Download pdf - Modeling the magnitude and spatial distribution of aesthetic impacts

IntroductionScenic beauty has been recognized as an important factor in forest management forat least the past four decades. For example, in the United States, the NationalEnvironmental Policy Act of 1969, the Multiple-Use Sustained-Yield Act of 1960and the National Forest Management Act of 1976 all require the US Forest Serviceto consider aesthetics along with wildlife, recreation, wilderness resources, and otherfactors as important components of national forest management. Legislative actionssuch as these, along with increased public concern for forest aesthetics, have encour-aged forest managers to consider the aesthetic impacts of their activities andprompted researchers to explore the topic of public perception of the aestheticimpacts of forest-management activities (Bishop and Karadaglis, 1997; Erdle, 1999).

Timber harvesting has greater impacts on forest aesthetics than virtually any otherforest-management activity. Clearcutting, where virtually all the trees in a relativelylarge contiguous area are removed, has especially dramatic aesthetic impacts. At leastin part because of these profound aesthetic impacts, clearcutting has become intenselycontroversial, and the economic, ecologic, sociologic, as well as aesthetic impacts ofthis practice have become the subject of heated debate (Kline and Armstrong, 2001;Limerick, 2002).

While recognizing the importance of all of the consequences of clearcutting, in thispaper we focus exclusively on the aesthetic impacts of the practice. Forest managershave long recognized the importance of addressing these aesthetic impacts, and theirmost frequent response has been to place clearcuts in areas with restricted viewsheds(Gustafson and Crow, 1998). Conventional GIS-based viewshed analyses are frequentlyused to find such locations. This strategy is obviously based on the intuitively sensibleidea that by limiting the area in which a clearcut is visible, the overall aesthetic impact

Modeling the magnitude and spatial distribution of aestheticimpacts

Denis J Dean, Alicia C Lizarraga-BlackardDepartment of Forest Sciences, 113 Forestry Building, Colorado State University, Fort Collins,CO 80523, USA; e-mail: [email protected] 27 November 2003; in revised form 21 December 2005

Environment and Planning B: Planning and Design 2007, volume 34, pages 121 ^ 138

Abstract. Timber-harvesting operations, especially clearcutting (that is, harvesting operations whereall of the trees in a given area are removed), have been criticized for many reasons, not least of whichis their unsightly appearance. Forest managers have recognized this, and have attempted to placeclearcuts in locations with limited viewsheds. In order to find such locations, forest managers havemade extensive use of standard geographical information system (GIS) viewshed operations. The useof conventional viewshed operations ignores the possibility that the aesthetic impacts of clearcutsmight be diminished by the screening effects of intervening vegetation, or the possibility that impactssimply decrease with increasing distance. In this study we found evidence that the aesthetic impacts ofclearcuts do in fact diminish in these ways. Photograph transects were performed around tenclearcuts. Each transect produced a series of pictures showing the clearcut from increasing distancesinto the surrounding forest. The Law of Comparative Judgments (LCJ) technique was used to developperceived-scenic-beauty rankings for each photograph. Statistical analyses showed aesthetic impactsdo in fact diminish with viewing distance through screening vegetation. A modified viewshed algo-rithm was then developed not just to identify areas where clearcuts are visible, but also to maplocalized aesthetic impacts of clearcuts. The approach presented here could be used to developsimilar models that map the aesthetic impacts of any proposed environmental modification.

DOI:10.1068/b30101

https://www.researchgate.net/publication/225602629_Simulating_Spatial_and_Temporal_Context_of_Forest_Management_Using_Hypothetical_Landscapes?el=1_x_8&enrichId=rgreq-bd8d71c6-3e9a-4eb5-b140-cb982aa54939&enrichSource=Y292ZXJQYWdlOzIzNTQxNTc3O0FTOjEwMzQwNTkzODgwNjc4NEAxNDAxNjY1MzA1MTI3

of the clearcut will also be limited. In effect, this approach uses the extent of theviewshed of a clearcut as a proxy measure of the overall aesthetic impact ofthe clearcut. Note that, although on the surface this simple approach seems intuitivelyappealing, it is based on a very nonintuitive implicit assumption. Specifically, itassumes that all areas within the viewshed of a clearcut are impacted equally by thecut.

It seems likely that the aesthetic impacts of clearcuts diminish with increasedviewing distance and as a consequence of the screening effects of vegetation betweenthe viewer and the clearcut. Research by Buhyoff and Wellman (1980) supports thehypothesis that the aesthetic impacts of clearcuts do in fact diminish as a result ofthese factors. If this hypothesis is true, merely identifying viewsheds is not an adequateway of measuring the aesthetic impact of clearcuts, because a clearcut with a large butlightly impacted viewshed might actually have less overall aesthetic impact than analternative clearcut with a smaller but highly impacted viewshed. A more completemeasure of aesthetic impact would differentiate between areas within the viewshed thatare heavily impacted and other areas that are less heavily impacted.

The purpose of this study is to investigate how screening vegetation affects themagnitude and spatial distribution of the aesthetic impacts caused by clearcuts, andto develop a GIS-based model to estimate these impacts across the landscape. It washypothesized that the perceived aesthetic impacts of a clearcut are at their maximumalong the edge of the cut, and that these impacts decrease as a function of distancefrom the cut. It was further hypothesized that the rapidity of this decrease is influencedby the amount and type of screening vegetation present. At some point, the impacts goto zero as the cut either becomes invisible or becomes an insignificant part of thevisible landscape. This basic hypothesis seems equally plausible in other situationswhere an aesthetically unappealing locationöindustrial or other development, othernatural-resource extraction operations (such as mining), and so forthöis sited withinan aesthetically pleasing environment that acts as a screen around the unappealinglocation.

An extensive body of literature exists describing how the relative perceived scenicbeauty of a series of landscapes can be measured (Arthur, 1977; Brown and Daniel,1986; Daniel and Boster, 1976; Hull et al, 1984; Ribe, 1990). These measurementtechniques typically involve individuals rating a series of photographs representingthe landscapes being compared. The end result of these comparisons are a series ofinterval-scale landscape preference ratings. In this paper, these techniques were used tomeasure the perceived scenic beauty of a series of landscape photographs taken atvarious distances, and through varying degrees of intervening vegetation, from clear-cuts. Regression analyses were performed on these rankings to determine if theexpected relationship existed between perceived scenic beauty and viewing distanceor extent of screening vegetation.

