Open Collections

UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Optimal path/neutral [i.e. neural] network approaches to modeling of forest road design for use in automated… Aron, Ionut Andrei 2003

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

Item Metadata

Download

Media
831-ubc_2004-0058.pdf [ 7.97MB ]
Metadata
JSON: 831-1.0074984.json
JSON-LD: 831-1.0074984-ld.json
RDF/XML (Pretty): 831-1.0074984-rdf.xml
RDF/JSON: 831-1.0074984-rdf.json
Turtle: 831-1.0074984-turtle.txt
N-Triples: 831-1.0074984-rdf-ntriples.txt
Original Record: 831-1.0074984-source.json
Full Text
831-1.0074984-fulltext.txt
Citation
831-1.0074984.ris

Full Text

OPTIMAL PATH / NEUTRAL NETWORK APPROACHES TO MODELING OF FOREST ROAD DESIGN FOR USE IN AUTOMATED GIS SYSTEMS by IONUT ANDREI ARON B.Sc, University TRANSILVANIA, Brasov, ROMANIA, 1992  A THESIS SUBMITTED IN PARTIAL FULFILMENT OF THE REQUIREMENTS FOR THE DEGREE OF MASTER OF FORESTRY in THE FACULTY OF GRADUATE STUDIES (Faculty of Forestry)  We accept this thesis as conforming to the required standard  THE UNIVERSITY OF BRITISH COLUMBIA September 2003 © lonut Andrei ARON, 2003  /  Library Authorization  In presenting this thesis in partial fulfillment 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.  Ion  at  Oec  Name of Author (please print)  Title of Thesis:  Degree:  (QtajKvuqj  K ^ s U r r &t  -tool  %  Date  P^W  I\UMr\\  For-eSlrVM  Wtfworfc  Year:  ^ ° Q ^  Abstract  A model that integrates the u s e of feed-forward Artificial Neural Networks a n d G I S t e c h n i q u e s w a s d e v e l o p e d for preliminary road d e s i g n in forestry. T h e model w a s built a n d tested with G I S data from M a l c o l m K n a p p R e s e a r c h Forest (approximately 5 0 0 0 ha).  Artificial Neural Networks having a variety of architectures w e r e trained using this d a t a , in conjunction with a s e v e r a l different learning parameters. O n c e the neural networks w e r e trained a n d t e s t e d , the k n o w l e d g e w a s u s e d to predict b a s e d on n e w data. T h e results of t h e s e predictions c o n s i s t e d in a numerical representation of the "likelihood" of a given cell to contain a road. T h e resulting v a l u e s w e r e represented in G I S format a s a n e w cost - surface. With the addition of this n e w s u r f a c e , a n optimal path function w a s run to trace the location of the road.  T h e Artificial Neural Network a p p r o a c h w a s c o m p a r e d to the c l a s s i c a l " S l o p e - C u r v a t u r e " a p p r o a c h , traditionally u s e d before in preliminary forest road d e s i g n with G I S . In all c a s e s , the Artificial Neural Network a p p r o a c h yields better results, both in terms of road length a n d spatial autocorrelation coefficient.  T h e results of this study s u g g e s t that the technique u s e d m a y be a r e a s o n a b l e method for determining the preliminary road location in forested a r e a s . T h e r o a d s g e n e r a t e d with the aid of Artificial Neural Networks h a v e a higher spatial autocorrelation with the original road a n d are a l s o shorter than the o n e s g e n e r a t e d with the S l o p e - Curvature cost surface. T h e route s e l e c t e d is not perfect from a n engineering point of view; h o w e v e r it still c a n provide g o o d information for the forest road-planner. T h e route s e l e c t e d by the c o m p u t e r s h o u l d be modified b a s e d on the field inspection a n d the knowledge a n d e x p e r i e n c e of the specialist.  Table of Contents Abstract  ii  List of T a b l e s  v  List of figures  vi  Acknowledgements  vii  1. Introduction  1  1.1.  B a c k g r o u n d and Rationale  1  1.2.  Objective of study  5  2. Previous research  6  2.1.  R o a d modeling  6  2.2.  Artificial Intelligence in Natural R e s o u r c e s  15  3. Study site  20  4. Data collection and preparation  22  4.1.  D a t a collection  22  4.2.  D a t a standardization  24  4.3.  G e n e r a t i n g the D E M a n d the raster d a t a s e t s  26  4.4.  Artificial Neural Network p r o c e s s i n g  28  5. Analytical Methods  30  5.1.  Training a n d testing the Artificial Neural Network  30  5.2.  Predicting on n e w data  37  5.3.  C r e a t i n g the cost surface  37  5.4.  T h e optimal path algorithm  38  5.5.  M o d e l calibration a n d testing  42  6. Results and Discussion  46  7. Conclusions and Recommendations  52  8. References  55  9. Appendices  58  iv  List of tables  Table 1  Existing a n d derived datasets u s e d in modeling the forest roads  24  Table 2  R o a d s m o d e l e d and criteria utilized to select the study datasets  34  Table 3  Input parameters for the neural network  36  Table 4  C u t - a n g l e v a l u e s for road H 5 0  51  Table 5  R o a d lengths table for all roads m o d e l e d  53  Table 6  S t r e a m c r o s s i n g s for all roads m o d e l e d  54  Table 7  Spatial autocorrelation coefficient v a l u e s for all roads m o d e l e d  55  Table 8  Spatial autocorrelation c o m p a r i s o n table for the 5 roads m o d e l e d  68  v  List of figures  Fig. 1  A n e x a m p l e of an Artificial Neural Network s i m p l e n o d e  19  Fig. 2  A n e x a m p l e of a simple Artificial Neural Network  20  Fig. 3  M a l c o l m K n a p p R e s e a r c h Forest O v e r v i e w M a p  22  Fig. 4  Location m a p for all roads m o d e l e d  35  Fig. 5  Architecture of the Neural Network u s e d in modeling the roads  38  Fig. 6  T h e algorithm for calculating the accumulative cost  42  Fig. 7  T h e algorithm for calculating the cost d i s t a n c e grid  43  Fig. 8  Back-link positions in the back-link grid  43  Fig. 9  T h e algorithm for calculating the cost allocation grid  44  F i g . 10  A n e x a m p l e of high spatial autocorrelation  46  F i g . 11  A n e x a m p l e of low spatial autocorrelation  46  F i g . 12  M a p of road H 5 0  63  Fig. 13  M a p of road H 9 0  64  F i g . 14  M a p of road H 1 1 0  65  F i g . 15  M a p of road A 4 2  66  F i g . 16  M a p of road LOO  67  Acknowledgments  I would like to a c k n o w l e d g e with gratitude several p e o p l e and organizations, w h i c h contributed greatly to this r e s e a r c h project and to my graduate e d u c a t i o n . First I would like to a c k n o w l e d g e my advisor, Dr. M i c h a e l Meitner, w h o s e advice and s u g g e s t i o n s in how to prepare the thesis project a n d c o m p i l e the data h a v e b e e n of great help.  Next I would like to thank the m e m b e r s of my committee, Dr. J o h n N e l s o n , Mr. P a u l L a w s o n and Dr. S t e v e n S h e p p a r d for their useful a d v i c e and a s s i s t a n c e throughout this study.  I a m a l s o very grateful to the M a l c o l m K n a p p R e s e a r c h Forest for providing the G I S dataset u s e d in this analysis and m u c h more other information related to this project.  vii  1. Introduction  1.1. Background and Rationale  The forest industry must be increasingly competitive on a global basis, driving the need to continually reduce costs. In light of this, forest companies strive to operate in a cost effective manner while balancing this with ecological and environmental concerns. One major component related to both of these concerns is the construction and use of forest road networks. Forest mangers are concerned with how forest-related activities can be conducted in the most economical way by planning an appropriate forest road network that takes into consideration timber transportation and road construction, as well as their environmental impacts. Approaches and methods to assist in the development and identification of current and future roading requirements continue to be a much needed area of research (Murray, 1998). One reason for this is the expectation that more comprehensive management planning will be carried out given the availability of more detailed and complete spatial information associated with today's forest regions. Another related reason is to ensure that the environmental and economical impact of forest management practices can be estimated. Thus, analytical methods and techniques are essential components of the management planning process associated with forest harvesting and transportation operations.  1  T h e planning of a road network to a c c e s s timber harvest sites is a difficult a n d time c o n s u m i n g task. Traditional m e t h o d s for d e s i g n i n g a forest road s y s t e m c o n s i s t e d largely of aerial photo interpretation a n d field r e c o n n a i s s a n c e . M o r e recently, forest e n g i n e e r s h a v e u s e d large-scale contour m a p s to select preliminary routes using dividers, a p r o c e s s knows a s route projection. T h e e n g i n e e r then c a l c u l a t e s the total construction, transport a n d road m a i n t e n a n c e cost for e a c h alternative. T h i s is both a n e x p e n s i v e a n d t i m e - c o n s u m i n g task given that the forest m a n a g e m e n t plan typically c o v e r s t h o u s a n d s of hectares a n d tens of t h o u s a n d s of timber entry points over a planning period of m a n y y e a r s . S e s s i o n s a n d Liu (1993) h a v e d i s c u s s e d the n e e d for tools a n d t e c h n i q u e s to a s s i s t in the d e v e l o p m e n t of road network s y s t e m s in forest m a n a g e m e n t , particularly within the context of a c c e s s i n g timber harvest sites. D e s i g n i n g a n d evaluating s e v e r a l alternative road networks is generally impractical, b e c a u s e of the complexity a n d t i m e - c o n s u m i n g nature of the task. P l a n n e r s are thus typically restricted to reviewing only o n e or two alternative networks, often only superficially (Walker a n d L o u g h e e d , 1989). Therefore there is a clear n e e d for a n automated method to a s s i s t forest m a n a g e r s a n d specialists in planning their forest road network. T h e introduction of G e o g r a p h i c Information S y s t e m s (GIS) represents a n important step forward in forest planning a n d modeling. T h i s technology provides a n e w m e a n s of integrating information in a w a y that helps us understand a n d a d d r e s s forest m a n a g e m e n t problems. G I S helps us to o r g a n i z e data a n d to  2  understand their spatial relationship, providing a b a s i s for m a k i n g more p r e c i s e a n d intelligent d e c i s i o n s .  T h e r e are s e v e r a l G I S tools that u s e raster G I S a n d Digital Terrain M o d e l s ( D T M s ) s u c h a s E S R I ' s Spatial A n a l y s t , E S R I ' s A r c G R I D toolbox, both of w h i c h h a v e predefined functions that c a n help the specialist in solving the problem of preliminary road d e s i g n . T h e s e of-the-shelf tools are b a s e d o n the u s e of shortest path algorithms. T h e s e algorithms attempt to minimize cumulative cost along a path between two or more cells in a two d i m e n s i o n a l array by calculating the accumulative cost over a surface a n d by c o m p e n s a t i n g for the actual d i s t a n c e that must be travelled taking into a c c o u n t horizontal a n d vertical factors influencing the total cost of moving from o n e location to another. T h e path with the least cumulative cost is then c o n s i d e r e d a s the optimal path for the road.  O n e d i s a d v a n t a g e of this a p p r o a c h is that the optimal path calculated by the computer d o e s not a l w a y s coincide with the path that would be c h o s e n by the specialist in the field. A n u m b e r of p o s s i b l e explanations for this d i s c r e p a n c y exist, s u c h a s : -  the underlying data is of insufficient a c c u r a c y or reliability to support the a n a l y s i s ;  -  the k n o w l e d g e of the specialist is not well represented in the optimal path m o d e l ;  -  the mathematical combination of all factors involved in road planning d o e s not a d d r e s s the i n t e r d e p e n d e n c e of t h e s e factors, w h i c h m e a n s  3  that a road location may be a nonlinear a n d n o n - s e p a r a b l e function of the combination of all t h e s e factors.  M y thesis a d d r e s s e s s o m e of t h e s e limitations by d e v e l o p i n g a m o d e l that integrates the u s e of traditional G I S cost - a n a l y s i s tools with the u s e of neural networks that have the ability to "learn" a n d to g e n e r a l i z e from limited information. T h e goal of this model is not to replace detailed ground survey a n d engineering plans but more intelligently direct a n d limit the possibilities that n e e d to be c o n s i d e r e d , a s well a s highlighting road locations that are not o b v i o u s but might be feasible. T h i s model indicates w h e r e s u c h survey should be undertaken, thus reducing the time a n d cost of determining the final road layout. M o r e o v e r , by allowing the u s e r to vary the input parameters a n d constraints, the m o d e l permits c o n s i d e r a b l e flexibility in the evaluation of alternative road networks.  4  1.2. Objective of Study  T h e objective of this thesis is to determine whether artificial neural networks c o m b i n e d with G I S t e c h n i q u e s might be useful in preliminary d e s i g n of roads in forestry. T o d o this, a c o m p u t e r m o d e l will be d e v e l o p e d by integrating G I S cost - a n a l y s i s tools s u c h a s shortest path algorithms with artificial neural networks that have the ability to "learn" a n d g e n e r a l i z e from limited information.  