To this point discussion has focussed on methods for measuring change in the forest itself, insofar as this can be detected from coarse resolution data. An alternative strategy is to switch focus from ecological change to describing the factors that drive ecological response; in this case human activity itself (Lesslie 1997).
Shifting emphasis in this way means that generalisations can be made about the level and extent of human intervention in ecosystems and, in turn, the exposure or vulnerability of these ecosystems (Lesslie 1997). It also alleviates difficulties associated with identifying and measuring key ecosystem phenomena, facilitates an explicit treatment of scale issues, improves prospects of acquiring suitable data, and removes the necessity to ascertain whether particular ecological outcomes are attributable to human activity or natural processes.
The analysis of spatial pattern in human activity in the landscape has a strong tradition in human geography, particularly from the 1930s to the 1970s, from which several key principles have emerged. Firstly, the spatial distribution of human activity reflects an ordered adjustment to distance. Of particular relevance is the notion of the attenuation of pattern or process with distance, as expressed in Tobler's 'first law of geography' which states that everything is related to everything else, but near things are more related than distant things (Tobler 1970). Secondly, human activity is generally located to minimise the 'frictional' effects of distance (Losch 1954). A related principle is the notion of accessibility or functional centrality. Finally, human activity agglomerates in settlements.
Generalisations concerning the spatial configuration of habitat (Noss & Cooperrider 1994) are encapsulated in the broader generalisation that the integrity of habitat is usually associated with spatial isolation from human activity. This broader principle also relates to more traditional geographical perspectives on the pattern of human activity in the landscape and the attenuation of pattern and process with distance. However, a reliance on spatial pattern to explain human interaction with ecosystems does have limitations. It involves, for instance, the necessary assumption that there is a direct relationship between the spatial location of human activity and its ecological effects. This precludes any satisfactory accounting for processes (e.g. hydrologic and atmospheric) where spatially distributed effects are non-linear or highly complex (e.g. multi-scale processes).
Spatially explicit indicators of ecological integrity, which are consistent with a strategy based on measuring isolation from human activity in the landscape, should have the following characteristics:
The index should be quantitative and the methodology should have the capacity to measure variation in exposure to human activity across the landscape.
Index values should derive directly from primary attribute data in a systematic and repeatable manner, and they should reflect the scale characteristics of primary data inputs.
The methodology should be transparent and as simple as possible. Notions of ecological integrity and naturalness (or isolation from human activity) are complex and difficult to deal with in a precise way. There is, therefore, little advantage gained from pursuing complex modelling techniques when these are likely to have deficiencies of a similar order of magnitude to simpler procedures. Complex modelling also has the disadvantage that it can become difficult to understand and interpret. Any modelling procedure in this area will have contentious aspects and it is far better for these to be explicit and well understood. Complex modelling also generally requires primary attribute data of accuracy and precision that is typically not available at global or regional scales. Global consistency and comparability is improved if the modelling procedure does not make demands for sophisticated primary attribute data.
The methodology should be amenable to elaboration in a staged and systematic manner. This is essential to enable the model to benefit from new or improved attribute data and knowledge. A capacity for elaboration may also be useful in local or regional situations where its possible to take advantage of additional or enhanced local attribute data, or where particular local factors have a known and significant impact on ecological integrity.
All other factors being equal, the core of a large natural area will be less exposed to (more isolated from) human activity than the core of a small natural area. The exposure of a natural area to human activity is therefore fundamentally related to its size.
Current interest in the size of natural areas, as a measure of relative isolation from human activity, is complemented by its long history in nature conservation science. Since the 19th century, for instance, it has been recognised that there is a link between the human-induced break-up of large natural areas and the extinction of species (De Candolle 1874). The importance of size in nature conservation science has, however, been promoted most strongly by island biogeography theory (MacArthur & Wilson 1967) and the species-area relationships of Preston (1962), which hold that larger areas typically support a greater diversity of habitats and contain more species and larger populations of individual species than smaller areas. Size is today similarly recognised as important in process-functional terms. Larger natural areas have the capacity to better accommodate change in larger-scale processes (eg shifts in climate patterns) and disturbance events (eg fire) (Forman 1996).