The results of these regression analyses were incorporated into a predictive GISmodel. This model extended existing GIS viewshed techniques to include calculatingviewing distance through screening vegetation before reaching a clearcut. These meas-ures could then be incorporated into the regression equations to estimate the relativeaesthetic impact of the clearcut at any point in the surrounding landscape.

Previous researchOver the past forty years, aesthetic preferences toward forested landscapes have beenexamined in many studies. In most of these studies there has been a differentiationbetween `near-view' and `vista' landscapes. Near-view scenes consist of forest land-scapes within 100 ^ 200 yards of an observer and depict the viewer's experiences

122 D J Dean, A C Lizarraga-Blackard

`within' the forest, whereas vista scenes involve distant landscape features such asmountains or overlooks. The nature of aesthetic reactions to these two types of land-scapes have been found to be quite different, and hence models developed for one typeof landscape should not be generalized to the other (Brown and Daniel, 1986). Underthis dichotomy, we focused exclusively on near-view landscapes and hence our resultsshould not be generalized to vista-type landscapes.

Several studies have related forest-inventory and forest-management attributes toperceived scenic beauty. Arthur (1977) found relationships between scenic beauty andtree species, tree size, amount of downed wood, tree density (number of trees per acre),and basal area (total square footage of the cross sections of all of the trunks of treesgrowing on a particular acre of ground), as well as a number of factors relating to thedesign of the photographs themselves (for example, lighting, position, detail). Usingnear-view models, Rudis et al (1988) found a positive association between scenic beautyand visual penetration through sawtimber-sized trees and a negative relationshipbetween scenic beauty and vision-obscuring foliage, twigs, and sapling-size trees.Ribe (1990) evaluated several physical factors such as downed wood, tree density,size, mortality, and species and found that tree mortality had the greatest impact onscenic beauty preferences. This finding was in agreement with earlier work by Buhyoffand Leuschner (1978) and Buhyoff et al (1982), who found strong relationships betweenthe amount of visible insect-caused tree mortality and scenic quality of both vista andnear-view landscapes. Hollenhorst et al (1993) confirmed this result by finding thatthe percentage of trees killed by gypsy moths was a significant predictor of near-viewforest scenic beauty.

Virtually all of the preceding studies have utilized one of two methods of quantify-ing scenic beauty. These methods are the Scenic Beauty Estimation Method (SBE)developed by Daniel and Boster (1976) and the Law of Comparative Judgment (LCJ)introduced by Buhyoff and Leuschner (1978), who built on the work of Thurstone(1927). Both methods produce indices based on perceptions of the scenic beautyof a series of photographs representing a landscape, and both methods have beenextensively tested and validated (Hull et al, 1984). The SBE technique involves a scalingprocedure where each respondent utilizes a scale (usually from 1 to 10) to subjectivelyrate the scenic beauty of a series of photographs. The LCJ method requires eachrespondent to compare all possible pairs of photographs and decide which photographin each pair is more scenic. The two procedures are described and compared in detailby Hull et al (1984).

In general, the SBE method is more efficient than the LCJ approach, in the sensethat each individual evaluating n photographs under the SBE method need only make nratings. Under the LCJ approach, the same n photographs require n (nÿ 1)=2 compar-isons of photograph pairs. As under most conditions the efficient SBE method hasbeen found to be just as accurate and reliable a predictor of scenic-beauty prefer-ences as the more labor-intensive LCJ approach, the SBE method has become thepreferred approach to scenic-beauty estimation (Daniel and Boster, 1976).

However, there is one exception to this general preference for the SBE approach.When the photographs being compared are very similar to one another (therebyrequiring the evaluators to make fine distinctions between photographs with similarscenic qualities), the use of all possible pairs of photographs in the LCJ approach offersdistinct advantages over the SBE method (David, 1988; Hull and Buhyoff, 1981). Inmany instances within this paper, evaluators were asked to compare the relative beautyof scenes that differed only by their distances from clearcuts. It was believed thatunder these conditions, the LCJ approach was more appropriate than the SBE method.Thus, the LCJ approach was used throughout this paper.

Modeling the magnitude and spatial distribution of aesthetic impacts 123

MethodsStudy areaClearcuts from two widely separate study areas were evaluated in this project. Fiveclearcuts were evaluated from the first study area; a southern yellow pine plantationlocated in the coastal plain of South Carolina (bottom portion of figure 1). The areawas intensively managed for the commercial production of pulpwood (this involves grow-ing relatively small trees suitable for the production of paper products). It consisted ofhomogeneous forest stands (that is, within each stand, there was very little variation inspecies composition, basal area, height, density, volume, or number of trees per acre).All stands had a very dense understory; that is, there were a great many tall grasses andshrubs growing underneath the trees. There was virtually no vertical relief in the area,with all slopes less than 2%. Two of the clearcuts from this area were surrounded byloblolly pine (Pinus taeda L.) stands, whereas the remaining three were surroundedby stands of hardwoods composed primarily of sweetgum (Liquidambar styracifluaL.), red maple (Acer rubrum L.), various oaks (Quercus spp.) and elms (Ulmus spp.),yellow-poplar (Liriodendron tulipifera L.), and hickories (Carya spp.).

ÿ10 0 10 20 30 40 km

Figure 1. Colorado (top) and South Carolina (bottom) study sites.


The second study site was located in the Colorado State Forest located in theRocky Mountain Front Range (top portion of figure 1). Vegetation was predominantlylodgepole pine (Pinus contorta) with interspersed quaking aspen (Populus tremuloides)and spruce/fir (Picea engelmanii, Abies lasiocarpa) stands. Understory in these standswas much less dense than that found in South Carolina. The topography includedsteeper slopes, ranging from 17% to over 40%. A total of five clearcuts in Coloradowere evaluated, with three surrounded by lodgepole pine stands, one by spruce/fir, andone by aspen stands.