5  2. Previous Research 2.1. Road modeling  A key c o m p o n e n t of integrated forest m a n a g e m e n t planning in C a n a d a is a road plan that s h o w s the network of forest roads to be constructed, maintained a n d d e c o m m i s s i o n e d over a period of time to allow a c c e s s to stands s c h e d u l e d for harvesting a n d m a n a g e m e n t . T o aid the forest m a n a g e m e n t planner, a n u m b e r of d e c i s i o n support s y s t e m s have b e e n d e v e l o p e d in recent y e a r s that take a d v a n t a g e of G e o g r a p h i c Information S y s t e m s , D a t a b a s e M a n a g e m e n t S y s t e m s a n d a d v a n c e d analytical m e t h o d s (e.g. Network, P e g g e r , R o a d v i e w ) . S o m e of t h e s e d e c i s i o n support s y s t e m s deal with the entire road network for a n extensive a r e a (e.g. Network) while others f o c u s o n individual roads or small clusters of roads over o n e or more periods of time ( P e g g e r , R o a d v i e w ) T a n (1992) identifies two primary s t e p s in the typical procedure of automated forest road locating: network formulating a n d road locating. T h e s e steps apply to both the more extensive road networks a s well a s for individual roads o r small cluster of roads. T h e s e s t e p s are: 1 - Network formulating 2 - R o a d locating Network formulating is c o n c e r n e d with determining the n o d e s a n d links contained in the network. H o g a n (1973) identifies links a s the existing r o a d s , roads to be constructed o r other m o d e s of transportation. N o d e s a r e c o n n e c t e d by links a n d they are the points through which the r e s o u r c e s of a m a n a g e m e n t unit must p a s s  6  for transportation to markets. T h e formulation of a network must b e performed with c a r e s o a s to include all feasible routes. Road locating is c o n c e r n e d with finding the most appropriate route for a road or a road network. F o r locating a road between a pair of points (origin destination), heuristic algorithms a n d shortest path algorithms (Minamikata 1  1984) c a n be applied in determining the road route of the least cost; while the d y n a m i c programming technique of D o u g l a s a n d H e n d e r s o n (1987) c a n locate the road route with the greatest benefit. T h e Traveling S a l e s m a n Algorithm (Wijngaard a n d R e i n d e r s 1985) a n d M i n i m u m S p a n n i n g T r e e m o d e l (Dykstra, 1984) are useful for finding a road network connecting a certain n u m b e r of points (e.g. centers of forest a r e a s ) with the minimum cost of road construction. T h e s e algorithms require s o m e expert instructions to e n a b l e t h e s e points to b e c o n n e c t e d by the road network. W h e n locating forest roads, it is c o m m o n that the points to b e c o n n e c t e d are u n k n o w n except for the given starting points. In s u c h c a s e s it would b e increasingly difficult to locate a forest road that would minimize the total cost of terrain transportation, road transportation, road construction a n d m a i n t e n a n c e . S o m e algorithms a r e reported in the literature (e.g. K o b a y a s h i 1984, N i e u w e n h u i s 1986) for estimating the optimum solution w h e n locating forest r o a d s . T h e road locating method presented by K o b a y a s h i (1984) u s e s a heuristic rule that begins with a given starting point a n d m o v e s iteratively to a  A rule of thumb, simplification, or educated guess that reduces or limits the search for solutions in domains that are difficult and poorly understood. Unlike algorithms, heuristics do not guarantee optimal, or even feasible, solutions and are often used with no theoretical guarantee (www.hyperdictionary.com) 1  7  neighbouring node to a c h i e v e the greatest benefit to cost ratio at e a c h iteration, until a satisfactory road s y s t e m is found. In this m e t h o d , if the routes b e t w e e n non-neighbouring n o d e s are not found before road locating, then it is difficult to c r o s s over extreme terrain. Alternatively, N i u e w e n h u i s (1986) initially d e f i n e s the p r o p o s e d routes for road construction between a n y pair of the n o d e s by the least cost path m e t h o d , a n d then e v a l u a t e s the alternative road routes by using the rule of m a x i m u m ratio of cell c o v e r a g e to cost. In this m e t h o d , timber transportation a n d forest s t a n d s are not c o n s i d e r e d a n d thus the cost benefit a n a l y s i s a n d spatial a n a l y s i s c o n c e r n i n g the road are limited. T h e N E T W O R K program ( S e s s i o n s , 1988) a d d r e s s e s the problem of optimization of logging transportation o v e r multiple periods of time. T h e m o d e l a n s w e r s the question: " W h i c h products should be hauled to w h i c h mills in order to m a x i m i z e discounted net r e v e n u e ? " T h e inputs for N E T W O R K c o n s i s t s of two lists of information: o n e list is of existing, p r o p o s e d or alternative links; the other list is of origins w h e r e material will enter the network, destinations w h e r e material will leave the network, the v o l u m e the activity will generate a n d the y e a r in w h i c h the activity will occur. T h e solution identifies the combination of routes that will be u s e d to provide the minimum total cost or m a x i m u m v a l u e . N E T W O R K c a n optimally solve problems involving only variable c o s t s but u s e s a heuristic, to find g o o d , feasible solutions for c a s e s with fixed a n d variable c o s t s ( S e s s i o n s a n d S e s s i o n s , 1988). E x a m p l e s of variable c o s t s are: transportation c o s t s , road surface rock replacement, etc, while fixed c o s t s might be road construction a n d m a i n t e n a n c e c o s t s s u c h a s : ditch a n d culvert cleanout, reconstruction c o s t s for  8  existing r o a d s , etc. N E T W O R K calculates the minimum cost or m a x i m u m value network by using a shortest path algorithm to s o l v e the variable cost problem p r o p o s e d by Ford (Ford and F u l k e r s o n , 1956). R e s u l t s for network planning p r o b l e m s that involve variable c o s t s will be optimal. F o r networks with variable a n d fixed c o s t s , the heuristic rules d o not guarantee a n optimal solution, but the solution will be feasible. N E T W O R K u s e s a n empirical formulation of a network, in w h i c h the u s e r defines n o d e s a n d links manually, a n d therefore several routes might be m i s s i n g d u e to errors in the formulation of the network. M o r e o v e r , the m o d e l doesn't take into a c c o u n t the s l o p e of the terrain, w h i c h is a d e c i s i v e factor in locating the forest roads, d u e to limitations in the m a x i m u m gradients that transport v e h i c l e s c a n negotiate.  In 1993, S e s s i o n s a n d Liu u s e d an improved version of N E T W O R K to d e v e l o p a model for preliminary planning of forest road s y s t e m s using Digital Terrain M o d e l s ( D T M ) . T h e method identifies feasible road s e g m e n t s , e v a l u a t e s their variable a n d fixed c o s t s a n d then determines the optimal set of road s e g m e n t s to be u s e d a n d the y e a r in which the roads are to be constructed. T h e solution procedure has three steps: identification of possible road s e g m e n t s o n the D T M a n d calculation of the c o s t s for e a c h road s e g m e n t , network a n a l y s i s to find the heuristic combination of road s e g m e n t s which minimizes the s u m of the construction, m a i n t e n a n c e a n d transport c o s t s , a n d  9  display the results of the a n a l y s i s o n the digital terrain m o d e l or contour m a p s o the transportation planner c a n review a n d verify the results in the field  T h e method t a k e s into a c c o u n t the m a x i m u m gradients that transport v e h i c l e s c a n negotiate, but environmental factors s u c h a s unstable s l o p e s or riparian m a n a g e m e n t a r e a s are not c o n s i d e r e d . After the road links h a v e b e e n g e n e r a t e d from the digital elevation m o d e l , a heuristic algorithm (Network I I ) is u s e d in order to identify the set of road s e g m e n t s , w h i c h minimize the s u m of construction, transport a n d road m a i n t e n a n c e c o s t s . T h i s algorithm s o l v e s a n iterative s e q u e n c e of shortest path p r o b l e m s , at e a c h iteration converting the fixed c o s t s into equivalent variable c o s t s . T h e method a s s u m e s that there are no existing r o a d s . H o w e v e r , existing roads c a n be incorporated in the model by identifying t h e m by the c l o s e s t grid point a n d adding them to the list of feasible road s e g m e n t s with z e r o construction c o s t s . T h e model d o e s not take into a c c o u n t soil information or s l o p e stability information in the creation of feasible road s e g m e n t s . With G I S , this information is b e c o m i n g more a c c e s s i b l e and could be u s e d a s a cell attribute to determine: If a road s e g m e n t is permissible H o w soil and ground s l o p e would affect cut a n d fill s l o p e s , a n d H o w existing vegetation would affect road construction c o s t s  10  T h e previous m o d e l s a d d r e s s the problem of l a r g e - s c a l e road d e s i g n (individual roads or small n u m b e r of roads in a limited area). T o a d d r e s s the problem of road network d e s i g n at a smaller s c a l e , R . M . N e w n h a m (1995) d e v e l o p e d a computer model n a m e d R O A D P L A N . T h e model divides the a r e a into cells, a s s i g n s them harvesting priorities a n d calculates a "roadability" factor b a s e d o n the n u m b e r of pixels classified a s water or m u s k e g . Further, the cells are g r o u p e d together in cutblocks b a s e d on m a x i m u m allowable a r e a a n d minimum allowable v o l u m e . Although R O A D P L A N c o n s i d e r s p o s s i b l e linkages b e t w e e n a potential node a n d all cells o n the existing network, the links that are finally s e l e c t e d are all straight lines. T o p o g r a p h y is not c o n s i d e r e d either in adjusting the construction c o s t s or in limiting the m a x i m u m s l o p e for the forest road. C A R P P ( C o m p u t e r A i d e d R o a d Projection P r o g r a m ) d e v e l o p e d by A n d e r s o n a n d N e l s o n (2003) c o n s i d e r s topography, stream c r o s s i n g s a n d other road d e s i g n parameters (such a s d i s t a n c e between s w i t c h b a c k s a n d junction angles) by a s s i g n i n g limits and penalties to e a c h link. P e n a l t i e s a d d cost to undesirable links s o that the algorithm will favour other links, a n d limits u n a c c e p t a b l e links, thereby restricting the network to only feasible links. H o w e v e r , the solutions are not optimal and there is variation c a u s e d by the order of landings a n d spatial r a n d o m n e s s of the node/link set. All of the existing m o d e l s that u s e G I S techniques are limited in terms of road location a c c u r a c y d u e to limitations in G I S data a c c u r a c y . With the advent of high-resolution D T M data, computer-aided road d e s i g n s y s t e m s have b e c o m e  11  more attractive, a s they provide quick evaluation of alternatives in a more systematic manner. O n e method for gathering high-resolution elevation data is with the u s e of airborne laser mapping technology s u c h a s L I D A R (Light Detection A n d Ranging). L I D A R elevation data has b e e n found to be a c c u r a t e to within 15 c m ( A h a m e d et a l . 2000). T h e model d e v e l o p e d by C o u l t e r (Coulter et a l , 2 0 0 1 ) u s e s L I D A R data to d e s i g n forest road s e g m e n t s , by estimating accurately a n d quickly the cut a n d fill earthwork that o c c u r s during forest road construction. T h e method d o e s not calculate the road layout but it u s e s user-defined centreline points a n d then s e a r c h e s alternate road locations to minimize earthwork while satisfying userspecified constraints and d e s i g n criteria. T h e method m a y significantly d e c r e a s e road d e s i g n time a n d effort, both in the field a n d in the office. A n e n g i n e e r in the field e q u i p p e d with a G l o b a l Positioning S y s t e m ( G P S ) unit would n e e d to m a k e far fewer field m e a s u r e m e n t s in order to c o m p l e t e road d e s i g n a n d earthwork calculations, m e a n i n g a greater n u m b e r of alternatives could be e x a m i n e d . If a n a r e a is c o v e r e d by a high resolution D T M , a field e n g i n e e r would n e e d only a s e r i e s of road centre points in order to calculate fairly accurate cut and fill v o l u m e s . At this point, the method applies to straight road s e g m e n t s a n d d o e s not d e a l with horizontal or vertical c u r v e s . Additionally, this model a l s o d o e s not take in-sloping, out-sloping, curve widening or turnouts into considerations.  