THE ACCURACY OF DEFLECTION-LINESDERIVED FROM DIGITAL ELEVATION MODELSbyDAVID ALEXANDER CHRISTIEB.S.F., The University of British Columbia, 1990A THESIS SUBMITTED iN PARTIAL FULFILMENT OFTHE REQUIREMENTS FOR THE DEGREE OFMASTER OF SCIENCEinTHE FACULTY OF GRADUATE STUDIESDepartment of ForestryWe accept this thesis as conformingto the required standardTIlE UNIVERSITY OF BRITISH COLUMBIAJune 1994© David Alexander Christie, 1994In presenting this thesis in partial fulfilment of the requirements for an advanceddegree at the University of British Columbia, I agree that the Library shall make itfreely available for reference and study. I further agree that permission for extensivecopying of this thesis for scholarly purposes may be granted by the head of mydepartment or by his or her representatives. It is understood that copying orpublication of this thesis for financial gain shall not be allowed without my writtenpermission.(Signature)___________________________Department of -The University of British ColumbiaVancouver, CanadaDate______________DE-6 (2/88)11AbstractLong-reaching skyline cable yarding systems have seen increased use within the BritishColumbia coastal forest industry. Deflection-line analysis, which estimates themaximum yarding distance and locates the harvest boundary, is the key component inplanning for skyline systems. Traditional deflection-line analysis involves field surveyswhich may be very difficult to perform in the terrain associated with skylines. As analternative, deflection-lines may be derived from Digital Elevation Models (DEMs).Concern regarding the elevational accuracy of the topographic forest planning mapsused to create the DEMs has limited their use for deflection-line analysis. Betterunderstanding of the magnitude and nature of elevational errors and their effect upondeflection-line analysis are needed before DEM-derived deflection-lines may be usedwith confidence.This study was performed in cooperation with Canadian Forest Products Limited(Canfor) in Woss, British Columbia (B.C.). Deflection-line analyses were performed forDEM-derived deflection-lines to test for error in yarding distance estimates. Errors inyarding distance estimates for DEM-derived deflection-lines were caused by interactionsbetween some or all of the following: the terrain shape (concavity/convexity), largeelevational errors and their location on the deflection-line, and the deflection-line length.While a majority of yarding distance estimates from DEM-derived deflection-lines werenot in error (70%), the erroneous estimates may result in costly planning errors.111Restricting the use of DEM-derived deflection-lines to the efficient pre-planning of fieldsurveys could help avoid these mistakes.A blunder was detected in one of the study cutblock maps. Distortions were discoveredin the maps for two other study cutbiocks where photogrammetrically derived andground surveyed maps had been joined through rubber sheeting. While random errorwas detected in the analyses, systematic error appeared to contribute more to both thegeneral level of elevational error and to the presence of large elevational errors.Different types of systematic error were detected, with at least some types evident in allof the deflection-line comparisons. Smoothing error was observed where terrainvariation had been reduced or eliminated, and positional errors were the most commonand influential systematic errors detected. The positional error of map features, andpositional error introduced using traditional surveying methods, may also affectoperational field surveying of deflection-lines, logging roads, and harvest boundaries.The presence of positional error and its subsequent effects upon harvest planning iseither not known or is ignored altogether.Detecting the presence of systematic error in topographic forest planning maps is thefirst step towards using DEMs confidently for deflection-line analysis. Further studiesinvolving the effects of positional error on DEM elevational error will allow the DEMsto be predicted and subsequently accounted for. Advances in map creation, computers,and Geographic Information Systems will allow for the acquisition and manipulation ofmore accurate digital elevation data now and in the future.AbstractTable of Contentsiv113.2.53.3 DEM3.3.13.3.23.3.33.3.43.3.53.3.63.3.74 Study Site5 Methods5.1 Cutbiock and Deflection-line5.2 Field Surveys5.3 Digital Elevation Models5.4 Estimating Yarding Distance5.4.1 Statistical Analysisx1466669121213151719202022252527293035viviiTable of Contents ivList of TablesList of FiguresAcknowledgements1 Introduction2 Objectives3 Background and Literature Review3.1 Operational Harvest Planning3.1.1 Terminology3.1.2 Skyline Yarding Systems .3.1.3 Deflection and Deflection-line3.2 Digital Elevation Models3.2.1 Terminology3.2.2 Elevational Data Sources3.2.3 The Regular Rectangular Grid3.2.4 The Triangulated Irregular NetworkAnalysisUsing DEMs in Operational Harvest PlanningAccuracyErrors: Blunders, Random, and SystematicSmoothing ErrorPositional AccuracyAge of Map DataSurveying QualityData RelevanceDEM Accuracy Standards and TestsSelection464646484950V5.5 DEM Elevational Error 525.5.1 Statistical Analysis 546 Results 586.1 Estimating Yarding Distance 586.2 DEM Elevational Error 607 Discussion 637.1 Estimating Yarding Distance 637.1.1 Deflection-line Length and Concavity 637.1.2 Large Elevational Errors 667.2 Elevational Error and Error Patterns 727.2.1 Blunders 737.2.2 Distortions Caused by Rubber Sheeting 757.2.3 Random Error 797.2.4 Systematic Error 837.2.5 Smoothing Error 837.2.6 Effect of Site 867.2.7 Positional Error 887.2.8 AgeofMapData 918 Conclusions 959 Recommendations 100Literature Cited 105Appendix A Sources of Error Affecting DEM Accuracy 110Appendix B Yarding distance estimates 115Appendix C Summaries of DEM Elevational Error 117Appendix D Glossary of Terms 121viList of TablesTable I Summary of cutbiock characteristics 44Table 2 Distribution of deflection-lines and settings within cutbiocks 45Table B Yarding distance (m) estimated for thirty-one deflection-line pairs . . 116Table Cl Summary of DEM elevational error (m) by cutbiock 118Table C2 Summary of DEM elevational error (m) by setting 119Table C3 Summary of DEM elevational error (m) by deflection-linecomparison 120viiFigure 1Figure 2Figure 3Figure 4Figure 5Figure 6Figure 7Figure 8Figure 9Figure 10Figure 11Figure 12Figure 13Figure 14Figure 15Figure 1678101116• . . 2436• 37• 38• 39• 40• 41• 42• 43656518List of FiguresCutbiock design showing settings, landings and yarding roadsDeflection of carriage below chordSkyline yarding system with carriage and payload of logsClearance of payload load-path and selection of boundary location.TIN DEM integrates important spot elevationsRegularly spaced grid points do not properly represent importantfeatures, such as peaks and passes.Deflection-lines overlayed onto a TIN DEMMap of cutblock locations within Canfor operating areaMap of AR1 cutblock and deflection-line locationsMap of AR6 cutbiock and deflection-line locationsMap of AT cutbiock and deflection-line locationsMap of CU cutbiock and deflection-line locationsMap of GC cutbiock and deflection-line locationsMap of NW cutbiock and deflection-line locationsMap of TR cutbiock and deflection-line locationsA positive elevational error in concave terrain resulted in theincorrect placement of the boundary for the NW-2 DEM-deriveddeflection-lineFigure 17 The concave shape of both deflection-lines for CU-2 1 providedadequate clearance for the entire length of the deflection-linesFigure 18 A large elevational error at the end of the GC-5 DEM-deriveddeflection-line resulted in an incorrect boundary location 67vii’Figure 19 Example of a large negative elevational error in the centre of thedeflection-line, causing an overestimate of yarding distance 68Figure 20 The deflection-line analysis piots for AT-202 show that thelanding was located properly on both plots despite the 33 metreunderestimation of yarding distance from the DEM-deriveddeflection-line (upper plot) 71Figure 21 Comparison plot for AR1-il shows divergence of deflection-lines,apparently due to a map blunder 74Figure 22 Comparison plots for AR1-12 shows divergence of deflection-lines, apparently due to a map blunder 74Figure 23 The well matched deflection-line pair for CU-20 is an example ofthe deflection-lines contained entirely within thephotogrammetrically derived portion of the map 78Figure 24 Comparison plot for CU-23 shows a large discrepancy inelevational error between the photogrametrically derived portionof the map (left side of plot) and the ground surveyed portion(right side) 78Figure 25 Deflection-line comparison plot for GC-5 indicates minor, randomvariation in elevations 80Figure 26 Elevational error histogram for GC-5 describes a normal curve 81Figure 27 Elevational error histogram for AT-200 is right skewed, indicatingthe presence of systematic error 81Figure 28 Comparison plot for AT-200 shows a well matched deflection-linepair, with the DEM-derived deflection-line tending to be higher inelevation 82Figure 29 Differing tree heights, due in part to variable sight conditions,have resulted in an even forest canopy which masked the terrainvariation from the photogrammetrist 84Figure 30 Past wind history created an even forest canopy which preventedthe photogrammetrist from detecting terrain variation 85Figure 31 Comparison plot for AT-202 indicates presence of positional error. . . 90ixFigure 32 Comparison plot for AT-203, which shares a common landingwith AT-202 (Figure 31), also indicates presence of positionalerror 90Figure 33 A PIP near deflection-line AR6- 1 helped to reduce the effects ofpositional error when the deflection-line location was transferredto the map 92Figure 34 Deflection-line AR6-3 was not located close to a PIP andtherefore experienced positional error when transferred to the map. . . 