Open Collections

UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

The extraction of landslides in a satellite image using a digital elevation model Donahue, John Patrick 1987

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

Item Metadata

Download

Media
831-UBC_1987_A7 D66.pdf [ 14.84MB ]
Metadata
JSON: 831-1.0097203.json
JSON-LD: 831-1.0097203-ld.json
RDF/XML (Pretty): 831-1.0097203-rdf.xml
RDF/JSON: 831-1.0097203-rdf.json
Turtle: 831-1.0097203-turtle.txt
N-Triples: 831-1.0097203-rdf-ntriples.txt
Original Record: 831-1.0097203-source.json
Full Text
831-1.0097203-fulltext.txt
Citation
831-1.0097203.ris

Full Text

THE EXTRACTION OF LANDSLIDES IN A SATELLITE IMAGE USING A DIGITAL ELEVATION MODEL By JOHN PATRICK DONAHUE 6.Sc., University of Massachusetts, 1983 A THESIS SUBMITTED IN PARTIAL FULFILMENT OF THE REQUIREMENTS FOR THE DEGREE OF MASTER OF APPLIED SCIENCE in THE FACULTY OF GRADUATE STUDIES (Department of Forestry) We accept this thesis as conforming to the required standard THE UNIVERSITY OF BRITISH COLUMBIA September, 1987 © John Patrick Donahue, 1987 97 In presenting this thesis in partial fulfilment of the requirements for an advanced degree at the University of British Columbia, I agree that the Library shall make it freely available for reference and study. I further agree that permission for extensive copying of this thesis for scholarly purposes may be granted by the head of my department or by his or her representatives. It is understood that copying or publication of this thesis for financial gain shall not be allowed without my written permission. Department of Fnrp<;try  The University of British Columbia 1956 Main Mall Vancouver, Canada V6T 1Y3 Date October 5. 1987 ABSTRACT Landslides in the landscape exhibit predictable properties of shape, structure and orientation. These properties are reflected to varying degrees in their depic-tion in a s a t e l l i t e image. Landslides can be isolated along with similar objects in a d i g i t a l image using d i f f e r e n t i a l and template operators. Extraction of the landslide features from these images can proceed using a logic-based model which draws on an appropriate object d e f i n i t i o n approximat-ing the depiction of the landslides in an edge-operated image and a d i g i t a l elevation model. An object extraction algorithm based on these concepts i s used in repeated t r i a l s to ascertain the effectiveness of this automated approach. A low resolution l i n e a r object d e f i n i t i o n (Fischler e_t a_L. , 1981) i s used to i s o l a t e candi-date pixel segments i n three enhanced images. These segments are c l a s s i f i e d as landslides or non-landslides according to their image pixel i n t e n s i t y , length, slope, and orientation. D i g i t a l elevation data i s used to evaluate slope and orien-tation c r i t e r i a . Results are compared to an inventory of landslides made using a e r i a l photographs. Study results indicate that 17% to 28% of landslides in the image are i d e n t i f i e d for t r i a l s that produce a commis-sion error rate of less than 50%. Commission errors are dominated by image objects related to roads and waste wood i i areas i n c l e a r c u t s . A higher r a t e of s u c c e s s f u l i d e n t i f i -c a t i o n was noted f o r l a n d s l i d e s which occurred w i t h i n 15 years of image a c q u i s i t i o n (24% to 32%), and was most ap-parent f o r the subset of that group which was l o c a t e d i n areas that were harvested more than 15 years before a c q u i s i -t i o n or were unharvested (29% to 38%). S u c c e s s f u l i d e n t i f i -c a t i o n s i n the t r i a l s are dominated by events g r e a t e r than 300 metres long and wider than 20 metres. The r e s u l t s sug-gest that the approach i s more r e l i a b l e i n unharvested areas of the image. The poor q u a l i t y of the d i g i t a l e l e v a t i o n data, s p e c i f i c a l l y a r t i f a c t s produced by the c o n t o u r - t o - g r i d a l g o r i t h m , was p a r t l y r e s p o n s i b l e f o r e r r o r s of commission and omission. The s i m p l i c i t y of the object d e f i n i t i o n used i s another f a c t o r i n e r r o r p r o d u c t i o n . The methodology i s not o p e r a t i o n a l , but r e p r e s e n t s a r e a l i s t i c approach to scene segmentation f o r resource management given f u r t h e r refinement. i i i TABLE OF CONTENTS I ABSTRACT i i II TABLE OF CONTENTS iv III LIST OF TABLES vi VI LIST OF FIGURES v i i V ACKNOWLEDGMENTS ix 1. INTRODUCTION 1 1.1 Landslide definition and process 2 1.2 Literature review: landslide inventory 3 1.3 Object extraction 6 1.3.1 Literature review: linear object extraction 7 1.3.2 A decision model for extraction of landslides 10 1.4 Research objectives 12 2. MATERIALS AND METHODS 14 2.1 Site description 14 2.2 Baseline landslide inventory for the study site . 14 2.2.1 Compilation 14 2.2.2 Data adaptation 16 2.3 Satellite imagery 19 2.3.1 Image description 19 2.3.2 Image band selection 19 2.3.3 Edge detection 22 2.3.3.1 Differential operator 22 2.3.3.2 Template matching 24 2.3.3.3 Threshold/template matching 28 2.4 Digital topography 31 3. IMPLEMENTATION 39 3.1 Introduction 39 3.2 Image evaluation program 40 3.3 Landslide image production and evaluation procedure 44 4. RESULTS AND DISCUSSION 50 4.1 Results 50 4.1.1 Commission errors 50 4.1.2 Success characterisation 57 4.2 Discussion 61 4.2.1 Commission errors 68 4.2.2 Identification successes and omissions .... 70 4.2.3 Baseline inventory 72 5. CONCLUSIONS 74 5.1 This study 74 5.2 Research directions 76 6. LITERATURE CITED 78 iv 7. GLOSSARY 82 8. APPENDIX A: L a n d s l i d e d e t e c t i o n i n an enhanced s a t e l l i t e image 84 A . l Procedure 84 A.2 R e s u l t s and c o n c l u s i o n s 85 9. APPENDIX B: Record of commission e r r o r s 89 10. APPENDIX C: Event c h a r a c t e r i s t i c s and i d e n t i f i c a t i o n 92 11. APPENDIX D: L a n d s l i d e template r e v i s i o n 110 v LIST OF TABLES 3.1 Tr i a l layout for landslide image production 46 4.1 Success and commission results for 14 t r i a l s 51 4.2(a,b) Causes of commission errors 58 4.3 Total number of events in each landslide category .. 62 4.4 Successful identifications for t r i a l D 63 4.5 Successful identifications for t r i a l M 64 4.6 Successful identifications for t r i a l K 65 4.7 Successful identifications for t r i a l N 66 4.8 Successful identifications for t r i a l E 67 D.l Success and commission results for study t r i a l s plus revised, eight direction template 112 vi LIST OF FIGURES 1.1 An example of a t r a n s l a t i o n a l s l i d e 4 1.2 Landsat TM image c o n t a i n i n g the study s i t e 11 2.1 L o c a t i o n map of the study s i t e 15 2.2 R e s u l t s from three t r a n s e c t s recorded across known l a n d s l i d e t r a c k s 21 2.3 A demonstration of the L a p l a c i a n operator used i n the study 24 2.4 The L a p l a c i a n edge image used i n the study 25 2.5 The Duda Road Operator as presented by F i s c h l e r et. a l . (1981) 27 2.6 The l a n d s l i d e operator developed f o r the study 27 2.7 The l a n d s l i d e template image used i n the study 27 2.8 Histogram of image band 3 of the study s i t e 30 2.9a-d A p o r t i o n of image band 3 of the study s i t e t h r e s h o l d e d at 4 p i x e l values 32 2.9e The thres h o l d / t e m p l a t e matching image used i n the study 33 2.10 S y n t h e t i c view of the d i g i t a l e l e v a t i o n model of the study s i t e 34 2.11 The slope adjustment operator used to estimate point slope values from the d i g i t a l e l e v a t i o n data 37 2.12 Code format f o r the f a l l l i n e v e c t o r f i l e 38 3.1 Pseudocode f o r segment i d e n t i f i c a t i o n 41 3.2 Pseudocode f o r segment l e n g t h a p p r a i s a l 42 3.3 D e f i n i t i o n s f o r v a r i a b l e s used i n F i g u r e s 3.1 & 3.2 43 3.4 Histogram of L a p l a c i a n image 47 3.5 Histogram of template matching image 48 4.1 T r i a l D ( L a p l a c i a n operator) 52 4.2 T r i a l M (Threshold/template o p e r a t o r ) 53 v i i 4.3 T r i a l K (Template o p e r a t o r ) 54 4.4 T r i a l N ( T h r e s h o l d / t e m p l a t e o p e r a t o r ) 55 4.5 T r i a l E ( L a p l a c i a n o p e r a t o r ) 56 4.6 Commission e r r o r A-4 59 4.7 Commission e r r o r B-16 60 A . l R e s u l t s : D e t e c t i o n r a t e of a l l l a n d s l i d e e v e n t s f o r 3 t h r e s h o l d s of the L a p l a c i a n image 86 A.2 R e s u l t s : D e t e c t i o n of l a n d s l i d e e v e n t s by the L a p l a c i a n o p e r a t o r by l a n d s l i d e type 86 A.3 P a r t 1 r e s u l t s : D e t e c t i o n of l a n d s l i d e e v e n t s by the L a p l a c i a n o p e r a t o r by l a n d s l i d e age 87 v i i i ACKNOWLEDGMENTS I would l i k e to thank my s u p e r v i s o r , Dr. Peter Murtha, and the members of my committee, e s p e c i a l l y Dr. Mark Sond-heim, whose ongoing guidance and patience was i n v a l u a b l e . Mark's f a m i l y was a l s o q u i t e kind to me on my frequent t r i p s to V i c t o r i a f o r c o n s u l t a t i o n . I a l s o wish to re c o g n i s e the able a s s i s t a n c e of the t e c h n i c a l s t a f f at the F a c u l t y of F o r e s t r y , namely Nedenia K r a j c i and Raoul Wiart, f o r t h e i r help and advice throughout. Tim Lee and Mark Majka, both with the Laboratory f o r Computational V i s i o n , were extremely h e l p f u l with programming advice and software development. F i n a l l y , I want to thank Paula West f o r her remarkable p a t i -ence and support dur i n g my tenure at UBC. T h i s t h e s i s was funded by the C.F.S. through the F a c u l t y of F o r e s t r y Block Grant. i x CHAPTER 1 INTRODUCTION This thesis describes the application of image enhance-ment techniques and a d i g i t a l elevation model (DEM) to an object detection and extraction problem in a s a t e l l i t e image. A methodology i s proposed which u t i l i s e s a l o g i c a l approach to the aggregation of scene, object, and a n c i l l a r y information, and a demonstration i s presented which i d e n t i -f i e s landslide features in the image using t h i s approach. The extraction of landslides from a s a t e l l i t e image i s an appropriate problem to investigate based on the need for an updatable inventory on forested lands, as well as an environmental monitoring system for water quality assess-ment. In Chapter One, a review of l i t e r a t u r e pertaining to the inventory of landslides on the landscape i s supplemented by a review of automated object extraction techniques for extracting linear objects in d i g i t a l images. The basis for the decision model i s also discussed. Chapter Two describes the ground truth data for the study as well as the imagery, enhancement operations, and elevation data which w i l l be employed in finding landslides. Chapter Three recounts the implementation of an automated system for extracting land-s l i d e s from enhanced s a t e l l i t e images. Results are presented in Chapter 4. A discussion of the results from the methodo-logy examines i t s success as compared to a conventional inventory, and possible reasons for the omission and commis-1 s i o n e r r o r s encountered. Conclusions are made i n Chapter 5, i n c l u d i n g some thoughts about f u t u r e work. Appendix A d e s c r i b e s the procedure and r e s u l t s from a p r e l i m i n a r y i n v e s t i g a t i o n i n t o the use of a low r e s o l u t i o n l i n e a r o b j e c t d e f i n i t i o n ( F i s c h l e r et a l . , 1981) to d e s c r i b e l a n d s l i d e s i n an enhanced s a t e l l i t e image. The r e s u l t s confirm the v a l i d i t y of such an approach, and helped to guide the design of the object e x t r a c t i o n system employed i n t h i s t h e s i s . Appendix B i s a c o m p i l a t i o n of commission e r -r o r s made i n s e v e r a l t r i a l s of the methodology, and Appendix C i s a catalogue of l a n d s l i d e event a t t r i b u t e s and t r i a l i d e n t i f i c a t i o n r e s u l t s . Appendix D explores the use of a l a n d s l i d e template which i s s l i g h t l y d i f f e r e n t than the one used i n the study. 1.1 L a n d s l i d e d e f i n i t i o n and process " L a n d s l i d e " i s a generic term f o r s e v e r a l types of mass movements i n the Coast Mountains of B r i t i s h Columbia and i n c l u d e s flows and s l i d e s (Varnes, 1958) as w e l l as t o r r e n t s (VanDine, 1985). I t i s used as such throughout t h i s work. The nomenclature i n d i c a t e s the process i n v o l v e d i n a mass movement. A s l i d e i m p l i e s one or many shear f a i l u r e s along r e c o g n i z a b l e zones or planes. The r o t a t i o n a l s l i d e presents a concave s u r f a c e , whereas the t r a n s l a t i o n a l s l i d e proceeds along a g e n e r a l l y l i n e a r plane. A t r a n s l a t i o n a l s l i d e o f t e n takes on the narrow, l i n e a r shape of the c l a s s i c mountain slope movement (Varnes, 1978:14). The r o t a t i o n a l 2 s l i d e , although commonly s m a l l e r , may give r i s e to l a r g e r , l i n e a r t o r r e n t events (VanDine, 1985). F i g u r e 1.1 i s an example of the t r a c k l e f t by a t r a n s l a t i o n a l s l i d e which was i n i t i a t e d by a road s i d e c a s t f a i l u r e . Flows are c h a r a c t e r -i s e d by a v i s c o u s f l u i d motion and a l a c k of s t r u c t u r a l s u r f a c e s between the mass and the s l o p e . According to Varnes, 'there i s a complete g r a d a t i o n from d e b r i s s l i d e s to d e b r i s flows, depending on water content, m o b i l i t y , and c h a r a c t e r of the movement'(1978:18). When such a flow occurs i n a channel or g u l l y , i t i s known as a d e b r i s t o r r e n t (Swanson and Swanston, 1977). 1.2 L i t e r a t u r e review; l a n d s l i d e i n v e n t o r y Methods f o r e v a l u a t i n g l a n d s l i d e hazard commonly r e -q u i r e an i n v e n t o r y of s l i d e s which have occurred a l r e a d y i n the area. Examples of these methods i n c l u d e those i n t r o d u c e d by Sondheim and R o l l e r s o n (1985), N e i l s e n and Brabb (1978), Wieczorek (1984), and Howes (1987). An i n v e n t o r y may be used i n combination with g e o l o g i c i n f o r m a t i o n , or on i t s own i n such schemes. I n v e n t o r i e s are commonly drawn from evidence obtained from f i e l d w o r k , a e r i a l photographs, and/or other remote sensing imagery. As such, the problem of i d e n t i f i c a -t i o n of slope movement evidence i n the landscape has been of i n t e r e s t to these and other workers. G r e s s w e l l , H e l l e r , and Swanston (1978) performed a predominately f i e l d - b a s e d i n v e n t o r y i n support of t h e i r study i n the Oregon Cascades. Methods f o r the i d e n t i f i c a t i o n 3 Figure 1.1. An example of a tr a n s l a t i o n a l s l i d e . Dimensions are approximately 200 metres long by 20 metres wide. 4 of l a n d s l i d e s on a e r i a l photographs have been e s t a b l i s h e d (Dishaw, 1967; Poole, 1969; Rib and L i a n g , 1978). Rice and Foggin (1971) used l a r g e s c a l e (1:5000) panchromatic a e r i a l photographs to estimate area a f f e c t e d by s o i l s l i p p a g e i n the San Dimas Experimental F o r e s t , C a l i f o r n i a . O'Loughlin (1973) used 1:30000 and 1:15000 panchromatic a e r i a l photo-graphs to i d e n t i f y l a n d s l i d e s . He reported small s c a l e c o l -our photographs to be of some value i n t h i s process, but found c o l o u r - i n f r a r e d photography to be of s u p e r i o r value because of i t s advantages i n d e t e c t i n g age and s p e c i e s d i f -f e r e n c e s i n stands and the strong bare s o i l / v e g e t a t i o n con-t r a s t p r o v i d e d . Rice e_t a_l. (1969), B a i l e y (1972), and Howes (1987) used a combination of a i r p h o t o a n a l y s i s and ground t r u t h i n mapping l a n d s l i d e i n v e n t o r y f o r s t a b i l i t y mapping p r o j e c t s . Although c o n v e n t i o n a l a e r i a l photography has been used e x t e n s i v e l y i n t h i s f i e l d , other remote sensing systems, i n c l u d i n g Landsat, have been i n v e s t i g a t e d . Gagnon (1975) st u d i e d the a p p l i c a b i l i t y of v a r i o u s systems i n unstable c l a y f o r m a t i o n s . He reported that Landsat MSS bands 6 and 7 were u s e f u l f o r study of c o n t r a s t s i n s o i l water content on a r e g i o n a l l e v e l and that thermal i n f r a r e d scanner imagery was u s e f u l f o r d e t e c t i n g s u r f a c e water c o n d i t i o n s when used i n c o n j u n c t i o n with i n f r a r e d black and white photography. T h i s type of imagery was a l s o used to guide f i e l d w o r k i n an unstable c l a y area by i d e n t i f y i n g seepage zones i n the area (Tanguay and Chagnon, 1972). Johnson ejt a l . (1977) success-5 f u l l y used 1:30000 c o l o u r - i n f r a r e d a i r p h o t o s to study vege-t a t i o n anomalies caused by s o i l moisture changes accompany-ing movement i n i n c i p i e n t s l i d e masses. According to A l f o l d i (1974), who worked i n s i m i l a r c l a y t e r r a i n , s a t e l l i t e imag-ery i s u s e f u l f o r i t s r e g i o n a l p e r s p e c t i v e , multi-band i n -t e r p r e t i v e c a p a b i l i t y , and i t s r e p e t i t i v e coverage upon which d e t a i l e d i n v e s t i g a t i o n may be based. Gimbarzevsky (1983) repo r t e d that approximately 40% of l a n d s l i d e s i n the Queen C h a r l o t t e I s l a n d s mapped from small s c a l e a e r i a l pho-tographs were v i s u a l l y d i s c e r n i b l e on MSS band 5, 6 and 7 c o l o u r composites. An a i r c r a f t mounted radiometer, using Landsat MSS bands, proved u s e f u l i n a d i g i t a l c l a s s i f i c a -t i o n of slope c o n d i t i o n ( e r o s i o n , d e p o s i t i o n , e t c . ) i n an a r i d r e g i o n (Pickup and Nelson, 1984). In that study, p l o t s of band r a t i o s f o r each p i x e l l o c a t i o n provided a c l a s s i f i -c a t i o n space f o r the parameters of i n t e r e s t . 