3D Visualization of Forest Cover ChangeHuman Impacts in Northeastern Kansas

Matt D. Dunbar
L. Monika Moskal
Mark E. Jakubauskas

While many studies have used visualizations to simulate natural landscapes, little work has been done to investigate their potential for illustrating land cover change using temporal data acquired from the real world. Using a combined approach of remote sensing, geographic information systems (GIS), and visualization techniques, the project described in this article demonstrates the potential of realistic computer visualizations for depicting the dynamic nature of forested environments. High-resolution digital imagery and aerial photography were classified using object-oriented methods. The re­sulting classifications, along with pre-existing land cover datasets, were used to drive the placement of vegetation in the visualized landscape, providing a more accurate representation of reality at various points in time. A special visualization software package was used to construct a variety of visualizations showing natural and human-driven forest cover change in two different ecological settings. Visualizations from Yellowstone National Park focused on the dramatic effects of the 1988 fires upon the lodgepole pine forest. In Kansas, visualization techniques were used to explore the continuous human-land in­ter­actions between 1941 and 2002 impacting the eastern deciduous forest and tallgrass prairie ecotone in the Midland, Kansas United States Geological Sur­vey (USGS) quadrangle. The resulting stills and animations demonstrate the flexibility and effectiveness of visualizations for representing patterns that change in both space and time, such as forest cover. Geovisualizations allow us­ers such as re­searchers, resource managers, and the pub­­lic to communicate findings and ex­plore new hypotheses in a clear, concise, and visually intuitive manner.

Visualization
Graphics, specifically data visualizations, are usually the simplest and most powerful means for communicating results. This important concept has been expanded to include the use of visualizations for representing geographical data, or geovisualizations. Geovisualizations can differ in their complexity and quality, from photorealistic products, to block structures with textures painted on them, to simple imagery draped over an elevation model. Most of our work at the University of Kansas deals with photorealistic visualizations that include lighting and atmospheric effects, de­tail­ed vegetation modeling, and complex animation ability. These realistic visualizations of natural landscapes, often called virtual worlds or virtual environments, provide a delivery tool for the results of environmental change studies and management plans, especially concerning forested environments. Geovisualizations can also be used to form hypotheses and explore data more effectively than traditional graphic representations.
Until recently, forest visualization efforts have focused primarily upon illustrating static concepts or possible outcomes of management actions. Visualizations can be enhanced by animating the static visualizations through time, benefiting from the fact that, perceptually, human vision is “hardwired” to detect motion. In addition to animation through time, visualizations can also use animation to move the viewer through a three-dimensional landscape. The combination of visualization and animation can provide a more effective representation of data describing changing land cover.

Software
Visualization and animation tools are still quite rudimentary in commercially available remote sensing and GIS software packages. By combining the functions of numerous pieces of software it is possible to demonstrate what a single geospatial package may someday be capable of producing. This project relied on five different types of software: remote sensing image analysis, GIS, image editing, video editing, and landscape visualization. After an ex­haust­ive search, 3D Nature’s Visual Nature Studio (VNS) was chosen as the most appropriate photo-realistic visualization software package for exploring forest rendering techniques at a variety of scales. Along with its lifelike rendering ability, VNS was selected for a number of other specific qualities:

-- Integration with georeferenced GIS datasets
-- Flexibility of land cover type development using “ecosystems” and “ecotypes”
-- Use of raster or vector formats to drive rendered vegetation components
-- Both motion and time-series animation ability.

Project Design
Before starting any visualization pro­ject it is important to consider the available data, the size of the study area, and the intended use of the resulting products. For Yellowstone, the visualizations attempted to accurately recreate the landscape and vegetation communities, as determined from multi-spectral imagery, at scales ranging from the en­tire park landscape down to individual stands of trees. The Kansas project used a more temporally rich but spectrally poor data set, in the form of six sets of panchromatic airphotos taken from 1941 to 2002. Due to the difference in data types, the Kansas visualizations were constructed with the goal of capturing the process of forest cover change with­in a much smaller study area by focusing more on animation techniques than ac­curate tree type representations. Tak­en individually, each static or animated vis­ual­ization product provides a unique method for displaying diverse aspects of forest cover change.