The systematic evaluation of natural areas for conservation, based on size, is first evident in the US early in the 20th century with the emergence of interest in wilderness and roadless areas. During the 1930s the US Forest Service prepared an inventory of available wilderness areas within National Forest, rating undeveloped roadless areas on the basis of size. The US Wilderness Act of 1964 set up a framework for wilderness identification and protection in the US that is still in place today. It requires relevant Federal land-holding agencies to identify and assess roadless areas.
Interest in the identification of large natural areas that have limited exposure to modern technological society is not restricted to the US. It has been a mainstream aspect of nature conservation assessment in other 'frontier' regions of the world where there is a relatively clear distinction between the presence and impact of human activity and natural ecological patterns and processes. This is especially the case in Australia, Canada and New Zealand. Notably, the concept has also been pursued in parts of the world where this distinction is not so clear and where there is continuing indigenous habitation. This includes countries such as Finland (Kajala & Watson 1997), South Africa (Elliot 1996), and Italy (Zunino 1995). Identification and assessment methodologies typically involve criteria that specify minimum size and shape characteristics and require a 'primary' vegetation cover, no urban, agricultural or other commercial land use, minimal constructed access and no permanent settlement. Detailed specifications vary on a regional basis and from study to study.
Several assessments of this type have been conducted at the global level. The first global 'reconnaissance-level' assessment of large areas with minimal impact and proximity to modern technological society was completed by McCloskey and Spalding (1989). Using 1:1,000,000 scale Jet and Operational Navigation Charts as a database, that study identified areas of at least 400,000 ha with no mapped human structures or roads. The global distribution of these areas is (Figure 15) is heavily dominated by arctic and desert regions and includes little area in forested zones.
A second global-level evaluation of human activity (Hannah et al.,. 1994) mapped relatively undisturbed (not simply roadless) natural areas, and reduced the minimum size of mapped 'undisturbed' areas to 100,000 ha. It defined undisturbed and two classes of non-natural areas as follows:
undisturbed - a record of primary vegetation and a very low population density (<10 persons per km2 density or <1 per km2 density in arid/semi-arid and tundra communities)
partially disturbed - a record of shifting or extensive agriculture, or other record of human disturbance; and
human dominated - a record of permanent agriculture or urban settlement, or where primary/potential vegetation is removed
Data were derived from a variety of information sources. Overall, the world was found to have around half of its total surface, but only 27% of its habitable surface undisturbed by man. A number of important forest areas, especially in the Indomalayan biogeographic realm, had no undisturbed area remaining and very little partially disturbed territory.
An evaluation of the status of the world's 'frontier forests' represents a third instance of a global-scale study involving the identification of relatively large natural areas (Bryant et al.,. 1997) (Figure 16). This assessment focuses solely on forest environments, distinguishing three classes of threat:
frontier forests under low or no threat - large intact forest ecosystems that are relatively undisturbed and large enough to maintain their biodiversity
frontier forests under medium or high threat - on-going or planned human activity (eg logging, agricultural clearing, mining)
non-frontier forests - secondary forest, plantations, degraded forest, and patches of primary forest.
The threat classification in this mapping exercise was drawn largely from expert opinion. No specific size threshold was applied in identifying frontier forests, although it was required that areas be of sufficient size to maintain biodiversity and to absorb large-scale disturbances.
These three global-scale spatial studies represent simple and useful assessments of the naturalness or integrity of ecosystems. Each study primarily relies on the assumption that isolation from the impacts and influences of human activity is a reasonable indicator of ecological integrity, although some reference is made to certain biophysical conditions in the latter two instances. In each case the accuracy and precision of results is dependent upon the quality of available suitable data of global extent.
A greater concern, particularly in the latter two studies, is that the analyses are not explicitly scaled; nor are they systematic and repeatable. This means that the precision of the mapping and the accuracy of attribute class allocations are not transparent and direct expressions of the data and the analytical procedure. It also means that they can not form the basis for any kind of consistent assessment programme to monitor change through time.