1.3 Object e x t r a c t i o n Object e x t r a c t i o n i s one approach to r e c o g n i s i n g pat-t e r n s of i n t e r e s t i n an image. P a t t e r n r e c o g n i t i o n i s a g eneral term which d e s c r i b e s the segmentation of a d i g i t a l image based on a set of c l a s s - d e f i n i n g c h a r a c t e r i s t i c s (Swain and Davis, 1978). When t h i s segmentation i s based only on s p e c t r a l r e f l e c t a n c e c h a r a c t e r i s t i c s of o b j e c t s i n the scene, i t i s c a l l e d a c l a s s i f i c a t i o n ( C o l w e l l , 1983). Object e x t r a c t i o n d e s c r i b e s the segmentation of an image using knowledge about the p h y s i c a l shape of image o b j e c t s as an a d d i t i o n a l d e f i n i n g c h a r a c t e r i s t i c . Swain and Davis 6 (1978) considered the subject of c l a s s i f i c a t i o n i n depth, while Rosenfeld and Weszka (1976) provided a c o n c i s e review of object e x t r a c t i o n . In object e x t r a c t i o n , a d e c i s i o n model based on know-ledge about the image gray l e v e l c o n d i t i o n s that an object e x h i b i t s i s used to i d e n t i f y object candidates as i n a c l a s s i f i c a t i o n . Shape i n f o r m a t i o n i s g e n e r a l l y i n t e g r a t e d i n t o the process i n r e g i o n growing, t r a c k i n g , or template matching systems (Rosenfeld and Weszka, 1976). The r e s u l t i s a segmentation which o u t l i n e s the o b j e c t ' s l o c a t i o n i n the image. Scenes c o n t a i n i n g o b j e c t s with w e l l - d e f i n e d gray l e v e l and shape c h a r a c t e r i s t i c s set a g a i n s t a p r e d i c t a b l e background provide s u i t a b l e input to r i g i d d e c i s i o n models. However, f l e x i b i l i t y i s c a l l e d f o r i n remotely-sensed scenes where gray l e v e l s may vary widely from expected values as a r e s u l t of shadows and atmospheric e f f e c t s . Furthermore, complex ob j e c t backgrounds can mimic s i m p l i s t i c object def-i n i t i o n s . 1.3.1 L i t e r a t u r e review: l i n e a r o b j e c t e x t r a c t i o n Techniques have been developed to recognise a wide range of f e a t u r e s i n v a r i o u s types of images. Workers i n v a r i o u s f i e l d s such as medicine and high energy physics r e p o r t the use of o b j e c t e x t r a c t i o n i n t h e i r work with imag-ery. However, t h i s review i s l i m i t e d to work i n the r e c o g n i -t i o n of l i n e a r o b j e c t s i n a e r i a l and s a t e l l i t e imagery, s i n c e t h i s a p p l i e s to the appearance of Coast Mountain l a n d -7 s l i d e s i n image radiance and shape. A good review of these works i s given i n Majka (1982). Majka (1982) d e s c r i b e d a technique which u t i l i s e s the primal sketch theory of image understanding (Marr and H i l -dreth, 1980). Road edges i n d i g i t i s e d a e r i a l photographs were d e f i n e d by the l o c a t i o n of the zero c r o s s i n g "bars" of the L a p l a c i a n of the image, and these bars were then e v a l -uated a c c o r d i n g to geometric c r i t e r i a . F i s c h l e r e_t a l . (1981) reviewed s e v e r a l image operat o r s i n developing an i n t e g r a t e d approach f o r i d e n t i f y i n g l i n e a r o b j e c t s . They c a t e g o r i s e d these o p e r a t o r s as Type I and Type II o p e r a t o r s . A Type I operator o f t e n misses the f e a t u r e i t i s intended to r e c o g n i s e , but commits very few f a l s e i d e n t i f i c a t i o n s (com-m i s s i o n s ) . A Type I I operator, a l t e r n a t e l y , a c c u r a t e l y i d e n -t i f i e s the intended f e a t u r e s , but makes s i g n i f i c a n t commis-s i o n e r r o r s as w e l l . T h e i r review of o p e r a t o r s , which i n -cluded t h r e s h o l d i n g , "edge" d e t e c t i o n and template matching, did not uncover one which s a t i s f a c t o r i l y minimised both types of e r r o r . T h e r e f o r e , a methodology was developed which organised the r e s u l t s of such o p e r a t i o n s a c c o r d i n g to ex-pected e r r o r s i n order to optimise the i d e n t i f i c a t i o n pro-c e s s . T a v a k o l i and Rosenfeld (1982) used a high l e v e l of reasoning to organise edge segments i n an a e r i a l scene i n t o b u i l d i n g and road o b j e c t s . There has been i n t e r e s t i n the r e c o g n i t i o n of o b j e c t s i n s a t e l l i t e images as w e l l . Bajcsy and T a v a k o l i (1976) used 8 s p e c t r a l and shape d e f i n i t i o n of roads to guide the l o g i c a l d i s c r i m i n a t i o n of roads i n a s a t e l l i t e image. A h i e r a r c h y of f e a t u r e o p e r a t o r s was e s t a b l i s h e d which f i r s t f i n d s s t r i p s w i t h the a p p r o p r i a t e geometry and s p e c t r a l c h a r a c t e r i s t i c s , then c o n n e c t s these s t r i p s a c c o r d i n g t o g e o m e t r i c c r i t e r i a , and f i n a l l y t h i n s the r e s u l t i n g " r o a d s " . They r e p o r t 95% r e c o g n i t i o n i n non-urban and 85% r e c o g n i t i o n i n urban a r e a s , a l t h o u g h they do not r e p o r t the r a t e at which commission e r r o r s are produced. E r r o r s were assumed t o be due to poor c o n t r a s t i n urban a r e a s and f a l s e i d e n t i f i c a t i o n s i n non-urban a r e a s . S p a t i a l p a t t e r n s i n a s a t e l l i t e image based on the t o p o g r a p h i c m o d u l a t i o n image ( E l i s a s o n et_ ajL. , 1981) have been used t o guide the d e l i n e a t i o n of r i v e r s ( H a r a l i c k et_ al_. , 1982). A " t o p h a t " o p e r a t o r was proposed by D e s t i v a l and LeMen (1986) t o e x t r a c t roads of v a r i o u s w i d t h s i n h i g h -er r e s o l u t i o n s a t e l l i t e imagery. T h i s i s a means of s i m u l t a -n e o u s l y t h r e s h o l d i n g s p e c t r a l edge and p h y s i c a l w i d t h i n f o r -m a t i o n . I t i s a precedent f o r the development of s p e c i f i c , t a s k - o r i e n t e d edge o p e r a t o r s based on the type of imagery and o b j e c t . S h i b a t a (1984) suggested the c o m p a t i b i l i t y of d i g i t a l e l e v a t i o n data w i t h d i g i t a l image d a t a . Other work i n t h i s f i e l d i n c l u d e s t h a t by B a j c s y and T a v a k o l i (1973), N e v a t i a and Babu ( 1 9 7 9 ) , and B a l l a r d ( 1 9 8 1 ) . These s t u d i e s c o n t a i n s e v e r a l common themes. They es-pouse the use of c o n v o l u t i o n and/or t e m p l a t e o p e r a t o r s t o e x t r a c t c o n t r a s t edge i n f o r m a t i o n i n the image. They make use of o b j e c t d e f i n i t i o n s based on image c h a r a c t e r i s t i c s and 9 r e a l world knowledge about the o b j e c t . F i n a l l y , model e f -f e c t i v e n e s s i s o f t e n evaluated by comparing r e s u l t s to a mapping of the o b j e c t ' s l o c a t i o n s i n the scene drawn from other sources. T h i s study w i l l f o l l o w these types of ap-proaches . 1.3.2 A_ d e c i s i o n model f o r e x t r a c t i o n of l a n d s l i d e s A d e s c r i p t i o n of an o b j e c t ' s expected r e p r e s e n t a t i o n i n the image i s preparatory to the design of a model which can s u i t a b l y detect occurrences of the o b j e c t . F i g u r e 1.2 shows the s a t e l l i t e image used i n the study. F i v e a t t r i b u t e s were chosen to d e s c r i b e l a n d s l i d e s : 1. Widths range from 5 metres to 50 metres. 2. Lengths range from a few metres to over 800 metres. 3. Composition i s mainly bare s o i l and rock or reveg-e t a t i o n s p e c i e s which c o n t r a s t with the surroundings. 4. They g e n e r a l l y f o l l o w the f a l l l i n e ( l i n e of s t e e p -est d e s c e n t ) . 5. They do not e x i s t to a s i g n i f i c a n t degree at s l o p e s below a c e r t a i n minimum (10 to 15 degrees). The d e p i c t i o n of l a n d s l i d e s i n an enhancement of the image and the DEM w i l l r e f l e c t the l i s t e d parameters. The image r e s o l u t i o n element ( p i x e l ) i s approximately 30 metres square, and l a n d s l i d e s i n the image are expected to i n v o l v e s i g n i f i c a n t numbers of mixed p i x e l s ( m i x e l s ) , with depicted widths of 1 to 3 p i x e l s . T h i s i s c o n s i s t e n t with the d e f i n i -t i o n of a low r e s o l u t i o n l i n e a r o b j e c t . Lengths of object r e p r e s e n t a t i o n s (segments) w i l l have an a r b i t r a r y lower l i m i t s p e c i f i e d i n the model ranging to an upper l i m i t of about 30 p i x e l s . Object composition should produce a gray 10 Figure 1.2. Landsat TM image of the study s i t e . Band 3 (red) i s shown in this rendition. The boundary of the study s i t e i s shown in black. 11 l e v e l c o n t r a s t which i s s u i t a b l e f o r d e t e c t i o n by an edge operator, which i n turn w i l l provide object c a n d i d a t e s . Slope and o r i e n t a t i o n a t t r i b u t e s w i l l be a s c e r t a i n e d from a DEM which i s r e g i s t e r e d to the image. S u f f i c i e n t i n f o r m a t i o n about each candidate w i l l be a v a i l a b l e i n the image and DEM such that a d e c i s i o n can be made about i t s source ( l a n d s l i d e or n o n - l a n d s l i d e ) . The model makes a d e c i s i o n about the source of an image segment based on parameter values a s s o c i a t e d with the l i s t e d a t t r i b u t e s . D i f f e r e n t values f o r these parameters w i l l be examined on a t r i a l b a s i s because of t h e i r expected v a r i a -b i l i t y f o r the spectrum of l a n d s l i d e s i n the image. No s i n -gle t r i a l i s expected to be optimum f o r a l l l a n d s l i d e s . A review of these t r i a l s w i l l examine the shortcomings of t h i s approach. The a c t u a l parameter values, and the reasoning behind them, are d i s c u s s e d i n more d e t a i l i n S e c t i o n 3.3. 1.4 Research o b j e c t i v e s T h i s study i s an examination of the e f f e c t i v e n e s s of an object e x t r a c t i o n d e c i s i o n model that i n t e g r a t e s d i g i t a l e l e v a t i o n data with a Landsat Thematic Mapper (TM) image. The d e t e c t i o n of l a n d s l i d e o b j e c t s provides the s c e n a r i o f o r a p p l i c a t i o n of t h i s methodology. The study o b j e c t i v e s are: 1. To develop and operate a d e c i s i o n model based on an o b j e c t d e f i n i t i o n . A v a r i e t y of enhanced images w i l l be used to provide candidate segments which f i t the d e f i -n i t i o n , and a d i g i t a l e l e v a t i o n model w i l l be employed to help s o r t the segments r e p r e s e n t i n g l a n d s l i d e s from the segments r e p r e s e n t i n g other, s i m i l a r o b j e c t s i n the scene. 12 2. To e v a l u a t e the e f f e c t i v e n e s s of the d e c i s i o n model by comparing study r e s u l t s to an a c t u a l l a n d s l i d e i n -ventory. T h i s e v a l u a t i o n w i l l determine which types of l a n d s l i d e s are e x t r a c t e d most r e l i a b l y . The erroneous i d e n t i f i c a t i o n s (commissions) w i l l a l s o be examined i n order to determine how they may be avoided i n f u t u r e a p p l i c a t i o n s of t h i s approach. 13 CHAPTER 2 MATERIALS AND METHODS 2.1 S i t e d e s c r i p t i o n The study s i t e c o n s i s t s of the Cascade, N o r r i s h , and Deroche creek watersheds as w e l l as an o u t l y i n g r e g i o n known 2 as H a t z i c woodlot and occupies approximately 150 km . The s i t e i s p r i m a r i l y u n d e r l a i n by g r a n i t i c r o c k s . Present l a n d -forms are s t r o n g l y i n f l u e n c e d by the l a t e s t ( F r a s e r ) g l a c i a -t i o n . These e f f e c t s i n c l u d e oversteepened v a l l e y s l o p e s that are o v e r l a i n with a mantle of g l a c i a l t i l l and c o l l u v i u m whose depth i n c r e a s e s downslope, and g l a c i a l and g l a c i a l -f l u v i a l t e r r a c e s i n the v a l l e y bottoms. I t i s l o c a t e d 65 km east of Vancouver, B r i t i s h Columbia, Canada, and i s d i r e c t l y east of Stave Lake and north of the F r a s e r River (see Fi g u r e 2.1). The s i t e i s d e s c r i b e d i n more d e t a i l by Howes (1987). 2.2 B a s e l i n e l a n d s l i d e i n v e n t o r y f o r the study s i t e 2.2.1 Compilation Howes (1987) used a s t a t i s t i c a l approach to a l a n d s l i d e i n v e n t o r y and t e r r a i n e v a l u a t i o n scheme on t h i s s i t e of the type i n t r o d u c e d by Sondheim and R o l l e r s o n (1985). A d e t a i l e d i n v e n t o r y of the l a n d s l i d e events (Howes, 1985b), as we l l as a t e r r a i n map (Howes, 1985a) was made f o r that study. The process r e s u l t e d i n a hazard r a t i n g map (Howes, 1985c). T h i s i n v e n t o r y c o n s i s t e d of l a n d s l i d e s evident on 1:15000 and 1:20000 black and white, v e r t i c a l photography from 1940, 14 F i g u r e 2.1. L o c a t i o n map of study s i t e 65 km east of Vancouver ( a f t e r Howes, 1987). S t u d y Area g § 15 1962, 1968, 1979, and 1981. Besides l a n d s l i d e l o c a t i o n , these photos provided i n f o r m a t i o n about the approximate date of l a n d s l i d e occurrence and of f o r e s t harvest i n the area. 2 A l l l a n d s l i d e s i n the i n v e n t o r y had a minimum area of 100 m and were i d e n t i f i e d by e s t a b l i s h e d techniques (Howes, 1981). A l a n d s l i d e i n v e n t o r y map was assembled by Howes (1987) from the photos using a 1:20000 base map (photo-enlarged from two 1:50000 NTS topographic sheets) and an e p i d i a s c o p e , a device f o r photo-to-map t r a n s f e r (Thomson, 1987). The in v e n -to r y i n c l u d e d d e t a i l e d i n f o r m a t i o n about s i t e c h a r a c t e r i s -t i c s , such as the year of the photography on which the l a n d -s l i d e f i r s t appeared, f o r each l a n d s l i d e . Twenty percent of a l l l a n d s l i d e s were f i e l d - c h e c k e d (Howes, 1987). 