12  "Off - the - Shelf" software, s u c h a s R o a d E n g , A u t o C A D , P e g g e r a n d R o a d V i e w , A r c V I E W and A r c l N F O ( E S R I 1992) c a n be u s e d in forest road planning a n d d e s i g n . R o a d E n g from SoftTree a s well a s A u t o C A D rely on s u r v e y d a t a collected in the field to generate terrain m o d e l s a n d very detailed e n g i n e e r e d road location a n d construction plans. T o u s e t h e s e programs, a road location h a s to be c h o s e n in a d v a n c e by the specialist, a n d accurate surveys must be carried out in the field. P E G G E R a n d R O A D V I E W d e v e l o p e d by R o g e r s (2001) are A r c V I E W e x t e n s i o n s that a s s i s t s in the initial road planning by automating the route projection p r o c e s s , a l s o known a s "pegging". T h e P E G G E R program is a tool for quickly identifying p o s s i b l e route location alternatives given g r a d e s specified by the user. T h e results are then displayed in R O A D V I E W , a road visualisation p a c k a g e for A r c V I E W . T h i s tool d o e s not evaluate additional environmental a n d e c o n o m i c constraints that must be c o n s i d e r e d by the professional forester s u c h a s soil types, hydrology, property lines a n d s l o p e c l a s s e s . M o r e o v e r , P E G G E R c a n be u s e d to d e v e l o p only one road s e g m e n t at the time a n d it cannot deal with more c o m p l e x road networks. A r c l N F O is a m u c h more powerful G I S p a c k a g e d e v e l o p e d by E S R I that c a n be u s e d in forest road planning. A r c G R I D is a raster or c e l l - b a s e d g e o p r o c e s s i n g toolbox that is integrated with A r c l N F O . In A r c G R I D , a planning a r e a c a n be divided into individual grids. E a c h grid could be a section of road  13  d e p e n d i n g on its surroundings (slope, earthwork, etc). T h e r e are three functions that c a n be u s e d for road location: C O S T D I S T A N C E , P A T H D I S T A N C E a n d C O S T P A T H . T h e s e functions are b a s e d on u s e of a shortest path algorithm. C O S T D I S T A N C E a n d P A T H D I S T A N C E are similar functions, but the P A T H D I S T A N C E function c o n s i d e r s horizontal a n d vertical factors. Horizontal factors determine the difficulty on moving from o n e cell to another while accounting for the horizontal e l e m e n t s that m a y affect the m o v e m e n t . T h e vertical factors determine the difficulty of moving from o n e cell to another while accounting for the vertical e l e m e n t s that m a y affect the m o v e m e n t . T o determine this vertical factor, the s l o p e between the F R O M cell a n d the T O cell is calculated from the v a l u e s defined in the input grid. U s i n g o n e of t h e s e functions, o n e c a n find the least cumulative cost from a cell (landing) to a s o u r c e cell (existing roads). T h e path from a cell to a s o u r c e cell is found by the C O S T P A T H function. If w e c o n s i d e r only the s l o p e factor, t h e s e functions c a n be u s e d directly for preliminary planning of forest roads. H o w e v e r , in forest road planning, considering only the g r a d e factor is not a d e q u a t e . W e must take into a c c o u n t s o m e other important factors s u c h a s hydrology, soil stability, e c o l o g i c a l factors, earthwork, etc. T h e goal of this study is to provide improvements to the w a y that cost-benefit a n a l y s i s and spatial a n a l y s i s are performed.  14  2.2. Artificial Intelligence in Natural Resources  G e o g r a p h i c Information S y s t e m s a s an aid to the d e c i s i o n making p r o c e s s provide excellent capabilities for understanding a n d p r o c e s s i n g spatial data, but very often this is overlooked in favour of the "presentation" of data. With the d e v e l o p m e n t of Artificial Intelligence (Al) b a s e d t e c h n o l o g i e s in recent y e a r s , new opportunities have e m e r g e d to e n h a n c e the tools u s e d to p r o c e s s spatial d a t a ( D e a d m a n a n d Gimblet, 1995). Artificial Intelligence t e c h n i q u e s h a v e t r e m e n d o u s capabilities to p r o c e s s real-world situation and simulate h o w the h u m a n brain p r o c e s s spatial data and therefore c o m b i n e d with G I S c a n be s u c c e s s f u l l y u s e d in making intelligent d e c i s i o n s in problems with a spatial c o m p o n e n t . Traditionally, Artificial Intelligence t e c h n i q u e s have b e e n u s e d in a r e a s s u c h a s : robotics, natural l a n g u a g e recognition, image recognition, detection of medical p h e n o m e n a , stock market predictions, etc but there are a few applications in Forestry a s well, s u c h a s : l a n d s c a p e planning, growth a n d yield m o d e l i n g , forest stand a g e calculation from satellite imagery, estimation of probability for landslides, etc. T h i s wide gamut of applications h a s c o n d u c t e d to the d e v e l o p m e n t of a large variety of new t e c h n i q u e s a n d methods that h a v e b e e n incorporated in desktop applications a n d thus making Artificial Intelligence increasingly available to r e s e a r c h e r s a n d specialists. T h e s e c o m m e r c i a l l y available software p a c k a g e s include Artificial Neural Networks ( A N N ) , G e n e t i c Algorithms, Expert S y s t e m s , Cellular A u t o m a t a etc.  15  S i n c e G I S s y s t e m s are data intensive a n d m a n y different types of data (e.g. remote s e n s i n g , topography, hydrology, etc) are u s e d by specialists in the d e c i s i o n making p r o c e s s , A l m a y prove useful in capturing expertise a n d representing k n o w l e d g e a s a n aid to d e c i s i o n m a k i n g . In light of t h e s e applications and in conjunction with the suitability of the structure of the input d a t a , A N N s w e r e c h o s e n a s the method to be investigated for applicability to the road planning problem.  A n artificial neural network, s o m e t i m e s referred to a s simply a Neural Net, differs from other computer algorithms in that it is not r u l e - b a s e d . Neural networks are applicable in virtually every situation in w h i c h a relationship (or pattern) between the input variables a n d output variables exists, e v e n w h e n that relationship is very c o m p l e x a n d not e a s y to articulate in c o m m o n statistical terms. Neural nets are u s e d in natural resource application d e v e l o p m e n t in a r e a s s u c h l a n d s c a p e planning, forestry, g e o l o g y a n d c a n be trained to r e s p o n d to a variety of input variables, of spatial or non-spatial form.  16  An artificial neural network is trained to recognize and generalize the pattern or the relationship between a set of inputs and outputs and consists of many simple processing units (or nodes) connected together.  y  Fig. 1 An example of an Artificial Neural Network simple node •  xi, X 2 , x  n  are inputs. These could be real numbers or Boolean values  depending on the problem. •  y is the output and is Boolean.  •  wi, w  •  T is the threshold and is real valued.  2  , w  n  are weights of the edges and are real valued.  The output y is 1 if the net input which is W ] X j + W2 X2 + ... + W X n  n  is greater than the threshold T. Otherwise the output is 0.  17  T h e behaviour of e a c h node is very s i m p l e , but together a collection of n o d e s c a n exhibit m a n y sophisticated behaviours and be u s e d to s o l v e c o m p l e x t a s k s . T h e knowledge in a N N is stored a s the strength of the interconnection weights (a numeric parameter), which is modified through a p r o c e s s called learning, using a training algorithm. T h e algorithmic function in conjunction with a learning rule is u s e d to modify the weights in the network in a n orderly f a s h i o n .  Fig. 2 A n e x a m p l e of a simple Artificial Neural Network  T h e most c o m m o n type of artificial neural network c o n s i s t s of three g r o u p s , or layers of units. T h e first layer is called the input layer a n d contains o n e n o d e (or neuron) for e a c h of the inputs in the training data file. T h e last layer is called the output layer a n d contains o n e neuron for e a c h of the network outputs. B e t w e e n the input a n d output layers are an arbitrary n u m b e r of hidden layers e a c h containing a n arbitrary n u m b e r of neurons. E a c h neuron is c o n n e c t e d to every other neuron in adjacent layers by a set of weights (see fig.2). T h e weights define the 'strength' of the flow of information from o n e layer to the next through  18  the network. E a c h weight c a n take on any positive or negative v a l u e . "Training" a neural network is simply the p r o c e s s of determining a n appropriate set of weights s o that the network accurately a p p r o x i m a t e s the input/output relationship of the training data.  Unlike most computer applications, a n Artificial N e u r a l Network is not " p r o g r a m m e d " , rather it is "taught" to give a n a c c e p t a b l e a n s w e r to a particular problem. Input a n d output v a l u e s are given to the network, initial weights are a s s i g n e d and then the N N repeatedly adjust t h o s e interconnection weights until it c a n s u c c e s s f u l l y produce output v a l u e s that match the original v a l u e s . T h i s weighted matrix of interconnections allows the neural network to learn a n d apply this k n o w l e d g e to n e w d a t a s e t s ( E a s s o n and Barr, 1996). In C h a p t e r 4.4 "Artificial Neural Network p r o c e s s i n g " , I will d i s c u s s into more detail the parameters utilized for the road location application.  A l t h o u g h there are simple c o m m e r c i a l N N tools available that a p p e a r a s simple to u s e a s a s p r e a d s h e e t , making the best u s e of t h e s e t e c h n i q u e s require s o m e understanding of the underlying algorithms. Therefore it is important to c o n s i d e r carefully how to m a p features of the problem to input/output layers, what (if any) hidden layers to have in b e t w e e n , what learning method to u s e a n d h o w to set parameters a n d initialize the weights.  19  3. Study site  The University of British Columbia Malcolm Knapp Research Forest was established in 1949 as a facility for research, education and demonstration in the practice of Forestry. The forest is situated about 60 kilometres East of Vancouver, British Columbia, at a latitude 49 18' N, longitude 122 35' W. Malcolm Knapp Research Forest - Overview Map  Fig. 3 Malcolm Knapp Research Forest Overview Map The forest is 5,157 hectares in size, with an average width (from west to east) of 4 km and an average length (from south to north) of 13 km. The area is typical lower coastal topography - mountain lakes, steep slopes and rock outcrops in the North and more gentle slopes of glacial till in the South. Elevation range from 20  sea level to 1025 meters. For more detailed information, please visit www.forestrv.ubc.ca/resfor/mkrf.html. The Research Forest has an extensive road network (over 100 km) covering the entire land-base, in a variety of terrain conditions and with various degrees of complexity. This, combined with an extensive GIS dataset, makes the Research Forest an excelent candidate for developing and testing the road model.  21  4. D a a t c o e lc t o i n and p r e p a r a o t in 4.1. Data collection  T h e dataset contains A r c V i e w a n d Arc/Info c o v e r a g e s a s well a s 0.5 m resolution ortho-photos n e e d e d for the project (Table 1).  T a b l e 1 Existing a n d derived d a t a s e t s u s e din modeling the forest roads  Dataset C o n t o u r lines  Existing  Derived  V  Lakes Forest c o v e r Soil B e a r i n g C a p a c i t y Streams  V  Roads Openings Slope  V  Side Slopes  V  Curvature  T h e actual contour lines have a 20 m interval a n d s o n e w contour lines (5 m interval) w e r e digitized using a n Arc/Info procedure in order to provide a more  22  a c c u r a t e s o u r c e for Digital Elevation M o d e l creation, n e c e s s a r y for a g o o d representation of the terrain in a road-modeling project of this level of detail.  T w o c o v e r a g e s w e r e digitized, e a c h using a different m e t h o d :  - 5 m contour lines w e r e digitized manually from 1:10,000 M Y L A R m a p s using the T R A C E module from A r c l N F O 7.2. T h i s c o v e r a g e is extremely important for the entire m o d e l , s i n c e it provides the b a s i s for the creation of a highly accurate Digital Terrain M o d e l ( D T M ) , which is further u s e d to generate s l o p e and curvature grids. T h e M Y L A R m a p s w e r e created from 1:20,000 air photo interpretation d o n e by hand o v e r 15 y e a r s a g o a n d are very a c c u r a t e c o m p a r e d to similar information in u s e e l s e w h e r e in B C ( L a w s o n , pers.com.)  - E c o l o g y m a p s w e r e digitized from geo-referenced s c a n n e d m a p s (Klinka 1975) using the o n - s c r e e n digitizing procedure.  After digitizing, elevation data w a s a d d e d to the contour lines a n d soil information to the e c o l o g y m a p s . T h e soil data w a s further p r o c e s s e d in order to calculate the soil bearing capacity for the entire a r e a . F o r more information o n soil bearing capacity calculation, p l e a s e s e e A p p e n d i x 3.  23  4.2. Data standardization  For this project, a set of GIS standards were developed and implemented. These standards refer to the data format, projection, storage precision, registration, file names, metadata and topology. Below is a short description of each of these standards. Format  Spatial data is stored in ARC/INFO coverage format. Storage Projection and Datum  The coverages are projected in the Universal Transverse Mercator (UTM) Zone 10, and the coordinates are in meters. The datum is North American Datum 83 (NAD 83 Canada).  Storage  Precision  Coverages are created in single precision coordinates (preserving about 7 decimals). This is sufficient for the dataset used in the analysis, as UTM projection coordinates can be stored in single precision with meter accuracy.  24  Data  registration  All the existing datasets w e r e subject to affine transformation (a warping p r o c e s s that p r e s e r v e s ratio of distance along the lines) in order to align t h e m with the new 0.5 m color ortho-photo to e n s u r e dataset c o n s i s t e n c y a n d accuracy.  Coverage  Names  E a c h c o v e r a g e ' s n a m e is c o m p o s e d of a descriptor referring to its thematic content or what kind of information it represents. T h e next (optional) c h a r a c t e r ' _ ' is simply a separator. T h e rest of the n a m e refers to the road to be m o d e l e d . T h i s convention is extended to the raster datasets a s well. S o m e e x a m p l e s of c o v e r a g e a n d grid n a m e s formed a c c o r d i n g to the n a m i n g s c h e m e would be: 'road_h50', 'Iake_h110' ' s l o p e _ a 4 2 \ etc. t  Metadata  E v e r y c o v e r a g e h a s metadata a s s o c i a t e d to it. M e t a d a t a is "data about data", providing information s u c h a s : w h o is responsible for the dataset, date of creation, data s o u r c e , spatial extent, a c c u r a c y , etc.  Clean Topology All c o v e r a g e s are t o p o l o g i c a l ^ c l e a n a n d correct a n d all polygon b o u n d a r i e s meet exactly.  