Site inventory procedureA photograph transect was established in the forest adjacent to each clearcut. Eachtransect was a straight line starting at the edge of the clearcut and moving into thesurrounding forest along a line extending out radially from the approximate center ofthe cut (figure 2). The specific orientation of the transect was randomly chosen;however, this process was frequently constrained because transects with certain orienta-tions were obviously obstructed by intervening terrain or caused the transect to passinto two or more stands with obviously differing visual characteristics (for example,a mature aspen stand and an immature spruce/fir stand). In these cases, transectorientation was randomly selected from the set of orientations that did not encounterobstacles or pass into multiple forest stands.

Starting at the edge of each clearcut, photographs of the cut were taken every 10 malong the transect until the clearcut was no longer visible. Once the cut was lost toview, an additional three photographs (covering 30 m) were taken to ensure that nosign of the clearcut was visible. A standard 35 mm camera with a 55 mm lens and ASA

Stand 7

Stand 8

Stand 3

Stand 4

Stand 2

Randomtransect

Stand 5

Stand 6

Stand 13

Stand 14

Stand 9

Hill

Azimuths disallowedowing to hill andmultiple stands

Stand12

Approx.center ofclearcut

Clear cut

Stand10

Figure 2. Layout of a photograph transect around a hypothetical clearcut.


400 print film were used. All photographs were taken at the highest depth of fieldpossible while maintaining a shutter speed of 1/60. All pictures were taken between8 AM and 5 PM to guarantee sufficient light and no overwhelming shadows. To eliminatepossible biases, every effort was made to ensure that no additional clearcuts, buildings,people, wildlife, or vehicles were photographed. In addition, photography was limited tothe summer season to prevent any bias due to seasonal changes (Buhyoff and Wellman,1980).

(c)

Figure 3. Examples of photographs from the South Carolina and Colorado study sites. (a) Photo-graph taken 0 m from edge of South Carolina clearcut; (b) photograph taken 20 m from edge ofSouth Carolina clearcut; (c) photograph taken 60 m from edge of Colorado clearcut; (d) photographtaken 140 m from edge of Colorado clearcut.

(a)


In addition to the photograph transects, field procedures also included taking treemeasurements at a randomly selected point along each transect line. These measure-ments were taken using common forest inventory techniques. A 0.10 acre plot wasestablished and the number of trees, average height of codominant trees (that is, thetrees that make up the canopy of the forest), and the diameter at breast height (DBH,a standard measure of tree girth) of all trees with DBH greater than 3 inches wasrecorded. Understory and ground cover were noted but were not quantifiably measured.

Aesthetic preference measurementThe LCJ technique was used to rate the scenic beauty of the photographs. LCJrequires a series of evaluators to select the more aesthetically pleasing photographin each of a series of photograph pairs. This preference information is used todevelop an interval-level scale of the relative attractiveness of each photograph.

(b)

(d)

Figure 3 (continued).


Six 466 inch color photograph prints were used to represent each transect. Eachset of six photographs started with the photograph taken at the edge of the cut andconcluded with a photograph taken from a point at which the cut was no longervisible. The remaining four photographs were randomly selected from all usablephotographs from each transect (a few photographs were eliminated from considera-tion because of poor focus, blurring, and so on). These photographs depicted theprogression along the transect from the edge of the clearcut to the point wherethe cut was no longer visible. Distances between photographs ranged from 10 m to50 m, and the maximum distance from the edge of the clearcut to the farthestphotograph ranged from 90 m to 180 m. Examples of photographs from representa-tive transects from both the South Carolina and Colorado study sites are shown infigure 3.

The photographs from each transect were arranged into ten separate photographalbums containing the six prints arranged in all possible pairs (fifteen pairs pertransect). The photographs were ordered in an optimum balancing routine (Ross,1934) in which each photograph appeared an equal number of times on the topand on the bottom of the page as well as being proportionally distributed from thebeginning to the end of the album.

Subjects for evaluating the photographs were Colorado State University studentsand faculty and other Colorado professionals (n � 110) varying in age, profession,interest, and ethnic background. Prior to rating the photographs, a standard set ofinstructions was read to the subjects; no additional information beyond these instruc-tions was provided. Each subject was presented with a photograph album and askedto indicate which photograph they preferred in each pair. They were allowed 5 ^ 8 s toselect their preferred photograph from each pair. To minimize the possibility of fatigue,each subject only evaluated six (three from each study area) rather than all ten albums.Thus, each individual rated a total of ninety pairs of photographs (fifteen photographpairs per transect 6 six transects).

Statistical analysisData from the evaluations of each photograph transect were analyzed separately. Theresults of each transect were analyzed using the methodology presented by Guilford(1954, pages 154 ^ 177). Only data from evaluators who were consistent in theirpreferences were used in the final analysis. Consistency was determined by examin-ing each evaluator's preferences. For example, if an evaluator preferred photograph Aover photograph B and photograph B over photograph C, this evaluator cannotprefer photograph C over photograph A. Kendall (1962) termed inconsistencies suchas these between three photographs c̀ircular triads'; when the number of photographsis larger than three, an inconsistency is referred to as a circular n-ad. Inconsistenciesmay be a result of guessing or may be caused by the fact that the photographs beingcompared are indistinguishable. In this paper, the photographs varied only in distancesfrom the clearcuts, and in some cases the photographs were almost identical. Althougheach transect was initially evaluated by sixty individuals, the number of consistentevaluations per transect ranged from twenty four to thirty nine.

Once the preference data from the consistent evaluators for a given transect wereidentified, they were used to develop an interval-level scale of the attractiveness ofeach photograph. This was accomplished by calculating the observed probabilitythat each photograph A would be preferred over photograph B for all pairs ofphotographs. Guilford (1954) terms the diagonally symmetric matrix showing allof these observed probabilities a dominance priority matrix. Following Guilford'sprocedures, the probabilities in the dominance priority matrix were then normalized


to produce Z scores for each probability, and these Z scores were then averaged toconstruct the estimated aesthetic ranking for each photograph.