2.2.2 Data a d a p t a t i o n One problem with these data i s that Howes' minimum s i z e 2 c r i t e r i a of 100 m was not r e a l i s t i c f o r the lower r e s o l u -t i o n TM imagery. A new minimum s i z e c r i t e r i a was developed based on the Landsat TM r e s o l u t i o n of 30 m: 1. Minimum le n g t h 150 m. 2. Minimum d i s t a n c e from another l a n d s l i d e 100 m. The f i r s t c r i t e r i o n was based on the assumption that a completely unvegetated s l i d e set at a diago n a l to the p i x e l g r i d would d i r e c t l y i n f l u e n c e p i x e l values i n a minimum s t r i n g of three p i x e l s l e n g t h . Other o r i e n t a t i o n s would i n c r e a s e p i x e l s t r i n g l e n g t h . The second c r i t e r i o n was based on an assumption of the scanner's a b i l i t y to r e s o l v e two 16 d i s t i n c t , p a r a l l e l l i n e a r o b j e c t s and i s considered a con-s e r v a t i v e estimate. L a n d s l i d e groups were developed accord-i n g l y . Given these c r i t e r i a , Howes' l i s t of 595 l a n d s l i d e s f o r the study area was reduced to 183 s i n g l e and combination "events" i n an event c a t a l o g u e . The catalogue c o n s i s t e d of 84 l a n d s l i d e groups that were cre a t e d a c c o r d i n g to the c r i -t e r i a d e s c r i b e d above, and which were recorded as s i n g l e events, as w e l l as 99 unique l a n d s l i d e events. They w i l l be r e f e r r e d to as events throughout the r e s t of t h i s paper to avoid c o n f u s i o n with the in v e n t o r y compiled by Howes. A r e g i o n of i n t e n s e l a n d s l i d e a c t i v i t y i n the Cascade Creek watershed (CA-48 to CA-64 i n Howes' i n v e n t o r y ) was e l i m i n a t e d from the study at t h i s p o i n t . T h i s r e g i o n con-t a i n e d a c o l l e c t i o n of approximately 20 l a n d s l i d e s which are predominately l e s s than 100m a p a r t . Any o b j e c t s detected i n t h i s area i n the course of the study were not considered i n any of the t a b u l a t i o n s . A l l 183 events i n the catalogue were c h a r a c t e r i s e d by the f o l l o w i n g a t t r i b u t e s : 1. Photo year of f i r s t appearance 2. Photo year of harvest of surrounding s i t e 3. Length 4. Width 5. Aspect 6. Type of event The f i r s t two parameters were obtained from Howes' i n v e n t o r y and the harvest map provided by him. Parameters 3 and 4 were 17 d e r i v e d from t h e i r appearance on the o r i g i n a l i n v e n t o r y map. T h i s was p o s s i b l e s i n c e the l a n d s l i d e s were d e p i c t e d on the map to s c a l e . The method i s j u s t i f i e d by the f a c t t h a t l e n g t h and w i d t h were r e q u i r e d i n o n l y 2 c l a s s e s each. The c l a s s boundary between l o n g and s h o r t events was 300 m, and the boundary between wide and narrow events was 20 m. These f i g u r e s d i v i d e the data i n t o a p p r o x i m a t e l y even s u b s e t s , and are c o n s i s t e n t w i t h p r e v i o u s e f f o r t s to c a t e g o r i s e s l o p e movement s i z e ( R o l l e r s o n , 1987). An average of t h r e e s e p a r a t e measures of w i d t h ( r e p r e s e n t i n g the e s t i m a t e d l o c a t i o n s of the e r o s i o n , secondary d e p o s i t i o n , and primary d e p o s i t i o n zones) was taken as the w i d t h of the e v e n t . For a grouped event, the dominant l a n d s l i d e i n the group ( y o u n g e s t , l a r g e s t ) was measured. Parameter 5 was a l s o taken o f f of the i n v e n t o r y map, and parameter 6 was i d e n t i f i e d f o r each event by Howes. Appendix C p r o v i d e s the l i s t s of t h e s e a t t r i b u t e s f o r a l l events c o n s i d e r e d . Other a t t r i b u t e s , such as r e v e g e t a t i o n s t a t e of the l a n d s l i d e and of c l e a r c u t s , are of i n t e r e s t , but are much more d i f f i c u l t t o q u a n t i f y o b j e c t i v e l y . The l i s t e d c h a r a c -t e r i s t i c s are a n a l y t i c a l , and may be used to i n f e r more s u b j e c t i v e p r o p e r t i e s of the e v e n t s . The l o c a t i o n of each event was t r a n s f e r r e d from the 1:20000 map to a s e t of x and y c o o r d i n a t e s u s i n g the T e r r a s o f t g e o g r a p h i c i n f o r m a t i o n system ( D i g i t a l Resource Systems, L t d , Nanaimo, BC) i n s t a l l e d at the F a c u l t y of 18 F o r e s t r y . T h i s l i s t i n g was exported to the VAX 11/780 at the UBC Laboratory f o r Computational V i s i o n (LCV), Department of Computer Science, and read i n t o a p l o t f i l e format compatible with the software o p e r a t i n g the Raster Technolo-gies One (RasterTech) image a n a l y s i s system i n s t a l l e d at the LCV. T h i s f i l e was read from the RasterTech as a standard 512 X 512 image f i l e f o r use i n the study r e p o r t e d i n Appen-dix A. 2.3 S a t e l l i t e imagery 2.3.1 Image d e s c r i p t i o n A Landsat-5 Thematic Mapper, d i g i t a l , seven band quarter-scene was acquired by the BC M i n i s t r y of Environment and Parks i n the summer, 1986 f o r use i n s t u d i e s i n t h i s r e g i o n . The image was obtained on J u l y 29, 1985 (Track 47, Path 26) and was c l o u d - f r e e f o r the study s i t e . A 512 X 512 window of t h i s image was obtained from the M i n i s t r y and t r a n s f e r r e d to the LCV f o r use t h e r e . T h i s window contained the e n t i r e study s i t e . 2.3.2 Image band s e l e c t i o n S i m p l i c i t y d i c t a t e d the s e l e c t i o n of one of the seven a v a i l a b l e s p e c t r a l bands f o r production of the edge images. The s e l e c t i o n was based on determining which band provided the g r e a t e s t c o n t r a s t a c r o s s the l a n d s l i d e t r a c k s . A s e r i e s of t r a n s e c t s was made across s e v e r a l known l a n d s l i d e s with v a r i o u s c h a r a c t e r i s t i c s i n the image. The r e s u l t s from three 19 of these t r a n s e c t s i s shown i n F i g u r e 2.2. These t r a n s e c t s r e p r e s e n t t h r e e d i s t i n c t f o r e s t c a t e -g o r i e s of i n t e r e s t through which the l a n d s l i d e s pass: p o s t -1968 h a r v e s t , pre-1968 h a r v e s t , and no h a r v e s t . T h i s i n f o r -mation was taken from a h a r v e s t h i s t o r y map p r o v i d e d by Howes. T h i s map shows the a r e a of new h a r v e s t e v i d e n t on each s e t of photos (1940, 1962, 1968, 1979, and 1981). P r e -1968 i n d i c a t e s t h a t an event or h a r v e s t f i r s t appeared on 1968 ( o r e a r l i e r ) p h otos, and post-1968 i n d i c a t e s i t f i r s t appeared on 1979 ( o r l a t e r ) p h o tos. Areas h a r v e s t e d p r e v i o u s to 1968 are assumed to have r e g e n e r a t e d s u f f i c i e n t l y such t h a t bare s o i l and s l a s h are masked. Note the s i m i l a r i t y i n appearance of the pre-1968 h a r v e s t and u n h a r v e s t e d t r a n -s e c t s . T h i s s i m i l a r i t y s u p p o r t s the a s s e r t i o n of s u f f i c i e n t r e g e n e r a t i o n i n t h e s e a r e a s and i s used as the b a s i s f o r g r o u p i n g t h e s e c a t e g o r i e s t o g e t h e r f o r the a n a l y s i s p r e s e n t -ed i n Chapter 4. A c c o r d i n g t o F i g u r e 2.2, TM band 3, c o r r e s p o n d i n g ap-p r o x i m a t e l y to r e d l i g h t , and band 2, c o r r e s p o n d i n g to green l i g h t , r e s u l t e d i n the h i g h e s t c o n t r a s t s t e p a c r o s s the l i n e a r o b j e c t s i n the unenhanced image. S t r o n g e r s t r i p i n g , shadows, and haze i n band 2 were d i s a d v a n t a g e s to i t s c a n d i -dacy. Band 3 was t h e r e f o r e s e l e c t e d as the source f o r a l l edge images. 20 Figure 2.2. Results from three transects recorded across known landslide tracks. Curves indicate pixel values at each transect location by band. 250 Pre-1968 Harvest Event 2 1 K 200-160-100-60-n r 2 8 4 -r 6 6 Transect Pixel 260-i * 200-| 160-°" 10-1 0 Post-1968 Harvest Event 161 ~[ 1 1 1 1 1 1 2 3 4 6 6 Transect Pixel 260 0 200 D J 150-1 1 0 0 °- 60-Unharve8ted Event 13 2 T ~T 1 8 4 6 6 Transect Pixel L e g e n d Bond 1 Bond 2 Band 3 Bond 4 Band 5 Band 7 Note: 'Pre-1968 harvest' indicates that the area around the landslide was f i r s t noted as harvested on the 1968 (or earlier) photos. Post-1968 indicates that harvest was f i r s t noted on the 1979 (or later) photos. 21 2.3.3 Edge d e t e c t i o n An "edge" i s the boundary between two regions of d i f f e r e n t , constant gray l e v e l (Davis, 1975). T h e r e f o r e , an edge o p e r a t i o n i s a d i g i t a l f i l t e r i n g process which, when performed on the image, emphasizes a s p e c i f i c category of c o n t r a s t edges (Swain and Davis, 1978). By d e f i n i t i o n , these o p e r a t i o n s produce new, enhanced images. Since t h i s study i s d i r e c t e d to d e t e c t i o n and e x t r a c t i o n of low r e s o l u t i o n l i n -ear o b j e c t s , the edge o p e r a t i o n s that are i n v e s t i g a t e d pro-duce images where these f e a t u r e s are emphasised. These images may then be used f o r the process of object e x t r a c -t i o n . They are r e f e r r e d to as enhanced images (or simply edge images) i n t h i s paper. Three d i v e r s e approaches were used to provide object candidates f o r the methodology. In g e n e r a l , these are d i f -f e r e n t i a l o p e r a t o r , template matching, and threshold/tem-p l a t e matching. 2.3.3.1 D i f f e r e n t i a l operator A d i f f e r e n t i a l operator i s a point o p e r a t i o n which pro-v i d e s a score based on a mathematical assessment of a s p e c i -f i c l o c a l c o n t r a s t c o n d i t i o n . The c o n d i t i o n i s recognised as an abrupt change i n gray l e v e l i n a c e r t a i n d i r e c t i o n ' from a p o i n t . The most u s e f u l of these, where scene knowledge i s l i m i t e d , i s an i s o t r o p i c o p e r a t o r . The L a p l a c i a n i s an example of such an operator (Rosenfeld and Weszka, 1976). A p r a c t i c a l study of the r e s u l t s of a 3X3 L a p l a c i a n o p e r a t i o n 22 showed that t h i s a l s o performed w e l l i n f i n d i n g and empha-s i z i n g high b r i g h t n e s s , low r e s o l u t i o n l i n e a r o b j e c t s i n an image. An example of t h i s phenomenon i s dep i c t e d i n F i g u r e 2.3. Consequently, an image was produced using t h i s L a p l a c -i a n f i l t e r on the raw image ( F i g u r e 2.4). I t should be noted that where the L a p l a c i a n has been mentioned elsewhere (Marr and H i l d r e t h , 1980; Majka, 1982), a Gaussian smoothing has i n v a r i a b l y been s p e c i f i e d . The purpose of t h i s smoothing i s to reduce the e f f e c t of image noise and to make the image s p e c i f i c f o r an intended c l a s s of edges. The s i n g l e p i x e l noise encountered i n Landsat imagery i s i n t e r p r e t e d as a c o n t r a s t edge by a 3X3 L a p l a c i a n f i l t e r . Since the l i n e a r o b j e c t s i n t h i s study are g e n e r a l l y d e p i c t e d on the image as 3 p i x e l s wide or l e s s , i t was determined that the Gaussian smoothing would produce a map-ping that was q u i t e c o n f u s i n g . In e f f e c t , the f e a t u r e s sought i n t h i s study c l o s e l y resemble the noise that the L a p l a c i a n i s sub j e c t t o , and e f f o r t s to suppress t h i s noise are c o u n t e r - p r o d u c t i v e . The purpose of the study i s , t h e r e f o r e , to examine the e f f e c t i v e n e s s of the methodology i n s o r t i n g the o b j e c t s from the noise i n such an image. Another d i f f e r e n t i a l operator, known as the Sobel gra-d i e n t , was considered f o r use i n t h i s study. L i k e the L a p l a -c i a n , i t i s an operator which c a l c u l a t e s a two-dimensional d e r i v a t i v e at each image point (McManis, 1987). U n l i k e the L a p l a c i a n , however, i t i s a maximum at the edge of a low 23 F i g u r e 2.3. A demonstration of the L a p l a c i a n operator used i n the study. Examples of a simple c o n t r a s t edge and a low r e s o l u t i o n l i n e a r are shown. a) Abrupt, gray s c a l e edge. 10 100 100 10 100 100 L a p l a c i a n value = 91. 10 100 100 b) High b r i g h t n e s s , low r e s o l u t i o n l i n e a r o b j e c t . 10 100 10 10 100 10 L a p l a c i a n value = 262. 10 100 10 L a p l a c i a n k e r n e l : -0.7 -0.5 -0.7 -0.5 4.8 -0.5 -0.7 -0.5 -0.7 r e s o l u t i o n l i n e a r o b j e c t , not at the c e n t r e . T h e r e f o r e , i t would produce a double bar at a low r e s o l u t i o n l i n e a r ob-j e c t . T h i s i d e a l was not r e a l i z e d , however, i n the complex Landsat image and attempts to t r a n s l a t e Sobel image l i n e a r r e p r e s e n t a t i o n s i n t o s i n g l e bar tokens were f r u s t r a t e d by the v a r i a b i l i t y i n image v a l u e s . For that reason, t h i s operator was not considered f u r t h e r . 2.3.3.2 Template Matching I f an o b j e c t has a known geometry, i t i s p o s s i b l e to c o n s t r u c t a template t h a t , when scanned over the image, w i l l 24 Figure 2.4. The Laplacian edge image used in the study. The 3X3 kernel shown in is in Figure 2.4 was used to calculate this image. .» I N -a . £<t ^ "4 -25 a s s i g n a s u i t a b l e score to occurrences of a gray scale/shape combination i n the image which f i t s the d e s c r i p t i o n . I f the d i r e c t i o n of the object i s not s p e c i f i e d , a s e r i e s of o r i e n -t a t i o n s to the template may be presented. One example i s the Duda Road Operator d e s c r i b e d by F i s c h l e r e_t a_l. (1981). T h i s operator looks f o r segments of three p i x e l ' s l e n g t h i n the two c a r d i n a l d i r e c t i o n s and the d i a g o n a l s . I t i s d e s c r i b e d as a Type I operator i n t h e i r terminology because i t i s prone to e r r o r s of omission. T h i s may be p a r t l y a t t r i b u t e d to i t s i n a b i l i t y to recognise segments not w i t h i n these d i r e c t i o n a l c o n s t r a i n t s (those three p i x e l segments which would be represented as a " j o g " , i n s t e a d of a l i n e a r a l i g n -ment). Two out of the four masks used i n t h i s approach are shown i n F i g u r e 2.5. A review of the c a l c u l a t i o n s a s s o c i a t e d with t h i s f i l t e r i s omitted here. Since i n f o r m a t i o n a n c i l l a r y to the image data ( i . e . topography) i s a v a i l a b l e i n t h i s study, the d i r e c t i o n a l l i m i t s which are evident i n the above example could be r e l a x e d i n an e f f o r t to reduce p o t e n t i a l omissions. A l a n d -s l i d e template r e f l e c t i n g t h i s concept i s shown i n F i g u r e 2.6. I t i s c o n c e p t u a l l y d i f f e r e n t from the road operator s i n c e c a l c u l a t i o n s f o r a l l d i r e c t i o n s are made during a s i n g l e pass of the f i l t e r . Note that the d i r e c t i o n a l con-s t r a i n t s are r e l a x e d i n comparison to the road operator s i n c e the f o u r , two p i x e l segments (which the operator examines at each p o i n t ) represent a l l p o s s i b i l i t i e s (see Appendix D f o r f u r t h e r d i s c u s s i o n of t h i s t o p i c ) . The con-26 F i g u r e 2.5. The Duda Road Operator as presented by F i s c h l e r et a l . (1981). T h i s i s an example of a template matching scheme. Note that a t o t a l of four o p e r a t i o n s i s r e q u i r e d to produce the f i n a l "edge" image. The score f o r the a2 p i x e l i s taken as the maximum of these four t r i a l s . b l b2 b3 b2 b2 b3 a3 a l a2 a3 a2 a l c3 c l c2 c3 c l c2 F i g u r e 2.6. The l a n d s l i d e operator developed f o r the study. Note that a l l four d i r e c t i o n s are considered during one pass. Appendix D addresses an o v e r s i g h t i n t h i s template. A B c D E SC0RE1 = L - (B+V)/2 + M - (W+C)/2 F G H I J SC0RE2 = G - (P+D)/2 + M - (U+E)/2 K L M N 0 SC0RE3 = H - (F+J)/2 + M - (K+0)/2 P Q R S T SC0RE4 = I - (B+T)/2 + M - (A+Y)/2 U V W X Y FINAL SCORE = MAXIMUM OF 4 SCORES. 27 t r a s t measurement i s taken at a one p i x e l d i s t a n c e , as i n the road operator, so that edge, or m i x e l , e f f e c t s near the boundary of the d e s i r e d l i n e a r are reduced. The highest score from the four d i r e c t i o n s i s taken as the output value at the c e n t r e . The r e s u l t i n g image used i n t h i s study i s shown i n F i g u r e 2.7. 