93xAcknowledgementsI wish to express my appreciation to both graduate supervisors I have had during thetime of my studies: first to Joe McNeel for initiating the study and for his enthusiasmand sense of humour; and second to Glen Young for taking on this role in the middle ofthe project and for his valuable insights. I would also like to thank the members of mycommittee: Val LeMay for statistical guidance and general advice on all thingsforestry; John Nelson for sharing his practical field experience and down to earthadvice; and Brian Klinkenberg for his GIS expertise and innovative recommendations.I would like to thank the Englewood Logging Division of Canadian Forest Products fortheir financial and technical support. In particular I want to express my gratitude toPhil Winkle for guiding me through some of the more difficult portions of my thesiswith advice, encouragement, and humour. Also with Canadian Forest Products I wouldlike to acknowledge John Argast, Tom Normand, Paul Nuttal, and Warren Jukes forassistance with field surveys and technical advice on planning for skyline systems.The Science Council of British Columbia provided a GREAT award which made thisproject possible. Digital Resource Systems provided TerraSoft and specialized software,and Dan Lemkow took valuable time to advise on the project. Craig Spears of Softreeprovided the Roadeng software and critical advice. Andy and Gordon Cooper ofECCCO Management performed the GPS and laser surveys and gave freely of theirknowledge.I am especially grateful for the support I have received from Astrid van Woudenbergthroughout my entire studies. Her encouragement and dedication were instrumental tothe completion of this study. I especially thank Astrid for her infectious sense ofhumour which has picked me up on so many occasions.Finally, I would like to thank my parents, Doug and Jean Christie, who have once againsupported me through my education with love, encouragement, and more support than ason could dare to ask for.I shall be telling this with a sighSomewhere ages and ages hence:Two roads diverged in a wood, and I -I took the one less travelled by,And that has made all the difference.Robert Frost - The Road Not TakenOf what avail are forty freedoms without a blank spot on the map?Aldo Leopold - A Sand County Almanac11 IntroductionChanging social, economic and terrain conditions have led to increased utilization oflong-reaching skyline cable yarding systems within the British Columbia coastal forestindustry. These systems present new challenges for operational harvest planners.Deflection-line analysis, used to check for clearance between suspended logs and theground, is the foundation of the planning process and especially critical for skylineplanning (Conway 1982). Deflection-line surveys become more difficult when dealingwith the long yarding distances and rough terrain associated with skyline systems. Boththe effectiveness and the productivity of the planning process may suffer. Also,adequate clearance is generally more critical for skyline systems than conventionalsystems such as highlead and grapple yarders.The prospect of using Digital Elevation Models (DEMs), pseudo-three dimensionalrepresentations of the terrain, for deflection-line analysis had been suggested as early as1974 (Burke 1974). Young and Lemkow (1976) developed a Digital Terrain Simulator(a DEM) that could be used for many aspects of forest operations planning, includingdeflection-line analysis. Limitations with speed and storage capabilities of the desktopcomputers used by remote forest operations had prevented the use of such models atthat time. Too often these limitations resulted in a compromise of either the accuracyof the data and/or the integrity of the models used to manipulate the data.2These conditions persisted until the rapid improvements in the abilities of personalcomputers in the late 1980s. Coincident with these improvements came the increasingsophistication of Geographic Information Systems (GIS), computerized mapping andanalysis systems. Integrated DEMs and GIS would allow deflection-line analysis toincorporate relevant information from the GIS database. This combination could helpmeet the increasing challenge for better planning for forest operations.While personal computer limitations have been reduced, concern regarding theelevational accuracy of the source data, primarily large scale topographic forest planningmaps, continues. Quantification of elevational accuracy is necessaiy before a DEM maybe used confidently for deflection-line analysis. If the accuracy is within acceptablelevels, then DEMs may be used. If the level of DEM accuracy is unacceptable then abasis is needed on which to set standards for the creation of suitable elevational data. Ifa technique can be developed to rectify or at least quantify and classify this elevationalerror, more planning would be possible using DEMs.This study was initiated by Canadian Forest Products Limited (Canfor) in Woss, BritishColumbia (B.C.). Canfor wished to assess the feasibility of using DEMs, developedfrom their 1:5000 scale topographic planning maps, as tools for deflection-line analysis.Field surveyed deflection-lines were compared with deflection-lines derived from thesame locations on DEMs. Deflection-line analyses were performed for each pair to seeif the yarding distance estimate for the DEM-derived deflection-line was influenced byelevational error. The elevational error between paired points on the field surveyed andDEM-derived deflection-line were quantified and analyzed. The nature and potentialcauses of the elevational error were investigated, analyzed, and discussed.Recommendations have been made regarding the best approach to using existingtopographic maps, as well as the acquisition of new elevation data, for deflection-lineanalysis.342 ObjectivesThis study investigated the elevational accuracy of deflection-lines derived from DigitalElevational Models (DEMs). The DEMs were developed from large scale topographicforest planning maps. Specific objectives were:1) to assess the accuracy of yarding distances estimated from DEM-deriveddeflection-lines;2) to quantify the level of elevational error of DEM-derived deflection-lines;3) to investigate the nature and potential causes of the elevational error of theDEM-derived deflection-lines;4) to develop recommendations for analyzing elevational error in DEM-deriveddeflection-lines, which are applicable to other maps, DEMs, and geographiclocations;5) to develop recommendations for the best use of DEM-derived deflection-lines,including recommendations for minimizing or predicting the systematic errorwhich affects the elevational error of the DEM-derived deflection-lines; and6) to develop recommendations for acquiring and using new elevational data forDEM-derived deflection-line analysis.563 Background and Literature Review3.1 Operational Harvest Planning3.1.1 TerminologyA cutbiock is the smallest individual unit in operational harvest planning. A cutblock isaccessed by a road or roads, which allow access for yarding equipment, logging trucks,and logging crews. A landing is a widening in the road, where the road is more or lesslevel and the skyline yarding machine has the best access to the logs in the cutblock.Yarding is the process of transporting logs from where they are felled within thecutblock, to the landing, where they are loaded onto logging trucks. The path the logsfollow while being yarded is called a yarding road. The area yarded from oneindividual landing is called a setting, and several yarding roads will usually be neededto reach all of the logs within that setting. One cutblock may contain several settings,as multiple landings are often needed for yarding the entire cutbiock (Figure 1).3.1.2 Skyline Yarding SystemsOperational harvest planning has become more demanding for coastal forest operationsin British Columbia. Two major interrelated forces have driven this change. The firstis the shift of harvesting into steeper, more rugged terrain. The second stems fromenvironmental concerns regarding the high densities of roads and ground disturbance7Figure 1 Cutbiock design showing settings, landings and yarding roadsassociated with conventional cable yarding systems (Sauder et a!. 1987). New, morerestrictive plans require the use of alternative harvesting systems that reduce these andother detrimental impacts of conventional yarding. For these reasons, a number ofBritish Columbia forest companies have acquired long-reaching, skyline cable yardingsystems. Skyline systems can help alleviate these problems while maintaining anacceptable level of productivity (Sears 1991, McNeel et al. 1991, Chittick 1991).A skyline is a wire rope suspended between two or more points (Conway 1982). Theskyline yarding system uses a carriage which moves along a skyline. This provides fullclearance for logs when they are yarded to the landing (Figure 2). A skyline systemcan avoid dragging logs on the ground during yarding which other conventional cableYarding roads Cutbiock- boundaryF.-Setting boundary8Figure 2 Deflection of carriage below chord.systems (highlead and grapple yarder) cannot. Fully suspended logs minimize soil andsite disturbance and concurrently reduce damage to logs and equipment. Thesedisturbances have been linked to soil erosion and landslides in sensitive areas (Sauder eta!. 1987). The skyline system also offers considerable flexibility for meeting otherharvesting objectives of protecting environmentally sensitive areas. For example, manycutbiocks contain creeks with critical riparian wildlife habitat. With the full suspensioncapability of the skyline system, suspended logs may be lifted over standing timber leftto protect this habitat.The long reach capabilities of the skyline system reduce the need for roads, particularlyin sensitive mid-slope areas with unstable soils (Sauder et a!. 1987, Hemphill 1991).9Road failures and subsequent landslides from mid-slope roads have been a major sourceof controversy between the forest industry and the environmental movement.Public concern regarding the aesthetic aspects of forest harvesting has prompted theMinistry of Forests to increase the emphasis on Visual Quality Objectives (VQO), ameasure of the alteration of the landscape due to human activity (Winkle 1992, Preus1992). Mid-slope roads are particularly undesirable since they create a stark contrast tothe natural surroundings. Slides caused by the construction of these roads exacerbatethe problem.3.1.3 Deflection and Deflection-line AnalysisThe one condition that is an absolute necessity for skyline systems in any situation isdeflection (Conway 1982). Deflection is the vertical distance, or sag, between thecarriage and an imaginary chord connecting the top of the skyline supports at either endof the skyline (Figure 3). The greater the deflection the heavier the payload of logs thatthe skyline system can support. This usually equates to higher production and loweryarding costs. As well, when dragging logs on the ground is avoided, site damage andequipment damage is minimized.Skyline systems are most effective for yarding over long distances and in ruggedterrain. For these same situations, it may be very difficult to obtain adequate deflectionnecessary for the desired payload of logs. The most important aspect of planning for10skyline systems is performing deflection-line analyses. Deflection-line analysis uses aprofile of the ground (a deflection-line) to determine if there is sufficient clearance ofpayload at the required deflection (Figure 4). This process is also an integralcomponent in determining the maximum possible yarding distance which in turn dictatesthe harvest boundary location. Insufficient clearance at a critical point on thedeflection-line will prevent achievement of the required deflection, and yarding will notbe possible beyond that point.Deflection-lines are obtained through surveys run in the proposed location of the skylineyarding roads. Stations, key points where elevations are determined, are situated atsignificant changes in the slope (10% or