Brian Orland
Rado Family Foundation/UGAF Professor of Geodesign
College of Environment and Design
University of Georgia
285 South Jackson Street
Athens, GA 30602

Dr. Richard Ready
Department of Agricultural Economics and Economics
Montana State University
P.O. Box 172920
Bozeman, MT 59717-2220

ABSTRACT

A study of non-hydrological benefits and citizen preference for green stormwater management strategies (Project 4) took an approach common in consumer choice research.  Traditionally, such studies have depended heavily on the use of survey instruments primarily using verbal protocols. The results of such research have led to considerable success in predicting market behavior. However, some of the attributes of natural resource “products” are not amenable to verbal representation. In the case of green infrastructure and stormwater management, some elements of the designs and outcomes include factors that are difficult to comprehend verbally, such as the biodiversity or design style of a setting. Such values might be better represented visually, as illustrations of the appearance of an area. Some other factors, such as extent of maintenance mowing may be easy to comprehend quantitatively but hard to visualize in terms of their impact on the beauty or acceptability of a residential setting.

As part of our integrated study of barriers to the adoption of green infrastructure practices, a study instrument was developed that used an extensive library of images to represent key variables related to anticipated landscape activities and the quality of typical residential settings. The image set comprised 3-D computer modeled imagery designed to reflect attribute levels derived from a fractional factorial experimental design. This paper describes the procedures followed to develop the study design, the creation of the image sets, the application within the consumer choice experiment, and discussion of the constraints and opportunities of using calibrated imagery in such applications.

Background

In their respective fields, discrete choice experiments and calibrated digital imagery have each proven to be powerful techniques, enabling a precision of prediction and estimation of human values hitherto impossible. Louviere, Anderson and colleagues have described the former and demonstrated the benefits of the approach (Louviere, 1988; Louviere and Woodworth, 1983). This paper focuses on calibrated digital images.

The idea of using images calibrated to known resource attributes to derive evaluations of the landscape is not a new one. Malm et al. (1981) used image processing techniques to develop images of pollution plumes in the Grand Canyon, based on the output of numerical models of atmospheric dispersion. Those images were used to derive human values for the impacts predicted on scenic resources in the Canyon. That study established the effectiveness of computer techniques but used computing resources beyond the means of typical natural resource agencies. Orland (1988) described the availability of micro-computer-based tools capable of the same range of image manipulations, but at much lower cost. The basic technique in use was to digitize photographic images and then use either editing or image processing software to make changes to the base image to represent anticipated changes. The use of those tools has evolved and become more sophisticated and has been used in a variety of settings for eliciting human responses to changes in the landscape (Orland 1993, 2005, 2015).

Key proponents of choice modeling and calibrated image methodologies were brought together at a 1991 workshop hosted at Quetico National Park by the Ontario Ministry of Natural Resources (OMNR). The powerful benefits which each would bring to the other were realized in several collaborations. The first was a study for the US Forest Service of the extent of contribution of scenery to the value of ski resort experiences in the Wasatch Mountains outside Salt Lake City (Daniel and Orland, 1992). The second studied the impacts of forest logging operations on the remote fishing experience in Northern Ontario (Orland, Daniel and Haider, 1993). A subsequent study, of boater preferences for beaches in the Grand Canyon (Stewart et al., 2003) has demonstrated the value of the approach for examining a wide range of human-landscape interactions.

At the core of each of these studies is the belief that some characteristics of the environment are critical to perceptions of that environment yet hard to express verbally.  In the case of the resort and recreation experiences above, it is hard to convey the contribution that scenery makes to the experience in words or numbers, but visually the contribution of snow-capped mountains, expansive forest or pristine canyon beaches is immediately apparent.  Other characteristics such as biodiversity, species mix or fuel loadings may have metrics that are readily understood but whose implications are not always clear until visualized.

Experimental Design Issues

 Project 4, Non-hydrological benefits and citizen preference for green stormwater management strategies, elicits the values that respondents have for a range of green infrastructure attributes:  extent of mowed turf, biodiversity, presence of water, formal vs naturalistic design, and cost to the homeowner.  The survey being developed asks respondents to project themselves into the situations of homeowners faced with responding to the need to contribute to improved water quality in the watershed encompassing their neighborhood.