The standard LCJ procedure just described produced what we termed `raw'aesthetic-quality rankings. These raw aesthetic-quality rankings were entirely consis-tent with one another within a single transect, but owing to their interval nature,were not comparable across transects. In order to make between-transect compar-isons, we standardized the raw scores so that the minimum ranking of each transectwas zero and the maximum ranking was one. Given the interval scale of aesthetic-quality rankings, this standardization was accomplished without losing any of theinformation contained in the original LCJ results. We termed these standardizedrankings.

Thirteen separate datasets were constructed from the standardized rankings andwere analyzed via regression analyses. The first ten datasets were specific to individualtransects; each of these datasets were based on only six observations (representing thesix photographs evaluated in each transect). Each of these datasets consisted of a singledependent variable (the standardized rankings of the photographs) and a single inde-pendent variable (the distance from the point where the photograph was taken to theedge of the clearcut). Two additional datasets were constructed by (1) aggregating allthe data from the South Carolina transects, and (2) aggregating all the data from theColorado transects. Each of these state-specific datasets were based on thirty observa-tions (six observations/transect 6 five transects per state). In addition to the dependentand independent variables found in the ten transect-specific datasets, the two state ^state specific datasets contained the forest-inventory measurements (tree density, DBH,average height of dominant and codominant trees, and a series of binary indicatorvariables used to identify species) measured along each transect as additional indepen-dent variables. Finally, an overall dataset was constructed by aggregating data fromall ten transects across both states; this dataset contained sixty observations. Thisoverall dataset contained the same dependent and independent variables as found inthe state-specific datasets.

Five intrinsically linear regression forms (true linear, squared, quadratic, exponen-tial, and logarithmic) were used to analyze each of the thirteen analyses just described.These forms were chosen because Buhyoff and Wellman (1980) and Hull and Buhyoff(1983) used them in previous studies that related perceived scenic beauty to forestcharacteristics and viewing distance. In addition, a nonlinear regression equation wasdeveloped for each of the thirteen datasets described previously. Thus, a total of seventy-eight regression results were produced (thirteen datasets 6 six regression forms perdataset).

GIS-based spatial modelingThe results of the regression analyses just described were used to develop a modeldesigned to estimate the aesthetic impacts of clearcuts on surrounding areas. Themodel was an extension of a modified GIS-based viewshed analysis technique pro-posed by Dean (1997). Standard viewshed analysis is binary; either the line of sightconnecting a viewpoint to a raster cell is completely unobstructed (meaning the rastercell is entirely visible from the viewpoint) or completely obstructed (meaning that theraster cell is entirely invisible from the viewpoint) (Burrough, 1986; Laurini andThompson, 1992; Tomlin, 1990). However, numerous authors have proposed nonbinaryalternative viewshed estimation systems. A partial list of these alternative techniquesincludes probabilistic approaches (which create a raster surface where each cell containsthe probability that it is visible from the viewpoint) (Fisher, 1992; 1994; Nackaerteset al, 1999), partial-cell approaches (which predict what proportion of each raster cell


is visible) (Sorenson and Lanter, 1993), and a system designed to account for thevision-obscuring effects of vegetation proposed by Dean (1997).

The modified viewshed-delineation system proposed by Dean (1997) classifiesraster cells into three categories: (1) completely visible (the line of sight from theviewpoint to the raster cells suffers from no obstructions, either topographic or vege-tative), (2) completely invisible (the line of sight is entirely obstructed by one or moreopaque topographic features), or (3) partially visible, meaning that the line of sight isnot obstructed by opaque terrain but does pass through some distance of vegetation.For raster cells falling into the third category, the modified viewshed technique alsocomputes the total length of the line of sight passing through vegetation.

Using the modified viewshed system just described, the results of the regressionanalysis could be used to estimate the aesthetic impacts of clearcuts. In concept, thismodel iteratively places a viewpoint on each cell in the DEM. If the line of sight fromthis viewpoint to the clearcut is completely obstructed [that is, falls into case (2)], theclearcut has no aesthetic impact on the viewpoint cell so the viewpoint is assigned anaesthetic-impact rating of zero. If the line of sight is completely unobstructed [that is,falls into case (1)], the viewpoint cell is assigned a maximum aesthetic-impact rating ofone. Finally, if the line of sight is partially obstructed [case (3)], the system computesthe distance that the line of sight travels through the intervening forest vegetation priorto reaching the clearcut. This distance estimate is then used in the regression equationdiscussed previously (along with the forest inventory measures of the forest throughwhich the line of sight is passing) to produce an estimated aesthetic impact rating forthe cell.

Results, validation, and discussionRelationship between aesthetic quality and viewing distance through forestThe ten transect-specific datasets were analyzed using the six regression models shownhere (Y � standardized scenic beauty ranking, X � distance to clearcut, b0 , b1 , andb2 are regression parameters, and A is the asymptote for the nonlinear regression):

Linear: Y � b0 � b1X ,

Square: Y � b0 � b1X2 ,

Quadratic: Y � b0 � b1X� b2X2 ,

Exponential: Y � b0 � b1 exp�X � ,Logarithmic: Y � b0 � b1 ln�X � ,Nonlinear: Y � A�1ÿ exp�b0 � b1X �� .All regression parameters for the five intrinsically linear models were estimated

using standard parametric least-squared techniques, whereas the nonlinear regressionparameters were estimated using the secant method (SAS, 1990). These analysesrevealed that the logarithmic model proved to be significant at the 0.05 level andproduced R 2 values greater than 0.7 for nine of the ten transects. The nonlinear,quadratic, square, and linear models were each significant at the 0.05 level for threetransects, and the exponential model was significant for one transect (table 1).

Similar analyses (using the same six regression forms) were also performed on thetwo datasets aggregated to the state level, but these analyses included the forestinventory measures as additional independent variables. A stepwise process was usedto identify the set of independent variables that (1) were each significant predictorsof the aesthetic ratings of the photographs and (2) collectively produced the highest


overall predictive power of all possible sets of independent variables. This processstarted with all independent variables in the model and then iteratively removed oradded variables to the model until no further removals or additions that improved themodel were available. The standardized rankings used as the dependent variables inthese analyses were adjusted so that the single photograph within each dataset with thehighest raw ranking was given a value of one (rather than the highest-ranked photo-graph from each individual transect being given a ranking of one) and the singlephotograph with the lowest raw ranking was given a value of zero.