more) along the deflection-line survey. TheFigure 3 Skyline yarding system with carriage and payload of logs.11.4for logsClearanceMaximum yarding distanceFigure 4 Clearance of payload load-path and selection of boundary location.elevations of these stations are the most critical for clearance.Deflection-line analysis may be performed by manual or automated methods. While themanual approach is more common, it is also very simplistic and restricted. Deflection-line analysis works best as an interactive process where several different combinationsof road, landing, and boundary locations may be investigated to find the best overallsolution. When using manual, field-based methods, the number of differentcombinations which may be analyzed is limited, especially for skyline cutblocks, andoverall planning consequently suffers.Harvestboundary12Automated methods, which require computers, may be used to perform very complexand comprehensive analyses. These automated methods may be combined withdeflection-lines derived from a Digital Elevation Model (DEM). This extremely rapidapproach to deflection-line analysis allows for a complexity of planning not possibleusing manual methods. While automated deflection-line analysis should never replacefield surveys, it should be used to pre-plan those surveys, ensuring that the most criticalareas are checked in the field. This is becoming essential as pressure for better forestpractices increases.3.2 Digital Elevation Models3.2.1 TerminologyDigital Elevation Models (DEMs) are pseudo-three dimensional computer models whichBurrough (1986) defined as any “digital representation of the continuous variation ofrelief over space. Webb (1990) described them as a “representation of a terrain surfaceconsisting of X, Y, Z coordinates stored in digital form.H The term Digital TerrainModel (DTM) is often used synonymously with DEM. Burrough (1986) distinguishedthe two by specifying that DEMs contain only elevational data and that the word terrainspecifies additional information (slope, aspect) about the landscape. The term DEM hasbeen used here since this study dealt solely with elevational data.13There are two distinct types of DEMs: the Triangulated Irregular Network (TIN) and theRegular Rectangular Grid (Grid). It is common practice to refer to DEMs and Grids asone and the same. From this, a comparison between DEMs and TINs is often madeincorrectly when the latter is actually a specific example of the former.3.2.2 Elevational Data SourcesPhotogrammetrically-measured contour maps are the most common form of modemtopographic maps (Petrie 1991). An overlapping pair of aerial photographs, called astereoscopic pair, is used to determine the elevation of ground features. This is possiblebecause of stereoscopic vision, which allows an observer to gain the impression ofdepth when viewing an object from two different viewpoints (Wolf 1980). This is thesame principle by which human eyes perceive the depth of an object. The apparentdisplacement of the object is called parallax, and the parallax difference between thebottom of an object and the top may be used to calculate the height of that object(Avery and Berlin 1985).Photo interpretation is the Metection, identification, description, and assessment ofsignificance of objects and patterns imaged on a photograph” (Wolf 1980). A photo-interpreter may use parallax differences to calculate the height of trees on a stereoscopicpair of aerial photographs (Loetsch and Hailer 1964). The same principles are utilizedin a more complex process to determine ground elevations at different points on thephotographs. When the area of concern is heavily forested it may be extremely difficult14to see the ground (Loving 1980). Openings where the ground is visible are used tomeasure both the ground elevation and the height of the adjacent trees. These treeheights are then used to estimate ground elevations where the ground is not visible.Aerial photography can be a relatively inexpensive method to obtain relevant data fortopographic mapping over a large area. The area to be mapped is flown in a series ofroughly parallel flight lines while vertical photographs are taken of the surface.Photographs are taken so that they overlap approximately 60 percent in the direction ofthe flight lines and 20 to 30 percent between flight lines (Avery and Berlin 1985). Thisprovides necessary stereoscopic coverage for the entire mapping area.A photo-interpreter may produce either analog or digital topographic maps from aerialphotographs. Analog maps are made directly from the aerial photographs withoutdigital information being stored, but maps must subsequently be converted to digitalform if they are to be used for automated deflection-line analysis. This process istedious, time consuming, expensive, and may introduce significant error into the digitalmap.This undesirable conversion process may be avoided altogether using recent techniqueswhich allow digital elevation data to be digitized directly from aerial photographs. Thisapproach has gained strong support with the advent of Geographic Information Systemsand DEMs that are supported on powerful new personal computers.153.2.3 The Regular Rectangular GridThe regular rectangular grid is the most common and most readily available form ofDEM (Burrough 1986). It consists of a regular rectangular grid containing Cartesiancoordinates in three-dimensions (Peucker et at. 1976). While there are many sources ofdata from which grids may be derived, digitized topographic contour maps aretraditionally the most commonly used (Maedel and Gaudreau 1989). The BritishColumbia government has recently created the Terrain Resource Information Mapping(TRIM) digital terrain maps. While these maps were created digitally and with amodern coordinate referencing system, they were also created at a scale of 1:20000 andwith a spot height densities appropriate for 20-metre contour intervals. The large scale(1:5000) topographic contour maps used in B.C. coastal forest planning typically havecontour intervals of 5 to 10 metres. It is unlikely that the TRIM maps could meet theaccuracy of these topographic planning maps.Grid DEMs are created by mathematically overlaying a grid of points onto the digitalcontour map and interpolating the grid point elevations from the neighbouring contours(Maedel and Gaudreau 1989). The many different algorithms used for interpolatingthese points tend to ‘smooth out’ important terrain features such as peaks, ridgelines,gulleys, rockbluffs, and saddles. If these terrain features are not properly represented,error in yarding distance estimation may occur during deflection-line analysis. Figure 5shows a scenario where a peak and adjacent ridge, indicated by spot elevations, werenot properly represented by regular grid points. Grid elevations are interpolated from16neighbouring contour and spot elevations and a weighting algorithm may be used togive more emphasis to the closest elevations. Unless a grid point falls directly on thespot elevation, the feature represented by the spot elevation will not be properlyrepresented by the interpolated grid elevations.The predominance of regular rectangular grid DEMs is largely due to the ease withwhich they can be handled by the computer (Burrough 1986). Creation, storage andmanipulation of elevational data is easily accomplished when it is in the regular gridformat. A variety of useful information may be derived from a grid DEM such ascontours, slope and aspect, hill shading, and automatic basin delineation. Grids are alsoeasy to conceptualize.17Problems associated with using regular grid DEMs in irregular terrain are welldocumented (Burrough 1986; Peucker et at. 1978). Foremost is their inadequacy indescribing irregular features within a regular framework. For example, a grid resolutionthat accurately displays the roughest terrain on a map results in great data redundancy inthe areas of lesser variation. This high density of data creates problems with storageand with the speed of generation, handling and manipulation of the DEM. Conversely,if resolution is decreased to reduce redundancy then accurate terrain representation iscompromised. While multiple-pass models exist which can increase grid density forareas of higher variation, a regular sampling pattern is still utilized and redundancy stilloccurs.3.2.4 The Triangulated Irregular NetworkThe Triangulated Irregular Network (TIN) model is based upon the philosophy that it isbetter to represent an irregular surface within an irregular data structure. In theirpioneering work, Peucker et a!. (1978) determined that it was necessary to have anelevation model where the locations of data points were dictated by the relief of thesurface being modelled. This model would also have to be computatively efficient.TINs represent surfaces through locating data points at key topological features such aspeaks, pits, passes, ridges and channels (Peucker et a!. 1976). This results in irregularlyspaced points which are connected by lines to form a continuous sheet of triangles.TIN models work best in areas with sharp breaks in slope where the edges of the18triangle may be aligned with those same breaks (Goodehild and Kemp 1990). Like thegrid model, the TiN model may also be used to derive other useful information.While TIN models may be created from contour data, proper selection of data pointsshould begin at the aerial photograph interpretation stage. Rather than deriving contoursfrom the photographs the interpreter should choose points that accurately represent all ofthe critical terrain features. Points are placed precisely on peaks, pits, and passes, andtriangle sides aligned very closely to ridge lines, break lines, and channels. The TINmodel thus allows data points to be concentrated in areas of complex relief while fewerpoints are collected from areas of smooth relief (Burrough 1986; Goodchild nd). Thisallows for both efficient and accurate modelling of areas with variable terrain. Figure 6Regularly spaced grid points do not properly represent importantfeatures, such as peaks and passes.19shows a TIN representing the same data set from Figure 5. The critical pointsrepresented by the spot elevations are incorporated directly into the TiN without anysmoothing of their elevations.3.2.5 Using DEMs in Operational Harvest PlanningThe potential of using DEMs for forest harvest planning had been suggested as early as1974 (Burke 1974). Young and Lemkow (1976) developed a Digital Terrain Simulator(a DTM) that could be used for many aspects of forest operations planning, includingdeflection-line analysis. Limitations with speed and storage capacities of desktopcomputers of that time prevented the efficient use of such models. These limitationsoften resulted in a compromise of the accuracy of the data andlor the integrity of themodels used to manipulate the data.These conditions persisted until rapid improvements in the abilities of personalcomputers (PCs) occurred in the late 1980s. PC-based