25  4.3.Generating the DEM and the Raster datasets  T h e model is raster b a s e d which m e a n s that a Digital Elevation M o d e l ( D E M ) w a s created a s a b a s i s for the elevation information a s well a s for generating other raster c o v e r a g e s u s e d in the a n a l y s i s , s u c h a s s l o p e , curvature, etc. A D E M provides a digital representation of a portion of Earth o v e r a two d i m e n s i o n a l surface. T h e basic data for a D E M is b a s e d o n terrain elevation o b s e r v a t i o n s that are derived generally from o n e of three s o u r c e s : digitized contours, photogrammetric data capture (including aerial photography a n d digital satellite imagery), a n d surveying. D E M s derived from h y p s o g r a p h y (topographic relief) o r contours are p e r h a p s the m o s t c o m m o n of all s o u r c e s . F o r this project a Triangulated Irregular Network (TIN) w a s created from 5m contour lines, s t r e a m s , lakes a n d roads, the last three integrated in the model a s break lines. B r e a k lines represent a discontinuity in the in the s l o p e of the surface a n d a r e u s e d for a better representation of the surface a n d to improve the display a n d a n a l y s i s of T I N . All the grids u s e d in the model have a 5 m resolution that e n a b l e s t h e m to better m o d e l the terrain conditions a n d a l s o relates to the road width, w h i c h in most c a s e s is 5 meters a s well. Increasing and/or d e c r e a s i n g this resolution w o u l d d e c r e a s e the m o d e l ' s a c c u r a c y by introducing interpolation a n d generalization errors.  26  F r o m the new T I N , a raster D E M w a s created a s well a s other grids, s u c h as: -  slope  -  side s l o p e s  -  curvature  S l o p e identifies the steepest downhill s l o p e for a location on a surface. T h e S l o p e c o m m a n d takes a n input surface raster and calculates a n output raster containing the s l o p e at e a c h cell. T h e lower the s l o p e v a l u e , the flatter the terrain is and the higher the s l o p e v a l u e , the steeper the terrain ( A r c G I S D e s k t o p Help). S i d e s l o p e s represent the s l o p e from o n e cell to all neighboring cells. Curvature represents the fragmentation of the terrain a n d it's calculated at e a c h cell center. A positive curvature indicates that the surface is upwardly c o n v e x at that cell while negative curvature indicates that the surface is upwardly c o n c a v e at that cell. A value of z e r o indicates that the surface is flat ( A r c G I S D e s k t o p Help).  27  4.4. Artificial Neural Network Processing  F o r a n A N N to be u s e d a s a tool to interpret spatial d a t a , the m a p information must be converted into patterns. T h e s e patterns consist of a value for e a c h input t h e m e at a given location. During the training of the A N N , the patterns must a l s o contain the value of the a c c e p t e d output v a l u e for e a c h location. O n c e the trained A N N h a s produced an interpretive result, the result must be converted b a c k into G I S format for the production of an output m a p , further u s e d in the analysis.  T h e following G I S layers w e r e u s e d to collect d a t a from: - slope - side - s l o p e s (slope to all neighboring cells) - lakes - streams - soil bearing capacity - curvature - roads T h e s t e p s required to translate the information, for e a c h input a n d output from A r c l N F O to a P C - b a s e d A N N are outlined below. T h e s e s t e p s begin after the data h a s b e e n automated a n d is in the form of A r c l N F O c o v e r a g e s . F o r a detailed list of c o m m a n d s u s e d , p l e a s e s e e A p p e n d i x 3.  28  (1) C r e a t e the Triangulated Irregular Network (TIN) from 5 m contour lines, including the s t r e a m s , lakes a n d roads a s break lines.  (2) G e n e r a t e the slope a n d side s l o p e grids from the T I N . T h e raster resolution c h o s e n w a s 5 m.  (3) C r e a t e a grid of points that will be u s e d to collect the information from all other layers. T h e points w e r e equally distributed at 5 m interval, coinciding with the centroids of the cells.  (4) Buffer the l a k e s , s t r e a m s and roads a s follows: -  l a k e s : 10m buffer  -  s t r e a m s : 2 0 m buffer  -  roads: 5 m buffer  (5) C o n c a t e n a t e the information from all the grids a n d polygon c o v e r a g e s into the point c o v e r a g e created at S t e p 3. In this step, a c u s t o m m a d e A r c M a c r o L a n g u a g e ( A M L ) program w a s u s e d in order to c o m b i n e all the attributes from all layers into a single G I S d a t a b a s e . T h e p r o c e s s w a s d o n e for the entire study a r e a , a n d the A M L program ran for 6 d a y s o n a P e n t i u m III - 9 0 0 M h z before the d a t a collection w a s c o m p l e t e d . T h e result is a d a t a b a s e with o v e r two million records a n d 15 variables, d a t a b a s e that w a s imported a n d stored in Microsoft A c c e s s format.  29  5. Analytical Methods 5 . 1 . Training and testing the Artificial Neural Network  Five c a s e - s t u d i e s w e r e undertaken to investigate the applicability of the model to preliminary forest road d e s i g n (see T a b l e 3). E a c h study had a n a r e a of approximately o n e s q u a r e kilometer, covering a variety of relief conditions a n d road configurations. After running the m o d e l , the resulting roads w e r e c o m p a r e d to the existing r o a d s o n the ground a n d a spatial autocorrelation coefficient w a s calculated in e a c h c a s e .  T a b l e 2 R o a d s m o d e l e d a n d criteria utilized to select the study d a t a s e t s Length (m) 1,065  Road complexity Simple  H110  1,151  Medium  3  H90  1,554  Complex  4  A42  715  Simple  Study 1  Road name H50  Terrain conditions  Training dataset Local  Steep 2  Local (over 30°) Local Flat  Local  (0-15°) 5  LOO  3,406  Complex  Medium  External  (15°-30°)  F o r e a c h road that w a s m o d e l e d , a s m a l l e r dataset (40000 - 6 5 0 0 0 records) w a s clipped from the d a t a b a s e created in step 5 to be u s e d in a n a l y s i s . T h e s m a l l e r dataset clipped in the previous step w a s then exported in A S C I I  30  format, using M S A c c e s s Export function, in order to be compatible with Q w i k N e t P r o f e s s i o n a l Neural Network software (http://qwiknet.home.comcast.net).  31  Fig.4 Location m a p for all roads modeled  A two layer back - propagation A N N with 5 neurons in e a c h hidden layer w a s u s e d for this project. T h e neural network implemented is a feed-forward multilayer perceptron, which simply m e a n s that it is m a d e up of s e v e r a l layers a n d the d a t a propagates forward through the network from input to output. T h e first layer is called the input layer a n d contains o n e node for e a c h of the inputs in the training data file. T h e last layer is called the output layer a n d contains o n e neuron for e a c h of the network outputs. B e t w e e n the input a n d output layers there is o n e hidden layer, containing five neurons. E a c h neuron is c o n n e c t e d to every other neuron in adjacent layers by a set of weights. T h e weights define the 'strength' of the flow of information from o n e layer to the next through the network. E a c h weight c a n take on a n y positive or negative v a l u e . "Training" a neural network is simply the p r o c e s s of determining a n appropriate set of weights s o that the network accurately approximates the input/output relationship of the training data. W h e n a grid cell is p r o c e s s e d , the v a l u e s a s s o c i a t e d with that cell are s c a l e d automatically by the software a n d b e c o m e the v a l u e s of the input n o d e s . D a t a scaling is e s s e n t i a l for neural networks to be a b l e to properly learn from training d a t a . T h e type of activation function c h o s e n for the output layer determines the range that the training data must be s c a l e d for. T h e s e s c a l e d v a l u e s are then p a s s e d to the n o d e s in the hidden layer c o n n e c t e d to the input n o d e s . H o w e v e r , rather than being p a s s e d directly, they are first multiplied by the weight a s s o c i a t e d with the c o n n e c t i o n s between the n o d e s . T h e receiving n o d e then a p p l i e s a sigmoid function to the s u m of v a l u e s they receive a n d the result is  33  p a s s e d to the n o d e s in the output layer. T h e following p a r a m e t e r s w e r e c o n s i d e r e d a s input for the neural network:  Table3. Input parameters for the neural network Description  Variable name Slope 2.  Slope_NO Slope_NW  Slope for the cell that is processed (percents) Variables 2 to 9 represent side slopes in percents as show in the diagram below:  Slope_WE Slope_SW Slope_SO Slope_SE Slope_EA  Slope_NE  Slope_NO  Slope_NW  Slope_WE  Slope  Slope_EA  Slope_SW  Slope_SO  Slope_SE  Slope_NE Presence or absence of a lake 10.  Lake  1 if a lake is present 0 if there is no lake Presence or absence of a stream  11  Stream  1 if a stream is present 0 if there is no stream  12  QF  13  Curvature  Soil bearing capacity Terrain curvature Presence or absence of a road  14  Road  1 if a road is present 0 if there is no road  34  T h e final neural network architecture is shown in the figure below:  Inputs  Hidden  layer  Output  Slope Slope_NO Slope_NW Slope_WE Slope_SW Slope_SO  Roads  Slope_SE Slope_EA Slope_NE Lakes Streams QF Curvature Fig.5 Architecture of the Neural Network used in modeling the roads  O n c e the architecture and learning method were c h o s e n , a n d the training parameters were set, the training s e s s i o n started with s a m p l e data (approximately 7 0 % of the entire dataset), s o the internal weights c a n start to adjust and learn patterns and trends in the data. Initially, the weights a s s i g n e d to each connection are set to random values. T h e weights are then adjusted b a s e d upon the training data by undergoing many iterations of a training function, such  35  a s the back-propagation. T h e back-propagation is the most c o m m o n l y u s e d training algorithm for neural networks. T h e weights (w) a r e updated a s follows:  Awij(t) = -ri  where  SE(t) Swij(t)  is the learning rate a n d  + aAwij(t -1)  a  is the m o m e n t u m .  T h e learning rate parameter controls the rate at w h i c h the neural network learns a n d the m o m e n t u m parameter controls the influence of the last weight c h a n g e o n the current weight update. Higher v a l u e s for t h e s e parameters lead to faster learning but this c a n c a u s e the network to be unstable a n d stop learning. T h e relationship learned by the N N must be tested before being u s e d in predictions, in order to determine the quality of the k n o w l e d g e . T e s t i n g of the trained N N requires that inputs, but not the outputs, for another portion of the training data be presented to the N N . T h e a c c e p t e d output is c o m p a r e d to the output p r o d u c e d by the N N to s e e if the N N ' s output is a c c e p t a b l e . If the output p r o d u c e d by the trained N N is correct within a c c e p t e d error r a n g e s ( R M S error < 0.1), the N N c a n be u s e d with n e w data. T h e testing dataset w a s represented by the remaining 3 0 % of the dataset.  36  5.2. Predicting on new data  After the net is trained, it w a s u s e d to predict b a s e d o n input data. T h e results of t h e s e predictions c o n s i s t e d in a numerical representation of the "likelihood" of a given cell to contain a road. T h e resulting v a l u e s range from 0 to 1, with 0 representing cells with the lowest likelihood to contain a road (for e x a m p l e a cell that is on a lake) and 1 representing cells with the highest likelihood to contain a road (for e x a m p l e a cell that is on a n existing road). T h e resulting v a l u e s w e r e represented in a G I S format a s a new cost s u r f a c e .  5.3. Creating the cost surface  T o calculate the cost value for e a c h cell, a new field w a s a d d e d to the d a t a b a s e and the cost w a s calculated using the equation [1]:  Cost =  [3] Likelihood  In other w o r d s , the cost of having a road on a specific cell is inversely proportional to the likelihood of that cell to contain a road (the higher the likelihood, the smaller the cost a n d vice versa). T h i s new variable w a s utilized in producing the cost surface on which the A r c l N F O optimal path function w a s run.  37  5.4. The optimal path algorithm  With the addition of this n e w resultant cost surface, an optimal path function w a s run to trace the optimal path between two or more cells. Instead of calculating the actual d i s t a n c e ( E u c l i d e a n distance) from o n e point to another, the function determines the shortest weighted d i s t a n c e (or a c c u m u l a t e d travel cost) from e a c h cell to the nearest cell in the set of s o u r c e cells. T h e weighted d i s t a n c e functions apply "distance" not in g e o g r a p h i c units but in cost units.  T h e optimal path function will return either the least cost path between the source(s) a n d destination(s) or a corridor of the least cost paths, respectively. In our c a s e , the s o u r c e cells represent the existing road network a n d the destination cells represent landings.  T r a c i n g the optimal path is a two-step p r o c e s s : - R u n the P A T H D I S T A N C E function - R u n the C O S T P A T H function  T h e P A T H D I S T A N C E function c a l c u l a t e s the a c c u m u l a t i v e cost o v e r a cost surface, a n d a l s o c o m p e n s a t e s for the horizontal a n d vertical factors influencing the total cost of moving from o n e location to another. T h e results from P A T H D I S T A N C E function are u s e d a s input in C O S T P A T H a n d C O R R I D O R grid functions. T h e s e functions return either the least cost path b e t w e e n the source(s)  38  a n d destination(s) or a corridor of the least cost paths, respectively. In our c a s e , the s o u r c e cells represent the existing road network a n d the destination cells represent landings. T h e P A T H D I S T A N C E function calculates the shortest weighted distance (or a c c u m u l a t e d travel cost) from e a c h cell to the nearest cell in the set of s o u r c e cells. T h e function creates three output grids: 1. the cost-distance grid 2. the cost-backlink grid 3. the cost-allocation grid  1. T h e first step in calculating the cost-distance  grid is to identify the s o u r c e  cells. T h e n the cost to travel to e a c h neighbour that adjoins a s o u r c e cell is determined. Next, e a c h of the neighbour cells is ordered from least costly to most costly in a list. T h e cell location with the least cost is then r e m o v e d from the list. Finally, the least-accumulative cost to e a c h of the neighbours of the cell that w a s just r e m o v e d from the list is determined.  