2.3.3.3 Threshold/template matching A simple image i n t e n s i t y t h r e s h o l d may be s u f f i c i e n t to d e l i n e a t e a f e a t u r e when that f e a t u r e i s p r i m a r i l y d i s t i n c t f o r i t s presence at a p a r t i c u l a r i n t e n s i t y range to the general e x c l u s i o n of other image f e a t u r e s . T h i s method has been employed f o r the e x t r a c t i o n of low r e s o l u t i o n l i n e a r o b j e c t s ( F i s c h l e r et_ <al. , 1981), and i s q u i t e f l e x i b l e i n i t s a p p l i c a t i o n . A histogram of image i n t e n s i t y values i s u s e f u l i n determining the gray l e v e l at which to t h r e s h o l d , e s p e c i a l l y i n an image dominated by two o b j e c t s (Weszka, 1978). Although a Landsat scene i s much more complex, a histogram may s t i l l guide a t h r e s h o l d i n g procedure. I f we assume that the image histogram mode (see F i g u r e 2.8) r e p r e s e n t s the dominant f e a t u r e i n the scene ( i n t h i s case old-growth f o r e s t ) , then p i x e l values i n l a n d s l i d e t r a c k s of the type under c o n s i d e r a t i o n should l i e to the r i g h t of that value, s i n c e they are g e n e r a l l y b r i g h t e r i n t h i s band due to s o i l exposure. Edge e f f e c t s from surround-ing v e g e t a t i o n c e r t a i n l y play a r o l e i n the s i g n a t u r e of these p i x e l s , and the l a n d s l i d e s occur i n both harvested and 28 Figure 2.7. The landslide template image used in the study. The raw s a t e l l i t e image was used in producing this image using the landslide template. 29 F i g u r e 2.8. Hi s t o g r a m of image band 3 of the study s i t e . Arrows i n d i c a t e t h r e s h o l d v a l u e s used t o produce the t h r e s h -o l d / t e m p l a t e matching image. 30000 -i P i x e l V a l u e 30 unharvested areas i n v a r i o u s s t a t e s of r e g e n e r a t i o n , produc-ing f u r t h e r e f f e c t s . T h e r e f o r e , four t h r e s h o l d s (of raw image p i x e l values) were s p e c i f i e d i n order to emphasize l a n d s l i d e s i n v a r i o u s c o n d i t i o n s : 25, 30, 40, 50 (mode = 20). The l a n d s l i d e template ( d e s c r i b e d above) was run on each thre s h o l d e d image i n order to i s o l a t e the l i n e a r ob-j e c t s i n the t h r e s h o l d e d images from o b j e c t s of other shapes. The r e s u l t i n g images were summed to produce a f i n a l , b i n a r y image f o r use i n the methodology (see Figure 2.9a-e). 2.4 D i g i t a l topography Topographic contours from the 1:20000 base map were d i g i t i s e d i n t o two f i l e s (due to software l i m i t a t i o n s ) , and were subsequently t r a n s l a t e d to a r e g u l a r 40m g r i d network (due to software l i m i t a t i o n s ) using the T e r r a s o f t system. T h i s g r i d was exported to LCV, where the g r i d was resampled to 30m to match the approximate p i x e l dimension of the TM scene, and thus could be manipulated as an image f i l e . D i s p l a y of the s i t e from a " s y n t h e t i c " view, as shown i n F i g u r e 2.10, i s an example of one such ma n i p u l a t i o n . T h i s image was produced using the " s y n t h e t i c " program at the LCV which assumes a Lambertian s u r f a c e and i n which the sun angle and l o c a t i o n may be s p e c i f i e d . o The s y n t h e t i c view, with the sun at 45 above the h o r i z o n and at the northwest, r e v e a l e d three problems with the gridded topography: a "stepped" e f f e c t on the h i l l -31 F i g u r e 2 . 9 a - d . The n o r t h w e s t q u a r t e r o f t h e s c e n e ( i m a g e band 3 ) t h r e s h o l d e d a t 4 p i x e l v a l u e s , a ) 25 b ) 30 c ) 40 d ) 5 0 . N o t e how d i f f e r e n t l i n e a r o b j e c t s a p p e a r a t d i f f e r e n t t h r e s h o l d s . 32 F i g ure 2.9e). The t h r e s h o l d / t e m p l a t e matching image used i n the s t u d y . T h i s i s a composite of the 4 t h r e s h o l d images (each reduced to l i n e a r o b j e c t s u s i n g the t e m p l a t e d e s c r i b e d i n S e c t i o n 2.3.3.2). 33 Figure 2.10. Synthetic view of the d i g i t a l elevation model of the stud^ s i t e . A Lambertian surface i s assumed and the sun i s at 45 above the horizon in the northwest. Note the "stair s t e p s " along the cardinal axes which would cause s i g -n i f i c a n t errors given a simple, point slope c a l c u l a t i o n . 34 slopes, a "T" appearance at the summits (not evident in Figure 2.10), and an abrupt edge at the mating of the two sections. The improper mate l i n e i s an a r t i f a c t of a regis-t r a t i o n problem at the d i g i t i s e r , and represents some loss of data and geometric i n t e g r i t y . Since minor geometric prob-lems can be accommodated in the image to map r e g i s t r a t i o n process and the loss of data i s r e s t r i c t e d to a small per-centage of the study area, t h i s problem did not necessitate re-entry of the data. The f i r s t two may be explained as a r t i f a c t s of the gridding routine that was used. A weighted average i s used in the routine for interpolating grid i n t e r -section values from grid/contour intersection values ( D i g i -t a l Resource Systems Ltd, 1987). This i s not a rigorous evaluation of the physical relationship of these points. The summit "T" i s caused by the consideration of contour i n t e r -sections along grid l i n e s only when grid point values are being interpolated and the improper weighting of these values in the interpolations. The "T" ramps follow these grid l i n e s . The "stepped" h i l l s l o p e i s of greater conse-quence to thi s study. It i s due to the improper weighting of cross-slope contour intersections in the interpolation pro-cess. Since no other gridding algorithm was available at the time of thi s study, a method was sought that would mitigate the adverse effect that the poorly gridded data would have on calculated slopes. Mode f i l t e r s of various sizes produced u n r e a l i s t i c a r t i f a c t s at ridge and valley locations and were 35 deemed u n s u i t a b l e . An " i n t e l l i g e n t " averaging f i l t e r was devised f o r producing a workable slope f i l e . The slope a l g o r i t h m which was used i s shown i n F i g u r e 2.11. I t i s based on the o b s e r v a t i o n that the steps i n the s y n t h e t i c view were between 1 and 2 p i x e l s long and always set a c r o s s the slope along the c a r d i n a l axes. The d i f f e r e n c e between A and B and between C and D was obtained, and the r a t i o of these d i f f e r e n c e s was taken as the tangent of an angle from which a f a l l l i n e azimuth (AZIMUTH) was d e r i v e d . The slope g r a d i e n t across each p i x e l was estimated by weighting the drop across four p i x e l s i n each c a r d i n a l d i r e c t i o n a c c o r d i n g to the AZIMUTH v e c t o r . Two coded r a s t e r f i l e s were crea t e d from t h i s e x e r c i s e . The f i r s t contained an i n t e g e r value at each p i x e l l o c a t i o n which represented the estimated drop i n metres across that p i x e l , and the second contained i n t e g e r values from one to nine which i n d i c a t e d the p i x e l through which the AZIMUTH vector passed ( i . e . the p i x e l which was d i r e c t l y d o w n h i l l of the one under c o n s i d e r a t i o n ) . The code format f o r t h i s second f i l e i s shown i n F i g u r e 2.12. These two f i l e s were the source of program a r r a y s DROP and DIRECTION r e s p e c t i v e l y (as d e s c r i b e d i n S e c t i o n 3.2). 36 F i g u r e 2.11. The s l o p e adjustment o p e r a t o r used t o e s t i m a t e p o i n t s l o p e v a l u e s from the d i g i t a l e l e v a t i o n d a t a . A c D B FALL LINE CALCULATION ANGLE - ARCTAN(A-B)/(C-D) (Where: C-D ? 0) I f C-D > 0 AZIMUTH = 90o + ANGLE I f C-D < 0 o AZIMUTH = 270 + ANGLE SLOPE CALCULATION o DROP = (|A-B|*( I ANGLE I/90 ) +o IC-D|*(1 - (|ANGLE|/90 ) ) ) / 4 DEFINITIONS: ANGLE i s an i n t e r m e d i a t e v a l u e f o r f a l l l i n e c a l c u l a t i o n AZIMUTH i s the azim u t h v a l u e of the f a l l l i n e v e c t o r . DROP i s the e s t i m a t e d drop a c r o s s the c e n t r e p i x e l i n metres . 37 F i g u r e 2.12. Code format f o r the f a l l l i n e v e c t o r f i l e . An i n t e g e r value from 1 to 9 at p i x e l X i n d i c a t e s the d i r e c t i o n of the f a l l l i n e a c c o r d i n g to t h i s code. 1 2 3 4 X 6 7 8 9 38 CHAPTER 3 IMPLEMENTATION The study was carried out in two parts. The f i r s t part demonstrates the usefulness of an edge enhanced image for detecting landslides using a linear segment object d e f i n i -tion without elevation information. Since i t was only i n -tended to examine the v a l i d i t y of using such images in an automated methodology, and does not constitute a viable method of meaningful image segmentation by i t s e l f , i t i s presented as an appendix (Appendix A) to this paper. The automated methodology presented here sorts the segments representing the landslides from the segments re-presenting other objects in three edge images by integrating a DEM into an automated technique. Results show the type of landslides which are sorted best, while indicating the shortcomings of this method in terms of data quality and model s i m p l i c i t y . 3.1 Introduction The Landsat image was registered to the DEM in three sections: northwest section, northeast section, and south section. Control points for thi s r e g i s t r a t i o n included v a l -ley road intersections, r i v e r crossings and points, and other features represented on the map. The three edge opera-tions as described in Chapter 2 were performed on this image. 39 Four c r i t e r i a f o r l a n d s l i d e d e t e c t i o n were s p e c i f i e d f o r each edge image: 1. P i x e l value above a s p e c i f i e d v a l u e . 2. Minimum segment le n g t h i n p i x e l s . 3. Minimum slope at each p i x e l . 4. Candidate segments o r i e n t e d along the f a l l l i n e . These c r i t e r i a were employed i n an image a n a l y s i s program w r i t t e n i n the C programming language at the LCV. 3.2 Image e v a l u a t i o n program Three f i l e s were necessary f o r program o p e r a t i o n : slope d i r e c t i o n (DIRECTION), slope magnitude (DROP) and the edge image under c o n s i d e r a t i o n (IMAGE). Three user s p e c i f i c a t i o n s were a l s o necessary to guide each t r i a l : image t h r e s h o l d v a l u e , minimum s l o p e , and minimum segment l e n g t h . Once input f i l e s were read i n t o program a r r a y s (DIRECTION, DROP, and IMAGE) and the edge image was t h r e s h o l d e d a c c o r d i n g to spec-i f i c a t i o n , program o p e r a t i o n commenced as two c o n s e c u t i v e e x e r c i s e s : segment i d e n t i f i c a t i o n and segment le n g t h ap-p r a i s a l . The pseudocode fragments i n F i g u r e s 3.1 and 3.2 h i g h l i g h t these two processes r e s p e c t i v e l y . F i g u r e 3.3 pro-v i d e s d e f i n i t i o n s f o r the terms used i n F i g u r e s 3.1 and 3.2. In segment i d e n t i f i c a t i o n , each candidate p i x e l (an edge image p i x e l whose value i s g r e a t e r than the s p e c i f i e d t h r e s h o l d ) i s evaluated a c c o r d i n g to i t s neighbours' v a l u e s . I f the neighbour which i s downhill from i t i s a l s o a c a n d i -date, then the p i x e l l o c a t i o n i s assigned a value of 4 i n 40 F i g u r e 3.1. Pseudocode f o r segment i d e n t i f i c a t i o n . i n i t i a l i z e array d e s i g n a t o r s r and c; begin loop, increment c by 1; i f (IMAGE[r][c] i s equal to 1) sol v e f o r r l and c l using DIRECTION[r][c]; i f ( I M A G E [ r l ] [ c l ] i s equal to 1 and DR0P[r][c] i s gre a t e r than minslope) s o l v e f o r upslope as an i n v e r s e of DIRECTION[r][c ] ; so l v e f o r r2 and c2 using upslope; i f (IMAGE[r2][c2] i s equal to 1 and DR0P[r2][c2] i s gre a t e r than minslope) set INTER[r][c] equal to 3; otherwise, set INTER[r][c] equal to 4; otherwise, set INTER[r][c] equal to 2; otherwise, set INTER[r][c] equal to 0; i f (c i s l e s s than the width of the image) r e t u r n to beginning of loop; otherwise, increment r by 1; i f ( r i s g r e a t e r than the le n g t h of the image) jump out of loop; otherwise, r e t u r n to beginning of loop; end . See d e f i n i t i o n s i n F i g u r e 3.3. 41 F i g u r e 3.2. Pseudocode f o r segment le n g t h a p p r a i s a l . i n i t i a l i z e a r r ay d e s i g n a t o r s r and c; begin loop, i n i t i a l i z e c o u n t e r l and counter2; increment c by 1 ; i f ( INTER[r][c] i s equal to 4) set s equal to r and set t equal to c; begin loop, increment c o u n t e r l by 1; sol v e f o r r l and c l using DIRECTION[s][t ] ; i f ( I N T E R [ r l ] [ c l ] i s equal to 3) set s equal to r l and t equal to c l ; r e t u r n to beginning of loop; otherwise, jump out of loop; set s equal to r and t equal to c; begin loop, increment counter2 by 1; i f ( c o u n t e r l i s g r e a t e r than or equal to minlength) set ( O P I X [ s ] [ t ] equal to 5); otherwise, set O P I X [ s ] [ t ] equal to 0; sol v e f o r r l and c l using DIRECTION[s][t ] ; set s equal to r l and t equal to c l ; i f (counter2 i s equal to c o u n t e r l ) jump out of the loop; otherwise, r e t u r n to beginning of loop; otherwise, set 0 P I X [ r ] [ c ] equal to 0; i f (c i s l e s s than the width of the image) r e t u r n to beginning of loop; otherwise, increment r by 1; i f ( r i s gre a t e r than the le n g t h of the image) jump out of loop; otherwise, r e t u r n to beginning of loop; end. See d e f i n i t i o n s i n F i g u r e 3.3. 42 F i g u r e 3.3. D e f i n i t i o n s f o r v a r i a b l e s used i n Fi g u r e s 3.1 and 3.2. IMAGE i s an array c o n t a i n i n g edge image i n f o r m a t i o n : 1 i n d i c a t e s edge image p i x e l greater than s p e c i f i e d t h r e s h o l d . 0 i n d i c a t e s edge image p i x e l l e s s than s p e c i f i e d t h r e s h o l d . DIRECTION i s an array with member values 1 to 9. The value i n d i -cates the p i x e l which i s d i r e c t l y d o wnhill of the present one. DROP i s an array i n d i c a t i n g metres of drop across the p i x e l . INTER i s an array f o r s t o r i n g segment i d e n t i f i c a t i o n r e s u l t s . 4 i n d i c a t e s segment beginning (top of l a n d s l i d e c a n d i d a t e ) . 3 i n d i c a t e s segment middle. 2 i n d i c a t e s segment end or p i x e l that has f a i l e d the t r i a l c r i t e r i a . 0 i n d i c a t e s no segment. r & c designate array member under c o n s i d e r a t i o n . r l & c l designate array member downhill from the present one. r2 & c2 designate array member u p h i l l from the present one. a & b are used to s t o r e array d e s i g n a t o r v a l u e s . minslope i s the designated minimum l a n d s l i d e s l o p e . upslope i s used to s t o r e the i n v e r s e of s l o p e [ r ] [ c ] , OPIX i s an array f o r s t o r i n g r e s u l t s . 5 i n d i c a t e s l a n d s l i d e i d e n t i f i c a t i o n success. 0 i n d i c a t e s l a n d s l i d e i d e n t i f i c a t i o n f a i l u r e . s & t are used to s t o r e array d e s i g n a t o r v a l u e s . minlength i s the designated minimum segment l e n g t h . c o u n t e r l i s used to count segment members i n i n i t i a l e v a l u a t i o n . counter2 i s used to count segment members during opix w r i t e . 43 the i n t e r m e d i a t e r e s u l t a r ray (INTER) which i n d i c a t e s a segment beginning. I f the neighbour d i r e c t l y u p h i l l i s a l s o a candidate, then the p i x e l l o c a t i o n i s assigned a value of 3 i n INTER which i n d i c a t e s the middle of a segment. Candi-date p i x e l s which f a i l the f i r s t t e s t are assigned a value of 2. Consequently, the lowest e l e v a t i o n member of a pro-cessed segment i s i m p l i e d by the appearance of a 2 downhill from a 3 i n INTER. INTER serves as the input to the segment l e n g t h ap-p r a i s a l process. I f a 4 i s encountered i n the a r r a y , a 3 i s sought at the dow n h i l l v e c t o r l o c a t i o n , and a 3 i s subse-quently sought downhill from that l o c a t i o n , and so on, while a counter t r a c k s the number of p i x e l s i n the segment. I f the segment i s as long or longer than the minimum s p e c i f i e d at the s t a r t , then each p i x e l i s w r i t t e n as a 5 to the output array (OPIX). A l t e r n a t e l y , each l o c a t i o n i s w r i t t e n as a 0. When completed, OPIX i s w r i t t e n as an image f i l e which serves as the v i s u a l r e s u l t of the t r i a l , or a " l a n d s l i d e image." 