|
Modeling Scenic Landscape Resources
An Integrated Approach at Articulating Color/Change Relationship in a High Elevation
Forest in Southern Utah
By Gary R. Clay, Ph.D. Scenic encounters don't only profit the individual. They additionally represent a multi-million dollar economic gain with regard to many states' recreational and tourist potential. Unfortunately, scenic quality is difficult to articulate as it relates to both the physical landscape and the responses of humans interacting with those settings. Assessing scenic resources, therefore, dictates an approach that first measures the variations in landscape composition, and then applies this knowledge to predict the public's reactions to those conditions. For land managers, the problem is essentially two-fold: (1) how can scenic amenities be measured and then geographically located? and, (2) how can the resultant data be integrated with other environmental variables into a comprehensive management effort?
Historically, data acquisition on public lands has emphasized physical variables such as plant species and water quality. Typically, ground reconnaissance has been the method of choice. More mechanized approaches (aerial photo-interpretation, satellite image processing) have recently been initiated to automate this process. In these efforts, field verification using global positioning (GPS) provides a ground truthing component, and insures that remotely sensed data has some relationship to more localized, ground conditions. The combined ground-aerial data can later be merged into a Geographic Information System (GIS) to catalog and spatially register the inventory.
Like their more-physical counterparts, scenic inventorying requires procedures for collecting information about the changes in their variety and extent. Recently, visual analysis has employed computer simulations to illustrate the potential effects of environmental change. These visualizations, which apply photographs as a base image, provide viewers with a means to "see" events or impacts before changes actually take place. Perceptual testing then utilizes those visualizations to assess the impact the simulated conditions might produce. While useful as a presentation tool, visualizations have traditionally been developed within an artistic rather than analytical framework. The research presented here applied a more quantitative approach, utilizing spectral analysis to document the color changes that have occurred in landscape scenes. Sampling techniques applied methods previously reserved for large scale image processing via remotely sensed imagery. The approach presented here further illustrated the potential for articulating color shifts at a smaller, ground-based orientation. MODELING SCENIC RESOURCES: A PILOT STUDY
A research program was formulated to satisfy the following: (1) the integration of GIS, digital elevation model (DEM), global positioning, and image processing technologies in the context of scenic resource management, (2) the testing of procedures for constructing a tree canopy model, using GIS contour data plus the field-estimated heights of surveyed trees, (3) the development of methods for extracting color samples of healthy and insect affected trees from portions of scanned ground photographs, (4) the testing of procedures to relate the extracted color/change data to field observed beetle activity levels, and (5) the merging of the ground photographs with the tree canopy model in specific observer-target view relationships.
A segment of a forested landscape within the Dixie National Forest in southern Utah was selected for analysis. An integrated approach was formulated to first measure and then spatially articulate changes in color relationships within portions of that forest condition. The site, a 22-acre parcel, is dominated by mature Engelmann spruce (Picea engelmannii ), is generally a north facing slope, and has an elevation range of between 9,650 feet (2941 meters) and 10,150 feet (3094 meters). At the time of data collection, the site and surrounding landscape was experiencing the effects of a spruce bark beetle (Dendroctonus rufipennis) event. The bark beetle, a natural mortality agent of mature spruce forests, produces a series of observable changes to the tree's color composition. These visible changes, and the modeling of the projected outcomes, were to become the overall focus of the research. DATA ACQUISITION
Within the study area, two related field surveys were conducted. The first measured the physical, biological and spatial characteristics of individual trees on site, while the second located a series of camera positions for a two-year photographic inventory.
Tree inventory. Approximately 5,000 trees within the study area were surveyed to compile the following data: (1) a tree identification number, (2) the distance in feet and degrees from a control point, (3) a tree species name, (4) the estimated height, and (5) the estimated elevation at ground level. Two techniques were employed to spatially locate each tree. First, control points were located and referenced to a UTM grid using a Trimble 4000SE Polycorder, generally known as a GPS unit. Next, two linear transacts of 15 points each were developed on the study area's east and west sides. Hand-held laser instrumentation was used to register these points to the GPS control points. The distance (in feet) and direction (degrees-minutes-seconds from north) from individual points was recorded. The surveyed tree data was later compiled on a micro-computer, using Quattro-pro.