Active accumulative cost cell list  0  rj  2.0  5.7  2.5 6.4  2.5 7.1  3.5  4.0  4.5  4.5  4.9  I Value=No Data 0  0  Source cells Neighbourhood cell to be added to active list  Fig 6. T h e algorithm for calculating the accumulative cost  39  T h e p r o c e s s is repeated until all cells on the grid h a v e b e e n a s s i g n e d a n accumulative cost. T h e grid d o e s not s h o w which s o u r c e to return to or how to get there. T h e accumulative v a l u e s are b a s e d on the cost unit specified o n the cost surface. 2. T h e cost- backlink grid with a value range from 0 to 8 that c a n be u s e d to reconstruct the route to the s o u r c e . E a c h value (0 through 8) identifies which neighbouring cell to m o v e into to get back to the s o u r c e . T h e s o u r c e cells are a s s i g n e d to 0 s i n c e they have r e a c h e d the goal (the source). If a cell on the back-link is a s s i g n e d 1, the cell immediately to its right will lead to the s o u r c e with the least costly path. T h e lower right diagonal cell will be a s s i g n e d 2, directly south would be value 3 and s o forth.  1 7 3 3 3 0  0 1 8 5  0 0 7 7  4  4  5  5  5  4  5 5 6  6 3  4 5  5  5 6 3  6  7  5  0  1  4  3  2  8  4 5 5  Fig 8: Back-link positions in the back-link grid  Fig 7: T h e algorithm for calculating the cost back-link grid  40  3 . T h e cost-allocation  grid identifies which cells will be allocated to which  s o u r c e , on the b a s i s of the lowest accumulative cost to reach the s o u r c e .  1 1 1 1 1 1 1 1 1 1 1 1 2 1 1 1 1 2 2 2 1 HI 2 2 2 2  2 2  2 2  2  2 2  2 2  F i g 9: T h e algorithm for calculating the cost allocation grid  In simple terms, the cost-distance grid represents h o w m u c h it would cost e a c h cell to reach a cell location from a source via the least-cost path. T h e costj  allocation grid defines, for e a c h cell, the least costly s o u r c e a n d the cost backlink identifies how to retrace the path to the s o u r c e from the destination location.  41  5.5. Model calibration and testing  T h e value of a prediction with a model is only a s g o o d a s the ability of the m o d e l to be effectively validated. T h e validation involves testing the m o d e l ' s predictive capabilities with information other than that u s e d in running the m o d e l . In other w o r d s , the road m o d e l e d h a s to be c o m p a r e d with the s a m e existing road from the field. Although the existing road might not be perfect from a n engineering point of view, it is the best k n o w l e d g e representation w e have with respect to that road. Therefore, the m e a s u r e of s u c c e s s for a m o d e l e d road is represented by the similarity of that road with the existing road from the field. T h e basic validation a p p r o a c h w a s to replicate a n existing road from the study a r e a a n d then to c o m p a r e the m o d e l ' s output to the existing road with respect to the length and spatial autocorrelation coefficient. T h e p r o c e s s has two p h a s e s : -  the visual p h a s e in which the newly m o d e l e d road is a s s e s s e d visually a n d in which a meaningful range of v a l u e s for the P A T H D I S T A N C E function parameters is established  -  the analytical p h a s e in which the length a n d the spatial autocorrelation coefficient w e r e calculated and c o m p a r e d for different runs a n d the best parameters w e r e c h o s e n to predict a new road in the study a r e a  42  Spatial autocorrelation is a measure of the similarity of objects within a n area. ( A r c G I S help file). A spatial object c a n have at least two types of attributes: 1. aspatial attributes, s u c h a s : elevation, slope, aspect etc) 2.  spatial attributes represented by the location of objects in a specified coordinate s y s t e m . From an object b a s e d view of spatial data, spatial autocorrelation measures the relationship between the difference of the aspatial attributes of objects with the distance between the objects.  In our c a s e the a-spatial attributes are represented by a B o o l e a n value (0 for cells with no road and 1 for cells with roads) a n d spatial attributes are represented by x, y coordinates for a specific object. Figure 10 presents two spatial objects (in this c a s e two roads) which are close together, have similar a-spatial descriptors and therefore are highly spatially correlated. Figure 11 presents an example of low spatial autocorrelation where objects with similar a-spatial descriptors are more far apart in s p a c e .  • 0. 08 Fig. 10: A n example of high spatial autocorrelation  Fig. 11: A n example of low spatial autocorrelation  43  G o o d c h i l d (1996) identifies two objects that are very c l o s e together a n d h a v e very different a-spatial attributes a s having a negative spatial autocorrelation. In the context of A r c l N F O G R I D module, the objects c o r r e s p o n d to cells a n d the a-spatial descriptors or attributes c o r r e s p o n d to cell v a l u e s .  T h e formula for calculating the C O R R E L A T I O N index is:  n  n  n  k  k  k  T h e general notation u s e d in correlation formulas a n d their G R I D interpretation are the following: n - the total n u m b e r of cells in a grid i - any cell on the first input grid j - a n y cell on the s e c o n d input grid Zj - the value of the attribute of cell i Zj - the value of the attribute of cell j  > - the m e a n value of the attribute of the first grid  z  Z j  - the m e a n value of the attribute of the s e c o n d grid  Cy - the similarity of i's a n d j's attributes :  z  ^  z  ^  44  In the terms of the a b o v e notation, spatial autocorrelation is simply a m e a s u r e of the attribute similarities in the set of cij with the locational similarities, a n d then s u m m i n g the results into a single coefficient ( G o o d c h i l d , 1986). O n c e the model is s u c c e s s f u l l y calibrated, it c a n be u s e d to project future roads in the study a r e a . B e c a u s e t h e s e predictions cannot be verified at the time, they must be understood in terms of probabilities. T h e predictions are thought to be a g o o d starting point for the specialist in developing a n e w forest road. E v e n if the route s e l e c t e d is not perfect from an engineering point of view, it still c a n provide g o o d information for the forest road-planning engineer. F o r preliminary planning the important thing is the relative u s e f u l n e s s of road routes, not the exact road layout. T h e route s e l e c t e d by the computer should be modified b a s e d o n the k n o w l e d g e a n d e x p e r i e n c e of the specialist a s well a s the exact terrain conditions s u r v e y e d in the field.  45  6. R e s u t s l and d s ic u s s o in s T h i s chapter presents the results for the 5 - c a s e study a n a l y z e d ( s e e T a b l e 2) E a c h study a d d r e s s e s a certain combination of factors that influence road d e s i g n a n d construction, including: terrain conditions, stream c r o s s i n g s , road complexity a n d configuration, etc. T o help evaluate the performance of the m o d e l , for e a c h road studied, the model w a s run on two cost s u r f a c e s : o n e generated by the neural network, which a c c o u n t e d for all 12 parameters (see T a b l e 3) a n d the other by a more traditional a p p r o a c h ( S e s s i o n s , 1993), w h i c h typically a c c o u n t s for only two parameters, s l o p e a n d curvature. T h e evaluation criteria u s e d for the p r o p o s e d road w a s the spatial autocorrelation coefficient calculated for the following combinations: 1. neural net road - existing road 2. slope-curvature road - existing road  H50 w a s the first road m o d e l e d . It h a s a relatively s i m p l e layout but w a s constructed on very steep terrain (Fig 12), making it suitable to test the m o d e l ' s performance in high s l o p e a r e a s . T h e s e c o n d road m o d e l e d w a s H90, with a more c o m p l e x d e s i g n , on very steep slope and with three s w i t c h b a c k s (Fig13). T h e road H110  is the third road m o d e l e d , in a n a r e a with m a n y s t r e a m s a n d  l a k e s with a moderate to steep s l o p e . It is c o m p o s e d of two s e g m e n t s that join together in a " Y " s h a p e , e a c h s e g m e n t linking a landing to the main road (Fig 14). T h e forth road m o d e l e d is A42, which is a very simple s e g m e n t of r o a d , o n a n a r e a that is not steep but is c r o s s e d by n u m e r o u s s t r e a m s (Fig. 15). T h e last  46  road m o d e l e d w a s LOO, o n the North side of L o o n L a k e , a n d it is c o m p o s e d of four s e g m e n t s of road linked together. T h i s experiment w a s c o n d u c t e d to test the applicability of the model in a r e a s with no roads a n d therefore no possibility to train the Neural Network. T h e training dataset utilized c o m e s from H 9 0 - H 1 1 0 a r e a (Fig 16), that h a s similar characteristics with the a r e a North of L o o n L a k e w h e r e the model w a s applied.  The Vertical Factor T h e P A T H D I S T A N C E function utilized in the model w a s run using different v a l u e s for the vertical factor ( V F ) parameter. T h i s parameter t a k e s into a c c o u n t the cost n e c e s s a r y to o v e r c o m e the slope between two cells. T o calculate the V F , the vertical angle or s l o p e is calculated between cells from the z v a l u e s a s s i g n e d to e a c h location a n d then the slope is correlated to a vertical factor o n a g r a p h . T h e r e m a y be a threshold angle s u c h that if the s l o p e e x c e e d s this a n g l e , then the cost is s o great that it b e c o m e s a barrier to travel. T h i s threshold is referred to a s the cut-angle.  T h e V F is a s s i g n e d to infinity w h e n the s l o p e  e x c e e d s the cut-angle. E x a m p l e s of the cut-angle v a l u e s a r e s h o w n in the table below: T a b l e 4 C u t - a n g l e v a l u e s for road H 5 0 Runs  Cut-angle (percents)  Cut-angle (degrees)  1  100  45  2  30  16.7  3  10  5.7  47  F o r H 5 0 , the model w a s run using a range of v a l u e s for the cut-angle, from 1% to 100%. A visual inspection of the m a p (Fig 12) s h o w s that the s m a l l e r the cuta n g l e v a l u e is, the c l o s e r the m o d e l e d road is to the original. T h e best fit is a c h i e v e d at a cut-angle value of 1 0 % , which is suitable for forest road d e s i g n . O n c e the appropriate parameters w e r e e s t a b l i s h e d , the model w a s run a n d the road length a n d spatial autocorrelation coefficient w e r e c a l c u l a t e d . F o r the other roads, the P A T H D I S T A N C E function w a s run without using the cutangle parameter and therefore without limiting the m a x i m u m s l o p e for the road. B y removing the cut-angle from the equation, other parameters than s l o p e (e.g. s t r e a m c r o s s i n g , lakes, networking, etc) b e c a m e more important in c h o o s i n g the optimal path for this road. H o w e v e r , this led to creating a p r o p o s e d road with s l o p e s greater than the m a x i m u m s l o p e a d m i s s i b l e for a forest road. B y introducing the cut-angle parameter in the P A T H D I S T A N C E function, there is a risk that the algorithm stops w h e n no viable alternatives (cells with s l o p e s m a l l e r than the cut-angle parameter value) c a n be found. In real life, w h e n situations like this o c c u r , the planner c a n prescribe blasting and/or cut a n d fill operations to o v e r c o m e the o b s t a c l e .  48  The road length T a b l e 5 s h o w s that the lengths of the neural network m o d e l roads are c l o s e r to their c o r r e s p o n d i n g existing road than the length of the slope-curvature model r o a d s . T h i s is important b e c a u s e the cost of building a n d maintaining a road is directly related to the length of that road.  Table5: R o a d lengths table for all roads m o d e l e d Road Name  Length (m) Original  Neural  S l o p e - Curvature  Network H50  1065  1059  1383  H90  1554  1652  2609  H110  1151  1004  1459  A42  714  856  976  LOO  3406  3401  8401  T h e neural network road length varies from 8 7 % of the original road in c a s e of H 1 1 0 to 1 2 0 % of the original in c a s e of road A 4 2 . T h e roads predicted with the slope-curvature model vary in length from 1 2 7 % for H 9 0 to 2 4 7 % for L O O .  49  Stream c r o s s i n g s B y visually inspecting all the m a p s , it c a n be s e e n that in all c a s e s the roads m o d e l e d with the neural network have fewer stream c r o s s i n g s than the roads m o d e l e d using slope-curvature (also s e e T a b l e 6). T h i s indicates that the neural network has learned from the training dataset that stream c r o s s i n g s are " e x p e n s i v e " a n d therefore a s s i g n e d a high cost to t h e m . B e c a u s e of the high cost, the optimal path algorithm, which is trying to minimize the total cost of the road, avoided those a r e a s .  T a b l e 6 S t r e a m c r o s s i n g s for all roads m o d e l e d Road  Stream crossings Original  Neural Network  Slope-Curvature  H50  0  0  0  H90  0  0  3  H110  0  0  6  A42  2  2  1  LOO  5  2  -  50  The spatial autocorrelation coefficient T h e spatial autocorrelation coefficient w a s calculated for all the m o d e l e d roads, to s h o w s the d e g r e e of similarity between the original r o a d , the neural network road a n d the slope-curvature road.  