Geographical InformationSystems (GIS) have allowed harvest planners to create and use their own complexDEMs. These GIS’s have employed mostly grid DEMs because of their ease ofhandling when using computers (Burrough 1986; Macdel and Gaudreau 1989). Thesemodels are poor representations of the ground surface and they have not achieved thefull potential of DEMs in forest harvest planning.20The TiN model provides a method by which surfaces may be represented withaccuracies which should be acceptable to the demands of harvest planning. Asdescribed above, in order to utilize TIN’ s to their fullest potential, they should becreated from data collected specifically for triangular representation. Creating a TINfrom contour maps or grid will add additional error. Petrie (1990) stated that theaccuracy of photogrammetrically derived contours is typically only one third that of spotheights measured directly from the same aerial photographs.Obviously remapping will not be immediately practical for many forest operations. Inthe interim, existing analog contour maps may be used so long as proper considerationis given to their limitations with respect to accuracy. Thus, when the opportunity arisesfor remapping, the operator will have an increased awareness of the accuracy issuespertaining to the use of DEMs in forest operations planning.3.3 DEM Accuracy3.3.1 Errors: Blunders, Random, and SystematicThe United States Geological Survey (USGS) classifies DEM errors into threecategories: blunders, random errors, and systematic errors (Caruso 1987). Blunders aregross errors that are usually easy to detect and therefore edit. It should be noted thatblunders which go undetected could have severe effects upon DEM accuracy and theaccuracy of resulting analysis.21Random error is lack of precision caused by measurement error. This could occur whenthe photo-interpreter makes an erroneous measurement of ground elevation for one pointon the aerial photograph. This error would be a ‘blip’ on the map, not consistent withthe surrounding errors. If these errors are truly random, then they will tend to cancelout with increased sample size. For example, a surveyor reading a compass makes arandom measurement error of one degree to the west. Since this error occurred bychance the more measurements the surveyor takes, the more chance the original errorwill be cancelled out by a random measurement error of one degree to the east.Systematic errors may be due to bias in measurements. For example, a photo-interpreter may consistently underestimate the height of trees on the aerial photographswhen trying to estimate the ground elevation. This would lead to a consistentoverestimation of ground elevations, spread evenly across the resulting map, whichwould be fairly easy to correct (Shearer 1991).If more than one factor is causing bias, then there may be an interaction effect on theerror. Tree height is somewhat dependent upon topography, with taller trees in valleybottoms and shorter trees on ridge tops. If the photo-interpreter assumes all the trees tobe the same height, then ground elevations may be overestimated in the valley bottomand underestimated on the ridge. This would result in systematic error spread unevenlyacross the map which takes more skill to detect and correct.22The presence of systematic error confounds attempts to measure random error (Li1991). This study was focused on the causes and nature of the systematic error. Therewere two major sources of systematic error: smoothing error and positional error.Appendix A contains a summary list of potential sources of error which may also affectthis study but which were beyond its scope.3.3.2 Smoothing ErrorSmoothing error occurs when natural terrain variation has been lost or ‘smoothed out’due to measurement error, interpolation patterns, or data transformations. Where thevariation is not represented properly by contours, grid points, or triangle edges, then thecontours, grid points, or triangle edges are positionally incorrect. Increased samplingcan reduce smoothing error by increasing the representation of terrain variability. Theuse of intelligent interpolation routines which incorporate local terrain characteristics topredict terrain variability is an alternative to increased sampling. While smoothing erroris a result of positional error, the effects of smoothing error will be consideredindependent since they have particular importance to this study.The causes of smoothing error may be considered in two steps. The first step includesthe processes involved in the creation of the analog source maps. K.C. Soel, Limited,created a significant portion of the 1:5000 scale topographic maps on Vancouver Islandincluding those for Canfor’s Englewood Division (Soel 1992). Soel claimed that thecreation of accurate contour maps from aerial photographs is a combination of the23practical field experience of the interpreters, combined with their ability to see theground through the forest canopy. Combs (1980) stated that success in photo-interpretation l largely depends on the training and experience of the interpreter,characteristics of objects to be studied, and the quality of the photographs being used.”Using contours to represent terrain relief will contribute to smoothing error. Bychoosing set contour intervals, detail located between those intervals is systematicallyexcluded. When contours are created from aerial photographs, a fixed elevation ischosen and that elevation is traced from the photographs to create the contour (LaPrade1980). When tracing an elevation through rough terrain, it is very difficult to accuratelycapture all of the detail. Sharp terrain breaks could be somewhat rounded. Transfer ofinformation from one media to another and drafting can also have an effect which willcontribute to smoothing error as well.The second step includes the conversion of contour maps to digital form and thesubsequent extraction of the deflection-lines. When cartographic lines, such ascontours, are represented “as sets of digitized points joined by straight line segments”(Veregin 1989), generalization error occurs. The degree of generalization error willnormally increase as line complexity rises (Burrough 1986). For example, when handdigitizing contours, there is a tendency to literally cut corners on the contours as terrainvariability increases. This generalization error in the contours will equate to smoothingerror in DEM-derived deflection-lines.24Additional elevation points could help alleviate the effects of smoothing error. Spotelevations could either be added for hill tops, ridges, and river or creek courses, or theycould be extrapolated from adjacent contours.The TIN creation process connects elevation points into a continuous sheet of triangles.There should be no loss of accuracy in this step because all of the original informationis retained. This also holds true for the deflection-line extraction procedure. Elevationsfor individual deflection-line points will be interpolated from their location on a trianglewhich should not add any new error (Figure 7).253.3.3 Positional AccuracyFor this study, positional accuracy refers to the horizontal or planimetric accuracy ofany feature represented on a map, analog or digital. Veregin (1989) defined positionalaccuracy as the accuracy of feature locations after transformations have been applied.If the process of measuring a feature for graphical and/or digital representation can beconsidered a transformation, then this definition supports the terminology used for thestudy.Positional error in a feature’s map location could exacerbate the elevational error of thatfeature (Veregin 1989). When the feature is not properly located on the DEM, theextracted elevations will most likely be in error. Positional error may be introduced inmost stages of map making and DEM creation and it is heavily dependent upon the ageof the data, the quality of surveying used to collect the data, and upon the relevance ofthe data to the intended use (Burrough 1986).3.3.4 Age of Map DataTie-points are features which are easily identified both in the field and on the map.Surveys are conducted from tie-points to deflection-lines, effectively ‘tying’ themtogether. This helps locate the deflection-line when it is transferred to the maps. Theaccuracy of tie-points was a very important aspect of this study.26The age of the data used to represent a dynamic feature on a map influences itsaccuracy as a tie-point. Water bodies such as lakes and creeks are most commonlyused. Over time the shore lines and banks may change. The older the map data, themore likely that this has occurred.Photo-identification-points (PIPs) are features which are easily identifiable in the fieldfor which the exact coordinates have been determined at the time of map making.Unfortunately, many of these features are subject to change with age. For example,lone trees in openings are often used for PIPs. These trees may die and fall over,rendering them useless as a PIP.Logging roads are quite often used as tie-points for deflection-lines. Logging roadsurveys are affected by the quality of the surveying methods used and the type ofequipment used as well. These are often performed a few kilometres at a time, over thespan of many years. Error in bearing or in length can be compounded over the years aseach new survey inherits any uncorrected error from previous surveys. This isdiscussed further in the next section.Age of the data can affect the orientation of a feature when transferred to a map.Magnetic north shifts over time, and if the compass declination is not adjustedaccordingly there is the potential to introduce error in orientation.27Another very significant consequence of data age comes from a change in majorreference systems. The North American Datum (NAD) was adopted as the standardcoordinate referencing system by Canada, the United States, and Mexico in 1913 (Pinch1990). The origin for the NAD was a marked point known as Meades Ranch, locatedin Kansas. In 1927 recomputations were begun to eliminate unacceptable errors in theNAD coordinates. This became known as the North American Datum of 1927(NAD27).As the NAD27 network was extended and densified over the next fifty years, there wasan accumulation of systematic errors (Pinch 1990). This systematic error, combinedwith the inherent limitations of the system led to the development of the NorthAmerican Datum of 1983 (NAD83). The NAD83 uses the centre of mass of the Earthas its origin, which makes it useful for global satellite positioning. Errors in theNAD83 coordinates are much smaller, less systematic, and insignificant compared toNAD27 coordinates. Converting coordinate data from the NAD27 to the NAD83involves transformations which add positional, and therefore, elevational error.3.3.5 Surveying QualityField surveys are prone to error, the level of which depends upon the precision andcondition of the equipment used. Traditional surveying methods for harvest planningutilize a nylon