3.3 L a n d s l i d e image production and e v a l u a t i o n procedure In a l l , 14 t r i a l s were run and 14 l a n d s l i d e images produced: 6 each f o r L a p l a c i a n and template edge images, and 2 f o r the threshold/template matching edge image. These t r i a l s are o u t l i n e d i n Table 3.1. Each t r i a l was evaluated f o r percent successes and percent commission e r r o r s before the next one was s p e c i f i e d . The study was l i m i t e d to the 44 c o n s i d e r a t i o n of 2 t h r e s h o l d s of the edge image (where ap-p l i c a b l e ) , 4 minimum slope v a l u e s , and 2 segment length v a l u e s . Two t h r e s h o l d values were determined from each edge image histogram, approximating the p i x e l value at 50% and 25% of the number of p i x e l s at the histogram peak. These histograms and corresponding t h r e s h o l d values are shown i n F i g u r e s 3.4 and 3.5. The t h r e s h o l d i n g was not r e l e v a n t f o r the threshold/template image, as i t was a l r e a d y a binary image. The 2 segment lengths of 4 and 5 were w i t h i n the range of minimum l a n d s l i d e l e n g t h under c o n s i d e r a t i o n (at l e a s t f o r c a r d i n a l o r i e n t a t i o n s ) . A value of 3 was r u l e d out as prone to commission e r r o r p r o d u c t i o n , as was the connec-t i o n of 2 p i x e l segments (as s p e c i f i e d i n Appendix A). Slope o o o o values of 10 , 15 , 20 and 25 could be considered f o r each op e r a t o r . T h i s i s based on world knowledge of a c t u a l l a n d -s l i d e occurrence where slop e s below a minimum value q u i c k l y reduce the d r i v i n g f o r c e behind a l a n d s l i d e (VanDine, 1985). Images were evaluated by v i s u a l v e r i f i c a t i o n of each segment i n the l a n d s l i d e image. The l a n d s l i d e image was d i s p l a y e d as an o v e r l a y on the s a t e l l i t e image, and image f e a t u r e s o r i e n t e d the user to the p o s i t i o n of the segments on a set of 1982, 1:40000 black and white v e r t i c a l a i r p h o t o s and on the l a n d s l i d e i n v e n t o r y map. I f a segment c o i n c i d e d i n p o s i t i o n and o r i e n t a t i o n to a l l or a p o r t i o n of a c a t a -logued l a n d s l i d e l o c a t i o n , i t was recorded as a s u c c e s s f u l i d e n t i f i c a t i o n . The i n c i d e n c e of a commission e r r o r was 45 Table 3.1 T r i a l layout f o r l a n d s l i d e image production T r i a l Operator Image t h r e s h o l d Minimum  slope (degrees) Segment  len g t h ( p i x e l s ) A L a p l a c i a n 5 25 4 B L a p l a c i a n 5 20 4 C L a p l a c i a n 5 20 5 D L a p l a c i a n 7 20 5 E L a p l a c i a n 7 15 5 F L a p l a c i a n 7 20 4 G template 2 20 4 H template 2 20 5 I template 2 25 5 J template 3 15 5 K template 3 20 5 L template 3 20 4 M thr./tem. — 15 4 N thr./tem. 0 10 4 Note: Slope i n degrees i s t r a n s l a t e d to drop across a 30 m p i x e l i n the model. 46 Figure 3 . 4 . Histogram of Laplacian image. Arrows indicate threshold values used. F i g u r e 3.5. H i s t o g r a m o f t e m p l a t e m a t c h i n g image. Arrows i n d i c a t e t h r e s h o l d v a l u e s u s e d . 4 8 r e c o r d e d s e p a r a t e l y . T r i a l s w i t h a commission r a t e of l e s s than 50% were deemed of i n t e r e s t s i n c e they are a s s o c i a t e d w i t h g r e a t e r user c o n f i d e n c e i n an o b j e c t i d e n t i f i c a t i o n system. I f commission e r r o r s were l e s s than 50% of the t o t a l number of i d e n t i f i c a t i o n s i n a t r i a l , then the appar-ent cause of the e r r o r , as i d e n t i f i e d from a i r p h o t o and image e v i d e n c e , was r e c o r d e d . The f o l l o w i n g r u l e guided the procedure f o r c h o o s i n g the parameters f o r subsequent t r i a l s on each edge image. For each t r i a l , i f the commission e r r o r s exceeded 50% of the i d e n t i f i c a t i o n s made, then the next t r i a l must not vary a parameter i n such a way t h a t an i n c r e a s e i n commission e r -r o r s c o u l d be e x p e c t e d . In t h i s way, the p r o c e s s was guided to f i n d the most r e l i a b l e o p e r a t i o n s . Two t r i a l s , each w i t h commission r a t e s of 49%, were grouped w i t h the t r i a l s over 49%. 49 CHAPTER 4 RESULTS AND DISCUSSION 4.1 Results Success and commission results for the 14 t r i a l s stud-ied are shown in Table 4.1. Trials with less than 50% com-mission rates were considered in a detailed analysis. Seven t r i a l s (D,M,K,N,E,I,J) f u l f i l l e d this requirement. Trials I and J, however, at 49%, were not considered for detailed analysis. The remaining 5 t r i a l s represent 2 each from the Laplacian and threshold images and 1 from the template image. Three properties of these t r i a l s are of interest in the evaluation of the methodology: what are the causes of the commission errors and what are the characteristics of the events that were and were not identified? Answers to these questions will indicate the usefulness of this ap-proach and point out where improvements could be made. The segments which indicate landslide locations for the 5 selected t r i a l s are displayed in black against the sa t e l l i t e scene in Figures 4.1 to 4.5. The location of each segment indicates a correct identification or a commission error. These images were reviewed in detail in order to categorise each segment. 4.1.1 Commission errors Commission errors were labeled and their probable cause indicated from photo evidence (see Appendix B). Of the 77 50 Table 4.1. Success and commission results for 14 t r i a l s . Rank by Rank by Tr i a l Operator Success Commission Commission Success % No. % No. 1 14 D Laplacian 18 33 35 18 2 11 M thr./temp. 23 42 35 23 3 9 K template 24 45 37 26 4 8 N thr./temp. 28 51 41 35 5 7 E Laplacian 28 51 41 36 6 12 I template 19 36 49 34 7 3 J template 36 67 49 65 8 6 F Laplacian 31 57 50 56 9 10 C Laplacian 24 45 51 47 10 4 H template 36 66 53 73 11 5 L template 34 62 55 76 12 1 G template 48 89 62 147 13 2 B Laplacian 37 68 70 158 14 13 A Laplacian 19 35 70 83 Trial references are in Table 3.1. Trials above horizontal line were considered in analysis. 51 F i g u r e 4.1. T r i a l D ( L a p l a c i a n o p e r a t o r ) . Northwest q u a r t e r of scene i s shown, w i t h i d e n t i f i c a t i o n s shown i n b l a c k . Compare the number of b l a c k segments here w i t h the w h i t e segments i n F i g u r e 2.5. 52 F i g u r e 4.2. T r i a l M (Threshold/template o p e r a t o r ) . Northwest quarter of scene i s shown, with i d e n t i f i c a t i o n s shown i n black. Compare black the number of black segments here with the white segments i n F i g u r e 2.10e. 53 F i g u r e 4.3. T r i a l K (Template o p e r a t o r ) . Northwest quarter of scene i s shown, with i d e n t i f i c a t i o n s shown i n black. Compare the number of black segments here with the white segments i n Fig u r e 2.8. 54 Figure 4.4. T r i a l N (Threshold/template operator). Northwest quarter of scene i s shown, with i d e n t i f i c a t i o n s shown in black. Compare the number of black segments here with the white segments in Figure 2.10e. F i g u r e 4.5. T r i a l E ( L a p l a c i a n o p e r a t o r ) . Northwest quarter of scene i s shown, with i d e n t i f i c a t i o n s shown i n black. Compare the number of black segments here with the white segments i n F i g u r e 2.5. 56 unique commission e r r o r s produced by the 5 s e l e c t e d t r i a l s , 34 were not c l e a r l y e x plained by photo evidence or were suspected of l a n d s l i d e a c t i v i t y a f t e r the i n v e n t o r y was per-formed. A subsequent f i e l d check i n d i c a t e d that 3 of these were l a n d s l i d e s which occurred i n the i n t e r v a l between i n -ventory and s a t e l l i t e image a c q u i s i t i o n , and one was an o l d , revegetated d e b r i s flow. The f i r s t 3 were added to the event catalogue f o r s t a t i s t i c a l purposes, and the f o u r t h was not added s i n c e Howes purposely omitted i t . Table 4.2a gives a breakdown of the commission e r r o r s that were made, i n c l u d i n g the harvest c o n d i t i o n of the area surrounding the commis-s i o n . Table 4.2b provides an a d d i t i o n a l e r r o r breakdown i n which commission e r r o r s are organised by f e a t u r e o r i g i n . Roads and l a n d i n g s are grouped with 'no evidence' and 'other' e r r o r s found i n r e c e n t l y harvested zones i n t o a ' d i s t u r b e d ' category, while e r r o r s due to c l e a r i n g s and ledge are grouped with 'no evidence' and other' e r r o r s found i n e a r l y harvest and o l d growth zones i n t o a ' n a t u r a l ' category. F i g u r e s 4.6 and 4.7 are examples of these e r r o r s . 4.1.2 Success c h a r a c t e r i s a t i o n The l a n d s l i d e a t t r i b u t e s e x t r a c t e d from the event c a t a -logue (see S e c t i o n 2.2.2) were used to examine the i d e n t i f i -c a t i o n r e s u l t s . Table 4.3 shows the number of l a n d s l i d e s which f a l l i n t o c a t e g o r i e s organised from these a t t r i b u t e s . These c a t e g o r i e s w i l l be used to help analyse the r e s u l t s . 57 Table 4.2a and b. Causes of commission errors. Two break-downs are presented: (a) specific cause of error, and (b) whether the cause is related to manmade (disturbed) or natural features in the landscape. Table 4.2a. Cause Harvest Total Percent pre-1968 post-1968 Number Road or landing 11 16 27 37 No evidence 6 5 11 15 Clearing 7 1 8 11 Ledge or talus 3 3 6 8 Other 6 15 21 29 Total 33 40 73 100 Table 4.2b. Cause Harvest Total Percent pre-1968 post-1968 Number Natural 22 5 27 37 Disturbed 11 35 46 63 Total 33 40 73 100 Note: The categories for Table 4.2a are outlined in Appen-dix B. Table 4.2b is described in the text. 58 Figure 4 . 6 . Commission error A -4. Large, ir r e g u l a r deposits of slash on the landscape were one contributor to commission error. 59 60 Categorical results for each t r i a l are examined in Tables 4.4 through 4.8. In these tables, four age and har-vest conditions are held constant against nine physical characteristics. The four include a l l events of a l l ages and a l l harvest conditions in order to produce an overview. Then a l l post-1968 events are examined, f i r s t in a l l harvest conditions in Section 2 of the tables, and then in post and pre-1968 harvest conditions in Sections 3 and 4 of the ta-bles. Note that unharvested areas have been grouped with the pre-1968 category, as discussed in Section 2.3.2. Northern and southern aspect locations are considered separately, and are designated by N and S in the tables. 4.2 Discussion From an operational standpoint, there are unacceptably high commission and omission rates. However, the results compare favourably to those of Gimbarzevsky (1983), where 40% of landslides present in a Landsat MSS scene were de-tected, because automated interpretation was not used in that study. Because of the range of input parameters (seg-ment length and slope, image threshold) that were used in the t r i a l s , no one operator proved to be superior to the others. This is substantiated by Table 4.1, where the three operators appear throughout the table. It seems that a l l three approaches to image enhancement are equally suited to providing candidate segments for this approach. Other compo-nents of the study, such as the method used for object ex-traction and the type and quality of data available are of 61 Table 4.3. Total number of events in each landslide category Event Type Aspect/ Char. a l l tor.S slide long short LW LN SW SN events flow Section 1 N S 72 55 17 42 30 16 26 12 18 114 80 34 75 39 29 46 12 27 Section 2 N S 39 29 10 21 18 10 11 8 10 59 43 16 38 21 18 20 8 13 Section 3 N S 10 6 4 4 6 0 4 1 5 17 12 5 10 7 3 7 3 4 Section 4 N S 29 23 6 17 12 10 7 7 5 42 31 11 28 14 15 13 5 9 Section 1: A l l events considered (total 186). Section 2: Events occurring after 1968 photos only (total 98) Section 3: Post-1968 events in an area harvested after 1968 (total 27). Section 4: Post-1968 events in an area harvested before 1968 or not harvested at a l l (total 71) LW- length > 300 m, width > 20m. LN- length > 300 m, width < 20m. SW- length < 300 m, width > 20m. SN- length < 300 m, width < 20m. 62 Table A.A. S u c c e s s f u l i d e n t i f i c a t i o n s f o r t r i a l D. A l l f i g -ures are % of t o t a l i n category Aspect/ Event Type Char. \ \ a l l tor.& s l i d e long short LW LN SW SN \ events flow S e c t i o n 1 N S 1A 11 2A 17 30 31 8 17 6 20 25 9 25 10 31 22 17 7 S e c t i o n 2 N S 23 17 AO 29 17 50 9 25 10 25 33 6 3A 10 39 30 13 8 S e c t i o n 3 N S 10 17 0 25 0 — 25 0 0 17 17 17 20 13 33 1A 33 0 S e c t i o n A N S 28 17 67 29 25 50 0 29 20 29 39 0 39 7 AO 38 0 11 S e c t i o n 1: A l l events considered ( t o t a l 186). S e c t i o n 2: Events o c c u r r i n g a f t e r 1968 photos only ( t o t a l 98). S e c t i o n 3: Post-1968 events i n an area harvested a f t e r 1968 ( t o t a l 27). S e c t i o n A: Post-1968 events i n an area harvested before 1968 or not harvested at a l l ( t o t a l 71) LW- l e n g t h > 300 m, width > 20 m. LN- l e n g t h > 300 m, width < 20 ra. SW- l e n g t h < 300 m, width > 20 m. SN- l e n g t h < 300 m, width < 20 ra. 63 T a b l e 4.5. S u c c e s s f u l i d e n t i f i c a t i o n s f o r t r i a l M. A l l f i g -u res a r e % of t o t a l i n c a t e g o r y A s p e c t / Char. a l l e vents t o r .& f l o w Event Type s l i d e l o n g s h o r t LW LN SW SN S e c t i o n 1 N 14 13 18 17 10 31 8 25 0 S 26 31 12 32 15 38 29 25 11 S e c t i o n 2 N 23 21 30 29 17 50 9 38 0 S 29 35 13 34 19 44 25 25 15 S e c t i o n 3 N 0 0 0 0 0 0 0 0 S 22 25 17 30 13 33 29 33 0 S e c t i o n 4 N S 31 31 26 39 50 9 35 36 25 21 50 47 14 23 43 20 0 22 S e c t i o n 1: A l l e v e nts c o n s i d e r e d ( t o t a l 186). S e c t i o n 2: E v ents o c c u r r i n g a f t e r 1968 photos o n l y ( t o t a l 9 8 ) . S e c t i o n 3: Post-1968 e v e n t s i n an a r e a h a r v e s t e d a f t e r 1968 ( t o t a l 2 7 ) . S e c t i o n 4: Post-1968 e v e n t s i n an a r e a h a r v e s t e d b e f o r e 1968 or not h a r v e s t e d at a l l ( t o t a l 71) LW- l e n g t h > 300 m, w i d t h > 20m. LN- l e n g t h > 300 m, w i d t h < 20m. SW- l e n g t h < 300 m, w i d t h > 20m. SN- l e n g t h < 300 m, w i d t h < 20m. 64 Table 4.6. Successful identifications for t r i a l K. A l l f i g -ures are % of total in category Aspect/ Char. a l l events tor .& flow Event Type slide long short LW LN SW SN Section 1 N S 14 31 11 34 24 24 17 29 10 33 31 38 8 24 17 25 6 37 Section 2 N S 21 39 14 44 40 25 24 40 17 48 50 44 0 25 25 25 10 62 Section 3 N S 0 22 0 33 0 0 0 20 0 25 33 0 14 0 33 0 25 Section 4 N S 28 45 17 48 67 36 29 39 25 57 50 47 0 31 29 20 20 78 Section 1: A l l events considered (total 186). Section 2: Events occurring after 1968 photos only (total 98). Section 3: Post-1968 events in an area harvested after 1968 (total 27). Section 4: Post-1968 events in an area harvested before 1968 or not harvested at a l l (total 71) LW- length > 300 m, width > 20 m. LN- length > 300 m, width < 20 m. SW- length < 300 m, width > 20 m. SN- length < 300 m, width < 20 m. 65 Table A.7. Successful identifications for t r i a l N. A l l f i g -ures are % of total in category Aspect/ Char. a l l events tor .& flow Event Type slide long short LW LN SW SN Section 1 N 19 20 2A 26 10 AA 15 25 0 S 32 38 21 AO 18 A5 38 25 15 Section 2 N 26 2A 30 33 17 50 18 38 0 S 36 A2 19 A5 19 56 35 25 15 Section 3 N 0 0 0 0 0 _ 0 0 0 S 22 25 17 30 13 33 29 33 0 Section A N S 3A Al 30 A8 50 18 Al 50 25 21 50 60 29 38 A3 20 0 22 Section 1: A l l events considered (total 186). Section 2: Events occurring after 1968 photos only (total 98). Section 3: Post-1968 events in an area harvested after 1968 (total 27). Section A: Post-1968 events in an area harvested before 1968 or not harvested at a l l (total 71) LW- length > 300 m, width > 20m. LN- length > 300 m, width < 20m. SW- length < 300 m, width > 20m. SN- length < 300 m, width < 20m. 66 Table A.8. S u c c e s s f u l i d e n t i f i c a t i o n s f o r t r i a l E. A l l f i g -ures are % of t o t a l i n category Aspect/ Event Type Char. \ \ a l l tor.S s l i d e long short LW LN SW SN \ events flow S e c t i o n 1 N S 22 20 29 26 17 44 15 25 0 29 36 12 37 13 34 40 17 11 S e c t i o n 2 N S 33 24 50 38 28 60 18 25 10 32 40 13 42 14 39 45 13 15 S e c t i o n 3 N S 20 17 25 0 33 — 0 0 40 17 17 17 20 13 33 14 33 0 S e c t i o n 4 N S 38 30 67 47 25 60 29 29 20 38 48 9 50 14 40 62 0 22 S e c t i o n 1: A l l events considered ( t o t a l 186). S e c t i o n 2: Events o c c u r r i n g a f t e r 1968 photos only ( t o t a l 98). S e c t i o n 3: Post-1968 events i n an area harvested a f t e r 1968 ( t o t a l 27). S e c t i o n 4: Post-1968 events i n an area harvested before 1968 or not harvested at a l l ( t o t a l 71) LW- l e n g t h > 300 m, width > 20m. LN- l e n g t h > 300 m, width < 20m. SW- le n g t h < 300 m, width > 20m. SN- l e n g t h < 300 m, width < 20m. 67 i n t e r e s t i n a d i s c u s s i o n of t h i s method. Because of the l i m i t e d scope of t h i s study, the d i s c u s s i o n i s l i m i t e d to a review of the ob s e r v a t i o n s and the proposal of hypotheses to e x p l a i n the t r i a l r e s u l t s . 