T a b l e 7 : S p a t i a l autocorrelation coefficient v a l u e s for all r o a d s m o d e l e d Road  Spatial autocorrelation coefficient  modeled  Original - Neural Network  Original - S l o p e - C u r v a t u r e  H50  0.40  0.12  H90  0.41  0.10  H110  0.30  0.08  A42  0.29  0.08  LOO  0.28  0.04  T a b l e 7 s h o w s that in all c a s e s the spatial autocorrelation coefficient b e t w e e n the original road and the neural network model is m u c h higher than the spatial autocorrelation coefficient calculated for the s l o p e curvature m o d e l . T a b l e 8 in A p p e n d i x 2 s h o w s a graphical representation of the spatial autocorrelation differences between the neural network model a n d the s l o p e curvature m o d e l .  51  7. Conclusions and Recommendations  T h e r e s e a r c h c o n d u c t e d for this thesis attempted to determine whether artificial neural networks c o m b i n e d with G I S t e c h n i q u e s might be useful in preliminary d e s i g n of roads in forestry. First, a cost surface w a s derived using a neural network algorithm (back-propagation) a n d then a n optimal path algorithm ( P A T H D I S T A N C E ) w a s run on the generated cost surface. Finally, the resulting roads w e r e c o m p a r e d to the s a m e roads generated by a slope-curvature model in terms of length, stream c r o s s i n g s a n d spatial autocorrelation T h e results indicate that forest roads c a n be predicted with a certain d e g r e e of a c c u r a c y , although the a c c u r a c y of the prediction m a y not be suitable for s o m e applications. In all c a s e s , the neural network model performed better than the slope-curvature m o d e l . T h i s technique m a y be useful for practical forestry in situations w h e n very a c c u r a t e G I S data is available a n d the cost of preliminary field work is prohibitively e x p e n s i v e . T h e utility of this method will i n c r e a s e o n c e high a c c u r a c y data collection t e c h n o l o g i e s b e c a m e cost effective a n d provide very a c c u r a t e terrain data for G I S modeling, (e.g. ' L I D A R - Light Detection A n d R a n g i n g - with approximately 1 meter horizontal a c c u r a c y a n d 0.3 meters vertical a c c u r a c y ) .  52  T h i s project is just a starting point for the c o m p u t e r modeling of forest road planning using G I S . T h e r e is m u c h room for improving a n d exploring. F r o m a n operational standpoint it would be useful to redo the a n a l y s i s using T R I M (Terrain R e s o u r c e Information M a n a g e m e n t ) data, provided by the Ministry of Environment, for an a r e a larger than the M a l c o l m K n a p p R e s e a r c h Forest. O n e of the d i s a d v a n t a g e s of this model is that the predicted roads might h a v e a very short turning radius, b e c a u s e the algorithm is b a s e d o n a cell-by-cell selection and therefore further studies are n e e d e d in order to a c c o m m o d a t e logging vehicle requirements. A n o t h e r shortcoming of the model is that it doesn't take into a c c o u n t earthwork calculations around horizontal c u r v e s a n d through vertical c u r v e s . T h i s , c o m b i n e d with a P A T H D I S T A N C E algorithm that d o e s n ' t u s e C U T A N G L E v a l u e s c a n lead to the creation of steep road s e g m e n t s that are not feasible from a n engineering point of view. Additionally, experiments could be performed to study the m o d e l ' s behavior in a r e a s with no road network, or w h e r e the road network is poorly d e v e l o p e d . T h i s will allow the integration of this model with other forest m a n a g e m e n t m o d e l s that are usually applied to large forested a r e a s over multiple harvest periods (e.g. F o r e s t P l a n n i n g Studio, C A R P P , etc). O n e limitation of the a p p r o a c h taken by this r e s e a r c h is that in the p r o c e s s of training the artificial neural network utilized only o n e algorithm (backpropagation). Future work might attempt to investigate this i s s u e , by testing different algorithms with different training parameters.  53  O n e of the most beneficial improvements would be to d e v e l o p a userinteractive and integrated s y s t e m , w h e r e the road locating procedure c a n be u s e d for drawing the road network a n d w h e r e the u s e r s would be allowed to m a k e c h a n g e s to the network thus d e s i g n e d .  54  8. List of references:  A h a m e d , K . M . , S . E . R e u t e b u c h , T . A . Curtis. 2 0 0 0 . A c c u r a c y of highresolution airborne laser data with varying forest vegetation cover. 2nd International  Conference  on Earth Observation  Information,  11-14 November  2000, Cairo,  and  Environmental  Egypt.  A n d e r s o n , A . and N e l s o n J . 2 0 0 3 - P r o c e d u r e s for projecting a n d evaluating forest road networks in strategic plans, Master thesis, University Columbia,  Faculty  of  of  British  Forestry  C h a r n i a k , E. and Mcdermott, D. 1985 - Introduction to Artificial Intelligence. Addison-Wesley  publishing  Company,  Massachusets  Coulter, E . D . , C h u n g W . , A k a y . A , S e s s i o n s , J . 2001 - F o r e s t road layout using a high resolution digital terrain model generated from L I D A R d a t a . In preparation University  for the First International of Washington,  Precision  Forestry  Symposium,  Seattle  C r a i g , R . F . 1992 - Soil M e c h a n i c s . Chapman  and Hall, London,  1994  D e a d m a n , P . B r o w n , R . D . Gimblett, H.R. 1993 - M o d e l i n g rural residential settlement patterns with cellular automata. Journal of Management,  Vol.37, pp:  Environmental  147-160  D o u g l a s R . A . a n d H e n d e r s o n , B . S . 1987. C o m p u t e r a s s i s t e d forest road route location. Proc of the Council on Forest Engineering High technology  in forest engineering,  10  1987. Syracuse,  th  annual  meeting.  New York. Pp:  201-  214  55  Dykstra, D P . 1984. Mathematical programming for natural r e s o u r c e m a n a g e m e n t . McGraw-Hill  Book Co., New York.  318p  E a s s o n , G . L . ,Barr D . J . - Integration of G I S a n d Artificial Neural Networks for Natural R e s o u r c e Applications. Proceedings Palm Springs,  California,  May 20-24  of the ESRI user  1996  E S R I , 1992 - C e l l b a s e d modeling with G R I D . ArclNFO Environmental  Systems  Research  conference  User's  Guide.  Institute  F o r d , L. R., F u l k e r s o n D. R. 1956 M a x i m a l Flow through a Network. Journal  of Mathematics,  vol. 8, pp.  Canadian  578-580  Gimblett, H.R. 1989 - D y n a m i c modeling in G I S - a cellular automaton a p p r o a c h to modeling the growth behavior of urban a n d rural development. GIS National  Conference  1989. Ottawa,  Canada  G o l d b e r g D. 1991 - G e n e t i c algorithms in s e a r c h , optimization a n d m a c h i n e learning. Addison-Wesley  publishing  Company,  Massachusets  G o o d c h i l d , M . F . 1 9 8 6 - S p a t i a l Autocorrelation. University Ontario,  London,  Ontario,  of  Western  Canada  H a m m o n d , C , Hall, D., Miller, S . and Swetik, P . 1992 - L e v e l I stability analysis (LISA) Documentation. US Department Service,  Table  of Agriculture,  Forest  5.5  H o g a n , J . D . 1973. U s i n g c o m p u t e r s in forest road location a n d d e s i g n . Canadian  Forest Industries  1993(7) pp.  34-37  K l i n k a , K 1975 E c o s y s t e m m a p s for University of British C o l u m b i a R e s e a r c h Forest, M a p l e R i d g e , B C , C a n a d a 56  K o b a y a s h i , H. 1984 P l a n n i n g s y s t e m for road-route locations in mountainous forests. Journal  of the Japanese  Forest Society  66(8) pp:  313-319  Kouichi Ichihara, T o s i m i T a n a k a , Isao S a w a g u c h i , Shuji U m e d a and K a t s u m i T o y o k a w a 1996 - T h e method for d e s i g n i n g the profile of forest roads supported by genetic algorithm. Can. J. For. Res. Vol.1, pp:  45-49  Murray, A T . 1998 - R o u t e planning for harvest site a c c e s s . Can. J. Vol.28, pp.  For.Res.  1084-1087  M i n a m i k a t a , Y . 1984 Effective forest road planning for forest operations a n d the environment. Proc. COFE/IUFRO University  of Maine,  Orono,  Conference,  Aug 11-18 1984 at  pp:219-224  N e w m h a n , R . M . 1995 - R O A D P L A N , a tool for d e s i g n i n g forest road networks. Journal  of Forest Engineering  Vol.6, No.2, pp:  17-26  N i e u w e n h i u s , M.A. 1986. D e v e l o p m e n t of a forest road location procedure a s a n integral part of a m a p b a s e d information s y s t e m . PhD Thesis, University Maine, Ann Arbor, Michigan  193p.  R o g e r s , L., S c h i e s s , P. 2001  - P E G G E R a n d R O A D V I E W - A n e w G I S tool  to a s s i s t e n g i n e e r s in operations planning. The international Logging  and 11  th  Pacific Northwest  Skyline  Symposium  of  Mountain  2001, pp.  177-182  S e s s i o n s , J . L i u , K . 1993 - Preliminary planning of road s y s t e m s using Digital Terrain M o d e l s . Journal of Forest Engineering,  Vol.9,  No.2  S e s s i o n s , J . 1987. A Heuristic Algorithm for the Solution of the V a r i a b l e a n d Fixed C o s t Transportation P r o b l e m . In Proc: The 1985 Symposium System Analysis  in Forest Resources.  Univ. of Georgia,  Athens,  pp.  on 324-336.  57  S e s s i o n s , J . a n d J . B . S e s s i o n s . 1988. A s c h e d u l i n g a n d network a n a l y s i s program for tactical harvest planning. In: Proceedings Mountain  Logging  Oregon.  December  and Pacific Northwest  Skyline  of the  International  Symposium,  Portland,  12-16, 1988. p. 71- 75.  S m i t h , David K. 1982 - Network Optimization P r a c t i c e : a computational guide. Ellis Horwood  Limited,  pp.55-92  S u i , D.Z. 1992 - A n initial investigation of integrating Neural Networks with G I S for spatial d e c i s i o n making. Proceedings  ofGIS/LIS  1992, San  Jose  California  T a n . J , 1992 - Planning a forest road network by a spatial data handlingnetwork routing s y s t e m . Acta Forestalia  Fennica  227,  1992  W a l k e r , H.D. and L o g h e e d , W . H . 1989 R o a d network d e s i g n in w o o d supply analysis. The Forestry  Chronicle,  December  1989,  pp:431-440  W a n g , F. 1992 - Incorporating a Neural Network into G I S for agricultural land suitability a n a l y s i s . Proceedings  ofGIS/LIS  1992, San Jose  California  W i j n g a a r d , P . J . M . and R e i d e r s , M . P . 1985 Optimisation of a forest road network. Netherlands  Journal  of Agriculture  Science  33 (1985) pp:  175-179  58  Appendix 1 - Maps  lBC,Jm»20O1 Fig. 12 M a p of road H50  60  Preliminary Road Layout for MKRF H90 Model 1  Model 2  U B C , July 2001  F i g . 1 3 M a p o f road H 9 0  Preliminary Road Layout for MKRF H 1 1 0 - Study 1 0 Neural Network Model  Slope - Curvature Model  Legend: Roads — —  H110 Original  Brisling Road Network  Contour Lines 5m interval  Length = 1150.70 m  Model4 (NN, noHF,VF=1.0) — —  Length = 1004.27 m  Streams  Model8 (SC, noHF, VF-1.0)  Lakes  — —  Projection UTM. Zone 10 Datum: NAD83 Canada  Length = 1459.39 m 0  80  160  320  'ISO  640 I Meters  lonut Aran, MF Candidate UBC. June 2001  Fig. 14 M a p of road H 1 1 0  62  P r e m il n ia r yR o a dL a y o u t for M K R F A42 Study 1  Legend:  Study 1  Contour Lines 5m Interval  Roads  Model 4(NN, noHF, Slope - 1.0) — —  Length = 714.57 m  Fig 15 M a p of road A 4 2  Study 2 Model 1(NN, noHF, Hcutangle - 6)  Length = 910.98 m  Length = 855 98 m  Model 8(SC, noHF, Slope - 1.0)  Streams A42 Original — —  Study 2  • Length = 98279 m  Model 5 (NN, noHF, Hcutangle - 6) — —  75  150  300  Length = 976.19 m  450  • Meters 600  Projection U T M , Zone 10 Datum: NAD 83 Canada  lonut Aron, M F Candidate UBC, June 2001  P r e m il n ia r yR o a dL a y o u tf o rM K R FL O O H110  Fig 16 M a p of road LOO 64  A p p e n d i x 2 - Spatial autocorrelation  Table 8 Spatial autocorrelation comparison table for the 5 roads modeled Road  H50  Original - Neural Network  A  /\  C = 0.12  C = 0.40  H90  \ C = 0.41  H110  Original - S l o p e Curvature  C = 0.10  Y If C = 0.30  C = 0.08  66  Road  Original - Neural Network  A42  C = 0.29  LOO  C = 0.28  k  Original - S l o p e Curvature  C = 0.08  A C = 0.04  67  A p p e n d i x 3: Soil bearing capacity  T h e ultimate soil bearing capacity (qf) is defined a s the least p r e s s u r e w h i c h would c a u s e s h e a r failure of the supporting soil immediately below a n d adjacent to a foundation ( R . F . C r a i g , 1992). T h e ultimate soil bearing capacity c a n be e x p r e s s by T e r z a g h i ' s formula:  q = -yBN cN DN l  f  r+  c+  q  [5]  Where:  Y - unit weight of a soil a n d represents the ratio between total weight a n d total volume B - width of footing (B = 0.80m) c - s h e e r strength parameter D - depth of footing (D = 0.10m) N, r  N, c  N  q  - bearing capacity factors  T h e v a l u e s f o r ^ and c have b e e n obtained from L e v e l I Stability A n a l y s i s D o c u m e n t a t i o n by H a m m o n d et al (1992) and the v a l u e s forAf,, N , c  N  q  have  b e e n extrapolated from a graph from C r a i g (1992), pp 3 0 5 .  68  A p p e n d i x 4: A r c l N F O functions and procedu  1. S L O P E - identifies the rate of m a x i m u m c h a n g e in z v a l u e from e a c h cell.  SLOPE(<grid> { D E G R E E | P E R C E N T R I S E } ) S L O P E ( < g r i d > , <z_factor>, { D E G R E E | P E R C E N T R I S E } ) Arguments <grid> - any valid combination of grids, s c a l a r s , n u m b e r s , operators a n d functions that p r o d u c e s a n output grid. { D E G R E E | P E R C E N T R I S E } - keywords specifying the units in w h i c h the v a l u e of s l o p e will be e x p r e s s e d . D E G R E E - with this option the inclination of s l o p e will be calculated in d e g r e e s . P E R C E N T R I S E - keyword to output the percent rise, a l s o referred to a s the percent s l o p e . <z_factor>- the number of ground x, y U N I T S in o n e surface Z U N I T . T h e <grid> Z U N I T S are multiplied by the specified <z_factor> to adjust the output grid Z U N I T S to another unit of m e a s u r e . T h e default v a l u e of the <z_factor> is 1. Higher z v a l u e s will result in a more e x a g g e r a t e d relief (surface) a n d thus in a more extreme s h a d i n g .  