-697 . As ra r , G . , F u c h s , M. , K a n e m a s u , E.T., and Hatf ield, J . L . , 1984. Est imat ing a b s o r b e d photosynthet ic radiation and leaf a rea index f rom spect ra l re f lectance in wheat . A g r o n o m y J . , 76: 300-306 . Banner , A , V . , 1986. Detect ion of bark beet le - a t tacked sp ruce us ing compu te r b a s e d image ana lys is . M . S c . T h e s i s . Facul ty of Forestry , Universi ty of Brit ish C o l u m b i a , V a n c o u v e r , Bri t ish C o l u m b i a , 145pp. Baret, R., and Guyo t , G . , 1991 . Potent ia ls and limits of vegetat ion ind ices for LAI and A P A R a s s e s s m e n t . R e m o t e S e n s i n g of Env i ronment , 35 :161 -173 . Fores t Heal th S u r v e y s Gu idebook , 1995. B C Ministry of Fo res t s and B C Env i ronment . B . C . Ministry of Fores ts , A s o c i o - e c o n o m i c ana lys i s of mounta in p ine beet le m a n a g e m e n t in Brit ish Co lumb ia , Fores t P rac t i ces B ranch , h t tp : / /www.for .gov.bc.ca/hfp/pubs/ in terest /mpbecon/ index.htm, N o v e m b e r 28 , 2000 . B C Ministry of Fores ts , h t tp : / /www. fo r .gov .bc .ca /PAB/News /Fea tu res /bee t les / l akes .h tm, N o v e m b e r 28, 2000 . B C Ministry of Fores ts , ht tp: / /www.for.gov.bc.ca/kamloops/distr ic t / l i l looet/ fh2.htm, N o v e m b e r 28, 2 0 0 0 B C Ministry of Fores ts , 2000 . Jus t the facts: A rev iew of si lv iculture and other forestry stat ist ics. B r a g a , R E. G . , D a S i l va . , and M a r a , R. L. F., 1991. Moni tor ing the g reen a r e a s a round Tucuru i (Brazi l) us ing ' T a s s e l e d C a p ' t ransformat ion. P roc . 2 4 t h International S y m p o s i u m on R e m o t e S e n s i n g of Env i ronment , R io de Jane i ro , Braz i l , V o l . 1: 255 -263 . Breui l , C , 2000 . A s s o c i a t e D e a n , G radua te S tud ies and R e s e a r c h , Facul ty of Forest ry , T h e Universi ty of Brit ish C o l u m b i a , V a n c o u v e r , (Pe rsona l Commun ica t i on ) . C h a v e z , P . S . , 1988. A n improved dark-object subtract ion techn ique for a tmosphe r i c scat ter ing correct ion of mult i -spectral data . R e m o t e S e n s i n g of Env i ronment , 24 :459 -479 . Churche r , J . J . , and M c L e a n , J . A . , 1984. V i sua l detect ion of beet le -a t tacked whi te s p r u c e t rees us ing aer ia l photography. P roc . R N R F s y m p o s i u m on the app l ica t ions of remote s e n s i n g to resou rces managemen t . Amer . S o c . of Photogrammet ry , Fa l l s C h u r c h , V i rg in ia , pp. 137-143 . C o h e n , W . B . , 1991. R e s p o n s e of vegetat ion ind ices to c h a n g e s in three m e a s u r e s of leaf water s t ress . Photogrammet r ic Eng ineer ing and R e m o t e S e n s i n g , 57 :195 -202 . 46 Detect ion of mountain p ine beet le infestat ions us ing Landsa t T M T a s s e l e d C a p Trans fo rmat ions C o h e n , W . B . , and S p i e s , T. A . , 1992. Est imat ing structural attr ibutes of Doug las - f i r /wes te rn hemlock forest s tands from Landsa t and S P O T imagery . R e m o t e S e n s i n g of Env i ronment , 41 i l -l y . C o h e n , W . B . , S p i e s , T. A . , and Fiore l la , M. , 1995 . Est imat ing the age and structure of forests in a mul t i -ownership l andscape of wes te rn O r e g o n , U . S . A . International Journa l of R e m o t e S e n s i n g , 16(4): 721 -746 . C o n g a l t o n , R . G . , 1991. A review of a s s e s s i n g the a c c u r a c y of c