4.2.1 Commission e r r o r s The vast m a j o r i t y of commission e r r o r s made by the 5 s e l e c t e d t r i a l s were confirmed as e r r o r s by photo and f i e l d checks. Many (39%) were segments of lo g g i n g roads or l a n d -i n g s . Since roads have the same approximate appearance (low r e s o l u t i o n l i n e a r o b j e c t ) i n the enhanced images as the d e s i r e d l a n d s l i d e t r a c k s , these commissions are a f a i l u r e of the DEM p o r t i o n of the l o g i c . I t i n d i c a t e s a c o i n c i d e n c e of DEM slope and f a l l l i n e v e c t o r f i l e e r r o r s that are e i t h e r inherent i n the data or due to r e g i s t r a t i o n problems. The low r e s o l u t i o n of the e l e v a t i o n data i s a c o n t r i b u t i n g f a c -t o r . I t i s u n c e r t a i n , however, whether the g r i d d i n g a l g o r -ithm problem d i s c u s s e d i n Chapter 2 had any bearing on these commission e r r o r s . Waste wood on the landscape i s another l o g g i n g f e a t u r e which c o n t r i b u t e d to the "no evidence and "other " c a t e g o r i e s of commission e r r o r s . These areas i n c l u d e d r i d g e s and g u l -l i e s where a great deal of s l a s h i s present (see F i g u r e s 4.8 and 4.9). The presence of waste wood could be an e x p l a n a t i o n f o r edges and short segment l i n e a r s i n the enhanced images. These areas are c h a r a c t e r i s t i c a l l y q u i t e confused i n image appearance, with frequent s k i d t r a i l s u s u a l l y o r i e n t e d down-68 slope and slash forming random patterns on the ground. These patterns likely create candidate objects that are oriented down the f a l l line. The quality of the DEM data probably has l i t t l e bearing on this commission error. Logging features constitute a significant majority of commission errors examined. In Table 4.2b, 63% of the errors from the 5 t r i a l s are attributed to man's ac t i v i t i e s . This result suggests the importance of logging features as the explanation of commission errors. The incidence of these errors would, therefore, decrease significantly i f the ap-proach was used in an unharvested areas. A less important source of error on this site (7%) was due to edges caused by talus and ledge outcrops in the land-scape. Because they indicate steep areas and thin soils, they are more likely to be found in unlogged areas. The addition of more contextual information (geology), and a more sophisticated approach to DEM information utilisation (gully locations, segment position on hillslope) could be used to refine the process and reduce this error. As discussed in Chapter 1, Fischler e_t _al. (1981) de-veloped an approach to combining the results from various edge operators based on the type of error each operator is likely to make. The level of commission error found in each of the t r i a l s in this study precluded the use of such a system in this study since there are no "Type I" operators from which to draw so-called "zero cost segments." 69 4.2.2 Identification successes and omissions The best overall detection rate for a l l of the t r i a l s was found in Section 4 of Tables 4.4 to 4.8. This segment of the landslide population is therefore of particular inter-est. This represents the set of more recent landslides which are in old growth forest or areas that were logged previous to 1968. Overall results are consistently greater than the Section 3 breakdown, although a few of the smaller categor-ies deviate from this generalisation. The comparatively small size of the Section 3 categories ( a l l but 2 contain less than 10 members) makes i t d i f f i c u l t to draw conclu-sions. However, one explanation for the above average suc-cess in Section 4 is the fact that the more recent land-slides are in an early stage of regeneration. This would result in a good spectral contrast with the surrounding forest in the image because of the vegetative and s o i l expo-sure contrast. The older surrounding forest causes a better contrast edge than recent clearcuts because the older trees are comparatively dark in this band, and s o i l exposure in the surrounding forest is minimal. Landslide geometry seems to be significant in Section 4. Long events are extracted at a higher rate in a l l but one case (south aspect, t r i a l 3). This is consistent with a greater opportunity for the model conditions to be satisfied in a longer event. Results for process type are less clear, and may reflect the small sample size of slide events. For example, slides on north aspects are consistently detected 70 th at or above the 50 percentile in Section 4. This is a relatively small category (6 events), of which 3 are younger than 1979, and a l l are characterised subjectively as unvege-tated by Howes (1985b). On the south aspect, however, the torrents are detected more often. This is consistent with expectations since the majority of torrents are in the long-er category, and that torrents, since they occur in chan-nels, revegetate more slowly than slides because of higher erosion activity. The characteristics of the omitted events are the oppo-site of the successes. The size of the omission error sug-gests improvements are needed to make this approach oper-ational. There is l i t t l e doubt that the "stairstep" effect encountered in the DEM was not fully solved by the slope operator that was employed. This would contribute substan-t i a l l y to the rejection of landslide image segments. A high accuracy DEM is a prerequisite for a system such as this one which requires fine pixel-to-pixel accuracy. The TRIM pro-ject (Terrain Resource Inventory Management) now underway in the BC Ministry of Environment and Parks promises to provide digital elevation for selected areas of the province in an irregular point format suitable for interpolation to a 25 metre grid. Given the high degree of accuracy and quality specified for this project, these data would prove quite valuable for the further investigation of this methodology. For these reasons, this study may be properly considered a preliminary one, with the absolute effectiveness of this 71 system given better data an open question. 4.2.3. Baseline inventory There are several problems with the evaluation of the methodology by comparison with this database. The f i r s t problem is whether the short category of landslides (<300 m) is defined appropriately. The specified segment lengths in the study (4 and 5 pixels) may be too long to detect the shorter events in the catalogue at some orientations to the grid, and this may prejudice the results downward in this category. Also, because landslides revegetate at differen-t i a l rates depending on morphologic section (erosion or deposition) (Miles e_t aj^. , 1984) , an event measured from a 1968 photo may present a much shorter edge segment in a 1983 satellite image as the deposition zone revegetates. These facts suggest that the "detectable" catalogue of events in the <300 m category used in this study may be optimistic. Another issue is the combination of landslides which are less than 100m apart into single "events". The advantage to doing this is that most of the landslides are accounted for, even i f i t is unlikely that many of them could be dis-cerned as individual landslides because of the low resolu-tion of the imagery. Because of their complex geometry, the low resolution linear object definition was expected to be insufficient to describe these events. However, the results suggest otherwise. The 84 'grouped' events represent 46% of the event catalogue, yet they comprised 50%, 50%, 60%, 53%, 72 and 47% of the t o t a l successful I d e n t i f i c a t i o n s for t r i a l s D, M, K, N, and E respectively. Other factors, such as an increased opportunity for the model c r i t e r i a to be f u l f i l l e d in the t r i a l s in these larger, more complex features may be an explanation for th i s outcome. 73 CHAPTER 5 CONCLUSIONS 5.1. This study The three edge operators presented in the study provide suitable candidates for low-resolution linear ob-jects in the Landsat TM image. No single operator proved superior to the others. A digital elevation model at a com-parable resolution provides valuable input to a reasoning structure for the extraction of landslide features. The implementation of such a structure in this study showed the shortcomings and advantages of such a system. A model based on a low resolution linear object defini-tion of landslide appearance in three enhanced images and a specific slope and orientation relationship to the DEM was implemented in an automated fashion, and the results were compared to a conventional inventory. The approach was sus-ceptible to commission errors (35% to 75% of identifications for a l l tr i a l s ) and identified from 17% to 47% of actual landslides in the landscape. Five t r i a l s of the methodology, chosen because of their favourable results with respect to commission errors, were examined in depth to assess the characteristics of their performance. Results from the five t r i a l s of interest ranged from 18% to 28% for successful identifications with a correspond-ing commission error rate of 35% to 41%. An examination of 74 these results showed that commission errors would be reduced on sites with less logging activity, and that certain clas-ses of landslides, specifically those in Section 4 of Tables 4.4 to 4.8, representing more recent landslides in older harvest or unharvested areas, were detected at the highest rate. Within this group, longer events (>300 meters) were detected more often than shorter ones. This is particularly true for events which measured wider (>20 metres) on the original inventory map. Torrents were detected more often than slides on south aspects, while the opposite was true on north aspects. However, the sample size for slides on north aspects was quite small (6), rendering the results for this breakdown less meaningful. These results are disappointing from an operational standpoint, but are comparable with previously reported results for discrimination (not extraction) of slope failure forms in a Landsat image of 40% of those mapped from a i r -photos (Gimbarzevsky, 1983). Better quality elevation data would provide a better idea of the absolute effectiveness of this approach. The results are more encouraging from a scene segmentation standpoint, as the vast majority of edge image candidate segments were eliminated from each scene. This is demonstrated by a visual comparison of the original enhanced images and their corresponding landslide images. 75 5.2. Research directions This study was not an exhaustive consideration of the usefulness of s a t e l l i t e imagery for detecting landslides. An interesting comparison could be made i f an experienced photo-interpreter i d e n t i f i e d the landslides he could see in the image without having previously seen the photos or i n -ventory. Even without the advantage of slope information, the interpreter would invariably produce better r e s u l t s than those seen in this study, because he can interpret scene context. Scene context, however, may be improved in the edge image of the s a t e l l i t e scene as well. Other low resolution linear objects may be suitable for extraction using similar techniques. For example, streams and roads are also empha-sized by the edge operators and have rules regarding their relationship to the elevation model. Streams are always at a low point in a perpendicular transect in the elevation model, while roads obey maximum slope rules. By contrast, hydro-electric l i n e s are persistent l i n e a r objects in the image, but have no rules regarding slope or elevation. A global segmentation of scene linears using these data could assign p r o b a b i l i t i e s to segment assignments that would help to eliminate landslide commissions that are a portion of longer road segments. This type of approach could use a variety of templates or operators with known c h a r a c t e r i s t i c s regarding s p e c i f i c l i n e a r objects in the image. Likelihoods 76 could be drawn from knowledge about the operator and object shape, geometry, relationship to DEM, and relationship to other l i n e a r s . This study has shown that this type of ap-proach may provide landslide inventories of a predictable r e l i a b i l i t y for s p e c i f i c classes of slope f a i l u r e s as well as for mapping other l i n e a r objects i n the landscape. With the increasing a v a i l a b i l i t y of d i g i t a l elevation data, the accurate segmentation of image features of interest to re-source managers w i l l benefit from techniques which success-f u l l y integrate t h i s data with imagery. 77 LITERATURE CITED Alfoldi, T.T. 1974. Regional study of landsliding in Eastern Ontario by remote sensing. Unpubl. MSc. thesis, Dept of C i v i l Eng., Univ. of Toronto. 79 pp. Bailey, R.G. 1972. Landslide hazards related to land use planning in Teton Nat'l Forest. U.S.F.S. Intermountain Region. 131 pp. Bajcsy, R. and M. Tavakoli. 1973. A computer recognition of bridges, islands, rivers, and lakes from satellite pictures. Proc. Mach. Proc. Conf., Purdue, IN. pp. 54-68. Bajcsy, R. and M. Tavakoli. 1976. Computer recognition of roads from sa t e l l i t e pictures. 2nd Int'l Conf. on Pat-tern Recognition. Ballard, D.H. 1981. Strip trees: A hierarchical representa-tion for curves. Comm. ACM, Vol. 24, No. 5. pp. 310-321. Colwell, R.N. 1983. Glossary. Manual for Remote Sensing. American Society of Photogrammetry, 16 pp. Davis, L.S. 1975. A survey of edge detection techniques. Computer Graphics and Image Processing, Vol. 4. pp. 248-270. Destival, I. and H. LeMen. 1986. Detection of linear net-works on sa t e l l i t e images. 8th Int'l conf. on Pattern Recog. Dishaw, H.E. 1967. Massive landslides. Photogram. Eng., Vol. 32, No. 6. pp. 603-609. Eliasson, P.T., L.A. Soderblom, and P.S. Chavez, j r . 1981. Extraction of topographic and spectral albedo informa-tion from multispectral images. Photogram. Eng. and Rem. Sens. Vol. 48, No. 11. pp. 1571-1579. Fischler, M.A., J.M. Tenenbaum, and H.C. Wolf. 1981. Detection of roads and linear structures in low resolu-tion aerial imagery using a multisource knowledge inte-gration technique. Computer Graphics and Image Proces-sing, Vol. 15. pp. 201-223. Gagnon, H. 1975. Remote sensing of landslide hazards on quick clays of Eastern Canada. Proc of the 10th Int'l Symp. Rem. Sens. Env., Ann Arbor, Mich, pp. 803-810. 78 Gimbarzevsky, P. 1983. Remote sensing of slope failures on the Queen Charlotte Islands. R.N.R.F. Symp. on the Appl. of Rem. Sens, to Res. Mgt., Amer. Soc. Photogram. pp. 409-413. Gresswell, S., D. Heller, and D.N. Swanston. 1979. Mass movement response to forest management in the Central Oregon Cascades. U.S.F.S. Res. Bull. PNW-84. 26 pp. Haralick, R.M., S. Wang, and D.B. E l l i o t t . 1981. Spatial reasoning to determine stream network from Landsat imagery. Proc. 6th Int'1 Conf. on Pattern Recog. pp. 502-516. Howes, D.E. 1981. Terrain inventory and geological hazards: Northern Vancouver Island. Assessment and Planning Div. Bull. 5, Terrestrial Studies Branch, B.C. Min. of Env. Howes, D.E. 1985a. Norrish-Cascade Terrain Map. Cart. Sec, Surveys and Res. Mapping Bran., B.C. Min. of Env. 1 sheet. Howes, D.E. 1985b. Norrish-Cascade Landslide Inventory Map. Cart. Sec, Surveys and Res. Mapping Bran., B.C. Min. of Env. 1 sheet. Howes, D.E. 1985c. Norrish-Cascade Clearcut Landslide and Introduced Landslide Debris Rating Map. Cart. Sec, Surveys and Res. Mapping Bran., B.C. Min. of Env. 1 sheet. Howes, D.E. 1987 (in press). A terrain evaluation method for predicting terrain susceptible to post-logging land-slide activity: a case study from the Southern Coast Mountains of British Columbia. B.C. Min. of Env. and Parks. 42 pp. Johnson, R.B., E.L. Maxwell, and J.A. McKean. 