2 . B U F F E R - c r e a t e s buffer polygons a r o u n d specified input c o v e r a g e features.  B U F F E R <in_cover> <out_cover> {buffer_item} { b u f f e r j a b l e }  {buffer_distance}  { f u z z y j o l e r a n c e } {LINE | P O L Y | P O I N T | N O D E } { R O U N D | F L A T } { F U L L | L E F T | RIGHT} Arguments <in_cover> - the c o v e r a g e containing features to be buffered. <out_cover> - the c o v e r a g e to be created. {buffer_item} - an item in the feature attribute table of <in_cover> w h o s e v a l u e is u s e d a s the feature's buffer distance. If a buffer table is u s e d , the {bufferjtem} functions a s a lookup item in {buffer_table}.  70  { b u f f e r t a b i e } - an I N F O lookup table which lists a buffer d i s t a n c e for e a c h {bufferjtem}. A {buffer_table} c a n be specified only if {bufferjtem} is specified. T h e {buffer_table} contains at least two items.  {bufferjtem} - defined the s a m e a s {bufferjtem} in the <in_cover> feature attribute table. T h e {buffer_table} must be sorted o n this item in a s c e n d i n g order. D I S T - the buffer distance for e a c h {bufferjtem} value. D I S T must be defined a s a numeric item (i.e., N, I, F, or B). A lookup table will c a t e g o r i z e item v a l u e s . T h e u s e of lookup tables is d e s c r i b e d in M a n a g i n g T a b u l a r D a t a , {bufferjdistance} - the distance u s e d to create buffer z o n e s around <in_cover> features w h e n {bufferjtem} or {bufferjtem} a n d {buffer_table} are not s p e c i f i e d . T h e default distance is 0.125 c o v e r a g e units. T h i s default buffer d i s t a n c e will be applied w h e n e v e r this argument is s k i p p e d with a '#', omitted, or a d i s t a n c e of 0 is s p e c i f i e d . T h e smallest buffer distance that c a n be c o m p u t e d is 0 . 0 0 0 0 0 0 0 5 c o v e r a g e units. Specifying a buffer distance below this threshold will result in a n empty output cover. F o r polygon features, if a negative buffer distance is u s e d , buffers will be generated on the insides of polygons. {fuzzy_tolerance} - the minimum distance between coordinates in the output c o v e r a g e . A value of 0 will not be a c c e p t e d . T h e default f u z z y tolerance is determined a s follows: 1) T o l e r a n c e v a l u e s are read from the c o v e r a g e T O L file if it exists. 2) If no T O L file exists a n d the width of the B N D is between 1 a n d 100, the tolerance is 0.002. 3) O t h e r w i s e , the tolerance is 1 / 1 0 , 0 0 0 of the width or height of the B N D , w h i c h e v e r is greater. {LINE | P O L Y | P O I N T | N O D E } - the feature c l a s s to be buffered: L I N E - arcs will be buffered. T h i s is the default option. P O L Y - polygons will be buffered. P O I N T - points will be buffered. N O D E - n o d e s will be buffered.  71  { R O U N D | F L A T } - for lines, the s h a p e of the buffer at the line end points. R O U N D will m a k e an end in the s h a p e of a half-circle. F L A T will construct rectangular line endings with the middle of the short side of the rectangle coincident with the end point of the line. { F U L L | L E F T | R I G H T } - for lines, the buffer m a y be generated on either side ( F U L L ) , or a "half buffer" may be generated to the topological left ( L E F T ) or right s i d e of a line ( R I G H T ) . S e e notes for a further explanation of the c o n c e p t of a half buffer.  3 . I N T E R S E C T - c o m p u t e s the geometric intersection of two c o v e r a g e s . O n l y those features in the a r e a c o m m o n to both c o v e r a g e s will be p r e s e r v e d in the output c o v e r a g e .  I N T E R S E C T <in_cover> <intersect_cover> <out_cover> { P O L Y | L I N E | P O I N T } { f u z z y j o l e r a n c e } {JOIN | N O JOIN}  Arguments  <in_cover> - the c o v e r a g e w h o s e polygon, line or point features will be intersected with the <intersect_cover>. <intersect_cover> - the overlay c o v e r a g e containing polygon features. <out_cover> - the c o v e r a g e to be c r e a t e d .  { P O L Y | LINE | POINT} - the <in_cover> feature c l a s s to be overlaid a n d preserved in the <out_cover>.  P O L Y - polygon-on-polygon overlay. T h i s is the default option. L I N E - line-in-polygon overlay. P O I N T - point-in-polygon overlay.  72  {fuzzy_tolerance} - the minimum d i s t a n c e between c o o r d i n a t e s in the output c o v e r a g e . B y default, the minimum f u z z y t o l e r a n c e v a l u e from input a n d intersect c o v e r a g e s is u s e d . It is determined a s follows:  1) T o l e r a n c e v a l u e s are read from the existing c o v e r a g e T O L file. 2) If no T O L file exists a n d the width of the B N D is between 1 a n d 100,  the  tolerance is 0.002. 3) O t h e r w i s e , the tolerance is 1 /10,000 of the width or height of the B N D , w h i c h e v e r is greater. A f u z z y tolerance value of 0 will not be a c c e p t e d by I N T E R S E C T . {JOIN | NOJOIN} - s p e c i f i e s whether all items in both the <in_cover> feature attribute and <intersect_cover> P A T will be joined into the output c o v e r a g e feature attribute table.  J O I N - all feature attribute items from both c o v e r a g e s will a p p e a r in the output c o v e r a g e feature attribute table. If a duplicate item is e n c o u n t e r e d , the item in the input c o v e r a g e will be maintained a n d the o n e in the join file will be d r o p p e d . T h i s is the default option and is u s e d u n l e s s N O J O I N is s p e c i f i e d .  N O J O I N - only the feature's internal n u m b e r (cover#) from the input c o v e r a g e a n d the intersect c o v e r a g e are joined in the output c o v e r a g e feature attribute table. T h i s option is useful in reducing the size of the output c o v e r a g e feature attribute table. T h e relate environment is then u s e d to relate to other tables containing the attributes of output c o v e r a g e features.  73  4. P A T H D I S T A N C E - calculates, for e a c h cell, the least-accumulative-cost distance over a cost surface from a s o u r c e cell or a set of s o u r c e cells while accounting for surface distance a n d horizontal a n d vertical cost factors.  P A T H D I S T A N C E ( < s o u r c e _ g r i d > , {cost_grid}, {surface_grid}, {horiz_factor_grid}, {horiz_factor_parm}, {vert_factor_grid}, {vert_factor_parm},  {o_backlink_grid},  {o_allocate_grid}, {max_distance}, {value_grid})  Arguments <source_grid> - a grid that identifies those cells from w h i c h a least-accumulativecost distance is calculated to e a c h cell. T h e input value types c a n be either integer or floating point. {cost_grid} - a grid defining the i m p e d a n c e or cost to m o v e planimetrically through e a c h cell. T h e value at e a c h cell location represents the cost per unit of surface distance for moving through the cell. E a c h cell location value is multiplied by the cell resolution (while a l s o c o m p e n s a t i n g for diagonal movement) to obtain the total cost of p a s s i n g through the cell. T h e v a l u e s o n the {cost_grid} c a n be integer or floating point, but they cannot be negative (you c a n n o t h a v e a negative cost). {surface_grid} - a grid identifying the z - v a l u e s at e a c h cell location. T h e v a l u e s are u s e d to calculate the actual surface distance that will be c o v e r e d w h e n p a s s i n g between cells. {horiz_factor_grid} - a grid defining the horizontal direction at e a c h cell. T h e v a l u e s on the grid must be integers ranging from 0 to 360 with 0 d e g r e e s being north, or towards the top of the s c r e e n , a n d increasing c l o c k w i s e . Flat a r e a s s h o u l d be given the value of - 1 . T h e v a l u e s at e a c h location will be u s e d in conjunction with the {horiz_factor_parm} in order to determine the horizontal cost incurred w h e n moving from a cell to its neighbours. {horiz_factor_parm} - defines the relationship between the horizontal c o s t factor a n d the horizontal relative moving angle (HRMA). T h e input parameter m a y be o n e of several keywords (and modifiers) identifying a defined horizontal factor  74  graph or an A S C I I file that creates a c u s t o m g r a p h . T h e g r a p h s are u s e d to identify the horizontal factor that will be u s e d in calculating the total cost for moving into a neighbouring cell. If the parameter is u s e d with a n y modifier, the string must be double quoted.  T h e format for the horizontal parameter is: "keyword { m o d i f i e M ... modifier_n}" If no keyword is s p e c i f i e d , the default horizontal parameter is B I N A R Y . In the explanations of the horizontal factor keywords a n d modifiers, two a c r o n y m s are u s e d : - H F s t a n d s for the horizontal factor defining the horizontal difficulty that is e n c o u n t e r e d in moving from o n e cell to the next. - H R M A stands for the horizontal relative moving a n g l e , which identifies the angle between the horizontal direction from a cell and the moving direction.  Horizontal-factor keywords: B I N A R Y - indicates that if the H R M A is less than the cut angle, the H F is set to the value a s s o c i a t e d with the zerofactor; otherwise it is infinity. F O R W A R D - e s t a b l i s h e s that only forward m o v e m e n t is a l l o w e d . T h e H R M A must be greater or equal to 0 and less than 90 (0 <= H R M A < 90). If the H R M A is greater than 0 and less than 4 5 d e g r e e s , the H F for the cell is set to the v a l u e a s s o c i a t e d with the zerofactor. If the H R M A is greater than or e q u a l to 4 5 d e g r e e s , then the side value modifier v a l u e is u s e d . T h e H F for a n y H R M A e q u a l to or greater than 90 d e g r e e s is set to infinity. L I N E A R - defines that the H F is a linear function of the H R M A . I N V E R S E J J N E A R - specifies that the H F is an inverse linear function of the HRMA. T A B L E - identifies that an A S C I I file will be u s e d to define the horizontal-factor graph u s e d to determine the H F s . T h e n a m e of the table is entered a s a modifier after a blank s p a c e following the keyword.  75  Modifiers to the horizontal keywords: Z E R O F A C T O R - e s t a b l i s h e s the horizontal factor to be u s e d w h e n the H R M A is 0. T h i s factor positions the y-intercept for any of the horizontal-factor functions. C U T A N G L E - defines the H R M A angle b e y o n d which the H F will be set to infinity. S L O P E - e s t a b l i s h e s the s l o p e of the straight line u s e d with the L I N E A R a n d I N V E R S E L I N E A R horizontal-factor keywords. T h e s l o p e is specified a s a fraction of rise over run (i.e., 4 5 percent s l o p e is 1/45, which is input a s 0.02222).  S I D E V A L U E - e s t a b l i s h e s the H F w h e n the H R M A is greater than or e q u a l to 4 5 d e g r e e s a n d less than 90 d e g r e e s for the F O R W A R D horizontal-factor keyword is s p e c i f i e d .  t a b l e _ n a m e - identifies the n a m e of the A S C I I table defining the H F . It is u s e d in conjunction with the T A B L E horizontal-factor keyword.  {vert_factor_grid} - a grid defining the z - v a l u e s for e a c h cell location. T h e v a l u e s are u s e d for calculating the s l o p e that is u s e d to identify the vertical factor incurred w h e n moving from o n e cell to another.  {vert_factor_parm} - defines the relationship between the vertical cost factor a n d the vertical relative moving angle ( V R M A ) . T h e input parameter m a y be o n e of several keywords (and modifiers) identifying a defined vertical factor graph or a n A S C I I file creating a c u s t o m graph. T h e g r a p h s are u s e d to identify the vertical factor that will be u s e d in calculating the total cost for moving into a neighbouring cell. If the parameter is u s e d with any modifier, the string must be double quoted. T h e format for the vertical parameter is: "keyword { m o d i f i e M ... modifier_n}" If no keyword is s p e c i f i e d , the default vertical parameter is S E C .  76  In the explanations of the vertical-factor keywords a n d modifiers, two a c r o n y m s are u s e d : - V F stands for the vertical factor defining the vertical difficulty that is e n c o u n t e r e d moving from o n e cell to the next. - V R M A stands for the vertical relative moving a n g l e , which identifies the s l o p e angle between the F R O M or p r o c e s s i n g cell a n d the T O cell.  Vertical-factor keywords: B I N A R Y - specifies that if the V R M A is greater than the low-cut angle a n d l e s s than the high-cut angle then the V F is set to the value a s s o c i a t e d with the zerofactor; otherwise it is infinity. L I N E A R - indicates that the V F is a linear function of the V R M A . S Y M J J N E A R - defines that the V F is a linear function of the V R M A in either the negative or positive side of the V R M A , respectively, a n d the two linear functions are symmetrical with respect to the V F (y) axis. I N V E R S E J J N E A R - indicates that the V F is an inverse linear function of the VRMA. S Y M J N V E R S E J J N E A R - identifies that the V F is a n inverse linear function of the V R M A in either the negative or positive side of the V R M A , respectively, a n d the two linear functions are symmetrical with respect to the V F (y) axis. C O S - defines the V F a s the c o s i n e - b a s e d function of the V R M A . S E C - identifies the V F a s the s e c a n t - b a s e d function of the V R M A . C O S - S E C - indicates that the V F is the c o s i n e - b a s e d function of the V R M A w h e n the V R M A is negative a n d the s e c a n t - b a s e d function of the V R M A w h e n the V R M A is nonnegative. S E C - C O S - specifies that the V F is the s e c a n t - b a s e d function of the V R M A w h e n the V R M A is negative and the c o s i n e - b a s e d function of the V R M A w h e n the V R M A is nonnegative. T A B L E - identifies that an A S C I I file will be u s e d to define the V F s . T h e n a m e of the table is entered a s a modifier after a blank s p a c e following the keyword.  