6.4 Typical survey page (image reproduced in the colour plate section)
The resultant images were incorporated into an on-line survey instrument (Figure 6.4) that was used to solicit public input in more than 200 face-to-face interviews. Image sets were shown as animations stepping viewers through five time-steps as shown. Each sequence of images was preceded by images of the real setting so that the validity of the base representation could be established, then viewers were asked to focus on the change in forest conditions as represented by the changing images. The viewer responses were remarkable for their degree of acceptance for these admittedly abstract images. The study design asked people to compare the acceptability of different forest management policies shown by the image sequences. Their responses were consistent with our expectations of
Applications in the forest landscape 99
the likely impacts of thinning and other management practices, suggesting that these calibrated images do meet the need for visualizations of this middle-ground of forest representations.
Conclusion
This section describes evolution in the representation of future landscapes and the unique difficulties represented by the middle-ground. Even given the immense effort described above to achieve a replicable but changing landscape over time, it is clear that much still needs to be done.
Two major issues arise that perhaps represent core questions for those involved in visualization. First is the necessity at the heart of any visualization to identify a spatial location for each object to be shown. While the complexity and scale of landscapes in this middle range, and the need to anticipate the locations of newly emerged trees, mitigate against achieving complete spatial data, every effort should be made to ensure that the landscape ‘behaves’ plausibly. Comparison between time-steps appears to require that the same viewpoint be represented at each step, so that growing trees remain in the same place through time and new ones clump or scatter as in nature.
The second issue is to consider whether the object of the visualization might be masked, or should be masked, by other scene artifacts. In the instance illustrated in Figures 6.2 and 6.3, base data was available for those shrubs and forbs present in 2001 but growth and development data for ground-cover and shrub species was not. Although the resulting images of those components were thus not accurate to the anticipated conditions, if such detailed information had been available the improved validity of that aspect of the visualization might well mask changes in the major vegetative component—
the trees. Groundlevel conditions at the site represented in those images in 2003 included five to six foot aspen seedlings—locations photographed two years previously had rapidly become visually impenetrable tangles of foreground foliage (Figure 6.5).
The concept of calibration is central to ensuring that images are useful to decision making or other judgments, but the experiences reported here indicate that there are critical questions to be addressed before use. In each example above, the trees rest in a matrix of other elements—geological, man-made, and natural—that are each changing at the same time as the object of the visualization. In most cases information will be incomplete and models will be poorly calibrated. In any case, our ability to predict the ephemeral impacts of weather, disease or fire is poor enough to render impossible any notion of accuracy in visualization of future events.
Nevertheless, in each of the cases described here the central issue was to represent the impact of change in just one element of a scene– the addition or subtraction of a street tree; the impact of insect damage on forest canopy; and the impact of tree growth and change in a forest landscape. Calibration exists to fit visual images to systematically altered representations of the future and our abilities to do so have improved markedly and continue to improve. If it is possible to limit the role of visualization to just that, then our major concern is to address the validity of the visual and spatial representation of the changing element of the landscape, and in that area good progress is being made.
6.5 Site conditions photographed in 2001 and2003
STUDYING THE ACC EPTABILITY OF FOREST MANAGEMENT PRACTICES USING VISUAL SIMULATION OF FOREST
REGROWTH
Ian D.Bishop, Rebecca Ford, Daniel Loiterton and Kathryn Williams
Introduction
Forest harvesting is a controversial topic in Australia. The Australian public has a special affection for the native Eucalpytus forest and the animals it supports. The practice of clearfelling, followed by high temperature burning and aerial sowing of eucalypt seeds (clearfell, burn and sow—CBS) is the most commonly used approach for wood production in tall wet eucalypt forest in south-east Australia (Florence 1996). This creates a short-term scene of apparent devastation within a harvest block—locally called a coupe—which is seen by some members of the public as a ruthless approach, insensitive to the values of sustainability, wildlife habitat and aesthetics. The public and private forest management agencies, on the other hand, see the practice as both the most efficient—in terms of cost of timber removal—and also among the safest and most environmentally appropriate practices.