Results from both states were very similar. For both states, five of the six regressionmodels were significant at the 0.05 level; only the exponential model was insignificant.The nonlinear model provided the highest pseudo R 2 of the five significant models,followed by the logarithmic model (table 1). A somewhat unexpected finding resultedfrom the stepwise independent-variable selection procedure. For each of the sixregression models investigated from both states (a total of twelve models), the onlyindependent variable that was retained in each of the final models was distance fromthe photograph to the edge of the clearcut; none of the forest-inventory measures weresignificant.

In retrospect, the reason why forest inventory measures were insignificant predic-tors of aesthetic preference was obvious. The South Carolina study site featured veryflat terrain. Given the flat terrain, lines of sight generally passed under the forestcanopy and thus were effected much more significantly by the understory vegetationthan they were by the overstory forest trees. As the forest inventory measures evaluatedin this paper pertained only to the overstory, it would have been surprising if thesemeasures were related to aesthetic preference.

In the Colorado study site, a subjective examination of the photographs shows thatunderstory still played the more obvious role but, given the rolling terrain, it iscertainly plausible that at least some lines of sight could have been impacted by theoverstory. The fact that none of the regression analyses described in table 1 found anyof the overstory inventory measures to be significant predictors of aesthetic impactimplies that if such overstory effects took place, they were not significant enough to beretained in the regression models.

Some of the stands in the Colorado study areas were composed of very denselypacked small trees (that is, `doghair' stands), and it seems likely that the density of treeboles in these stands could impact even lines of sight passing under the forest canopy.However, no single inventory measure could identify these doghair standsöidentifyingthese stands required evaluation of the combined values of all three inventory measures.Thus, the Colorado dataset was analyzed again using a stepwise regression procedurethat considered as potential independent variables not only distance and the overstoryinventory measures evaluated previously, but also all four possible interactions of thecontinuous inventory measures (trees per acre 6 height, trees per acre 6 mean DBH,height 6 mean DBH, and trees per acre 6 height 6 mean DBH). Once again, allsix forms of the regression model were evaluated. In each of these six analyses thethree-way interaction of trees per acre 6 height 6 mean DBH was found to be asignificant predictor of aesthetic preference, but the addition of this interactionvariable did not noticeably improve the R 2 values of the resulting regression models.The linear model showed the greatest improvement, with its R 2 value increasingfrom 0.3114 without the interaction variable to 0.3202 with the interaction variableöan increase of only 0.0088. Given the very small magnitude of this improvement,it was decided to remove the forest inventory measures from all of the remaininganalyses.


The same six regression models were applied to the overall dataset created bycombining data from all ten transects. Once again, the standardized rankings in thisfinal dataset were adjusted so that the single photograph with the highest raw rankingwas given a value of 1 and the photograph with the lowest raw ranking was given avalue of 0. The results from this final dataset were very similar to those obtained withthe previous two datasets: five of the six models produced significant results at the 0.05level, with the nonlinear and logarithmic models producing the highest R 2 values. Onlythe exponential model produced insignificant results.

Regression model validationThe results shown in table 1 indicate that the logarithmic and nonlinear regressionsgenerally outperformed (in terms of R 2 and p-values) the other regression models. Inaddition, table 1 indicates that models aggregated at the state level tended to outper-form models aggregated across both states; this may be because of the obviousdissimilarities in the screening capabilities of the forests in the two states. Thesefindings implied that in the second phase of this study, the predictive model designedto estimate the aesthetic impacts of clearcuts should be based on data aggregated atthe state level and logistic and nonlinear regression forms should be considered.However, prior to developing such a predictive model, the predictive power of thesetwo regression models was evaluated using a pair of bootstrapping processes.

Each bootstrapping process involved removing the data from one transect from astate level database. This produced a reduced dataset containing information from fourtransects. Logarithmic and nonlinear regression analyses were then used to produceregression results from this reduced dataset. These reduced regression results were thenused to derive estimated aesthetic rankings for photographs taken at each of the sixdistances from clearcuts contained in the removed transect. These estimated aestheticrankings were then compared with the observed standardized rankings contained inthe removed transect, and differences between actual and predicted rankings werenoted. This process was then repeated by removing a different transect from theoriginal database, until each transect had been removed once.

Table 1. Results of the regression analyses, for each individual transect, each study site, and bothstudy sites combined.

NCRa n Exponential Linear

R 2 p R 2 p

Transect 1 14 6 0.2548 0.3072 0.7700 0.0216Transect 2 21 6 0.1615 0.4297 0.5153 0.1082Transect 3 15 6 0.9519 0.0009 0.5514 0.0909Transect 4 27 6 0.0001 0.9902 0.1987 0.3756Transect 5 14 6 0.3127 0.2486 0.6101 0.0666Transect 6 22 6 0.2358 0.3289 0.8437 0.0097Transect 7 28 6 0.2826 0.2778 0.9245 0.0022Transect 8 29 6 0.0092 0.5631 0.4938 0.1194Transect 9 27 6 0.0352 0.7220 0.7171 0.0334Transect 10 25 6 0.0692 0.6146 0.4655 0.1354CO only 98 30 0.0158 0.5076 0.3114 0.0014SC only 124 30 0.0581 0.1996 0.4135 0.0001CO and SC 222 60 0.0312 0.1770 0.3203 0.0001

aNCR � number of consistent ratings. This is the number of individuals who consistentlyrated the photographs in each transect. This should not be confused with the sample size (n).


The results of the bootstrapping analysis are shown in table 2. The logarithmicregression model resulted in an average mean difference (observed value minus pre-dicted value) of 0.0674 for the Colorado study site (differences range from ÿ0.4476 to0.4170) and an average mean difference of 0.0898 for the South Carolina study site(range ÿ0.4621 to 0.3040). The nonlinear regression model resulted in an average meandifference of ÿ0.0008 for the Colorado site (range ÿ0.3657 to 0.4125) and ÿ0.0038for the South Carolina site (range ÿ0.3192 to 0.2729). It is interesting to note that sixof these eight extreme values were recorded at zero distance from the clearcut; if thesezero-distance estimates are ignored, the overall performance of both models improvedsignificantly.