chain, hand-held compass, and a clinometer. The rigorous conditions ofthe harvest surveys can affect the equipment precision. The nylon chain can stretch28after being repeatedly wetted. The compass and clinometer can lose precision whendropped or struck against hard objects. If these instruments are not checked foraccuracy, they can introduce positional error to the survey data. Positional error fromlogging road surveys can accumulate as new surveys are performed based upon theerroneous locations of existing logging roads.Human error in surveying may result in the displacement of the feature being surveyed.If a surveyor has a tendency to read bearings incorrectly in the same direction, then biaswill be introduced. The survey will diverge from the intended bearing, which is thebearing with which the feature will be represented on the map. Elevations will beextracted from the intended location on the DEM, and will appear to be in error whencompared to the field surveyed elevations. This will be more evident in more variableterrain.Elevational data is obtained through some type of survey, usually aerial photographicsurveys. The quality of the equipment used, the atmospheric conditions, and thecondition of the film can all influence the positional accuracy of map features.Presence of haze when the aerial photographs were taken will make it more difficult forthe photo-interpreter to see the ground, compared to aerial photographs taken when theair was clear. This could result in incorrect estimation of the ground elevation whichcould, in turn, result in positionally incorrect contours. While these errors are likely tooccur, they will have much less influence on elevational accuracy than field surveyingerror.293.3.6 Data RelevanceAnother factor which can lead to positional error is the relevance of the data to thepurpose for which they are applied. Roads and creeks were often drafted on maps morefor representational purposes than for accuracy. The roads may be broken to fit sidehillcreek crossings, with little concern for the accuracy of the road in between thecrossings. Using these features as tie-points will likely add positional error, whichcould be significant when very high accuracy is required.Sidehill creeks, often used as tie-points, may not be visible on aerial photographs. Thelocation of the creek is indicated by the presence of a small valley running down thesidehill, evident in the elevations of the forest canopy. The exact location of the creekchannel will be estimated by the photo-interpreter and then represented by a sharp lineon the map. When road plots are transferred to the planning maps the creeks are oftenused to locate the roads. Any positional error in the location of the creek will beinherited by the road.The description accompanying a PIP can often be very ambiguous. For example, a PIPmay be described as being in the Northwest corner of cleared area.” The corner of thiscleared area may have appeared to be a sharp line to the photogrammetrist. On theground the boundary of the cleared area could be transitional over many metres makingthe exact corner very difficult to locate. To make matters worse, the cleared area maysince have been extended from the corner. The PIP will then be located somewhere30along the edge of the cleared area, perhaps impossible to locate. When accuracies ofless than five metres are desired these types of tie-points are not very useful.3.3.7 DEM Accuracy Standards and TestsThompson (1956) recognized early that a map’s accuracy was not independent of thepurpose for which it was to be used. Thompson (1960) also discussed many cases inwhich topographic maps were being used in which an accuracy is presumed that wasnever intended.’ This is still relevant today when most of the existing topographicmaps were created before the advent of small, powerful, and extremely precise GISprograms. Converting these maps to DEMs for use in more demanding operations maynot be feasible. Ultimately, new elevational data should be collected specifically forDEMs at an accuracy level suitable to the new use. Since this will not always bepossible, old topographic maps may be used, provided their limitations are taken intoconsideration. Essential knowledge of these limitations may be gained through testingmaps with methods appropriate to their new intended use.There are several accuracy standards in existence for both topographic maps and DEMs.In the United States, the definitive standard is the National Map Accuracy Standard(NMAS) which is applied to the USGS topographic map series (Veregin 1989). Testsare based upon “a comparison of at least 20 well-defined map points relative to asurvey of higher accuracy.” For vertical accuracy, NMAS states that not more than 1031percent of the elevations tested should be in error of more than one-half the contourinterval (Thompson 1988).In British Columbia, the Surveys and Resource Mapping Branch of the Ministry ofEnvironment, Lands, and Parks uses accuracy standards set out under the NorthAmerican Treaty Organization’s (NATO) specifications for 1:5000 and 1:2500 scaletopographic maps and DEMs (M0ELP 1992). For 1:5000 scale “ninety percent of alldiscrete spot elevations and DEM points shall be accurate to within 1.25 metre”. Whilethis value is statistically derived, the error is assessed using the same pass/fail methodas the NMAS.Both the NMAS and the NATO tests are intended for compliance to an accuracystandard and provide virtually no information as to the magnitude of individual errors(Veregin 1989). There can be a significant difference in accuracy between one mapwith ninety percent of sampled elevations just under one-half contour interval in errorand another map with ninety percent of the points free from error altogether. Similarly,a rejected sample elevation could be in error of just over one-half of the contourinterval and be considered the same as one that is, for example, in error by three timesthe contour interval.The American Society of Civil Engineers has developed the Engineering Map AccuracyStandard (EMAS) for large scale maps (Veregin 1989). The EMAS is designed to be32application specific and checks error data against maximum acceptable limits as set bythe map user. EMAS tests both the sample mean errors and the standard deviation.An alternative statistic that has been used in the United States is called the Root MeanSquare Error (RMSE) (Gustafson and Loon 1981):RMSE = ± (e12/n) RMSE Root Mean Squared Errore = elevational error of the ith test pointn = number of test pointsThe RMSE is preferred to the 90% criteria “since it is completely unambiguous andnot easily subject to misuse.” The RMSE is equivalent to the standard deviation whenmean error is equal to zero. RMSE is often equated with standard deviation for allcases without making proper distinctions.Another option is Koppe’s formula which recognizes the relationship between verticalerror and terrain slope (Gustafson and Loon 1981). A contour that is positionallyincorrect by 5 metres will cause less vertical error on a 5 percent slope than on an 80percent slope. The Koppe formula has incorporated an increasing tolerance for verticalerror as terrain slope increases.Gustafson and Loon (1981) noted that most map accuracy standards assume thatblunders and systematic errors have been eliminated from map error and that the33residual random error is normally distributed. In reality, systematic error will always bepresent to some degree, especially in areas that are difficult to map.Thompson (1960) singled out terrain covered with tall, dense, coniferous forests as areaswhich caused difficulties in mapping. The problem of accuracy testing in heavilyforested areas is dealt with in much the same way: it is avoided. For example, a studyon DEM accuracy testing in Great Britain stated that “not every [test] point wasmeasured because a certain number fell in a woodland area or on some other unsuitablefeatures” (Li 1991).Existing accuracy standards often state that points tested must be clear and well defined.For example, the NATO standards used in B.C. specif’ that stated elevational accuracies“relate to ground not sufficiently obscured by vegetation or other features to causesignificant error” (M0ELP 1992).Finally, it is possible to have a deflection-line derived from a DEM that has anacceptable mean error but that has one or more single large errors which willsignificantly affect the analysis. For example, the photo-interpreter may fail to detect aridge due to the nature of the forest cover. The ridge would not be plotted on thetopographic map and subsequently not represented by the DEM. Deflection-lineanalysis from field surveys will identify the ridge as the yarding boundary. Analysisperformed using the DEM could result in a longer yarding distance estimate due to the34absence of the ridge. This could lead to significant equipment, log and site damage, aswell as lower productivity.For the purposes of deflection-line analysis, traditional statistics may not be sufficient toadequately assess the elevational accuracy of DEM-derived deflection-lines. The meanerror indicates a trend in the elevational error, but it does not represent any individualfeature on the deflection-line which may cause problems. Large individual errors maybe masked by the mean error through the cancelation of positive and negative errors.354 Study SiteField surveys were performed on northern Vancouver Island, British Columbia, incooperation with the Englewood Logging Division of Canadian Forest Products (Canfor)Ltd. The Englewood Logging Division, located in Woss, operates under Tree FarmLicence (TFL) 37 and Forest Licence (FL) A19233 in the Nimpkish Valley region(Figure 8). The dominant landforms in the region are of glacial origin with someintermixed volcanic influence (intrusions). The Nimpkish region is within the CoastalWestern Hemlock biogeoclimatic zone.Seven proposed skyline cutblocks (individual harvest units) were selected for study,three in the TFL and four in the FL (Figure 8). These cutbiocks were chosen primarilydue to their scheduling for harvest and secondarily for representation of the forest andterrain types within which skyline systems operate. A total of thirty-one deflection-lineswere located at potential landings within the seven cutbiocks (Figures 9-15). Generalcharacteristics of each cutbiock are shown in Table 1. Table 2 shows the distribution ofindividual deflection-lines and settings within each cutblock. All cutblocks were locatedin old growth forest types.36Figure 8 Map of cutbiock locations within Canfor operating area.Leg end/•1-4IFrjI‘%4-I/•1SI¶ —10 kmS/Vn couvenIs londCutbiock ® RGLokesRiuersTFL Boundary — — —FL BoundôryRoecjs37Figure 9 Map of AR1 cutbiock and deflection-line locations.-, —7 / (/ I - ( — ----——-. —S,/_/ 131 !