1977. Application of colour density enhancement of aerial photography to the study of slope st a b i l i t y . Proc 15th Ann. Eng. Geol. Soils Eng. Symp., Idaho State Univ., Pocatello. pp. 199-216. Majka, M.S. 1982. Reasoning about spatial relationships in the primal sketch. unpubl. MSc thesis, Dept. of Comp. Sci., Univ of Bri t . Col. 85 pp. Marr, D. and E. Hildreth. 1980. Theory af edge detection. Proc. of the Royal Soc. of London, B-207. pp. 187-217. McManis, C. 1987. Low-cost image processing. Byte, Vol. 12, No. 3. pp. 191-196. 79 Miles, D.W.R., F.J. Swanson, and CT. Youngberg. 1984. Effects of landslide erosion on subsequent Douglas-fir growth and stocking levels in the Western Cascades, Oregon. Soil Sci. Soc. Amer. Jour., Vol. 48. pp. 667-671. Neilson, T.H. and E.E. Brabb. 1977. Slope-stability studies in the San Francisco Bay region, California. Reviews in Eng. Geol., Vol. 3, Geol. Soc. Amer. pp. 235-243. Nevatia, R. and K.R. Babu. 1979. Linear feature extraction and description. Proc. of the 6th Int'l Joint Conf. on A r t i f i c i a l Intelligence, pp. 639-641. O'Loughlin, C.L. 1973. An investigation of the stability of steepland forest soils in the Coast Mountains of S.W. British Columbia. Unpubl. Phd. thesis, Dept. of Forest-ry, Univ. of Brit. Col. 147 pp. Pickup, G. and D.J. Nelson. 1984. Use of Landsat radiance parameters to distinguish s o i l erosion, stability, and deposition in arid Central Australia. Rem. Sens. Env., Vol. 16, No. 3. pp. 195-209. Poole, D.H. 1969. Slope failure forms; their identification, characteristics and distribution as depicted by selec-ted remote sensor returns. Proc. 6th Int'l Symp. on Rem. Sens, of the Env., Vol II, Ann Arbor, MI. pp. 927-965. Rib, H.T. 1978. Recognition and identification. in Land-slides: Analysis and Control. TRB Special Report 176. pp. 34-80. Rice, R.M., E.S. Corbett, and R.G. Bailey. 1969. Soil slips related to vegetation, topography, and soils in South-ern California. Water Resources Research, Vol. 5. pp. 647-659. Rice, R.M. and G.T. Foggin. 1971. Effects of high intensity storms on s o i l slippage in mountainous watersheds in Southern California. Water Resources Research, Vol. 7, No. 6. pp. 1485-1496. Rollerson, T.P. 1984. Terrain stability study, T.F.L. 44. Land Use Planning Advisory Team, Woodlands Division, Macmillan Bloedel Ltd., Nanaimo, B.C. 86 pp. Rollerson, T.P. 1987. Personal communication. Rosenfeld, A. and J.S. Weszka. 1976. Picture recognition, in Digital Pattern Recognition (K.S. Fu, ed.), Comm. and Cyber. Series Vol. 10, Springer-Verlay, N.Y. pp. 135-165. 80 Shibata, T. 1984. Application of digital terrain model in remote sensing. 7th Int'l Conf. on Pattern Recog. pp. 404-406. Sondheim, M.W. and R.P. Rollerson. 1985. Quantitative defi-nitions of stability classes as related to post-logging clearcut landslide occurrence. Proc. 9th B.C. Soil Sci. Workshop, B.C. Min. of Env. Swanson, F.J. and D.N. Swanston. 1977. Complex mass movement terrains in the Western Cascades, Oregon. Reviews Eng. Geol. Vol. 3, pp. 113-124. Swain, P.H. and S.M. Davis. 1978. Remote Sensing: The Quan-titative Approach. McGraw-Hill, N.Y. 396 pp. Tanguay, M.G. and J.Y. Chagnon. 1972. Thermal I-R imagery at the St. Jean Vianney landslide. Proc. 1st Canadian Symp. on Rem. Sens. pp. 387-402. Tavakoli, M. and A. Rosenfeld. 1982. Building and road extraction from aerial photographs. I.E.E.E. transac-tions SMC-12, Jan-Feb. pp. 84-91. Thomson, B. 1987. Personal communication. VanDine, D.F. 1985. Debris flows and torrents in the South-ern Canadian C o r d i l l e r a . Can. Geotech. Jour., Vol. 22. pp. 44-68. Varnes, D.J. 1958. Landslide types and processes. in Landslides and Engineering Practice, H.R.B. Spec. Report 29. pp. 20-47. Varnes, D.J. 1978. Slope movement types and processes, in Landslides: Analysis and Control, T.R.B. Spec. Report 176. pp. 11-33. Weszka, J.S. 1978. A survey of threshold selection techni-ques. Comp. Graph. and Image Proc, Vol. 7. pp. 259-265. Wieczorek, G.F. 1984. Preparing a detailed landslide-inven-tory map for hazard evaluation and reduction. Bull, of Assoc. of Eng. Geol., Vol. 21, No. 3. pp. 337-342. 81 GLOSSARY Affine transform: A lin e a r transformation of one coordinate system into another in order to achieve r e g i s t r a t i o n in scale and position between two data sets. C l a s s i f i c a t i o n : A mapping based on a decision or set of decisions about the class assignments to an input image based on the measurements taken from a set of image features. DEM: D i g i t a l Elevation Model. D i g i t a l Elevation Model: Elevation measurements of region stored in an image compatible, raster format. Edge detection: The perception of the presence and location of abrupt changes in gray>levels within a d i g i t a l image. Gaussian d i s t r i b u t i o n : Analogous to the normal d i s t r i b u t i o n . , Lambertian surface: An i d e a l , perfectly d i f f u s i n g surface which r e f l e c t s energy equally in a l l d i r e c t i o n s . Laplacian: An i s o t r o p i c , two-dimensional derivative of an image region performed about each discrete image point ( p i x e l ) . Low resolution l i n e a r object: A lin e a r image region of high brightness values width a depicted width of one to three p i x e l s . Object extraction: A method for e x p l i c i t l y extracting items from a picture by invoking shape and gray l e v e l know-ledge about the object. The output of the process i s necessarily a mapping of the object's location. Pattern recognition: The automated process by which uniden-t i f i e d patterns in an image can be c l a s s i f i e d into a limited number of discrete classes through comparison with other class-defining patterns or c h a r a c t e r i s t i c s . Region growing: An approach to object extraction by the addition of points, or the merging of subregions, i f appropriate acceptance conditions are s a t i s f i e d . Registration: The process of geometrically aligning two or more sets of image data such that resolution elements representing a s p e c i f i c ground area are d i g i t a l l y or v i s u a l l y superimposed. 82 Segmentation: The division of an image into regions based on a set of c r i t e r i a . Template matching: An approach to object extraction in which a set of prototypes for an object class is compared with the image. Areas approximating the prototype ac-cording to a set of pre-selected similarity c r i t e r i a are written to an output image at a distinctive score. Thematic Mapper: A sensor launched in March, 1984 on Land-sat-5 that senses reflected electromagnetic radiation in seven bands at approximately 30 metres resolution (thermal band 120 metres). These bands range from v i s i -ble blue to thermal infrared. Thresholding: The segmentation of an image based on a gray level which defines the upper boundary of one class and the lower boundary of another. TM: Landsat Thematic Mapper. Tracking: A type of region growing where one begins at a point lying on an edge or curve, and successively accepts neighbouring edge or curve points until the entire object, or the entire curve, has been traversed. Zero crossings: The contours where the output of an isotro-pic operator passes through zero. 83 APPENDIX A LANDSLIDE DETECTION IN AN ENHANCED SATELLITE IMAGE Procedure The Laplacian image described in Chapter 2 was regis-tered to the digital landslide inventory using a three point affine transform. Registration points were in valley loca-tions having approximately the same aspect and elevation. The histogram of this image (Figure 3.8) was used to develop three threshold values for study: pixel value 6, 10, and 14. Photographs were taken of f u l l resolution image quarters on a rendition of the image exported to the MacDonald Dettwiler Meridian (PC) system at the Faculty of Forestry. The images for these photographs were displayed as follows: Laplacian image values above the specified threshold displayed in green at pixel value 100, and the image rendition of the digitised landslide inventory displayed as red at pixel value 100. Photographic slides of the described images were pro-jected on poster paper, and the locations of the landslides were outlined in red. Although overlap between Laplacian edges and the landslide inventory was expected to be evident by a yellow colour, this effect was hindered by problems with registration in certain portions of the image. Some degree of subjective interpretation was necessary to verify successful detection in these areas. 84 A simple object definition was employed involving strings of "on" pixels of a certain minimum length. Connect-ed strings of "on" pixels of three-or more which coincided with a landslide's estimated location and direction was defined as a success. If a string of two connected "on" pixels was separated from another by one "off" pixel, and this line-up coincided with a landslide location as defined above, this was also considered a success. These successes were recorded on the poster paper and later related to their corresponding event for tabulation of results. Results and Conclusions The results are shown graphically in Figures A.l to A.3. Event detection is considered in terms of two c r i t e r i a : event type (for events representing landslide groups, the type of the dominant and most recent landslide in that group) and event age (year of f i r s t appearance on photo-graphy). The total success achieved by the three thresholds is not surprising, and closely parallels an apparent in-crease in image noise as shown in Figure A.2. It is inter-esting to note that those events identified as primarily torrents are significantly more detectable than slides, as shown in Figure A.3. This may be due to the smaller sizes common in the slide category, or slower regeneration rates in the torrent tracks. When the detected events are consid-ered in terms of age, as shown in Figure A.4, we see that younger events, which are commonly in an early stage of regeneration ( i f at a l l ) , are detected more often than the 85 F i g u r e A . l . R e s u l t s : D e t e c t i o n r a t e o f a l l l a n d s l i d e e v en t s f o r 3 t h r e s h o l d s o f the L a p l a c i a n image . 100-j 9 0 -8 0 -Threshold Value F i g u r e A . 2 . R e s u l t s : D e t e c t i o n o f l a n d s l i d e e v e n t s by the L a p l a c i a n o p e r a t o r by l a n d s l i d e t y p e . 86 Figure A.3. Results: Detection of landslide events by the Laplacian operator by landslide age. 100-1 9 0 -8 0 -1940 1962 1968 1979 1981 Photo Year older events. Two major conclusions were drawn from the results. These conclusions were used to guide the implementation of the thesis. A significant proportion (>50%) of certain clas-ses of events were indicated by this edge operator. These include the combined category of events, events less than 15 years old, and torrents regardless of age which were a l l detected at or above the 50 percentile in the lowest thresh-old image. Higher threshold images yielded more disappoint-ing results. However, the low threshold edge Image presents a great deal of commission error. Therefore, more informa-8 7 tion, such as the addition of elevation data to a model, i s necessary to approach the high l e v e l of detection in the low threshold image while reducing the high l e v e l of apparent commission. This conclusion should be the same for the tem-plate matching and the threshold/template matching images since the world objects which produce a l l of the low resolu-tion l i n e a r objects in these images ( i . e . roads, streams, hydroline rights of way) are a constant. Another conclusion i s that the image-to-map r e g i s t r a -tion using the simple 3 point a f f i n e transform was found to be unsatisfactory. Therefore, a piecewise approach to regis-t r a t i o n was employed i n thesis to solve l o c a l r e g i s t r a t i o n problems using this same transformation when registering the elevation data to the images. This investigation has provided a demonstration of the absolute d e t e c t a b i l i t y of landslides using a linear segment d e f i n i t i o n in an edge-operated image. The results showed that detection was greater than 50% for landslides less than 20 years old, but that the image contained far too much commission error to make this approach suitable for extrac-tion of landslides. Therefore, i t was concluded that the addition slope information was necessary for sorting the various candidate segments. 88 APPENDIX B RECORD OF COMMISSION ERRORS These errors were catalogued and their apparent cause documented from photo evidence. Where photo evidence did not indicate the cause, or where there was suspicion of subse-quent mass movement, a f i e l d check was performed. The des-cription of each column in the table is described below. 1: indicates reference number for segment. 2: indicates harvest condition of surrounding area: A: pre-1968 harvest or unharvested. B: post-1968 harvest. 3: indicates origin of the error: N: error caused by naturally occurring feature. D: error caused by feature arising from man's dis-turbance of the landscape. 4: indicates t r i a l s in which the segment occurred. 5: indicates total number of t r i a l s in which segment occurred 6: * f i e l d check was performed. 7: comments on cause of error. 89 Appendix B: 1 2 3 Column Numbers 4 5 6 7 A 1 A N 1 2 4 5 4 * A 2 A N 1 3 5 3 A 3 A N 1 3 5 3 A 4 B D 1 1 * A 5 B D 1 3 2 * A 6 B D 1 3 5 3 * A 7 A D 1 3 5 3 A 8 A N 1 5 2 * A 9 A N 1 5 2 * A 10 B D 1 3 5 3 * A 12 B D 1 2 3 4 5 5 A 13 A N 2 4 2 * A 14 A N 2 4 2 A 15 B D 2 4 2 A 16 B D 2 5 2 A 17 B D 2 4 2 A 18 A D 2 4 2 A 19 A N 2 3 5 3 A 20 B D 2 1 * A 21 B D 2 4 2 * A 22 B N 1 2 4 3 * A 23 A N 2 3 4 3 A 24 A D 2 3 4 5 4 A 25 A N 3 1 A 26 B D 3 1 * A 27 B D 2 1 A 28 B D 3 5 * A 29 A D 3 1 A 30 A D 3 1 A 32 A N 3 1 * A 34 A N 3 1 A 35 B D 1 3 5 A 36 A N 4 1 A 37 B N 4 1 * A 38 A N 4 1 A 39 B D 4 1 A 40 A D 4 1 A 41 A N 4 1 A 42 B D 4 1 A 43 B N 4 1 * A 44 B N 4 1 A 45 A D 3 4 A 46 B D 4 1 A 47 B D 4 1 * A 48 A N 1 5 A 49 B D 5 1 * A 50 B 5 1 * A 51 A D 5 1 A 52 B D 5 1 * Vegetated gully Clearing Clearing Gully/slash Slash area Hydroline/road cut Road No evidence No evidence Gully/slash Hydroline No evidence Clearing Road Road Road Road Ledge Vegetated gully Vegetated gully Talus slope Clearing Road Talus slope No evidence Clearing Landing/slash Road Landing No evidence Ledge Road Vegetated gully Ledge Stream Hydroline Road Stream Road Ledge Stream Road Road Eroded gully Clearing Gully/slash Torrent Road No evidence 90 Appendix B: 1 2 3 Column Numbers 4 5 6 7 A 54 A 5 1 * A 55 B 1 2 3 4 5 5 B 1 A N 1 1 B 4 B D 1 5 2 * B 5 B D 1 5 2 B 6 B D 1 2 B 7 A 1 3 5 3 * B 8 A D 1 2 4 5 4 B 9 A N 1 1 * B 10 B D 2 4 5 3 B 11 A D 2 4 2 B 12 B D 2 3 4 3 B 13 A N 2 4 2 B 14 B N 2 4 2 * B 15 A N 2 4 5 3 * B 16 B D 2 4 2 B 17 B D 2 4 2 B 18 B D 3 5 2 * B 20 A N 3 1 B 21 B D 3 1 * B 22 B D 3 1 * B 23 B D 3 1 * B 24 A D 3 5 B 25 B D 4 1 B 26 B D 5 1 B 27 B D 5 1 B 29 B D 5 1 B 30 A N 5 1 * Old debris flow Torrent Clearing Ridge/slash Road Road Slide Road No evidence Road Road Road Stream Stream Eroded gully Landing Eroded gully Ridge/slash Clearing No evidence No evidence No evidence Road Road Road Road Road No evidence 91 APPENDIX C EVENT CHARACTERISTICS AND IDENTIFICATION 92 The events identified from Howes' inventory (Howes, 1985b) are listed and described in the following tables. Three events (204, 300, 301) occurred between the time of the inventory and this study. The table is divided into four sections. Selected columns are described below. - Number: Event number for reference. - Description: Howes' landslide identification codes. More than one may be listed for each event. - Photo year: Photos on which the event's major constituent f i r s t appeared. - Harvest: Photos on which harvesting f i r s t became apparent in the area surrounding the event. In the case of multiple harvest years surrounding one event, the primary one is taken. - Type: S slide, T torrent. - Operator codes (e.g. T2S20L5): T2 image threshold value of 2. o S20 minimum segment slope of 20 . L5 minimum segment length of 5 pixels. 1 indicates that the event was extracted by the operation. 0 indicates that the event was not extracted. 