Landscape Level Visualizations
Visualizations covering the largest possible area, the landscape level, are used primarily to provide an overview of the study region and show the general spatial arrangement of landscape elements. The landscape level visualizations for Yellowstone used VNS to demonstrate the image-based visualization approach of draping imagery data over a digital terrain model. Six re­mote­ly sensed image datasets at a variety of spatial resolutions, from 30m Landsat Thematic Mapper imagery to sub-meter digital camera imagery, were draped over an 80m digital elevation model (DEM) using texture-mapping techniques. Finally, GIS vector layers, such as the Yellowstone National Park boundary and text labels, were inserted to aid in interpretability. Still renders were gen­erated from this project and an animation was created using a pre-defined camera flight path in the form of a vector GIS layer. The flight path was chosen to highlight each of the imagery types available for the park, demonstrating their coverage extent, spatial resolution, and spectral characteristics. The geo­visualizations, stills and animations, help to orient those unfamiliar with the study area, illustrate properties of re­motely sensed data, and pro­vide a general sense of land cover structure.

Stand Level Visualizations
Projects designed to display the structure of a functional unit of land cover are termed stand level visualizations. The next stage of visualization used in the Yellowstone effort was actually a mixed-scale approach, us­ing vegetation objects to visualize forest structure be­tween a landscape and stand level of detail. The visualizations highlight land­scape characteristics such as the spatial ar­rangement of stand types, stand structure, and land cover change. The focus of the landscape/stand level visualizations was on three successional stages of the lodgepole pine forest: post-fire seed­ling/regeneration; young successional forest, with a dense, even-age can­opy; and the mature forest stage. VNS represents trees, snags, deadfall, ground cov­er, and other vegetation types using image objects taken from the real world. Objects are either placed in­dividually on the landscape or grouped to­gether in associations call­ed “ecotypes.” Each ecotype consists of groups of image objects, each with their own height range and density specifications. At the land­scape/stand lev­el, where only general land cov­er classes are known, GIS polygon coverages were used to drive the placement of ecotypes upon the landscape.
Landscape metrics and spatial analysis are becoming widely used in many aspects of ecological assessment and resource management. By quantifying the landscape before and after the 1988 fire using various landscape metrics, comparisons can be made between the temporal representations of the landscape. Text-based summary tables can quantify differences in metrics representing change in the forested landscape, but indicating the specific location or nature of the change is often difficult. Visualizations support landscape metrics analysis by making information more accessible to forest managers, ecologists and the public. Figure 3 provides a comparison between the traditional method for representing landcover change and more realistic geovisualizations. While the stills show a snapshot of a stand level view, animations created for this project further acquaint the viewer with the landscape structure by means of motion simulating flight and ground-based movements.

Plot Visualizations
The most detailed stage of visualization is the plot level, which highlights unique structural characteristics of a specific forest plot. More precise changes in forest composition and structure are presented at this scale. The plot level visualization in this study demonstrates the ability of object-oriented feature extraction to describe the position and size of individual trees and other landscape components. In contrast to the use of GIS polygons at the stand level, exact locations for trees, snags, and deadfall were known at the plot level. For these visualizations, GIS point coverages were used to place individual image objects, such as trees and standing dead snags of various heights upon the terrain.
Two 200m2 study sites were selected for the plot level visualizations. The first site was located in a regenerating forest while the second was a mature forest. While a GIS provides the means to catalog and display the spatial position of point features representing trees in the plot, only someone intimately familiar with both the forest ecology and GIS symbology can comprehend what is physically on the ground (Figure 4). In contrast, the visualized representation of these plots clearly communicates the forest structure, including species/object type, location and size, in a simple yet powerful manner (Figure 5).
In an effort to improve upon the plot and stand level visualizations, the park was revisited in 2002 to collect image objects specific to Yellowstone and photographs documenting various land cover classes. These datasets allowed the refinement of the ecotypes developed for the various land cover types created to more exactly match the living and non-living vegetation objects specific to Yellowstone (Figure 6).