When evaluating these deviations it is helpful to recall that all rankings werestandardized to range from zero to one, so an average difference of 0.09 indicatesthat average differences covered 9% of the possible range of standardized rankings.Clearly, both the logarithmic and nonlinear models produced estimates that onaverage were fairly close to observed values, but again on average the nonlinearmodel outperformed the logarithmic model.

Aesthetic impact modelGiven that they outperformed all of the other regression models, both the logarithmicand nonlinear regression equations were used in building a GIS-based aesthetic impactmodel. As the regression models failed to include anything other than distance as anindependent variable, there was no need to include any forest inventory measures inthe aesthetic-impact model.

The aesthetic-impact model was tested for logical consistency by applying it toseveral artificial datasets based on simple elevation surfaces. The overall results pro-duced by the model seem intuitively reasonable, and cell-by-cell examination of theseresults showed the expected patterns: raster cells immediately adjacent to the clearcutsshowed the greatest level of impact, and the magnitude of impacts decreased as viewingdistance increased. Raster cells where the clearcut was completely hidden by interveningterrain showed no aesthetic impact.

Table 1 (continued).

Square Quadratic Logarithmic Nonlinear

R 2 p R 2 p R 2 p R 2 p

0.5484 0.0923 0.8679 0.0742 0.8175 0.0133 0.8932 0.09930.3607 0.2074 0.6422 0.2221 0.9107 0.0031 0.8962 > 0:10.7644 0.0227 0.8909 0.1591 0.0902 0.5630 0.9779 0.02180.0537 0.6586 0.6473 0.1103 0.7502 0.0257 0.9148 >0.10.4232 0.1618 0.7436 0.1482 0.8736 0.0063 0.8622 >0.10.7121 0.0347 0.8710 0.1503 0.7735 0.0209 0.8722 >0.10.7573 0.0242 0.9598 0.0301 0.8065 0.0151 0.9661 0.07650.2942 0.2662 0.7267 0.1175 0.9495 0.0010 0.9682 0.02400.4757 0.1297 0.9840 0.0023 0.7053 0.0364 0.9301 0.07200.2550 0.3069 0.8694 0.0329 0.9508 0.0009 0.9674 0.04800.1451 0.0378 0.5520 0.0001 0.5188 0.0001 0.9947 0.01830.2300 0.0073 0.5054 0.0006 0.7786 0.0001 0.9976 0.01220.1750 0.0009 0.4309 0.0001 0.5482 0.0001 0.9983 0.0014


Table 2. Bootstrap results (observed rating ÿ predicted rating) for (a) Colorado and (b) SouthCarolina.

Bootstrap results at (distance from clearcut in meters):

0 10 20 30 40 50 60 70 80

(a) Colorado study siteLogarithmic modelTransect 1 0.0193 0.0264 0.0230

Transect 2 0.2839 0.1139 0.0686

Transect 3 ÿ0.3178 0.0542 0.1874 0.0700 0.0648

Transect 4 ÿ0.4476 ÿ0.0942 ÿ0.0012 0.1059 0.3503

Transect 5 ÿ0.0631 ÿ0.1208 0.1082

Mean ÿ0.1051 ÿ0.0117 0.1069 0.0344 0.0764 0.0648 0.3503

Minimum ÿ0.4476 ÿ0.1208 0.0264 ÿ0.0012 0.0230 0.0648 0.3503

Maximum 0.2839 0.1139 0.1874 0.0700 0.1082 0.0648 0.3503

Nonlinear modelTransect 1 0.1346 ÿ0.0356 ÿ0.1161Transect 2 0.4125 0.1394 ÿ0.0882Transect 3 ÿ0.2271 0.0841 0.1678 0.0154 ÿ0.0461Transect 4 ÿ0.3657 ÿ0.1042 ÿ0.0751 0.0109 0.2497

Transect 5 0.0467 ÿ0.1329 ÿ0.0187Mean 0.0002 ÿ0.0034 0.0661 ÿ0.0299 ÿ0.0530 ÿ0.0461 0.2497

Minimum ÿ0.3657 ÿ0.1329 ÿ0.0356 ÿ0.0751 ÿ0.1161 ÿ0.0461 0.2497

Maximum 0.4125 0.1394 0.1678 0.0154 0.0109 ÿ0.0461 0.2497

Mean deviation for the logarithmic model � 0.0674.Mean deviation for the nonlinear model � ÿ0.0008.

(b) South Carolina siteLogarithmic modelTransect 1 ÿ0.4621 0.2130 0.1203

Transect 2 ÿ0.0701 0.2504 0.1326 0.2068 0.0206

Transect 3 0.0865 0.0319 ÿ0.0010 0.1685

Transect 4 ÿ0.3584 0.0596 0.3040 0.1399 0.1532

Transect 5 ÿ0.0112 0.1045 0.1499 0.1530

Mean ÿ0.1631 0.0596 0.1289 0.2130 0.1510 0.1326 0.1557 0.1140 0.1532

Minimum ÿ0.4621 0.0596 0.0319 0.2130 ÿ0.0010 0.1326 0.1203 0.0206 0.1532

Maximum 0.0865 0.0596 0.2504 0.2130 0.3040 0.1326 0.2068 0.1685 0.1532

Nonlinear modelTransect 1 ÿ0.3192 0.0745 ÿ0.0098Transect 2 0.0960 0.1010 ÿ0.0378 0.0406 ÿ0.1414Transect 3 0.2729 ÿ0.1414 ÿ0.1778 0.0073

Transect 4 ÿ0.2115 ÿ0.0054 0.1587 0.0031 0.0232

Transect 5 0.1650 ÿ0.0618 ÿ0.0250 ÿ0.0074Mean 0.0006 ÿ0.0054 ÿ0.0341 0.0745 ÿ0.0147 ÿ0.0378 0.0113 ÿ0.0472 0.0232

Minimum ÿ0.3192 ÿ0.0054 ÿ0.1414 0.0745 ÿ0.1778 ÿ0.0378 ÿ0.0098 ÿ0.1414 0.0232

Maximum 0.2729 ÿ0.0054 0.1010 0.0745 0.1587 ÿ0.0378 0.0406 0.0073 0.0232

Mean deviation for the logarithmic model � 0.0898.Mean deviation for the nonlinear model � ÿ0.0038.