The study design uses a paired-comparison approach whereby each attribute level is balanced in a partial factorial design. Table 1 indicates the association of attribute levels used in each comparison set.

Set	Design	BioDiversity	Water	Mowed	Cost
1	Med	High	Always	30%	High
	Low	Med	Intermediate	0%	Low
2	Med	Low	Intermediate	0%	Low
	Low	Med	No Water	70%	High
3	Med	Med	Intermediate	30%	Low
	Low	Low	Always	0%	High
4	High	Med	Always	30%	Low
	High	Low	No Water	100%	High
5	Low	High	No Water	0%	Low
	High	Low	Intermediate	100%	High
6	Med	Med	No Water	0%	High
	Low	Low	Intermediate	70%	Low
7	High	Low	No Water	30%	High
	Med	High	Always	70%	Low
8	High	High	Intermediate	30%	High
	Med	Low	No Water	70%	Low
9	High	Low	No Water	30%	Low
	Low	High	Intermediate	0%	High
10	Low	High	No Water	30%	Low
	High	Med	Always	70%	High
11	Low	Med	Always	30%	High
	High	High	No Water	70%	Low
12	High	Low	Always	100%	Low
	Med	Low	Intermediate	70%	High

Table 1:  Study design

Verbal or numeric survey descriptions of environmental features can be generated automatically, drawing verbal phrases from a look-up-table to fill the requirements of a fractional factorial experimental design used to develop the choice pairs for the survey instrument (Anderson et al., 1994; Haider, 2002). This process works well with words which serve then as abstractions of the kinds of conditions represented by the independent variables in the study. For instance, the words “100 yard buffer strip” stand for any combination of conditions that can buffer place A from place B by about one hundred yards. The precise configuration or components need not be specified for the reader to have a mental image of what is intended.

This situation is quite different when using pictures that immediately make the mental image concrete. The same 100-yard buffer must be shown as vegetation, or not; as one species, many, or a mix; as a particular density or texture; on a realistic surface; and in the context of a surrounding matrix of residential or commercial development… or forest. The consequence is that, while a verbal phrase can be used over as a surrogate for a general concept of landscape attributes, a picture implies a specific location and thus cannot stand as a surrogate for multiple situations.

Creating the Visual Instrument

A previous study (Daniel and Orland, 1992) used assemblies of photographic images in a format closely analogous to a sentence of words. A study of ski resort values used images of the context scenery and visitor facilities to represent typical conditions of hypothetical resorts. Visitor facilities were represented by collages of three images, of inside, outside, and people photographs. In that case the idea of the sentence was followed quite closely. Images were assembled into pages so that the whole page constituted a ski resort image. Individual images were pulled from a library to insert in a video presentation format in different combinations to suit the requirements of the design specification. However, the image component was used to represent only two independent resort attributes — scenic context and resort “lifestyle”. At the time of that study a significant limitation was the need to rely on photographs to represent different levels of attributes in the study design.  As a result, there was no inherent metric for “scenic” or “lifestyle”. To overcome this limitation, images were scaled to verify their validity in representing the two variables.

The recent evolution in 3-dimensional computer modeling accompanied by advanced rendering algorithms, however, presented some opportunities to overcome those challenges. In this case a greater number (four) of attributes were represented visually but more significant was the interaction of those attributes in the visual display.  Using 3-dimensional modeling software it would be possible to take a single residential development setting and systematically create all possible combinations of the attributes required by the study experimental design.

There were several types of software potentially useful to this project including:  Visual Nature Studio, Maya, Revit, LandFX, CommunityVIZ, and Vue.  Discussions with visualization practitioners suggested that a combination of CommunityVIZ and Visual Nature Studio would provide the ability to produce both the photorealistic images, as well as timely production of data-driven alternate scenarios.  As a result, CommunityVIZ and Visual Nature Studio were chosen for the project.

Although it would be possible to create the entire visualization project using a hypothetical location, the project used an existing neighborhood, Butterfly Acres in Lititz, Pennsylvania in order to ground the project. Raw spatial data for that site was combined with LiDAR data and other public GIS data available from pasda.psu.edu, to create a dataset to test and learn the new software.  The result of this process was a Visual Nature Studio image which replicated a photograph taken at the selected study site (See Fig. 1)

Figure 1: Visual Nature Studio representation of study site

The study design dictated that four attributes would be represented visually – design, biodiversity, water and mowing.  Table 2 describes the attribute levels for design, biodiversity and water.