Photographic inventory. A photographic inventory was conducted at the southern Utah site. The objective was to acquire photography on the same date and time for both years using identical camera/target relationships. Before the photographic activities, a series of camera positions were staked out in quadrants to the north of the study site. Two referencing techniques were employed. First, five points were located using the GPS techniques applied during the tree inventory. The remaining points were then referenced from these points using the mentioned laser equipment. Positional data would later be converted to metric equivalents and input into the study's GIS. The photographs were taken using three camera orientations: South 30_ East, South, and South 30_ West. A tripod mounted 35mm camera with a 50mm lens, a UV Haze filter, and Ecktachrome 100 slide film was employed for all photography. For future image processing, a Macbeth Color Checker chart was positioned in each photograph. DATA DEVELOPMENT
GIS development. Sections of three USGS 7.5 minute topographic maps were initially digitized using Autocad 12.0 to create a drawing file with the following layers: (1) 40 foot contours, (2) selected spot elevations, (3) roads, (4) streams, (5) break and ridge lines, and (6) selected tree edges. The layers were separated, saved as individual *.dxf files, transferred to a UNIX-based Arc/INFO environment, and converted into point and line GIS coverage (Figure 1). Applying the UTM coordinate system established during inventory, each coverage was assigned a northing and easting registration. Data from the study area's tree survey and the GPS photographic positions were additionally converted to GIS point coverage, applying the mentioned UTM coordinate system. A series of corrections to the tree inventory data reduced the initial tree list to 4473 individually attributed trees. Digital elevation model (DEM) development
The GIS coverage was used to produce a series of 3D surfaces that articulated the ground plane plus the study area's tree canopy. A modeling strategy was implemented whereby the contours were employed to create the ground surface outside the study area, and the individual at-ground tree elevations were used to generate the ground surface inside the study area. The 3D form of each tree was then delineated by extending each tree's ground elevations, using the estimated tree heights to create a composite canopy for selected trees. The peripheral ground plane was maintained so only selected trees were displayed at their extended height. A series of ground-canopy models were constructed to simulate a spreading pattern of beetle activity across the study area. An initial surface delineated a small central cluster of Engelmann spruce which was designated the starting point of a beetle event (Figures 3, 4, 5, 6). Perspective views were generated of this surface, applying the camera/target positions acquired during the photographic survey. A second surface was then generated, showing an enlarged central core of spruce. This simulated a spreading pattern of beetle activity from the central tree mass.
Photographic digitization and pre-processing. Thirty-eight slides representing several scenic camera-target relationships were sent to Kodak laboratories for digitization and archival storage on a Photo CD master disk. Upon return, the scanned images were extracted and translated to a 24-bit *.tif file format in Adobe Photoshop. A series of pre-processing operations prepared the imagery for future analysis. The images were then normalized to re-distribute the color ranges within those scenes. Normalization produced an image collection with comparable color relationships, thus ensuring that the extracted color data would be a function of the tree's color characteristics, and not due to time/date issues or photographic processing. Normalization applied the mean color values from pixel samples extracted from either portions of each image's color chart, or from objects (rocks, signs) located in each image. The sample data was input into a statistical program (SPSS) for regression analysis. After regression analysis, the images were manipulated in an image processing software (ERDAS) to apply the regression equations to normalize them to the control image. Extracting color signatures indicative of a bark beetle event
Equating color shifts to environmental events requires a mechanism to discriminate color variation, and an understanding of the conditions indicative of those color changes. To document the color variations produced by a beetle event, Forest Service personnel were asked to identify on-site the locations, extent, and estimated age of an existing beetle impact. After this documentation, photographs of these zones were sampled using ERDAS to obtain color data indicative of beetle activity in the following three stages: (1) healthy or visibly unaltered trees, (2) a 1-2 year beetle activity level, and (3) a 3-5 year beetle activity level (Figure 2). Samples were extracted from the three ranges, and mean color values were computed per sample. From this data, corrective multipliers were generated for: (1) the healthy, non-visible beetle levels and the 1-2 year beetle level, and (2) the healthy, non-visible beetle levels and the 3-5 year beetle level. These multipliers represented the statistical changes between the RGB levels of the unaffected and affected tree samples. In the visualization component, these multipliers were applied to images to simulate color variation representative of the different beetle activity levels. FINAL SCENIC VISUALIZATION
Final scenic visualization addressed three related issues: (1) the association of the three incremental color ranges to field-assessed levels of beetle activity, (2) the registration of those color ranges in forested portions of scanned images, and (3) the final simulation of incremental beetle activity. A two-stage strategy was developed, whereby the 1-2 year multiplier was first applied to a central core of spruce trees. Later, this core was re-processed to simulate a 3-5 year insect pattern, and an expanded fringe area was given a 1-2 year beetle color definition.
Spectral Association. This effort applied the RGB corrective multipliers to selected images portions (only spruce trees) to simulate to the color relationships of the different beetle levels. An initial sampling effort identified the spectral range for the only spruce population. Image processing later isolated this spruce range from other trees and peripheral site elements. The 1-2 year corrective multipliers were then applied uniformly to this range to simulate the color characteristics in the overall canopy. The second stage, dealing with the 3-5 year beetle level, proved to be problematic as both the colors and their relationships to each other could not be fully represented by applying a uniform multiplier. A different strategy was, therefore, developed whereby the 3-5 year multipliers were applied using a stepped approach to better represent the changing statistical relationship between red, green, and blue data from the 3-5 year samples. This combined set of multipliers were then applied to the overall tree canopy, using the image processing techniques applied to the 1-2 year beetle simulation.