lass i f i ca t ions of remotely s e n s e d data . R e m o t e S e n s i n g of Env i ronment , (37) 35 -46 . C o p p i n , P. , K. Nackaer ts , K., Q u e e n , L., and Brewer , K., 2000 . Opera t iona l monitor ing of g reen b i o m a s s c h a n g e for forest managemen t . In p ress "Pho tog rammet r i c Eng ineer ing and R e m o t e S e n s i n g " . (Persona l Commun ica t ion ) C o p p i n , P. R., and Bauer , M . E . , 1996. Digital c h a n g e detect ion in forest e c o s y s t e m s with remote sens ing imagery. R e m o t e S e n s i n g R e v i e w s , Vo l . 13 , pp. 207 -234 . Cr ist , E . P . , and Kauth, R. J . , 1986 .The T a s s e l e d C a p De-Myst i f ied . Pho tog rammet r i c Eng ineer ing and R e m o t e S e n s i n g , 52(1): 81-86 . Cr ist , E. P., Laur in, R., and C i c o n e , R . C , 1986. Vege ta t i on and so i ls information con ta ined in t ransformed themat ic mappe r data . P r o c e e d i n g s of I G A R S S ' 8 6 s y m p o s i u m , 1465-70. Cr ist , E . P . , and C i c o n e , R . C , 1984. App l ica t ion of the T a s s e l e d C a p concep t to s imula ted Themat i c M a p p e r data. Pho togrammet r i c Eng ineer ing and R e m o t e S e n s i n g , 50 (3): 343 -352 . G o w a r d , S . N . , and Huemmr i ch , K .F . , 1992. Vege ta t ion c a n o p y P A R abso rp tance and norma l i zed di f ference vegetat ion index: an a s s e s s m e n t us ing S A I L mode l . R e m o t e S e n s i n g of Env i ronment , 39 :119-140 . Hal l , F . G . , St rebal , D. E., N i c k e s o n , J . E. , and G o e t z , S . J . , 1991a . Rad iomet r i c recti f ication: T o w a r d a c o m m o n radiometr ic r e s p o n s e a m o n g mult i -date, mul t i -sensor images . R e m o t e S e n s i n g of Envi ronment , (35): 11-27. Hal l , F . G . , Botk ins, D.B. , S t rebe l , D. E., W o o d s , K .D. , and G o e t z , S . J . , 1991b. La rge s c a l e patterns of forest s u c c e s s i o n s determined by remote s e n s i n g . Eco logy , 72 :628 -640 . Hal l , P. M. , M c L e a n , J . A . , and Mur tha , P .A . , 1983 . Dye - laye r densi ty ana lys i s identif ies new Doug las fir beet le at tacks. C a n a d i a n Jou rna l of Fo res t R e s e a r c h , 13(2): 279 -282 . Hal l , P . M . , M c L e a n , J . A . , and Mur tha, P .A . , 1981 . R e m o t e sens ing of doug las fir t rees newly infested by bark beet les. P roc . Eighth B ienn ia l W o r k s h o p on C o l o r Aer ia l Pho tography , A m e r i c a n Soc ie ty of Photogrammetry , Fa l l s C h u r c h , V i rg in ia , pp. 91-98 . Harr is , J . W . E . , D a w s o n , A . F., and G o o d e n o u g h , D., 1978. Eva luat ion of Landsa t da ta for forest pest detect ion and d a m a g e appra isa l su r veys in Brit ish C o l u m b i a . C a n a d i a n Fo res t Se rv i ce , Pac i f i c Fores t R e s e a r c h Cen t re , P u b . R e p . B C - X -182 , 12 pp. Hobbs , A . J . , 1983. Ef fects of air photo s c a l e on early detect ion of mounta in p ine beet le. M . S c . T h e s i s , Dept. Forestry / R e m o t e sens ing , Univ. of Bri t ish C o l u m b i a , V a n c o u v e r , Brit ish C o l u m b i a , 137 pp. 47 Detect ion of mounta in pine beet le infestat ions us ing L a n d s a t T M T a s s e l e d C a p Trans fo rmat ions Hobbs , A . J . , and Mur tha, P.A. , 1984. V i sua l interpretation of four s c a l e s of aer ia l photography for early detect ion of mounta in pine beet le infestat ions. P r o c . R N R F s y m p o s i u m on the app l ica t ions of remote sens ing to resources managemen t . A m e r i c a n Soc ie ty of Pho togrammet ry , Fa l l s C h u r c h , V i rg in ia , pp. 433-444 . Horler, D. N. H. and Ahe rn , F. J . , 1986. Forest ry information content of Themat i c M a p p e r Da ta . International Journa l of R e m o t e S e n s i n g , V o l . 