Forestry Tasmania has established a silvicultural systems trial (Hickey et al. 2001) within the Warra Long Term Ecological Research (LTER) site to undertake scientific research into the consequences of alternative forest management practices. The site is an area of wet eucalypt forest with a very dense understorey. The dominant species, and main timber tree, is the stringybark (Eucalyptus obliqua). The major understorey species are dogwood (Pomaderris apetala), myrtle (Nothofagus cunninghamii) and silver wattle (Acacia dealbata). Several other species occur in smaller numbers but are economically important: e.g. leatherwood (Eucryphia lucida) for bee keepers, celery top pine (Phyllocladus aspleniifolius) for boat builders and joiners.
From 1998 to 2003, sections of forest have been harvested to proscribed patterns and the distribution of seed fall, germination and regrowth monitored. Among the harvest and regeneration treatments being rigorously assessed are the following.
Applications in the forest landscape 101
• Clearfell, burn and sow: the area cleared is typically about 60 ha, the burn is hot.
• Dispersed retention: a percentage of individual eucalypt trees are retained for a full rotation for fauna habitat and natural seed supply. The slash is partially-cleared using a low-intensity burn.
• Aggregated retention: islands of undisturbed forest are retained for a full rotation for habitat, seed supply (all species) and aesthetics. A low-intensity burn is used. In order to present the public with an understanding of the full harvest and management sequence—as distinct from the emotive view of a burnt scar—we have created animation sequences covering 200 years of forest life. This is typically two harvest cycles.
Preliminary simulation and assessment
In order to determine which elements of the forest environment are the most important to simulate, and which features of these elements require accurate portrayal, we created an initial set of still images. These were done quickly using simplified versions of the procedures described below. The forest elements included in these initial simulations were based on site observation of and advice from professional foresters. These preliminary simulations were shown to 18 people recruited from organizations with a range of interests in forest management (forest industry, minor species timber users, conservation groups and people from organizations with no formal position on forest management).
Before visiting field sites, participants rated the acceptability of the simulated forest management systems. At each of the corresponding field sites, participants completed a brief questionnaire. They rated the acceptability of the harvesting system used at the site and listed their reasons. Participants then rated the accuracy of the simulation and described any differences between simulation and site that were relevant to their acceptability judgment. The trial simulations did quite poorly on the level of accuracy or realism and there was relatively low correlation between the acceptability judgments made from the simulations and the field sites. Descriptions of differences between the simulations and the field sites highlighted areas for improvement. From this we identified a number of necessary improvements in the simulation process:
• careful rechecking of density and height estimates used;
• less uniformity in the regrowth pattern;
• use of textures better representing the species under regrowth conditions;
• more understorey plants;
• more stumps and logs and other harvest waste—blackened as appropriate;
• bare areas along snig (tree removal) tracks.
Realistic simulation process
Based on our initial experiences and the on-site assessments, we began a new round of simulation development. A key element in the process was the selection of the rendering software. Several packages, which could potentially assist with the model development and rendering process, were available. These included SDstudio Max, Bryce and the
public domain programs Forester and VTP. None of these, however, gave us the level of control we needed to generate a time series of images in which:
• trees were randomly located but remained in the same location at the next time step;
• several different textures were available for each species and were randomly allocated to individuals of that species;
• the textures could be changed at a specific age corresponding to a change in the growth habit of the species;
• the crossed-planes upon which the textures were pasted were randomly rotated—but retained that rotation as they grew;
• within each age class a level of height variation was randomly distributed;
• we could reduce the tree density further from the viewpoint to contain rendering times.
To achieve complete control over the rendered model we chose to use the public domain renderer POV-Ray. This is wholly controlled by text files which describe the scene in the POV-Ray modelling language. We then developed a Visual Basic program to generate the text files.
Determination of forest species mix
Existing data from previous vegetation surveys were used to generate tables of likely species density at selected ages (1, 3 10, 25 and 89 years). Growth formulae have also been developed for the most common species and used to compute the heights of the plants at the key ages (Table 6.1).