93 SECTION NUMBER DESCRIPTION PHOTO YEAR HARVEST TYPE NW 1 CA-1A -> IB 68 79 S NW 2 CA-03 68 62 T NW 3 CA-06 79 62 T NW 4 CA-08 -> 09 79 62 T NW 5 CA-12 -> 13 62 62 T NW 6 CA-15 -> 19 79 62 T NW 7 CA-20 81 62 S NW 8 CA-22 79 62 T NW 9 CA-24 -> 30 81 62 T NW 10 CA-36 -> 37 68 62 S NW 11 CA-39 79 62 S NW 12 CA-40 79 62 T NW 13 CA-41 79 62 T NW 14 CA-42 79 62 T NW 15 CA-43 -> 45 68 62 T NW 16 CA-46 -> 46A 79 62 T NW 17 CA-47 68 62 T NW 18 CA-65 -> 67 79 68 T NW 20 CA-68 -> 69 79 68 T NW 21 CA-70 81 79 T NW 22 CA-72 79 79 S NW 23 CA-73 79 62 T NW 24 CA-73A 81 0 T NW 25 CA-74 -> 74B 79 62 T NW 26 CA-77 62 62 T NW 27 CA-78 79 62 T NW 28 CA-79 -> 80 79 62 T NW 29 CA-81 79 0 T NW 30 CA-82 79 0 T NW 31 CA-83 -> 85 81 62 T NW 32 CA-86 -> 87 79 62 T NW 33 CA-88 79 62 T NW 34 CA-92 -> 92A 79 62 T NW 35 CA-94 79 68 T NW 36 CA-99 -> 100 79 68 T NW 37 CA-101 -> 104 79 79 T NW 38 CA-111 ->116 79 68 T NW 39 CA-117 -> 118 79 68 T NW 40 CA-119 -> 120 79 79 T NW 41 CA-121 -> 122 79 0 T NW 42 CA-121A 79 79 T NW 43 CA-123 -> 124 81 68 T NW 44 CA-125 79 68 T NW 45 CA-126 -> 127 68 68 T NW 46 CA-128 79 0 T NW 47 CA-129 79 0 T NW 49 NR-107 79 79 T NW 50 NR-108 79 79 T NW 51 NR-110 62 79 T NW 52 NR-111 62 79 T NW 53 NR-112 ->115 62 0 T NW 54 CC-02 -> 03 40 0 T NW 55 CC-04 -> 05 40 0 T 94 SECTION NUMBER DESCRIPTION PHOTO YEAR HARVEST TYPE NW 56 CC-08 -•> 12 62 0 T NW 57 NR-194 -> 198 79 79 T NW 58 NR-196 79 79 S NW 59 NR-222 -> 223 79 62 T NW 60 NR-224 79 62 T NW 61 NR-225 -> 226 79 62 T NW 62 NR-228 -> 231 79 62 T NW 63 NR-205 -> 206 81 62 T NW 64 NR-207 -> 208 62 62 T NW 65 NR-209 68 62 S NW 66 NR-187 62 62 T NW 67 NR-182 -> 183 79 79 T NW & NE 68 NR-144 -> 150 62 0 T NW & SE 70 NR-142 62 0 S NW & SW 71 NR-132 -> 133 62 0 T NW 174 CA-10 62 62 T NW 175 CA-11 62 62 T NW 176 CA-14 68 62 S NW 177 CA-32 79 62 F NW 178 CA-35 79 62 S NW 179 CA-75 -•> 76 79 62 T NE 72 NR-238 -> 243 68 62 T NE 73 NR-245 62 62 F NE 74 NR-250 -> 250A 68 62 T NE 75 NR-254 81 68 F NE 76 NR-255 79 62 S NE 77 NR-256 -> 260 79 68 S NE 78 NR-262 79 68 T NE 79 NR-263 -> 266 79 68 T NE 80 NR-268 -> 268A 79 68 T NE 81 NR-270 -> 271(273) 62 68 S NE 82 NR-272 79 68 S NE 83 NR-275 -> 278 68 62 T NE 84 NR-279 -> 280 81 62 T NE 85 NR-217 79 0 F NE 86 NR-281 -> 282 81 62 T NE 87 NR-283 81 62 S NE 88 NR-291 -> 292 62 83 T NE 89 NR-293 -> 294 62 62 T NE 90 NR-295 -> 296 62 62 T NE 91 NR-297 68 62 T NE 92 NR-298A 81 68 F NE 93 NR-298 -> 299 81 81 T NE 94 NR-300 68 62 F NE 95 NR-302 81 62 T NE 96 NR-303 81 79 T NE 97 NR-304 81 62 S NE 98 NR-306 62 0 S NE 99 NR-309 -> 311 62 0 T NE 100 NR-313 -> 317 68 0 T NE 101 NR-323 62 0 T NE 102 NR-325 62 0 T NE 103 NR-326 68 62 S 95 SECTION NUMBER DESCRIPTION PHOTO YEAR HARVEST TYPE NE 104 NR-328 -> 329 62 0 T NE 105 NR-331 68 0 T NE 106 NR-332 -> 333 62 83 T NE 107 NR-333B -> 333C 62 0 T NE 108 NR-337 62 0 S NE 109 NR-347 -> 351 68 0 T NE 110 NR-355 -> 356 81 81 T NE 111 NR-178 62 79 T NE 112 NR-179 62 79 T NE 113 NR-210 -> 212 81 62 T NE 114 NR-213 -> 214 81 62 S NE 115 NR-174 62 0 T NE 116 NR-151 -> 152 79 68 T NE 117 NR-154 -> 163 79 79 T NE 118 NR-164 79 79 S NE 119 NR-165 79 79 T NE 120 NR-166 68 79 T NE 121 NR-167 62 0 S NE 122 NR-169 -> 170 62 0 T NE 123 NR-171 -> 171A 68 0 T NE 124 NR-175 81 62 T NE 200 NR-168 62 0 T NE 201 NR-261 81 68 S NE 202 NR-70 81 83 S NW 203 CA-01 79 62 S NW 204 CA-02A 83 62 T SW 127 WL-09 40 0 F SW 128 WL-10 -> 11 40 0 F SW 129 WL-12 40 0 S SW 130 NR-27 -> 28 40 0 S SW 131 NR-29 -> 30 40 0 S SW 132 NR-31 40 0 S SW 133 NR-32 -> 34 40 0 S SW 134 NR-48 -> 49 (45) 79 62 S SW 135 NR-62 -> 63 81 79 S SW 136 NR-76 79 83 S SW 137 NR-77 -> 79 81 83 S SW 138 NR-81 79 83 S SW 139 NR-84 81 83 F SW 140 NR-85 -> 86 79 79 T SW 141 NR-88 -> 89 79 79 T SW 142 NR-90 -> 91(93,94) 40 79 T SW 143 NR-92 79 79 S SW 144 NR-121 62 0 T SW 145 NR-122 -> 124 68 0 T SW 146 NR-128 -> 128A 68 0 T SW 147 NR-129 79 62 T SW 148 NR-131 62 0 T SW 149 NR-404 -> 405 68 62 T SW 150 NR-406 62 62 F SW 151 NR-412 40 0 S SW 152 NR-414 40 0 T SW 153 NR-414A 68 62 S 96 SECTION NUMBER DESCRIPTION PHOTO YEAR HARVEST TYPE SW 154 NR-423 40 0 S SW 155 NR-425 -> 428 68 0 T SW 156 NR-430 40 0 T SW 180 NR-65 79 79 S NE 181 NR-125A 62 0 T NE 182 NR-127 68 0 T NE 184 NR-157 62 0 S SE 157 NR-448 -> 450 81 68 T SE 158 NR-363 -> 368A 68 62 T SE 159 NR-372 -> 374 79 79 S SE 160 NR-375 81 62 S SE 161 NR-377 62 62 T SE 162 NR-380 62 62 S SE 163 NR-381 -> 383(393) 81 62 S SE 164 NR-387 62 62 T SE 165 NR-395 68 62 T SE 166 NR-396 62 62 S SE 167 DE-01 -> 07 79 79 T SE 168 DE-08 -> 09 79 79 T SE 169 DE-11 62 62 S SE 170 DE-12 -> 13 62 62 S SE 171 DE-15 62 79 F SE 172 DE-16 62 79 S SE 173 DE-21 62 0 T SE 183 NR-140 -> 141 62 0 T 97 NO. ASPECT LENGTH WIDTH L CLASS W CLASS 1 2 800 10 2 1 2 2 800 40 2 2 3 2 800 60 2 2 4 2 800 40 2 2 5 2 900 20 2 1 6 2 900 40 2 2 7 2 400 40 2 2 8 2 400 20 2 1 9 2 900 30 2 2 10 2 200 20 1 1 11 4 800 60 2 2 12 4 1000 40 2 2 13 4 800 60 2 2 14 4 1000 20 2 1 15 3 500 20 2 1 16 3 800 60 2 2 17 3 800 60 2 2 18 3 500 60 2 2 20 4 700 20 2 1 21 4 200 10 1 1 22 4 200 20 1 1 23 1 400 40 2 2 24 4 150 10 1 1 25 1 500 40 2 2 26 4 400 10 2 1 27 4 400 20 2 1 28 4 600 50 2 2 29 4 200 10 1 1 30 3 600 30 2 2 31 4 200 10 1 1 32 4 500 10 2 1 33 4 300 40 1 2 34 3 600 20 2 1 35 3 500 10 2 1 36 3 800 10 2 1 37 3 400 20 2 1 38 3 600 20 2 1 39 3 800 10 2 1 40 3 700 20 2 1 41 4 300 60 1 2 42 4 500 10 2 1 43 4 300 30 1 2 44 4 300 30 1 2 45 4 200 20 1 1 46 1 500 30 2 2 47 4 300 30 1 2 49 2 900 10 2 1 50 2 1000 10 2 1 51 2 600 10 2 1 52 2 600 10 2 1 53 4 1100 10 2 1 54 3 2200 40 2 2 55 3 400 40 2 2 98 NO. ASPECT LENGTH 56 4 1600 57 2 1000 58 2 200 59 2 200 60 2 500 61 2 300 62 2 900 63 2 800 64 2 800 65 3 200 66 4 800 67 2 300 68 2 1400 70 2 1000 71 2 400 174 2 800 175 2 300 176 2 200 177 2 200 178 2 300 179 4 500 72 3 1000 73 2 300 74 2 800 75 2 400 76 2 600 77 3 1300 78 3 600 79 3 600 80 3 300 81 4 1400 82 3 200 83 3 500 84 4 400 85 1 300 86 4 200 87 4 200 88 3 600 89 3 800 90 3 600 91 2 600 92 2 200 93 3 1000 94 3 150 95 3 400 96 3 200 97 3 400 98 3 1000 99 3 1400 100 4 600 101 3 500 102 1 400 103 1 150 WIDTH L CLASS W CLASS 10 2 1 10 2 1 10 1 1 20 1 1 30 2 2 20 1 1 30 2 2 20 2 1 20 2 1 10 1 1 30 2 2 10 1 1 20 2 1 180 2 2 10 2 1 20 2 1 20 1 1 20 1 1 30 1 2 10 1 1 50 2 2 20 2 1 30 1 2 20 2 1 30 2 2 20 2 1 30 2 2 20 2 1 20 2 1 20 1 1 30 2 2 40 1 2 20 2 1 10 2 1 40 1 2 10 1 1 40 1 2 20 2 1 20 2 1 30 2 2 30 2 2 10 1 1 30 2 2 20 1 1 20 2 1 10 1 1 30 2 2 10 2 1 40 2 2 20 2 1 30 2 2 30 2 2 30 1 2 99 NO. ASPECT LENGTH 10A . 1 AOO 105 2 600 106 1 600 107 1 1000 108 3 2000 109 A 2A00 110 A AOO 111 A 300 112 A 800 113 2 1600 114 2 500 115 1 800 116 2 200 117 2 800 118 2 300 119 2 AOO 120 2 200 121 1 300 122 1 300 123 2 800 12A 1 600 200 2 800 201 3 300 202 1 150 203 2 150 20A 2 800 127 A 200 128 A 200 129 A 200 130 1 150 131 2 300 132 2 AOO 133 2 300 13A 2 300 135 2 AOO 136 2 200 137 2 200 138 1 150 139 2 200 1A0 1 1200 1A1 2 300 1A2 1 600 1A3 1 200 1AA 3 1000 1A5 A 200 1A6 3 200 1A7 2 300 1A8 2 AOO 1A9 A 200 150 A 200 151 3 200 152 3 500 153 3 200 WIDTH L CLASS W CLASS 20 2 1 10 2 1 20 2 1 10 2 1 10 2 1 10 2 1 10 2 1 10 1 1 10 2 1 AO 2 2 50 2 2 20 2 1 20 1 1 20 2 1 30 1 2 30 2 2 10 1 1 30 1 2 10 1 1 10 2 1 10 2 1 10 2 1 20 1 1 20 1 1 20 1 1 AO 2 2 10 1 1 50 1 2 80 1 2 30 1 2 50 1 2 30 2 2 30 1 2 30 1 2 AO 2 2 10 1 1 20 1 2 30 1 2 30 1 2 10 2 20 1 20 2 20 1 10 2 20 1 10 1 30 1 20 2 10 1 20 1 20 1 20 2 10 1 100 NO. ASPECT LENGTH 154 3 200 155 4 400 156 4 300 180 2 300 181 3 800 182 3 200 184 2 600 157 4 1600 158 3 400 159 4 300 160 1 200 161 4 500 162 4 200 163 4 800 164 4 600 165 4 400 166 4 300 167 1 800 168 2 500 169 2 200 170 2 300 171 2 500 172 2 300 173 3 200 183 2 1100 WIDTH L CLASS W CLASS 30 1 2 30 2 2 20 1 1 50 1 2 10 2 1 10 1 1 10 2 1 10 2 1 20 2 1 30 1 2 20 1 1 30 2 2 10 1 1 30 2 2 20 2 1 20 2 1 10 1 1 20 2 1 10 2 1 20 1 1 20 1 1 20 2 1 20 1 1 20 1 1 30 2 2 101 NO. T2S20L5 T25S10L4 T3S20L5 T3S20L4 T2S25L5 T3S15L5 1 0 0 1 1 1 1 2 1 1 1 1 1 1 3 1 1 1 1 1 1 4 1 1 1 1 0 1 5 0 1 0 1 0 1 6 1 1 1 1 1 1 7 0 0 0 0 0 8 0 1 0 0 0 0 9 0 1 0 1 0 1 10 1 1 1 1 0 1 11 1 1 1 1 1 1 12 0 0 0 0 0 13 1 1 1 1 1 1 14 1 1 0 0 0 0 15 1 0 0 0 0 0 16 1 0 1 0 0 1 17 0 1 0 0 0 1 18 0 0 0 1 0 1 20 1 0 0 0 0 0 21 0 0 0 0 0 0 22 0 0 0 0 0 0 23 0 1 0 0 0 1 24 0 0 0 0 0 0 25 1 1 1 1 1 1 26 0 0 0 0 0 0 27 0 0 0 0 0 0 28 1 1 1 1 1 1 29 1 1 1 1 1 1 30 0 0 0 0 0 0 31 0 0 0 0 0 0 32 0 0 0 0 0 0 33 0 1 0 0 0 0 34 0 0 0 0 0 0 35 0 0 0 0 0 1 36 0 1 0 0 0 1 37 1 1 1 1 0 1 38 0 0 0 0 0 0 39 0 1 0 0 0 0 40 0 0 0 0 0 0 41 0 0 0 0 0 0 42 0 0 0 0 0 0 43 0 0 0 0 0 0 44 0 0 0 0 0 0 45 0 0 0 0 0 0 46 1 0 1 1 1 1 47 1 1 1 1 1 1 49 0 0 0 0 0 0 50 0 0 0 0 0 0 51 0 0 0 0 0 1 52 0 0 0 0 0 1 53 0 0 0 0 0 1 54 1 0 0 0 0 0 55 0 0 0 0 0 0 102 NO. T2S20L5 T25S10L4 T3S20L5 T3S20L4 T2S25L5 T3S1 56 0 0 0 0 0 0 57 1 0 0 1 0 1 58 0 0 0 0 0 0 59 1 1 1 1 1 1 60 1 1 1 1 1 1 61 1 1 1 1 1 1 62 1 1 1 1 1 1 63 0 1 0 0 0 1 64 1 0 1 1 0 1 65 0 0 0 1 0 0 66 0 0 0 0 0 0 67 1 0 0 0 1 0 68 1 0 0 1 1 0 70 0 0 0 0 0 0 71 1 1 0 1 1 1 174 1 0 1 0 0 0 175 1 0 0 0 0 0 176 0 0 0 0 0 0 177 0 0 0 0 0 0 178 0 0 0 0 0 0 179 0 0 0 0 0 0 72 0 1 0 0 0 1 73 0 0 0 0 0 0 74 1 1 1 1 1 1 75 0 0 0 1 0 1 76 0 0 0 0 0 0 77 0 1 0 0 0 0 78 1 0 1 1 0 1 79 1 0 1 1 1 1 80 1 0 1 1 0 1 81 1 1 0 1 1 1 82 0 0 0 1 0 0 83 1 1 1 1 1 1 84 0 0 0 0 0 0 85 1 1 1 1 0 1 86 0 0 0 0 0 0 87 1 1 1 1 1 1 88 0 0 0 0 0 0 89 0 0 0 1 0 1 90 1 0 1 0 0 0 91 1 1 1 1 1 1 92 1 0 0 0 0 0 93 1 1 1 1 0 1 94 1 0 0 0 0 0 95 0 0 0 0 0 0 96 0 0 0 0 0 0 97 0 0 0 1 0 0 98 0 0 0 0 0 0 99 0 0 0 0 0 0 100 0 0 0 0 0 0 101 0 0 0 0 0 0 102 0 0 0 0 0 0 103 0 0 0 0 0 0 103 NO. T2S20L5 T25S10L4 T3S20L5 T3S20L4 T2S25L5 T3S1] 104 0 0 0 0 0 0 105 0 0 0 0 0 0 106 1 0 1 1 1 1 107 1 1 0 0 1 0 108 0 1 0 1 0 1 109 1 0 0 0 0 1 110 0 0 0 0 0 0 111 0 0 0 0 0 1 112 1 0 0 0 0 1 113 0 0 0 0 0 0 114 0 1 0 0 0 1 115 0 0 0 0 0 0 116 1 0 1 0 0 0 117 1 1 0 1 0 0 118 1 1 0 1 1 1 119 0 0 0 0 0 0 120 0 0 0 0 0 0 121 0 0 0 0 0 0 122 0 0 0 0 0 0 123 0 0 0 0 0 0 124 1 1 0 1 1 1 200 1 1 0 1 1 0 201 1 0 1 1 1 1 202 1 0 0 0 0 0 203 1 0 1 1 1 1 204 1 1 1 1 0 1 127 0 0 0 0 0 0 128 0 0 0 0 0 0 129 0 0 0 0 0 0 130 0 0 0 0 0 0 131 1 1 1 1 1 1 132 0 0 0 0 0 0 133 0 0 0 0 0 0 134 1 0 1 1 0 1 135 0 0 0 0 0 0 136 1 0 1 0 0 1 137 1 0 1 1 0 0 138 0 0 0 0 0 0 139 1 0 1 1 1 1 140 0 0 0 1 0 0 141 1 0 1 0 0 0 142 1 1 1 1 0 1 143 0 0 0 1 0 0 144 0 1 0 0 0 0 145 0 0 0 0 0 0 146 0 0 0 0 0 0 147 0 0 0 0 0 0 148 0 1 0 0 0 0 149 0 0 0 0 0 0 150 0 0 0 0 0 0 151 0 0 0 0 0 0 152 0 0 0 0 0 0 153 0 0 0 0 0 0 104 NO. T2S20L5 T25S10L4 T3S20L5 T3S20L4 T2S25L5 T3S15L5 154 0 0 0 155 0 0 0 156 0 0 0 180 0 0 0 181 0 0 0 182 0 0 0 184 1 0 0 157 0 0 0 158 1 1 1 159 0 0 0 160 0 0 0 161 0 1 0 162 0 0 0 163 0 0 0 164 0 0 0 165 0 0 0 166 0 0 0 167 0 0 0 168 0 0 0 169 0 0 0 170 1 0 0 171 0 0 0 172 0 1 0 173 0 0 0 183 0 0 0 0 0 0 0 0 0 0 0 0 1 0 1 0 0 0 0 0 0 0 1 0 0 0 0 1 1 1 0 0 0 0 0 0 1 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 1 0 1 0 0 0 0 0 1 0 0 0 1 0 0 105 NO. T7S20L5 T25S15L4 T7S20L4 T5S25L4 T7S15L5 T5S2 1 0 0 1 0 0 2 1 1 1 1 1 1 3 1 1 1 1 1 I 4 1 1 1 0 1 I 5 1 1 1 1 1 I 6 1 1 1 1 1 1 7 0 0 0 0 8 1 1 1 1 1 I 9 0 1 1 0 0 1 10 0 1 1 0 1 I 11 1 1 1 1 1 I 12 0 0 0 0 13 1 1 1 1 1 I 14 0 0 1 0 0 1 15 0 0 1 0 0 I 16 0 0 1 1 0 I 17 1 1 1 0 1 1 18 0 0 0 0 0 0 20 0 0 1 0 1 1 21 0 0 0 0 0 0 22 0 0 0 0 0 0 23 0 1 0 0 1 0 24 0 0 0 0 0 0 25 1 1 1 1 1 1 26 0 0 0 0 0 0 27 0 0 0 0 0 0 28 1 1 1 1 1 1 29 1 1 1 1 1 1 30 0 0 0 0 0 0 31 0 0 0 0 0 0 32 0 0 0 0 0 0 33 0 1 0 0 0 0 34 0 0 0 0 0 0 35 0 0 0 0 0 0 36 0 1 0 0 1 0 37 1 1 1 0 1 1 38 0 0 0 0 1 0 39 0 0 0 0 0 0 40 0 0 0 0 0 0 41 0 0 0 0 0 0 42 0 0 0 0 0 0 43 0 0 0 0 0 0 44 0 0 0 0 0 0 45 0 0 0 0 0 0 46 1 0 1 1 1 1 47 0 1 0 0 0 0 49 0 0 0 0 0 0 50 0 0 0 0 0 0 51 0 0 0 0 1 0 52 0 0 0 0 1 0 53 0 0 1 0 0 0 54 0 0 0 0 0 0 55 0 0 0 0 0 0 106 NO. T7S20L5 T25S15L4 T7S20L4 T5S25L4 T7S1 56 0 0 0 0 0 57 0 0 0 1 0 58 0 0 0 0 0 59 0 1 1 0 0 60 1 1 1 0 0 61 0 1 1 1 1 62 1 1 1 0 1 63 1 0 1 1 1 64 1 1 1 1 1 65 0 0 0 1 0 66 0 0 0 0 0 67 0 0 0 0 0 68 0 0 1 1 0 70 0 0 0 0 0 71 0 0 1 1 0 174 0 0 0 0 0 175 0 0 0 0 0 176 0 0 0 0 0 177 0 0 0 0 0 178 0 0 0 0 0 179 0 0 0 0 0 72 0 0 0 0 0 73 0 0 0 0 0 74 1 1 1 1 1 75 0 0 0 0 1 76 0 0 0 0 0 77 0 0 0 0 0 78 1 0 1 1 1 79 1 0 1 0 1 80 1 0 1 0 1 81 0 0 0 1 0 82 0 0 0 0 0 83 0 1 1 0 1 84 0 0 1 1 0 85 1 1 1 0 1 86 0 0 0 0 0 87 1 1 1 1 1 88 0 0 0 0 0 89 0 0 0 0 0 90 0 0 0 0 0 91 1 1 1 1 1 92 0 0 1 0 1 93 1 1 1 0 1 94 0 0 0 0 0 95 0 0 0 0 0 96 0 0 0 0 0 97 0 0 0 0 0 98 0 1 0 0 0 99 0 0 0 0 0 100 0 0 0 0 0 101 0 0 0 0 0 102 0 0 0 0 0 103 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 1 1 0 0 0 0 0 0 0 0 0 0 0 0 0 107 NO. T7S20L5 104 0 105 0 106 0 107 0 108 0 109 0 110 0 111 0 112 0 113 0 114 0 115 0 116 0 117 0 118 1 119 0 120 0 121 0 122 0 123 0 124 0 200 0 201 0 202 0 203 0 204 1 127 0 128 0 129 0 130 0 131 1 132 0 133 0 134 0 135 0 136 0 137 0 138 0 139 0 140 1 141 0 142 1 143 0 144 0 145 0 146 0 147 0 148 1 149 0 150 0 151 0 152 0 153 0 T25S15L4 T7S20L4 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 1 1 1 1 0 0 0 0 0 0 0 0 0 0 1 1 1 1 0 0 0 1 0 0 1 1 0 0 0 0 0 0 0 0 1 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 1 0 0 1 1 0 0 0 0 0 0 0 0 0 0 1 1 0 0 0 0 0 0 0 0 0 0 T5S25L4 T7S15L5 0 0 0 1 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 1 1 0 0 0 0 0 0 0 0 0 0 1 1 1 0 1 0 0 1 0 0 0 1 0 0 0 0 0 0 0 0 1 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 1 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 1 1 0 0 0 0 0 0 0 0 0 0 T5S20L4 1 0 0 0 0 0 0 0 0 0 0 1 1 1 1 0 0 0 0 0 1 1 0 1 1 1 0 1 0 0 1 0 0 0 0 0 0 0 1 1 0 1 0 0 0 0 0 1 0 0 0 0 1 108 NO. T7S20L5 T25S15L4 T7S20L4 T5S25L4 T7S15L5 T5S2< 154 0 0 0 0 0 0 155 0 0 0 0 0 0 156 0 0 0 0 0 0 180 0 0 0 0 0 0 181 0 0 0 0 0 0 182 0 0 0 0 0 0 184 0 0 1 1 0 1 157 0 0 0 0 0 1 158 0 1 0 0 1 1 159 0 0 0 0 0 0 160 0 0 0 0 0 0 161 0 0 0 0 1 1 162 0 0 0 0 0 0 163 0 0 0 0 0 0 164 0 0 0 0 0 1 165 0 0 0 0 0 0 166 0 0 0 0 0 0 167 0 0 1 1 0 0 168 0 0 0 0 0 0 169 0 0 1 0 0 1 170 0 0 0 0 0 0 171 0 0 0 0 1 0 172 0 1 0 0 0 0 173 0 0 0 0 0 0 183 0 0 1 0 0 0 109 APPENDIX D LANDSLIDE TEMPLATE REVISION An oversight was discovered in the configuration of the landslide template after the study was completed. The four directions examined in the template (see Figure 2.6) are insufficient for detecting certain object termini. A l l eight pixels surrounding the centre should be considered in order to make a rigorous examination of these objects. A preliminary investigation of the effectiveness of this revised approach was made. The resulting enhanced im-ages from the template/image and template/threshold opera-tions were remarkably similar to those used in the study. Five t r i a l s , two using the template/image and three using the template/threshold approaches, were made. The t r i a l s were examined for number of successful identifications and commission errors, and the results tabulated with the study results in Table D.l. They are indicated by the double l e t -ter symbols. The results are comparable, and indicate some improve-ment in landslides detected for identical t r i a l s with the original template images, especially in the template/thresh-old approach. However, an increase in both successful ident-ifications was accompanied by an increase in commission errors. This is consistent with the results from the en-hanced images used in the study, which indicates the persis-tence of commission errors in this approach to object 110 extraction. It also suggests that the study template, though not i d e a l , provides suitable candidate objects for this system. Succeeding studies are advised to use the eight direction template. I l l Table D.l. Success and commission results for study t r i a l s plus the revised, eight direction template. Rank by Rank by Tr i a l Operator Success Commission Commission Success % No. % No. 1 19 D Laplacian 18 33 35 18 2 15 M thr./temp4. 23 42 35 23 3 13 K template4 24 45 37 26 4 14 BA template8 24 44 37 26 5 9 N thr./temp. 28 51 41 35 6 8 E Laplacian 28 51 41 36 7 10 BB thr./temp8. 25 46 42 34 8 3 BC thr./temp8. 36 67 45 55 9 16 BD template8 20 38 48 35 10 17 I template4 19 36 49 34 11 4 J template4 36 67 49 65 12 7 F Laplacian 31 57 50 56 13 12 C Laplacian 24 45 51 47 14 5 H template4 36 66 53 73 15 6 L template4 34 62 55 76 16 11 BE template8 24 45 63 64 17 1 G template4 48 89 62 147 18 2 B Laplacian 37 68 70 158 19 17 A Laplacian 19 35 70 83 Definitions: Template4 - original, four direction template used in study. Thr./temp4. - four direction template used on four thresholds and the results summed. Template8 - eight direction template. Thr./temp8. - eight direction template used on four thresholds and the results summed, o Tr i a l BA - minimum slope 15Q, minimum length 5 pixels T r i a l BB - minimum slope 20 o, minimum length 4 pixels T r i a l BC - minimum slope 15 .-, minimum length 4 pigels T r i a l BD - minimum gray level 4, minimum slope 20., minimum length 5 pixels o Tr i a l BE - minimum gray level 3, minimum slope 20 , minimum length 5 pixels Other t r i a l references are in Table 3.1. 112 

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:
http://iiif.library.ubc.ca/presentation/dsp.831.1-0097203/manifest

Comment

Related Items