7 (3): 4 0 5 - 4 2 8 . Huete , A . R., 1988. A So i l Ad jus ted Vegeta t ion Index (SAVI ) . R e m o t e S e n s i n g of Env i ronment , 25 :295 -309 . J a c k s o n , R. D., 1983. Spect ra l ind ices in n -space . R e m o t e S e n s i n g of Env i ronment , 13 :409 -421 . J e n s e n , J . , 1996. Introductory digital image p rocess ing : a remote sens ing perspect ive . S e c o n d edit ion. Prent ice Hal l , 316p. Jus t i ce , C O . , T o w n s h e n d , J . R . G , Ho lben , B .N . , and Tucke r , C . J . , 1985. Ana l ys i s of the phenology of g lobal vegetat ion us ing meteoro log ica l satel l i te data . International Journa l of R e m o t e S e n s i n g , 6:1271 -1318. Kauth , R. J . , and T h o m a s , G . S . , 1976. T h e T a s s e l e d C a p - a g raph ic descr ipt ion of the spect ra l -tempora l deve lopmen t of agricultural c rops a s s e e n by Landsa t . P roc . of s y m p . on M a c h i n e P r o c e s s i n g of Remote ly S e n s e d da ta , Pu rdue Univers i ty , W e s t Lafayet te, Indiana, 21 J u n e -1 July, 4 B 4 1 - 4 B 5 1 . Kauth , R . J . , L a m b e c k , P . F . , R i c h a r d s o n , W . R . , T h o m a s , A . , and Pen t land , A . P . , 1979. Fea tu re extract ion app l ied to agricultural c rops a s s e e n by Landsa t . P r o c . of the Techn i ca l s e s s i o n s , T h e L A C IE S y m p . , N A S A J o h n s o n S p a c e Cente r , Hous ton , T e x a s , ( l l ) :705-721. Kneppeck , I. D., and A h e r n , F . J . , 1988. A c o m p a r i s o n of i m a g e s f rom a pushb room s c a n n e r with normal co lor aer ia l photography for detect ing scat tered recent coni fer mortality. Pho togrammet r i c Eng ineer ing and R e m o t e S e n s i n g , Lauver , C . L., and Whist ler , J . L., 1993. A h ierarchical c lass i f icat ion of Landsa t T M imagery to identify natural g rass land a reas and rare s p e c i e s habitat. Pho tog rammet r i c Eng ineer ing and R e m o t e S e n s i n g , vo l .59 (5): 627 -634 L i l lesand, T. M. , and Kieffer, R .W. , 2 0 0 0 . R e m o t e S e n s i n g and Image Interpretation, Fourth Edit ion, J o h n Wi ley & S o n s , N e w York . Major, D. J . , Baret , F., and Guyot . , G . , 1990. A ratio vegeta t ion Index adjusted for soi l br ightness. International Journa l of R e m o t e S e n s i n g , 11(5): 727 -740 . M c M u l l e n , L. H., Safranyik , L , and Linton, D. A . , 1986. S u p p r e s s i o n of mounta in p ine beet le infestat ions in lodgepole pine forests. C a n a d i a n Fo res t S e r v i c e Pac i f i c Forestry Cen t re , Inf. R e p . No . Brit ish C o l u m b i a - X - 2 7 6 . Moran , S . M . , J a c k s o n , R. D., Slater, P . N . , and Teil let, P . M . , 1992. Eva luat ion of s impl i f ied p rocedures for retrieval of land sur face ref lectance factors f rom satel l i te s e n s o r output. R e m o t e S e n s i n g of Env i ronment , (41): 169-184. M O E P , h t tp : / /www.env.gov.bc .ca/main /newsre l / f i sc9900/ february / ib418.h tm, N o v e m b e r 28 , 2 0 0 0 48 Detect ion of mounta in pine beet le infestat ions us ing Landsa t T M T a s s e l e d C a p Trans fo rmat ions Mur tha , P .A. , and C o z e n s , R., 1985. C o l o r infrared photo interpretation and ground su r veys to eva luate sp ruce beet le attack. C a n a d i a n Journa l of R e m o t e S e n s i n g , 11(2): 177-187 . Mur tha. , P .A. , 1985. Photo interpretation of s p r u c e beet le a t tacked sp ruce . C a n a d i a n Journa l of R e m o t e S e n s i n g , 11(1): 93 -102 . Mur tha , P .A. , and Fournier , R.A. , 1992. Va ry ing ref lectance pat terns in f luence photo interpretation of d e a d tree c rowns. C a n a d i a n Journa l of R e m o t e S e n s i n g , 18(3): 167-173 . Mur tha. , P .A. , and Wiart, R. J . , 1989a . P C - B a s e d digital ana lys is of mounta in p ine beet le current-a t tacked and non-at tacked lodgepo le pine. C a n a d i a n Jou rna l of R e m o t e S e n s i n g , V o l 1 5 (1): 7 0 -76. Murtha. , P .A . , and Wiart , R. J . , 1989b. C lus te r ana lys i s of pine c rown fol iage patterns aid identif ication of mounta in pine beet le current-attack. Pho togrammet r i c Eng ineer ing and R e m o t e S e n s i n g , Vo l . 55 (1): 83-96 . Mur tha. , P .A. , and Wiart , R. J . , 1987. P C - B a s e d digital image ana l ys i s for mounta in pine beet le g reen attack: prel iminary results. C a n a d i a n Jou rna l of R e m o t e S e n s i n g , V o l . 13(2): 92 -95 . Murtha. , P .A. , Bortolot, Z . , and Thurs ton , J . , 2 0 0 0 . A Landsa t T M spect ra l unmix ing mounta in pine beet le attack-fract ion probabil i ty map in the V a n d e r h o o f Fo res t District, Bri t ish C o l u m b i a , P roc . of R e m o t e S e n s i n g S y m p o s i u m (on C D ) . Pat te rson, M . W . and Y o o l , S . R., 1998. M a p p i n g f i re- induced vegetat ion mortality us ing Landsa t 5 Themat i c M a p p e r data : A compar i son of l inear t ransformat ion techn iques . R e m o t e S e n s i n g of Env i ronment 65 :122 -132 . Pr ice , K .P . , and J a k u b a u s k a s , M . E . , 1998. Spec t ra l re t rogress ion and insect d a m a g e in lodgepo le pine s u c c e s s i o n a l forests. International Journa l of R e m o t e S e n s i n g , V o l . 19(8) :1603-1611. Pr i ce , K .P . , S t e p h e n , L .E . , Nel l is , M.D. , L e e , R .Y . , and B o y c e , R., 1997. Mapp ing land c o v e r in a High P la ins a g r o - e c o s y s t e m using a mult i-date Landsa t Thema t i c M a p p e r mode l ing app roach . T ransac t i ons of K a n s a s A c a d e m y of S c i e n c e s , 100(1/2): 21 -33 . Q i . J . , C h e b o u n i , A . , Huete , A . R., Kerr , Y . H . , and So roosh ian . , 1994. A modi f ied soi l ad jus ted vegetat ion index. R e m o t e S e n s i n g of Env i ronment , 48 :119 -126 . Renc j , A . N . , and Neme th , S . , 1985. Detect ion of mounta in pine beet le infestat ion us ing Landsa t and s imula ted themat ic mapper data. C a n a d i a n Journa l of R e m o t e S e n s i n g , 11(1): 50-58 . R i c h a r d s o n , A . J . , and W e i g a n d , C . L . , 1990. C o m p a r i s o n of two mode ls for s imulat ing the so i l -vegetat ion compos i te ref lectance of a deve lop ing cotton canopy . International Jou rna l of R e m o t e S e n s i n g , 11(3) :447-459. R i cha rdson , A . J . , and W i e g a n d , C . L . , 1977. Dist inguishing vegetat ion f rom soi l background information. Pho togrammet r i c Eng ineer ing and R e m o t e S e n s i n g , 4 3 : 1 5 4 1 - 1 5 5 2 . Richter, R., 1997. Cor rec t ion of a tmospher i c and topograph ic ef fects for high spat ia l resolut ion satell i te imagery. International Journa l of R e m o t e S e n s i n g , 18(5) :997-1110. 