Table 2 (continued).

90 100 110 120 130 140 150 160 170 180

0.1414 0.0560 0.1452

0.0580 0.0375 ÿ0.06420.3387

0.4170

0.1806 0.1909 0.0906

0.0997 0.3121 0.0560 0.0375 0.1909 ÿ0.0642 0.1179

0.0580 0.1806 0.0560 0.0375 0.1909 ÿ0.0642 0.0906

0.1414 0.4170 0.0560 0.0375 0.1909 ÿ0.0642 0.1452

ÿ0.0134 ÿ0.0997 ÿ0.0058ÿ0.1310 ÿ0.1585 ÿ0.2592

0.2097

0.3161

0.0455 0.0592 ÿ0.0379ÿ0.0722 0.1904 ÿ0.0997 ÿ0.1585 0.0590 ÿ0.2592 ÿ0.0219ÿ0.1310 0.0455 ÿ0.0997 ÿ0.1585 0.0592 ÿ0.2592 ÿ0.0379ÿ0.0134 0.3161 ÿ0.0997 ÿ0.1585 0.0592 ÿ0.2592 ÿ0.0058

ÿ0.0178 0.2270 0.3108

0.0444

0.1411 0.1812

0.2287

0.0635 0.1237

0.0578 0.1762 0.1812 0.2270 0.3108

ÿ0.0178 0.1237 0.1812 0.2270 0.3108

0.1411 0.2287 0.1812 0.2270 0.3108

ÿ0.1387 0.1180 0.2060

ÿ0.1107ÿ0.0127 0.0334

0.1041

ÿ0.0896 ÿ0.0264ÿ0.0879 0.0389 0.0334 0.1180 0.2060

ÿ0.1387 ÿ0.0264 0.0334 0.1180 0.2060

ÿ0.0127 0.1041 0.0334 0.1180 0.2060


Predicted aesthetic impacts produced using the logarithmic equation were similarto those produced using the nonlinear equation.When the raster maps produced by thetwo equations were subtracted, it was found that in general, the nonlinear modelproduced slightly higher aesthetic-impact estimates than the logarithmic model atvery small distances from the clearcut, but tended to produce lower-impact estimatesat larger distances.

Once the model had been tested on these artificial study areas, it was applied toreal-world situations involving clearcuts from both the Colorado and South Carolinastudy sites but not used in the model-building process. A set of typical results fromthese analyses can be seen in figure 4.

It is interesting to view the results shown in figure 4 as if they were comparisons ofproposed clearcuts. If you were using the standard viewshed approach to evaluate thecuts, you would conclude that cut A was the least objectionable (because it hasthe smallest viewshed), followed by cut C, with cut B being the most objectionable(that is, it has the largest viewshed). However, the aesthetic-impact model indicatesthat the large viewshed of clearcut B is mostly lightly impacted aesthetically, whereasthe smaller viewshed of cut C is almost-entirely heavily impacted. This fact is reflectedin the sum of impacts values shown in the figure. These values are simply the sumacross all raster cells of the predicted aesthetic-impact values. Recall that a cell that ishighly impacted has a large predicted aesthetic value whereas a cell that is notimpacted at all has an aesthetic-impact value of zero. Thus, higher sums of impactsvalues reflect clearcuts that have greater overall impacts. Using this model, you wouldstill conclude that A is the least objectionable clearcut, but then you would find cut Bto be the next best alternative, with cut C being the least desirable.

Clearcut A Clearcut B

Clearcut C

Area � 95.13 haSum impact � 46 432

Aesthetic impact

High Low

Clearcut

Area � 916.83 haSum impact � 385 870

Area � 733.86 haSum impact � 539852

Figure 4. Results of aesthetic-impact estimation model when applied to real-world datasets.


ConclusionsIn this paper we have demonstrated that it is possible to combine the models that havebeen developed to quantify aesthetic preferences with the spatial modeling capabilitiesof GIS. The resulting hybrid has unique advantages that are not present in eitherconventional aesthetic preference models or off-the-shelf GIS software. It seems likelythat a fully functional version of the hybrid model developed here could be of consid-erable worth to forest managers. Furthermore, it seems plausible that approachessimilar to those used here could be adopted to other aesthetic impact situationsöforexample, the impacts of the wind turbines investigated by Bishop (2002). In fact, thereis no conceptual reason why the model developed here could not be extended tosituations involving urban or other man-made environments. The only assumptionmade here is that an aesthetically unappealing site is located within a more appealingsurrounding environment, and that the effectiveness of the surrounding environmentat screening the unappealing site increases as one moves further away from the site.As long as this assumption is met, a model of the same basic type as that presentedhere could be used to estimate the impact of any aesthetically unappealing site.

At least two issues remain to be resolved before a fully functional aesthetic-impactmodel could be constructed. First, we looked only at near-view impacts; this wouldobviously have to be married with a model that estimates impacts over longer viewingdistances to produce estimates of total aesthetic impacts. Fortunately, there is no reasonto believe that the techniques used in this study could not be extended to longer viewingdistances. A cutoff distance (probably equal to the maximum length of transects likethose used in this study) could be established, and a model identical to the one devel-oped here could be used to estimate aesthetic impacts within this distance. Beyond thiscutoff, photographs of clearcuts could be taken from longer distances and used in the SBEor LCJ approach to measure aesthetic preferences over vista-scale landscapes. Regressionanalyses could then be used to quantify the relationship between aesthetic quality andthese longer viewing distances, and these regression results could be incorporated intoGIS-based models to estimate aesthetic impacts over vista-scale landscapes.