Attribute	Low	Med	High
Manmade-naturalistic design (low = Natural)	– Individual plants planted irregularly – plant shape and height varied by 25% – Plant species mixed “naturally” throughout	– Individual plants planted irregularly – plant shape and height varied by 10% – Plants grouped in obviously manmade forms	– Individual plants on an obvious grid or in rows. -Plant shape and height consistent – Plants grouped in obviously manmade forms
Biodiversity (low = Monoculture)	Richness: 1 species each,            grasses, shrubs and trees Evenness: Even within trees, even within shrubs, even within grasses	Richness: 3-5 species grasses, shrubs and trees Evenness: Even within trees, even within shrubs, even within grasses	Richness: 10 – 20 species grasses, shrubs and trees Evenness: Even within trees, even within shrubs, even within grasses
Water	Water only visible during storm event.	Water visible for 24-48 hours.	Water always visible.

Table 2:  Attribute levels in study design

The use of visualization to convey key aspects of green infrastructure—design, biodiversity, water presence and extent of mowing—is driven by the need to communicate to respondents what is meant by each of those terms.  Each unique contribution of attributes must, then, be represented in a visualization of the neighborhood “place” where respondents are required to make their choices of preferred green infrastructure treatment.

Figure 2 illustrates three of twenty-four attribute combinations for a single “scene” in a Low Density residential neighborhood setting.

Figure 2:  Three landscape treatments at a single neighborhood location

As a precursor to creating the visualizations it was necessary to establish an appropriate scale for each attribute—what constitutes high, medium and low for a given attribute and how is that represented visually.  Figures 3 illustrates an array of two attributes: water presence on the vertical axis and biodiversity on the horizontal. Figure 4 illustrates biodiversity left to right, design formality top to bottom.

Figure 3:  Attribute combinations, biodiversity left to right, water presence top to bottom

Figure 4:  Attribute combinations, biodiversity left to right, design formality top to bottom

In addition to direct project needs there is significant educational potential in using these images to illustrate for non-expert audiences what is to expect as a physical outcome when a professional or scientist says “this will enhance the biodiversity of the location.”  All project images will eventually be available via an on-line selection tool enabling exploration of landscape attributes one at a time.  That interface is in early stages of development, on-hold until the requirements of the survey design have been completed and the survey deployed.

Survey respondents are asked to project themselves into a situation where they are making personal choices regarding their preferred surroundings.  In order to facilitate visualization of the green infrastructure implications for a neighborhood like their own, the survey allows them to pick one of three neighborhood types that most closely resembles where they live (Figure 5).

Figure 5.  Neighborhood density – high, medium, low

To that end, three complete sets of imagery have been developed, showing all study design attribute levels at each of the three neighborhood densities. Once respondents have chosen the neighborhood design they most closely associate with the survey proceeds, drawing images showing only that density.

Using a single image to represent a neighborhood offers its own challenges.  To best represent everyday experience the visualization would depict ground-level views from within the neighborhood.  However, a single ground-level viewpoint is unable to convey an overall impression of the area and disproportionately represents foreground aspects of the landscape setting over more distant ones, or those obscured by intervening buildings and vegetation.  To address this concern, each neighborhood “place” is represented by three images: one aerial oblique to provide an impression of the overall context, one ground level view looking along the green space in the development and one looking between the homes into the heart of the green space.  The combination is repeated for each study set as shown in Figure 6.

Low density

Medium density

High density

Figure 6:  Three images represent a single “place”

The development of the full array of images required by the study design has been a challenge in achieving the appropriate combination of attributes and developing equivalent representations for each of the three neighborhood types, for the three-view combinations representing each landscape place.  Figure 7 illustrates the assembly of a “key” view plus two “supporting” views to more completely represent the context.

Figure 7:  Image triplet compiled for survey

While design, biodiversity, water presence and extent of mowing are communicated in terms of their everyday impacts via the visualizations, the images may not necessarily be understood by respondents to represent the study variables.  Concept such as “biodiversity” and “standing water” may either be unfamiliar in terms of their visual implications, or may be interpreted differently by different respondents.  For this reason we have chosen to be somewhat redundant in repeating the attribute descriptions verbally in the on-line survey format – “Diversity in Species – High”, Standing Water – Always”, “Percentage of Green Space Mowed – 30%” (Figure 8).