77  Modifiers to the vertical-factor keywords: Z E R O F A C T O R - e s t a b l i s h e s the vertical factor to be u s e d w h e n the V R M A is 0. T h i s factor positions the y-intercept of the specified function. B y definition, the zerofactor is not applicable to any of the trigonometric vertical functions ( C O S , S E C , C O S _ S E C , or S E C _ C O S ) . T h e y-intercept is defined by t h e s e functions. L C U T A N G L E - defines the V R M A angle below which the V F will be set to infinity. H C U T A N G L E - defines the V R M A angle a b o v e which the V F will be set to infinity. S L O P E - establishes the slope of the straight line u s e d with the L I N E A R and I N V E R S E J J N E A R vertical-factor keyword. T h e slope is specified a s a fraction of rise over run (e.g., 45 percent slope is 1/45, which is input a s 0.02222).  t a b l e _ n a m e - identifies the n a m e of the A S C I I file defining the V F . It is u s e d in conjunction with the T A B L E vertical-factor keyword.  {o_backlink_grid} - the n a m e of the output back-link grid. T h e back-link grid contains v a l u e s from 0 through 8, which is a c o d e identifying the direction of the next neighbouring cell (the s u c c e e d i n g cell) w h e n retracing (from the destination to the source) the least-accumulative-cost path from a s o u r c e cell to a destination.  If the path is to p a s s into the right neighbour, the cell will be a s s i g n e d the v a l u e ' 1 ' , '2' for the lower-right diagonal cell and continuing c l o c k w i s e . T h e value '0' is reserved for s o u r c e cells.  {o_allocate_grid} - the n a m e of the output cost-allocation grid. T h e output grid defines for e a c h cell, the z o n e on the <source_grid> that a c h i e v e s the m i n i m u m cost distance or the least-accumulative cost in order to reach the cell.  {max_distance} - defines the threshold that the a c c u m u l a t i v e - c o s t - d i s t a n c e v a l u e s cannot e x c e e d . If an accumulative-cost-distance value e x c e e d s the  78  {max_distance}, the output value for the cell will be N O D A T A .  The  {max_distance} defines the extent to which the accumulative-cost d i s t a n c e s are computed.  {value_grid} - an optional input grid that identifies the z o n e value that should be u s e d for e a c h cell on the <source_grid>.  T h e value defined by the {value_grid}  for e a c h s o u r c e cell will be a s s i g n e d to all cells that are allocated to the s o u r c e cell location in terms of the minimum-cost distance. T h e default z o n e v a l u e for a s o u r c e cell is the value on <source_grid>.  T h i s grid is particularly important if the s o u r c e grid w a s created by T E S T or a B o o l e a n operator that will only output 1s, Os and N O D A T A , or if alternative v a l u e s or z o n e s are to be u s e d instead of the existing o n e s o n the s o u r c e input  Default V a l u e s for Horizontal F a c t o r Modifiers  Keywords  Zerofactor  Cutangle  Slope  Sidevalue  Binary  1.0  45  -  -  Forward  0.5  4 5 (fixed)  -  1.0  Linear  0.5  181  1/90  Inverse linear  2.0  180  -1/90  Default V a l u e s for Vertical F a c t o r Modifiers  Keyword  Zerofactor  Lcutangle  Hcutangle  Binary  1.0  -30  30  L i n e a r 1.0  -90  90  1/90  Symjinear  1.0  -90  90  1/90  Invjinear  1.0  -45  45  -1/45  -45  45  -1/45  S y m _ i n v e r s e _ l i n e a r 1.0  Slope Cospower  79  Cos  -90  90  1.0  Sec  -90  90  Cos sec  -90  90  1.0  1.0  S e c cos  -90  90  1.0  1.0  1.0  5. C O S T P A T H - p r o d u c e s a n output grid that records the least-cost path(s) from s e l e c t e d cell(s) in the input <fromcell_grid>, or from interactive selection on the display, to the c l o s e s t s o u r c e cell defined within the <accumcost_grid> in terms of cost distance.  C O S T P A T H ( < f r o m c e l l _ g r i d > , <accumcost_grid>, <backlink_grid>, { B Y C E L L | BYZONE | BYLAYER}) C O S T P A T H ( < * > <accumcost_grid>, <backlink_grid>, { B Y C E L L | B Y L A Y E R } , {verify_grid})  Arguments:  <fromcell_grid> - a grid that identifies those cells from w h i c h the least cost path is determined to the least costly s o u r c e . T h e grid c o n s i s t s of cells w h i c h are to be c o n s i d e r e d in the C O S T P A T H calculations having valid v a l u e s ('0' is a valid value), a n d the remaining cells must be a s s i g n e d to N O D A T A .  <accumcost_grid> - the n a m e of a cost distance grid to be u s e d to determine the least cost path from the <fromcell_grid> cell locations to a s o u r c e . T h e <accumcost_grid> w a s created with the C O S T D I S T A N C E (or by the C O S T A L L O C A T I O N or C O S T B A C K L I N K functions) or P A T H D I S T A N C E functions. T h e <accumcost_grid> stores, for e a c h cell, the minimum a c c u m u l a t i v e cost distance o v e r a cost surface from e a c h cell to a set of s o u r c e cells.  80  <backlink_grid> - the n a m e of a cost back link grid u s e d to determine the path to return to a s o u r c e via the least-cost path. F o r e a c h cell in the b a c k link grid, a value identifies the neighbour that is the next cell on the least accumulative cost path from the cell to a single or set of s o u r c e cells.  { B Y C E L L | B Y Z O N E | B Y L A Y E R } - a keyword defining the m a n n e r in w h i c h the v a l u e s and z o n e s on the <fromcell_grid> or from the interactive input will be interpreted in the cost path calculations.  B Y C E L L - for e a c h cell with valid v a l u e s on the <fromcell_grid> or from the interactive input, a least-cost path is determined and s a v e d on the output grid of the C O S T P A T H function. With the B Y C E L L keyword, e a c h cell of the <fromcell_grid> input is treated separately, and a least-cost path is determined for e a c h 'from' cell.  B Y Z O N E - for e a c h z o n e on the <fromcell_grid> or from interactive input, a single least-cost path is determined and s a v e d o n the output grid of the C O S T P A T H function. With the B Y Z O N E keyword, for e a c h z o n e , the least-cost path begins from the cell with the lowest cost distance weighting in the z o n e .  B Y L A Y E R - for all cells on the <fromcell_grid> input, the least cost path is derived from the cell with the minimum of the least cost paths to s o u r c e cells.  <*> - allows for the interactive (graphical) entrance of the from cells. A dialogue is entered and the user is prompted for the points from w h i c h to calculate the least cost path. T o enter a point, the cursor is positioned o v e r the desired location, and the appropriate button on the m o u s e is clicked (the appropriate button is specified in the dialogue). If the point is not the o n e d e s i r e d , a s e c o n d button is d e p r e s s e d . T o stop entering points, the appropriate button o n the m o u s e specified by the dialogue is d e p r e s s e d . O n l y the locations of single cells c a n be c h o s e n with the <*> option w h e n identifying the from cells in w h i c h cost  81  paths are to be calculated. There is a maximum of 100 cells or picks with the cursor.  {verify_grid} - defines the grid whose cell values will be printed for verification when interactively entering the input locations with the <*> argument. Once a cell is chosen (by depressing the appropriate button on the mouse), the value for the cell location on the {value_grid} will be displayed on the screen. If the cell is the one that is desired, another cell can be chosen or the session can be ended. If the value that is printed to the screen is not the cell that is desired, the cell input can be canceled by depressing the appropriate button on the mouse. The value that is printed to the screen is not used in the processing of the COSTPATH function.  82  AML procedure for side slope calculation /* Calculates the side slope for a cell /* /* /* /*  ul ! up ! ur  /*  .1  i  le 'cell! ri  /* /*  .1  I*  i  II ! lo ! Ir !  I*  .1  i  I* I* Upper left &if [exists slope_ul -grid] &then kill slope_ul all /* Upper &if [exists slope_up -grid] &then kill slope_up all /* Upper right &if [exists slope_ur -grid] &then kill slope_ur all I* Right &if [exists slope_ri -grid] &then kill slope_ri all /*Lower right &if [exists s l o p e j r -grid] &then kill s l o p e j r all /* Lower &if [exists s l o p e j o -grid] &then kill s l o p e j o all  /* L o w e r left &if [exists s l o p e j l -grid] &then kill s l o p e j l all /* Left &if [exists s l o p e j e -grid] &then kill s l o p e j e all  DOCELL s l o p e _ u l = ( a b s ( h 5 0 d e m j n t ( - 1 , -1) - h 5 0 d e m j n t ( 0 , 0))) / 1 . 4 1 / 5 * 100 s l o p e _ u p = ( a b s ( h 5 0 d e m j n t ( 0 , -1) - h 5 0 d e m j n t ( 0 , 0))) / 5 * 100 s l o p e _ u r = ( a b s ( h 5 0 d e m j n t ( 1 , -1) - h 5 0 d e m j n t ( 0 , 0))) / 1 . 4 1 / 5 * 100 slope_ri = ( a b s ( h 5 0 d e m j n t ( 1 , 0) - h 5 0 d e m j n t ( 0 , 0))) / 5 * 100 s l o p e J r = ( a b s ( h 5 0 d e m J n t ( 1 , 1) - h 5 0 d e m j n t ( 0 , 0))) / 1 . 4 1 / 5 * 100 s l o p e _ d o = ( a b s ( h 5 0 d e m j n t ( 0 , 1) - h 5 0 d e m j n t ( 0 , 0))) / 5 * 100 s l o p e j l = ( a b s ( h 5 0 d e m j n t ( - 1 , 1) - h 5 0 d e m j n t ( 0 , 0))) / 1 . 4 1 / 5 * 100 s l o p e j e = ( a b s ( h 5 0 d e m j n t ( - 1 , 0) - h 5 0 d e m j n t ( 0 , 0))) / 5 * 100 END  AML procedure for data collection  AUTHOR  /* A  /* /*Original C o d i n g :  ESRI  /* /*N  NAME  /* /*GRIDSPOT70.AML /"Copyright 1995, Environmental S y s t e m s R e s e a r c h Institute, Inc. /* /*P  ^  PURPOSE  /* /* T h i s A M L sets a n item in a point c o v e r a g e P A T equal /* to the value of a grid cell at the corresponding location. It is intended /* to be similar to the L A T T I C E S P O T c o m m a n d except that it d o e s not /* perform interpolation. T h i s A M L requires A R C / I N F O 7.0 or higher.  /* /*U  USAGE  /* I* G R I D S P O T <grid_name> <point_cover> {spot_item} /* /*V  VARIABLES  /* /*  L o c a l variables:  /* /* grid_name N a m e of the grid to take cell v a l u e s from /* ptcov  N a m e of point c o v e r a g e to which cell v a l u e s are a d d e d  /* s p o t j t e m  N a m e of P A T item to which cell v a l u e s are a d d e d . If  /*  omitted, the default item n a m e is S P O T .  /* o l d $ m e s s a g e s  S a v e setting of & m e s s a g e s  I* o l d $ d i s p l a y S a v e setting of D I S P L A Y /* o l d $ e c h o  S a v e setting of & e c h o  /* /*  G l o b a l variables: N O N E  /* /*C  CALLS  /* /* N o n e . /* /*==================DISCLAIMER================================ / * Y o u m a y u s e , copy, modify, m e r g e , distribute, alter, reproduce and/or /*create derivative works of this A M L for your o w n internal u s e . A l l /*rights not specifically granted herein are reserved to E S R I . /* / * T H I S A M L IS P R O V I D E D " A S - I S " W I T H O U T W A R R A N T Y O F A N Y K I N D , EITHER / * E X P R E S S O R IMPLIED, INCLUDING, B U T N O T LIMITED T O , T H E IMPLIED /*WARRANTIES OFMERCHANTABILITY AND FITNESS F O R A PARTICULAR PURPOSE, /*WITH R E S P E C T T O T H E A M L . /* / * E S R I shall not be liable for any d a m a g e s under any theory of law /*related to your u s e of this A M L , e v e n if E S R I is a d v i s e d of the /*possibilites of s u c h d a m a g e . T h i s A M L is not supported by E S R I . /*============================================================  & a r g s g r i d _ n a m e ptcov s p o t j t e m &severity &error &routine bailout  /*  Argument Checking  86  &if [show program] ne ' A R C &then &return T h i s ami must be run from A R C  &if [null %grid_name%] &then &return U s a g e : G R I D S P O T <grid> <point_coverage> {spot_item}  &if  A  [exists % g r i d _ n a m e % -grid] &then  &return % g r i d _ n a m e % is not a grid.  &if [null %ptcov%] &then &return U s a g e : G R I D S P O T <grid> <point_coverage> {spot_item}  &if  A  [exists % p t c o v % -point] &then  &return % p t c o v % is not a point c o v e r a g e .  &if [null %spot_item%] &then &sv s p o t j t e m S P O T  & s o l d $ m e s s a g e s [show & m e s s a g e s ] & m e s s a g e s &off  /* C h e c k whether s p o t j t e m exists in the point c o v e r a g e P A T  &if [iteminfo %ptcov%.pat -info % s p o t J t e m % -exists] &then &type W A R N I N G : Existing item % s p o t J t e m % in % p t c o v % . P A T will be recalculated. &else  additem %ptcov%.pat %ptcov%.pat % s p o t J t e m % 4 12 f 3  /* N o w go into Arcplot a n d get the v a l u e s  & s v old$display [show display] display 0  ap /* D e c l a r e and o p e n a c u r s o r to read and write to the P A T  cursor ptcur d e c l a r e % p t c o v % points rw cursor ptcur o p e n  &s o l d $ e c h o [show &echo] & e c h o &off  /* Start a loop to go through the P A T , find the cell v a l u e at e a c h /* point location, and write it to the S P O T item in the P A T .  & d o &while %:ptcur.aml$next% &s temp [show cellvalue % g r i d _ n a m e % [show select % p t c o v % point 1 xy]] &if [type %temp%] = 1 &then &s :ptcur.%spot_item% -9999 &else & s :ptcur.%spot_item% % t e m p % cursor ptcur next &end  q  /* quit from arcplot  &call exit &return /* T o calling ami /* /*  Routine Exit  /* &routine exit &if [variable old$display] &then  88  display % o l d $ d i s p l a y % &if [variable o l d $ m e s s a g e s ] &then &messages %old$messages% &if [variable old$echo] &then &echo %old$echo% & return /* /*  Routine Bailout  /* &routine bailout &severity &error &ignore &call exit &return; &return &error Bailing out of gridspot.aml  

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.831.1-0074984/manifest

Comment

Related Items