/ Th/ FI(// II —i i1II(/1/t:1 — I 2L4\ \\ -7 )) I I / — ‘—-— J 1 1 ——-Th I I\ I (\\ \___\‘b ‘_i 1’2 1 235\ \\ N\ \___LegendDeflection-line30.5 m contourIs II 7.7 m contourRoodCreek38Figure 10 Map of AR6 cutbiock and deflection-line locations.Legendoriect1on—I meC COn tOM7.7 m COflteu*Rad = = =Creek — -I100 metresFigure 11 Map of AT cutbiock and deflection-line locations.Legend39100 metresI DePIection- me30.5 m contotr7.7 rn contourReedCreek40Figure 12 Map of CU cutblock and deflection-line locations.100 metres LegendDerection-I me30.5 m contour7.7 rn cont.ourRoodCreekGround survey [ZZ E..,41Figure 13 Map of GC cutbiock and deflection-line locations.100 METRESIr 1LegendOePectiori— me30.5 m cortour7.7 m contourRoodCreek —Ground survey r42Figure 14 Map of NW cutbiock and deflection-line locations.IrLegend100 metresDerIection-ire30.5 m contour7.7 m contourRoeciCreek —River43Figure 15 Map of TR cutbiock and deflection-line locations.100 metresILegend4Oerlection-I me30.5 m contour7.7 m contourRoedCreek —44Table 1 Summary of cutbiock characteristics.Cutbiock Slope 1) Elevations Brokenness 2) Forest Type 3)(%) (m)AR1 30 90 - 200 M CWHb1, CWHb2(0-146) H - creek canyon Hw, Cw,BaAR6 50 180-500 M CWHb1,2(2-120) H - upper slopes Hw, Ba, (Cy ,Ss)AT 46 400 - 800 S - lower flat area CWHb1,2(0-104) L - slope and ridge Hw, Ba, Cw, CyCU 47 600 - 850 M CWHb2, MHa(11-72) Cw, Hw, Ba, (Cw)GC 46 200 - 550 M CWHa1(0-13 1) Df, Hw, (Cw)NW 44 150 - 550 M - upper slope CWHb1(0-96) L - valley bottom Hw, Ba, (Cw)TR 55 350 - 700 H CWHb2(2-150) Hw, Ba, Cw, Cy1) average slopes found from deflection-line surveys (range in brackets)2) S - SmoothL - LowM - ModerateH-High- terrain is basically flat, no impediment to traverse.- gently undulating, rock outcrops, no impediment to traverse.- significant rock bluffs, creek canyons, difficult to traverse.- cliffs, sheer sided creek canyons, difficult to impossible totraverse.3) Biogeoclimatic Units (Nuszdorfer et al. 1985)MHa - Mountain Hemlock, Maritime ForestedCWHa - Coastal Western Hemlock, Maritime, dryCWHb1 - Coastal Western Hemlock, Windward Submontane Maritime, wetCWHb2 - Coastal Western Hemlock, Windward Montane Maritime, wetDominant Tree Species (Watts 1983)Ba - Abies amabilis (Dougl.) Forbes (amabilis fir, Pacific silver fir, balsam)Cw - Thuja plicata Donn (western red cedar)Cy - Chamaecyparis noolkatensis (D. Don) Spach (yellow cedar, cypress)Fd - Pseudotsuga menziesii (Mirb.) Franco var. menziesii (Douglas-fir)Hw - Tsuga heterophylla (Raf.) Sarg. (western hemlock - may be mountain hemlockin upper elevations - not specified)Ss - Picea sitchensis ( Bong.) Carr. (sitka spruce - adjacent to main creek, AR69 1)45Distribution of deflection-lines and settings within cutblocks.Cutblock 1) Setting Deflection -linesAR1 AR1-Si AR1-ilAR1-S2 AR1-12AR6 AR6-S1 AR6-1AR6-S2 AR6-2, AR6-3AR6-S3 AR6-4AR6-S4 AR6-5AR6-S5 AR6-la, AR6-2aAT AT-Si AT-200, AT-201AT-S2 AT-202, AT-203CU CU-Si CU-20, CU-21CU-S2 CU-22, CU-23GC GC-L1 GC-i, GC-2, GC-3GC-S2 GC-4, GC-5GC-S3 GC-7, GC-8GC-S4 GC-iONW NW-Si NW-i, NW-2TR TR-S1 TR-i, TR-4TR-S2 TR-2, TR-3Cutblock names have been shortened for clarity and for consistency with settingand deflection-line names. The full names are listed below:AR1 AR16OAR6 AR691AT AT295BCU CU7GC GC6NW NW74TR TR38Table 2465 Methods5.1 Cutblock and Deflection-line SelectionThirty-one deflection-lines were located in the seven proposed skyline cutbiocks by theEnglewood engineering staff. The first step involved visual inspection of thetopographic maps to identify potential landings for the skyline machine. Landings werelocated on established, traversed or proposed road locations depending on the area andthe state of its development. Deflection-lines started from these landings and weresurveyed on constant bearings. The bearings were chosen in areas which, through thevisual inspection, were potential problem areas for clearance. All landings located onsurveyed roadway were surveyed to the nearest established roadway.5.2 Field SurveysField surveys were conducted using Canfor’s standard method of tight-chaining with a50-metre nylon chain, hand-held compass and clinometer. Survey stations were locatedat changes in terrain slope of ten percent or more, and at significant features such ascreeks and traversed and established roads. Foresights and backsights were taken withthe compass to identify and eliminate bearing errors due to human error or magneticanomalies. The nylon chain had tags set at increments 1.0 metres, with tags at 0.1metre for the first metre. Length measurements were taken in such a way as to ensurethat they were accurate to within the 0.1 metre markings. The compass had a minimum47increment of 2 degrees. The clinometer had increments of 1 percent for slopes of 0 to70 percent and 2 for slopes of 70 to 150 percent.Tie-point surveys were conducted to locate the deflection-lines. Initial tie-points wereestablished or surveyed roads and junctions, creeks and junctions, as well as PIPs whichwere located from the original air photo mapping. Potential tie-points were firstselected on maps and then attempts were made to locate them in the field. It becameevident that, for the most part, the map representation of these tie-points contained ahigh degree of positional error and most of them were found to be unreliable.Furthermore, some tie-points were extremely hard to find in the steep and heavilyforested terrain of the study cutblocks.In an attempt to improve tie-point accuracy, a Trimble 4000SE Land Surveyor (TrimbleNavigation Ltd. 1992) Global Positioning System (GPS) in conjunction with a Criterion(Laser Technologies Inc. 1992) hand held laser surveyor were tested in the GCcutbiock. The GPS was used to locate four well distributed tie-points and then the laserwas used to traverse the deflection-lines and tie them to the tie-points.All of the cutblock maps were in NAD27 coordinates and any recent additions to thedata, from sources such as GPS, were in NAD83. To bypass the problems associatedwith transferring between these two datums, GPS survey data were integrated by fittingthem to the map features, independent of the reference system. This avoided thenecessary complex transformations.48Deflection-line and tie-point survey notes were compiled using ROADENG (SoftreeTechnical Systems 1992) forest engineering software. The deflection-line and tie-pointlocations were plotted from ROADENG using a HP DraftMaster 1-drum plotter(Hewlett-Packard Company 1987). The plots were used to transfer deflection-linelocations to the maps.5.3 Digital Elevation ModelsCanfor provided mylar map copies for each of the seven cutblocks. These maps wereof 1:5000 scale with a 7.62-metre (25-foot) contour interval. The original mylar basemaps were compiled in 1973 at a scale of 1:4800 (1 inch = 400 feet) and later photo-reduced to 1:5000 scale. The base maps were primarily photogrammetrically-measured.Although small portions of the GC and CU cutblock maps were field surveyed, theanalysis focused on the photogrammetric portion of the maps.Contours and relevant features were hand digitized into TerraSoft version 10.03 (DigitalResource Systems Ltd. 1992) map files using an Altek Model AC3O (GentianElectronics Ltd. 1987) digitizing tablet. Contours were stream digitized, with pointsrecorded every 0.3 millimetres (1.5 metres ground scale) along the contours. Sincesmoothing error had already existed in the maps, additional effort was made to keep thecross hairs of the digitizer puck in the centre of the contour at all times. While this wasoften difficult, especially in the most variable terrain, the addition of more smoothingerror was likely kept to a minimum. Contour points were thinned using a weeding49corridor of 1.0 metres to reduce data redundancy (Digital Resource Systems Ltd. 1992).Deflection-line locations were digitized from the mylar maps into the correspondingmap files.Triangulated Irregular Networks (T]Ns) were created for each map. The TIN modelwas chosen over the grid model due to its superior representation of terrain features.Specialized software extracted deflection-line data by draping the location of thedeflection-lines onto the DEM. The software allowed the horizontal spacing of theelevation points to be set at one metre to conform with the precision of the fieldsurveys. An ASCII file was produced in the form of cumulative horizontal and verticalvalues referenced to the origin of the deflection-line.5.4 Estimating Yarding DistanceDeflection-line analysis was performed on the deflection-line pairs to estimate theyarding distance. These distances were compared to see if the DEM-derived deflectionline caused any erroneous estimations. The Terrain module of ROADENG (SoftreeTechnical Systems 1992) was used for the deflection-line analysis. Cableconfigurations, machine specifications, type of analysis, calculation parameters, andother options were selected to conform to Canfor’s procedures.Analyses for deflection-line pairs were performed keeping most, but not all, optionsconstant. Within given ranges, different payload weights and equivalent mid-span50deflections were allowed for analysis of each deflection-line pair. The DEM-deriveddeflection-line was analyzed independent of the analysis of the field surveyeddeflection-line and vice versa. This was done to mimic an operational deflection-lineanalysis in which a forest engineer would be using one type of deflection-line, or theother, but not both.For example, a midsiope bench, evident on a field surveyed deflection-line, may proveto be an impassable boundary for yarding when performing deflection-line analysis.Since reducing payload or mid-span deflection cannot overcome this obstacle, theseoptions would not be pursued. If the bench were less evident on the correspondingDEM-derived deflection-line, reducing payload and mid-span deflection may provideclearance over the bench, causing an erroneous yarding distance estimate. In anoperational use of DEM-derived deflection-lines, the forest engineer would not have thefield surveyed deflection-line with which to compare, and would therefore make theerroneous estimation.5.4.1 Statistical AnalysisDeflection-line analyses were performed using the DEM-derived and field surveyeddeflection-line pairs to estimate yarding distances. These distances were then comparedto see if the DEM-derived deflection-lines give significantly different results. Thehypothesis for this experiment was:51H1o: There is no significant difference between yarding distance estimatedusing paired DEM-derived and field surveyed deflection-lines.H1a: There is a significant difference between yarding distance estimated usingpaired DEM-derived and field surveyed deflection-lines.Failure to reject the null hypothesis indicates that the elevational error of DEM-deriveddeflection-lines does not affect yarding distance estimates. Rejection of the nullhypothesis shows that the elevational error of DEM-derived deflection-lines does affectyarding distance estimates.The errors in yarding distance estimates were tested using Systat 5.2 (Systat 1992). Theerrors were checked for normalcy using the Lilliefors Test, a modification of theKolmogorov-Smirnov test used for non-standardized data (Systat 1992). A significancelevel of oc=O. 1 was used. This large alpha value (0.1) was used to reduce the chance ofa Type 2 error, accepting the null hypothesis when it is false (Walpole 1982). Inpractical terms, the test was designed to reduce the chance of indicating that the yardingdistance estimates from the DEM-derived deflection-lines were the same as those fromthe field surveyed deflection-lines, when they were not. Since the error displayed anon-normal distribution, a sign-test was used to test the mean.525.5 DEM Elevational ErrorElevations were sampled using deflection-lines, which are ground profiles. Gossard(1976) lists profiles as an objective means for analyzing elevational errors, while Kellieand Bryan (1981) found no statistical differences between elevational error estimatesbased upon profile data and on randomly located point data. Kellie and Bryan locatedthe profiles randomly, unlike the methodology in this study, in an attempt to estimatethe general elevational accuracy of the DEM. The systematic selection of deflection-line locations was considered to be within the bounds of the study, since the studyfocussed on accuracy issues related to DEM-derived deflection-lines.The ASCII files were loaded into a spreadsheet for manipulation and comparison withfield surveyed deflection-line data. Macros were created to eliminate extraneous dataand to interpolate and extract points that aligned with the survey stations. This allowedthe field surveyed and DEM-derived deflection-line elevations to be compared atcommon horizontal points. These deflection-line pairs were plotted together as a visualcheck for blunders in the elevations of the DEM-derived deflection-line. Whendeflection-lines indicating blunders were detected, the DEMs and topographic mapswere checked to determine the source of the blunder. If the blunder could not becorrected, analyses were performed with and without the affected deflection-lines andthe results compared.53The field surveyed and DEM-derived deflection-line elevations were given the sameelevation, or calibrated, at one station. This station was chosen as the one where thephoto-interpreter could obtain the best measurement of the ground elevation. Thiswould be the station on the DEM-derived deflection-line with the most accurateelevation. This was usually a creek crossing, if it was sufficiently visible on the aerialphotographs. If there was no major creek, or if the creek was hard to see on the aerialphotographs, a ridge-top or an open rock-bluff was used.Deflection-line pairs were then compared at individual stations to obtain elevationaldifferences. The field surveyed elevations were the controls, and any differencesbetween them and the DEM elevations were considered to be errors. The error waspositive when the DEM elevation was higher than the field surveyed elevation.Conversely, the error was negative when the DEM elevation was lower than the fieldsurveyed elevation.It had been assumed that the elevations of the field surveyed deflection-lines weremeasured without error. Although this was not possible, most accuracy tests specifythat DEM elevations should be tested against a survey of higher accuracy. Since thefield surveyed elevations should be, on average, more accurate than those of the DEMderived deflection-lines, this criteria of accuracy testing had been met. However, whenconsidering the results, the possible confounding effects of field surveying error shouldbe considered.545.5.1 Statistical AnalysisElevational errors were grouped by individual deflection-line, setting, cutbiock, and byentire study area. Characteristics common to these groupings might have influenced thenature and/or magnitude of the elevational errors. For example, some of the cutbiockshad distinctly different forest types when compared to other cutbiocks. These differentforest types may have had differing effects upon the photo-interpreters ability toaccurately measure ground elevations. These differences may have led to varying levelsof error, or different patterns of systematic error, depending upon the forest type.Error was grouped by settings to test the quality of the tie-points used to locate thelandings within each setting. The landing within each setting served as a common tie-point for all the deflection-lines contained within that setting. Some settings containedonly one deflection-line, a result of the limited resources available for manuallysurveying skyline deflection-lines. Error grouped by deflection-lines was used to assessthe elevational accuracy of the deflection-lines themselves and the comparison plotswere used to detect and illustrate the different error patterns which may exist in DEMsand DEM-derived deflection-lines.The hypothesis for this section of the study was:H2o: There is no significant difference between the elevation of paired pointson the DEM-derived and field surveyed deflection-lines.55H2a: There is a significant difference between the elevation of paired points onthe DEM-derived and field surveyed deflection-lines.The statistical testing determined if the mean error was significantly different from zero.This was intended to show if the elevations of the DEM-derived deflection-linesadequately represented the elevations of the field surveyed deflection-lines. Whileforest planners may be willing to accept mean errors which are not equal to zero, it wasnot possible to pick one mean error level acceptable for every deflection-line.Therefore, error testing was restricted to detecting mean errors which were significantlydifferent from zero, allowing forest planners to judge how much error may acceptable.Existing accuracy standards had various inadequacies which made them unacceptablefor testing the elevational accuracy of DEM-derived deflection-lines. Statistically basedstandards assume that the error is normally distributed; an assumption not made in thisstudy. Other standards stated that ground points tested for elevational error must beclear and well defined, a situation not often found in the heavily forested areas plannedfor harvesting. Finally, some standards required only ninety percent of the errors tomeet an acceptable level. The largest errors, which were the most critical for estimatingyarding distance, were therefore ignored. This was not acceptable for assessing theelevational accuracy of DEM-derived deflection-lines.Data from the spreadsheet comparisons was imported into Systat 5.2 (Systat 1992) inorder to perform the statistical analyses. Histograms were created for visual inspection56of the distribution of the errors. The data were then checked for normality using theLilliefors Test, a modification of the Kolmogorov-Smirnov test used for nonstandardized data (Systat 1992). A significance level of x=0. 1 was used for all tests(results at cc=0.05 are included). The large alpha value (0.1) was used to reduce thechance of a Type 2 error, accepting the null hypothesis when it is false (Walpole 1982).In practical terms, the test was designed to reduce the chance of indicating that theDEM-derived elevations were the same as the field-surveyed elevations when they werenot.For samples with normal distributions, a paired t-test was used. For samples with non-normal distributions, a sign-test was used. The sign test is less efficient than the t-testin that it does not utilize as much information (Walpole 1982). In this study, if thenumber of negative errors was approximately equal to the number of positive errors, thesign test indicated that the mean error was not significantly different from zero. Resultsof this test could have been misleading since it did not consider the size of eachindividual error. A deflection-line could have had an equal number of positive andnegative elevational errors, but the positive errors were greater in absolute magnitudethan the negative errors. This would have been indicative of some type of systematicerror which could have had a very significant effect upon the yarding distanceestimates. Also, the severity and type of error in the yarding distance estimates mayhave varied, depending upon whether individual elevational errors were positive ornegative. Neither the sign test nor the paired t-test provided this information. To avoidthese scenarios, results of hypothesis tests were evaluated in conjunction with57histograms, comparison plots, and basic descriptive statistics (mean error, mean absoluteerror, maximum negative and positive errors, the range of the error, and the standarddeviation). As well, on the ground experience gained from field surveying eachindividual deflection-line added valuable insight to the evaluation.Mean and mean absolute errors were used as a guide to the relative levels of systematicand random error in the elevations of the DEM-derived deflection-lines. If both themean error was low (less than two metres) and mean absolute error was low (less thanfour metres) then there was a low level of both types of error. If the mean error waslow and the mean absolute error was high then the error was mostly random. If themean error was high but very similar to the mean absolute error (within two metres)then the error was mostly systematic. If the mean error was high and the mean absoluteerror more than two metres higher then there was both random and systematic errorpresent. The size of the mean and mean absolute errors used to detect the presence ofthe different error patterns were determined by comparing deflection-lines whichdisplayed obvious error patterns with deflection-lines which displayed no obvious errorpatterns.586 ResultsVisual inspection of both deflection-line comparison plots for the AR1 cutbiock, AR1-11 and AR1-12, showed an apparent blunder. Both deflection-line pairs divergedsteadily indicating a possible error in the contour interval of the source map. Since thecause of this blunder could not be determined conclusively (and corrected), analyseswere conducted with and without the cutbiock for comparison.Portions of both the GC and the CU cutbiock maps had been created from field surveyswhile the rest were photogrammetrically derived. The GC deflection-lines were alllocated within the photogrammetric portions of the maps so they were not affected. ForCU approximately one half of deflection-lines CU-22 and CU-23 were located withinthe field surveyed map. It was possible that significant positional error could haveoccurred at the boundary between the two different mapping techniques. This errorcould not have been accounted for and therefore would have had an unknown influenceupon the elevational error. Due to this uncertainty, analyses were also conducted withand without these two deflection-lines.6.1 Estimating Yarding DistanceError in yarding distance estimates was deemed positive when the estimate from theDEM-clerived deflection-line was longer (an overestimation) and negative when theestimate was shorter (an underestimation). Twenty-two of thirty-one (7 1%) deflection-59lines produced the same yarding distance