Registration of Color/Change Relationships. Using the GIS data structure to isolate only the spruce trees, a starting point for a hypothetical beetle event was located in the center of the study area. From that location, projections were made as to extent and distribution of a typical event as it might spread across the site. A surface model depicting a central core of Engelmann spruce was used to generate a series of perspective views to correspond to the previously processed site images. Each perspective mimicked the camera lens, focal plane, and geographic extent of the corresponding image. These 3D perspectives were further divided to include one scene with a central core of Engelmann spruce, and a second scene which extended that core to the east and west. The views (generated in workstation Arc/Info) were saved and transferred to a PC platform for future use in Adobe Photoshop.
The images that had been processed using the corrective multipliers were then merged with their associated DEM perspective views (Figure 7). This created a series of composite images, with the DEM perspectives superimposed over the simulated insect damage. In this way, the three-dimensional tree data was positioned in the processed site photographs. For each view, the following composites were generated: (1) one showing the central core superimposed over the 1-2 year simulation, (2) one showing the enlarged core superimposed over the 1-2 year simulation, and (3) one showing the central core superimposed over the 3-5 year simulation. The new composites were saved for use in the final visualization phase.
Final Simulation of Incremental Beetle Damage. The composite images were transferred to a PC for the development of the final visualizations. The original unprocessed base scenes were used as the basis for this effort. The strategy was to maintain the integrity of the base images while modifying only those portions that were defined by the perspectives as being areas of simulated change. The process can be summarized as follows: (1) an irregular boundary was constructed around the desired trees for both the 1-2 year and 3-5 year image composites, (2) the boundary was copied to a clipboard, (3) the images with the two simulated levels of beetle activity were opened, and the boundary was pasted in its appropriate location, (5) using this boundary, the portions of infested trees were cut out and saved to a clipboard, (6) the clipped trees were inserted into the appropriate scenes, (7) the 3-5 year beetle image portion was positioned on top of the 1-2 year beetle simulation zone (Figure 8). SUMMARY
The final visualizations were applied to a program of perceptual testing. Generally, the simulated environments produced response patterns comparable to those obtained when using unedited slides of similar conditions. While not conclusive, the results lend some support for applying visualizations as surrogates of actual or projected scenic conditions. Testing results also provided limited support for the application of these techniques in other assessment efforts in similar forested conditions.
Regarding the methods applied in the research, several procedural advantages were documented. First, the approach provided a convenient, reproducible process for transferring data to forested portions of scanned photographs. While the technique cannot be considered a direct, two-way exchange between map and photographic data, the work illustrated a level of precision that extended beyond the earlier, more artistic efforts. The combined GPS/GIS approach also added spatial accuracy to the merging process, and showed the potential for exchanging data between source files with diverse view relationships and ground orientations. The research further illustrated the utility of applying a coordinate system as a referencing device, and the inherent potential for integrating scenic data with other peripheral material such as satellite data or existing resource inventories. Applying this approach, data connectivity can be achieved with a minimal amount of corrections or modifications. In the Utah example, the projected patterns of beetle impact were controlled through GIS referencing, which provided an accurate three-dimensional representation of the affected trees' distribution patterns. Attribute data (species type, height, vigor) were arranged in the GIS, and then displayed to illustrate spatial trends and within all or segments of the study area.
In previous research, environmental change was simulated by applying stand data that estimated the percentage compositions per-species. During image merging the emphasis was on overall terrain form and spatial patterns, not on the subtlety of individual or tree clusters. The work presented here attempted to articulate individual tree definition, which was a requirement of a near-view study. Future in-stand assessment work might require a mechanism to merge the GIS data with photographs that emphasizes a stronger individual tree relationship. New research will need to explore the accuracy of the merging procedures at different camera-target distances to document the utility of the extended methods. Future visualization activities, should also address issues of color correction and balancing to substantiate the perceptual findings derived from scenic simulations. Additionally, improvements in sampling techniques might be achieved by integrating remotely sensed data with ground-acquired information. This would facilitate data acquisition from multiple view angles by applying bidirectional reflectance factors to associate the sampled color material acquired from different orientations. About the Author:
Gary R. Clay is a professor of Landscape Architecture at Californial Polytechnic State University in San Luis Obispo, Calif. His research interests are large-scale environmental modeling, environmental perception, and scenic assessment techniques. His long term research goal is to merge the quantitative image processing techniques outlined in this paper with broader-scale environmental inventories acquired from satellite imagery, aerial videography, and other remote sources. He may be reached via e-mail: [email protected]
Back
|