Figure 8:  Survey attribute levels as both visual and verbal representations

Survey development

Image development work completed in year four contributed to the survey used in Project 4 and the survey was approved by the Penn State Institutional Review Board.  To pretest the instrument respondents were recruited from community groups.  This method resulted in unacceptably low response rates. After consulting Penn State’s Survey Research Center, an external vendor, Knowledge Networks, was chosen to implement the survey using their respondent panel. Survey implementation is scheduled in early 2016.  The full range of images specified by the study design were completed, for each of three different housing density types.   The visual attributes design, biodiversity, water presence and extent of mowing have been communicated in terms of their everyday impacts via visualizations, the images may not necessarily be understood by respondents to represent the study variables.  Biodiversity, standing water, percentage of green space mowed and annual homeowner association fee are conveyed as verbal questions. The final survey format (Figure 9) was pilot-tested and evaluated and deployed in the Project 4 survey.

 

Figure 9:  Survey display layout

References

Anderson, D.A., Wiliams, M., Haider, W. and Louviere, J.J., 1994. Efficient Experimental Designs for the Study of Remote Tourists’ Destination Choices. Decision Support 2001, Toronto: American Society for Photogrammetry and Remote Sensing.

Baker, F.A., and Rabin, D., 1988. Using computer graphics to assess the visual impact of limb rust in Ponderosa pine. Utah Science Utah Agricultural Experiment Station, Logan: UT. 49(4):98-102

Daniel, T.C., 1994. Identifying and Scaling Factors Affecting Tourists’ Experiences. Decision Support 2001, Toronto: American Society for Photogrammetry and Remote Sensing.

Daniel, T.C. and Orland, B., 1992. Principal Investigators, Visual Simulations for Recreation Choice Modeling. US Forest Service Co-op Agreement.

Haider, W., 1994. Assessing Timber and Remote Tourism Trade-offs in Northen Ontario. Workshop description in Proceedings, Decision Support 2001, Toronto: American Society for Photogrammetry and Remote Sensing.

Haider, W. 2002. Stated Preference & Choice Models – A Versatile Alternative to Traditional Recreation Research. Monitoring and Management of Visitor Flows in Recreational and Protected Areas. Conference Proceedings ed by A. Arnberger, C. Brandenburg, A. Muhar 2002, pages 115-121

Louviere, J.J. 1988. Analyzing Decision making — Metric Conjoint Analysis. Sage Publications Inc. Newbury Park, CA. 95pp.

Louviere, J.J. and G. Woodworth. 1983. Design and analysis of simulated consumer choice or allocation experiments: an approach based on aggregate data. Journal of Marketing Research, 20, pp350-367.

Malm, W.,Kelley, K.,Molenar, J., & Daniel, T.C., 1981. Human Perception of Visual Air Quality (Uniform Haze). Atmospheric Environment, 15(10/11), 1875-1890.

Orland, B., 1988. Video-imaging: A Powerful Tool for Visualization and Analysis. Landscape Architecture, 78, (4), 78-88.

Orland, B., 1993. Synthetic landscapes: a review of video-imaging applicatons in environmental perception research, planning, and designs. In D. Stokols & R. Marans (Eds.), Environmental Simulation: Research and Policy Issues. Plenum: New York. 213-252.

Orland B., T.C. Daniel, and W. Haider. 1995. “Calibrated Images: Landscape visualization to meet rigorous experimantal design considerations” in Proceedings Decision Support – 2001: Volume 2 eds. J.M. Power, M.

Orland, B.  2005. “Calibrating” images to more accurately represent future landscape conditions in forestry.  Chapter 6 In: I. Bishop and E. Lange, (eds.) Visualization in Landscape and Environmental Planning. London: Taylor and Francis.  104-111.

Orland, B. 2015. Persuasive new worlds: Virtual technologies and community decision-making, Landscape and Urban Planning, (142) 132-135.  http://dx.doi.org/10.1016/j.landurbplan.2015.08.010.

Strome, and T.C. Daniel. Bethesda, Maryland: American Society for Photogrammetry and Remote Sensing.