estimates. For all thirty-one deflection-linesthe mean error in yarding distance estimates was -4.1 m and the range of error was -54.1 m to 127.7 m. Two errors were positive and seven were negative. The mean errorwas not significantly different from zero using a sign test (cc = 0.1). The inherentinefficiency of the sign test should be taken into account when considering these results.Of the four deflection-lines which were eliminated, only AR1-12 displayed an error inyarding distance estimate and this error was the largest from all thirty-one deflection-lines, at 127.7 m. When CU-22, CU-23, AR1-1 1, and AR1-12 were eliminated, themean error in yarding distance estimates was significantly different from zero at cc = 0.1but not significantly different from zero at cc = 0.05 (p0.07). Once again the sign testwas used, and the results analyzed accordingly. The mean error was -9.4 m, and rangedfrom -54.1 m to 7.8 m. Nineteen of twenty-seven deflection-line pairs (70%) producedthe same yarding distance estimates. More detailed information is presented inAppendix B.Longer deflection-lines had more tendency towards erroneous yarding distance estimatesthan did shorter deflection-lines. The four longest deflection-lines had erroneousyarding distance estimates, and six of the nine deflection-lines with different estimationswere from the eight longest deflection-lines. Considering the error in yarding distanceestimates as a percentage of the deflection-line length, the average error percentage was2.4% for all deflection-lines and 8.2 % for the nine incorrect yarding distance estimates.606.2 DEM Elevational ErrorDescriptive and inferential statistics were calculated for the elevational differences atindividual stations for all pairs of the thirty-one deflection-lines. The mean error for alldata (n=675) was 1.4 m, which was significantly different from zero for cx=0. 1. Themean absolute error was 5,8 m. The error ranged from -22.9 m to 26.8 m and thestandard deviation was 7.5 m. Elimination of AR1-1 1 and AR1-12 made virtually nodifference to the mean error, range, or standard deviation. The same was true whenCU-22 and CU-23 were eliminated. Additional results for AR1 and CU, by cutblock,setting, and deflection-line are tabulated in Appendix C.All of the following results are for the twenty-seven deflection-lines remaining after theelimination of CU-22, CU-23, AR1-1 1, and AR1-12. The mean error for all data(n=594) was 1.6 m, which was significantly different from zero for x0. 1. The meanabsolute error was 5.3 m, the error range was 49.2 m (-22.4 m to 26.8 m), and thestandard deviation was 6.9 m. Data were also analyzed by cutblock, setting, and byindividual deflection-line to identifS’ patterns and attempt to isolate the error sources.Detailed results, including the statistical tests used for individual analysis, are tabulatedin Appendix C.The mean error by cutblock ranged from -1.0 to 5.8 m. Three cutblocks had positivemean errors and three had negative mean errors. The mean absolute error ranged from614.4 to 7.0 m. The smallest range of error was 29.2 m and the largest was 42.7 m. Thestandard deviations ranged from 5.7 m to 8.8 m.The mean error was significantly different from zero for AT, GC, NW, and TR (ccO.1).AT had the highest overall error with the largest mean error, mean absolute error,positive error, range, and standard deviation. CU had the lowest overall error with thesmallest mean error, mean absolute error, range and standard deviation.The mean error by setting ranged from -2.2 to 8.4 m and thirteen of the eighteensettings had positive mean errors (72%). The mean absolute error ranged from 3.5 to12.0 m. The smallest range of error was 15.4 m and the largest was 42.7 m. Thestandard deviation ranged between 4.5 m and 14.2 m.The mean error was significantly different from zero for five of eleven settings. One ofthese was GC-S 1, which had the lowest overall error, and which had the smallest meanabsolute error, range, and standard deviation. Another setting with a mean errorsignificantly different from zero was AT-S2 which had the highest mean error, meanabsolute error, maximum positive error, range, and standard deviation.For the error analyzed by deflection-line, the mean ranged from -4.1 to 7.4 m. Elevenof twenty seven (41%) deflection-lines had mean errors that were significantly differentfrom zero (x0. 1). Nineteen deflection-lines (70 %) had positive mean errors. Themean absolute error ranged from 2.7 to 9.5 m. The smallest range was 9.2 and thelargest was 42.3. The smallest standard deviation was 2.8 and the largest was 10.4.GC-2 had the overall lowest error with the smallest mean absolute error, range, andstandard deviation. TR-4 had the overall highest level of error with the largest meanabsolute error, range, and standard deviation.62637 Discussion7.1 Estimating Yarding DistanceStatistical analysis had shown that the error in yarding distances, estimated from DEMderived deflection-lines, was not significantly different from zero (cx= 0.1) when thethirty-one deflection-lines were tested and was significantly different from zero (x= 0.1)when CU-22, CU-23, AR1-1 1, and AR1-12 were eliminated. Conclusions were notmade based upon these results since a sign test was used for the analysis. It was morevaluable to investigate the deflection-lines which produced erroneous yarding distanceestimates and to discuss the cause of those errors.7.1.1 Deflection-line Length and ConcavityOf the thirty-one deflection-lines analyzed, nine (29%) displayed error in yardingdistance estimates, and most of these nine were of the longest deflection-lines. Thisshows that error in yarding distance estimates was more likely to occur as deflectionline length increases. There were several possible reasons for this, one being that thelonger the deflection-line the more chance that elevational error sufficient to influenceyarding distance estimates would occur. A more likely cause was an interactionbetween deflection-line length, the elevational error, and the concavity of the deflectionline.64For the skyline system considered in this study, adequate skyline deflection wasobtainable only on primarily concave terrain surfaces. Problems with concavitygenerally occur at the beginning and/or end of the deflection-line where it rolls overonto a ridge or similar feature. These were generally poor areas for clearance due tothe proximity to either the landing or the boundary. Relatively small errors in elevationon the DEM-derived deflection-line can cause problems with obtaining adequatedeflection and subsequently cause problems with yarding distance estimates.Evidence of this can be seen in the deflection analysis plots for the five longestdeflection-lines with incorrect yarding distance estimates; NW-i, AT-203, NW-2, AT-202, and AT-200. Adequate clearance was not obtainable for the entire length of thesefive deflection-lines. For example, deflection-line NW-2 had problems with deflectionin the convex terrain around 300 metres (Figure 16). When this occurred and adequatedeflection could not be obtained for the entire deflection-line, a small elevational erroron the DEM-derived deflection-line caused an incorrect yarding distance estimate.By comparison, CU-21 provided adequate clearance for the entire length of both thefield surveyed and DEM-derived deflection-lines. It was therefore easier to obtain thesame yarding distance estimates for the deflection-line pair. This had to do with theposition of the deflection-line relative to the terrain (Figure 17). CU-2 1 wassymmetrically located with respect to the valley so that it could best take advantage ofthe terrain concavity. NW-2 on the other hand started at the bottom of a valley,continuing up the side of a ridge and into an area of convex terrain.—200—250— 3DDFigure 1665Figure 17 The concave shape of both deflection-lines for CU-21 providedadequate clearance for the entire length of the deflection-lines.—1500LaU’o a 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 00 0 in 0 WI 0 IL, 0 WI 0 WI 0 It) 0 WI 0 It) 0 WIIn — — CN CN fl fl q WI WI <0 10 N N CO 0) 0) oi1111111 11111111111111111111111111111111111111111111 11111111111111111111111111111A positive elevational error in concave terrain resulted in the incorrectplacement of the boundary for the NW-2 DEM-derived deflection-line.r IIJLJIJ950900t600—550500- In .- C.I CN Ffl rfl ‘ ‘ In Wi <0 (0 N N 0) 0) (7,r i....i....i....i....i....i....i....i ....i....i66Of the four deflection-lines which were eliminated, only AR1-12 displayed an erroneousyarding distance estimate, an overestimate of 127.7 metres. This was the longest errorfor all the deflection-lines and AR1-12 was the shortest deflection-line with an incorrectyarding distance estimate. AR1 -12 was primarily convex, and the elevational errorswere of a size which may have not influenced yarding distance estimated on concaveterrain. The other deflection-line eliminated from cutbiock AR1, AR1-1 1, hadelevational errors of similar magnitude, but it was on concave terrain and did notexperience error in yarding distance estimates. The two deflection-lines eliminated formCU, CU-22 and CU-23, were much longer than AR1-12 but did not experience error inyarding distance estimates. These two deflection-lines were similar to CU-21 (Figure17) in that they were symmetrically located in concave terrain. These results emphasizethe important relationship between elevational errors, and their location with regard tothe terrain, and their effect upon yarding distance estimates.7.1.2 Large Elevational ErrorsWhile the mean error gives a good indication of the general trend in the elevationalerror of a DEM-derived deflection-line, large individual errors are much more relevantwhen analyzing potential problems with yarding distance estimates. A large error mayindicate a significant local feature such as a ridge that was not represented properly onthe DEM. This appears to have been the problem for the three remaining deflectionlines which had incorrect yarding distance estimates. In Figure 18 it can be seen that a67missing rock bluff on the DEM-derived deflection-line had affected the yarding distanceestimate, resulting in a difference of -35.4 metres.A small mean error and standard deviation could hide the presence of a few largeindividual errors. Only one large error is needed to cause an erroneous yarding distanceestimate. If a large negative error occurs at the end of a deflection-line, anunderestimation may result such as occurred for GC-5. Conversely, a large positiveerror at the end of the deflection-line could lead to an overestimation.A large negative error towards the centre of the deflection-line could cause anoverestimation of yarding distance (Figure 19). Conversely, a large positive errorFigure 18 A large elevational error at the end of the GC-5 DEM-deriveddeflection-line resulted in an incorrect boundary location.—3O0250200—l0C<‘I5001,C‘ DCwC C C C C C C C C C C C C C CC C Lfl c ‘r c tt C Lfl C C Lfl C W CLfl — - CN C Ui In