Non-Animated Visualizations
The first visualizations developed from the Kansas forest dataset were static images displaying the extent of forest cover at specific dates corresponding to the aerial imagery. An object-oriented classification approach, provided by Definiens’ eCognition, was used to classify all six dates of air photos into forest/tree cover, non-forest, roads/building structure, and water classes for each date. Appropriate ground cover and vegetation image objects were combined to mimic a generic forest and grassland vegetation cover for this region as well as water and roads. To complete the models, the classified forest cover data sets for all six years were brought into VNS, where the appropriate vegetation classes were linked to the VNS ecotypes. After selecting an appropriate view angle, still visualizations of the forest cover through time were rendered by cycling through the classification dates (Figure 7).
The non-animated visualization of multi-temporal land cover data results in individual stills representing a single date of forest cover derived from air photos. Multiple dates of still renderings can be viewed side by side allowing comparisons of forest structure through time. The stills are based on the idea that forest cover is more easily understood when it is displayed as a collection of 3D tree objects. The resulting products are similar in appearance and provide the same benefits as the landscape/stand scale visualizations from Yellowstone. This style of geovisualization provides a more familiar and interpretable way of comparing multiple dates of land cover than a traditional GIS Polygon view (Figure 8).

Animated Visualizations
Visualizations using animation provide a greater sense of the process of land cover change recorded by multi-temporal imagery. By representing real world time in years with animation time in seconds, this display method represents change in the same way that we are familiar with viewing it in the real world. The animated visualizations for Kansas were created using a different approach than the static renderings. This was done because the resulting product needs to describe the change between years, rather than the before and after snapshots produced by the still frames. Within a GIS, classification comparisons were created between each consecutive date pair of forest cover data. This process reduced the six dates classified imagery to five change comparison images. In VNS, instead of using static ecotypes, as with the still visualizations, the changing land cover classes were replaced with animated ecotypes to match the classification comparison results. Once the classified comparison data and animated ecotypes were brought together in VNS, animations were created by selecting a camera location and defining the time interval to be rendered.
Animated visualization for the Kansas study area used the same classified data sets as the static visualizations, but recreated the changing forest cover as a dynamic process rather than individual moments in time. By animating the appearance or removal of trees in much the same fashion that forest cover would change in the real world, viewers of these animations can experience the events of the landscape history of this area. Animated visualizations can be used as a stand-alone data exploration tool for drawing conclusions regarding the nature of the changing forest cover. Used in connection with quantitative patch structure measurements, the animations provide a visual reference in a qualitative format. To demonstrate this concept, the final Kansas animation was amended to include graphs of several landscape metrics (area, number of patches, average patch size, and total edge), which are displayed in an animated fashion that progresses in time with the visualization (Figure 9).

Conclusions
The examples demonstrate the use of photorealistic 3D computer visualizations for illustrating various types of forest cover change in two distinctly different environments at several spatial scales. Any remote sensing based re­search resulting in forest compositional or structural classification, forest modeling, or forest management plans could benefit from the ability to more clearly relay results to intended audiences. It is also reasonable to assume that the visualization methods demonstrated here are not limited to forestry research and could be effectively ap­plied to any geo­spatial data set.

Acknowledgements
This project was conducted at the Kansas Applied Remote Sensing (KARS) Program (Edward A. Martinko, Director). The Yellowstone National Park research described in this paper was funded by the National Aeronautics and Space Administration (NASA) Earth Science Enterprise Food and Fiber Applications of Remote Sensing (FFARS). The work in Northeastern Kansas was funded by the NASA Carbon Sequestration Program—Forest Cover Change Project. The Principal Investigator for both projects is Dr. Mark Jakubauskas and the visualization effort has been supported by Dr. Jerome Dobson, both of whom are at the KARS Program and the University of Kansas Geo­graphy Department. The authors would like to thank David Guinotte, Researcher at the KARS Program, for the long hours he has spent classifying the Kansas data set.

About the Authors
Matt Dunbar is a Graduate Re­search Assistant at the Kansas Applied Remote Sensing (KARS) Program in the Geography Department at the University of Kansas.
L. Monika Moskal is an Assistant Professor in the Geography, Geology and Planning Department at Southwest Missouri State University (formerly with the KARS Program, University of Kansas).
Mark Jakubauskas is an Assistant Research Professor in the KARS Program and is a Courtesy Assistant Professor in the Department of Geo­graphy at the Uni­­versity of Kansas.

Back