49 Detect ion of mounta in pine beet le infestat ions us ing Landsa t T M T a s s e l e d C a p Trans fo rmat ions S a c h s . D.L., So l l ins , P., and C o h e n , W . B . , 1998. Detect ing l a n d s c a p e c h a n g e s in the interior of Brit ish C o l u m b i a f rom 1975 to 1992 using satell i te imagery. C a n a d i a n Jou rna l of R e m o t e S e n s i n g , V o l . 28 :23-36 . S c i e n c e wri ters gu ide to Landsa t -7 , h t tp : / / l andsa t .gs fc .nasa .gov /announcements /L7_Wr i te rs_Gu ide .h tm l , N o v e m b e r 28 , 2000 . Sel ler , P . J . , Tucker , C . J . , Col la te, G . J . , Los , S . , Jus t i ce , C O . , Daz l i ch , D. A . , and Randa l l , D. A . 1994. A global 10*10 NDVI da ta set for c l imate s tud ies. Par t 2 - the ad jus tment of the NDVI and generat ion of g loba l f ie lds of terrestrial b iophys ica l parameters . International Jou rna l of R e m o t e S e n s i n g , V o l . 15 :3519-3545 . Sho re , T. L., Saf rany ik , L , and Lemieux , J . P . , 2000 . Suscept ib i l i ty of lodgepo le pine s tands to the mounta in p ine beet le: test ing of a rating sys tem. C a n a d i a n Jou rna l of Fo res t R e s e a r c h , Vo l . 30 : 44-49 . Sho re , T. L., and Safrany ik , L., 1992. A s y s t e m for rating the suscept ibi l i ty and risk of lodgepole pine s tands to mounta in pine beet le. For. C a n . P F C - F i l e R e p . 5203 .17pp . Si ro is , J and A h e r n , F . J . , 1988. A n invest igat ion of S P O T H R V Data for detect ing recent mounta in pine beet le mortality. C a n a d i a n Jou rna l of R e m o t e S e n s i n g , 14(2): 104-108 . Taylor , S . P., 1998. A compar i son be tween aer ia l ske tch mapp ing and L a n d s a t survey techn iques for detect ing a t tacked t rees by mounta in pine beet le and Douglas- f i r beet les ( B C M O F internal report). T h e D ispa tch , V o l . 1-4, C L M A and N F P A E m e r g e n c y T a s k Fo rce , www.moun ta inp inebee t le . com, N o v e m b e r 28 , 2000 . T h o m p s o n , D. R., and W e h m a n e n , O . A . , 1980. Us ing Landsa t digital da ta to detect moisture s t ress in co rn - soybean growing regions. Pho togrammet r i c Eng ineer ing and R e m o t e S e n s i n g , 46 :1087 -1093 . Unger , L., 1993. Mounta in P ine Beet le . Forest ry C a n a d a , Fo res t Insect and D i s e a s e Survey , Fores t Pes t Leaf let No . 76, 7p. 50 

Cite

Citation Scheme:

        

Citations by CSL (citeproc-js)

Usage Statistics

Share

Embed

Customize your widget with the following options, then copy and paste the code below into the HTML of your page to embed this item in your website.
                        
                            <div id="ubcOpenCollectionsWidgetDisplay">
                            <script id="ubcOpenCollectionsWidget"
                            src="{[{embed.src}]}"
                            data-item="{[{embed.item}]}"
                            data-collection="{[{embed.collection}]}"
                            data-metadata="{[{embed.showMetadata}]}"
                            data-width="{[{embed.width}]}"
                            async >
                            </script>
                            </div>
                        
                    
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:
https://iiif.library.ubc.ca/presentation/dsp.831.1-0074962/manifest

Comment

Related Items