An obvious area where we failed in this paper was in relating changes in aestheticpreferences over distance to forest inventory measures. It is certainly possible thatother overstory inventory measures not investigated in this paper might produce betterresults. However, based on subjective reviews of the photographs used in this paper,relating aesthetics to understory vegetation measures seems much more promisingthan any measure of overstory characteristics.

Finally, it remains to be seen if the more efficient SBE method could be used in thissort of study rather than the LCJ approach adopted here. It is undeniably true that theSBE method is more efficient than the LCJ system and, in many other areas, the SBEmethod has been found to be just as precise and accurate as the LCJ approach. If thisis found to be true for the transect sampling used in this study, the SBE method wouldbe a very attractive alternative to the more labor-intensive LCJ system.

Despite these limitations, the model developed in this paper fulfilled its purpose ofdemonstrating the feasibility of producing spatial aesthetic-impact models. The resultspresented here provide a starting point for further research in the development oflinked LCJ, SBE, and GIS models designed to quantify and analyze the commutativeaesthetic effects.

ReferencesArthur L M, 1977, ``Predicting scenic beauty of forest environments: some empirical tests'' Forest

Science 23 151 ^ 160Bishop I D, 2002, `̀ Determination of thresholds of visual impact: the case of wind turbines''

Environment and Planning B: Planning and Design 29 707 ^ 718


Bishop I D, Karadaglis C, 1997, `̀ Linking modelling and visualisation for natural resourcesmanagement'' Environment and Planning B: Planning and Design 24 345 ^ 358

BrownTC, Daniel T C,1986, `̀ Predicting scenic beauty of timber stands''Forest Science 32 471 ^ 487Buhyoff G J, Leuschner WA, 1978, `̀ Estimating psychological disutility from damaged forest

stands'' Forest Science 24 424 ^ 434Buhyoff G J,Wellman J D, 1980, `̀ Seasonality bias in landscape preference research'' Leisure

Science 2 181 ^ 190Buhyoff G J,Wellman J D, Daniel T C, 1982, `̀ Predicting scenic beauty for mountain pine beetle

and western spruce budworm damaged forest vistas'' Forest Science 28 827 ^ 838Burrough PA, 1986 Principles of Geographical Information Systems for Land Resource Assessment

(Oxford University Press, Oxford)Daniel TC, Boster R S,1976Measuring LandscapeAesthetics:The Scenic Beauty EstimationMethod

research paper 167, USDA Forest Service Rocky Mountain Research Station, Fort Collins, CODavid H A, 1988 The Method of Paired Comparisons 2nd edition (Charles Griffin, London)Dean D J, 1997, `̀ Improving the accuracy of forest viewsheds using triangulated networks and

the visual permeability method'' Canadian Journal of Forest Research 27 969 ^ 977Erdle T A, 1999, `̀ The conflict in managing New Brunswick's forests for timber and other values''

Forestry Chronicle 75 945 ^ 954Fisher P F, 1992, `̀ First experiments in viewshed uncertainty: simulating fuzzy viewsheds''

Photogrammetric Engineering and Remote Sensing 58 345 ^ 352Fisher P F, 1994, ``Probable and fuzzy models of the viewshed operation'', in Innovations in GIS

Ed. M F Worboys (Taylor and Francis, London) pp 161 ^ 175Guilford J P, 1954 Psychometric Methods (McGraw-Hill, New York)Gustafson E J, CrowT R, 1998, `̀ Simulating spatial and temporal context of forest management

using hypothetical landscapes'' Environmental Management 22 777 ^ 787Hollenhorst S J, Brock S M, Freimund WA, Twery M J, 1993, `̀ Predicting the effects of gypsy

moth on near-view aesthetic preferences and recreation appeal'' Forest Science 39 28 ^ 40Hull BR, Buhyoff G J, 1981, `̀ On the law of comparative judgement: scaling with intransitive observers

and multi-dimensional stimuli''Educational and Psychological Measurement 41 1083 ^ 1089Hull B R, Buhyoff G J, 1983, ``Distance and scenic beauty: a nonmonotonic relationship''

Environment and Behavior 15 77 ^ 91Hull B R, Buhyoff G J, Daniel T C, 1984, `̀ Measuring of scenic beauty: the law of comparative

judgement and scenic beauty estimation procedures'' Forest Science 30 1084 ^ 1096Kendall M G, 1962 Rank Correlation Methods 3rd edition (Hafner, NewYork)Kline J D, Armstrong C, 2001, `̀Autopsy of a forestry ballot initiative: characterizing voter support

for Oregon's measure 64'' Journal of Forestry 99 20 ^ 27Laurini R, Thompson D, 1992 Fundamentals of Spatial Information Systems (Academic Press,

San Diego, CA)Limerick P N, 2002, `̀ Forestry and modern environmentalism: ending the cold war'' Journal of

Forestry 100 46 ^ 50Nackaerts K, Govers G, Orshoven J V, 1999, `̀Accuracy assessment of probabilistic visibilities''

International Journal of Geographical Information Science 13 709 ^ 721Ribe R G, 1990, `̀A general model for understanding the perception of scenic beauty in northern

hardwood forests'' Landscape Journal 9 8 ^ 101Ross RT,1934, `̀ Optimum orders for the presentation of pairs in themethod of paired comparisons''

Journal of Educational Psychology 25 370 ^ 382Rudis VA, Gramann J H, Rudell E J,Westphal J M, 1988, `̀ Forest inventory and management

based visual preference models of southern pine stands'' Forest Science 34 846 ^ 883SAS, 1990 SAS/STAT Users Guide,Version 6,Volume 2 4th edition (SAS Institute, Cary, NC)Sorensen P A, Lanter D P, 1993, `̀ Two algorithms for determining partial visibility and reducing

data structure induced error in viewshed analysis''Photogrammetric Engineering and RemoteSensing 59 1149 ^ 1160

Thurstone L L, 1927, `̀A law of comparative judgement''Psychological Review 34 273 ^ 328Tomlin C D, 1990 Geographic Information Systems and Cartographic Modeling (Prentice-Hall,

Englewood Cliffs, NJ)

ß 2007 a Pion publication printed in Great Britain