Moreover, the selection of particular size thresholds to identify places that are isolated from the impacts and influences of human activity can be questioned. Exposure to human activity is not simply a matter of presence or absence, it is a matter of degree.
Figure 16 Forest Frontiers Map
Recognition of problems with the use of fixed size thresholds, qualitative terms and a lack of repeatability gave rise, in the early 1980s, to new quantitative indicator-based approaches to the identification of remote natural areas. For example, emphasis switched from the identification of wilderness areas based on qualitative thresholds to the 'continuum' concept of wilderness (Lesslie and Taylor 1985), involving the quantitative measurement of relative variation in remoteness from human activity across the landscape (Kirkpatrick & Haney 1980; Lesslie et al., 1988). This type of approach underpinned the Australian government's National Wilderness Inventory (Lesslie & Maslen 1995) and the production of similar remote and natural lands databases elsewhere (e.g. Husby 1995).
The Australian wilderness study places emphasis on measuring the extent to which points in the landscape are remote from, and undisturbed by, the influence of modern technological society. It does so by quantitatively measuring variation in remoteness and naturalness across the landscape using four indicators:
remoteness from settlement (remoteness from places of permanent habitation);
remoteness from access (remoteness from established access routes);
apparent naturalness (the degree to which the landscape is free from the presence of permanent structures associated with modern technological society); and,
biophysical naturalness (the degree to which the natural environment is free from biophysical disturbance caused by the influence of modern technological society).
The two remoteness indices and the apparent naturalness index are based on a measurement of Euclidean distance between each point in the landscape and ordered classes of settlement and infrastructure. Variations in exposure to different types of settlement and infrastructure features are accommodated through a weighting and distance-decay regime whereby more prominent feature types (such as highways or commercial centres) are accorded greater influence than less prominent types (such as vehicle tracks or residences). In this way, distance-based measures represent spatial pattern that is specific to the location of individual landscape features, allowing for the attenuation of levels of technological activity according to the distance from the feature and its prominence or likely influence. The use of distance measures in this context is not based on any empirical relationship between distance and the flow of resources associated with types of technological activity.
The biophysical naturalness index is rated according to the intensity of land use in areas where the primary vegetation structure is essentially intact. Land use, in this context, refers to activity that is not confined to defined physical structures (settlement and infrastructure) and includes a variety of forms of spatially distributed resource procurement activity, such as timber production and livestock grazing. Land use intensity is rated on the basis of historical records or of land use likelihood modelling.
The distribution of wilderness quality across Australia, or Australian Wilderness Index (AWI) was obtained by linear un-weighted combination of the four component measures as illustrated in Figure 17.
This type of approach has a number of advantages over conventional mapping methods. Of particular relevance to the question of periodic assessments and monitoring is the fact the analysis is quantitative and repeatable. Estimates of isolation or exposure to human activity produced by the analysis are a direct expression of the data and the modelling that is applied. This means that the scale of the analysis can be explicitly matched to the accuracy and precision of data inputs. GIS-based application of the model, which effectively automates the analysis, also promotes flexibility so that new primary attribute data can readily be introduced and the analysis repeated or manipulated in a variety of ways.
The Australian wilderness index measures environmental exposure or isolation from human activity in the landscape in terms of Euclidean distance from classes of settlement and infrastructure, along with a rating of land use intensity. No attempt is made to represent exposure or isolation in ways that are more exact or 'real'. The notion of exposure to, or isolation from human activity may be elaborated in two ways: 1) the refinement of the spatial, distance-based aspects, and 2) enhanced calibration of the intensity of human activity, taking account of data relating to appropriation and use of resources.
One obvious way in which the spatial, distance-based component of isolation may be enhanced is through use of more refined measures of accessibility. The accessibility of places in the landscape is not simply a function of Euclidean distance from access points and the quality of that access. Accessibility is also dependent, for example, on terrain. For instance, a forest that is located in rugged terrain at a given distance from a road is generally less accessible to timber harvesting than a forest that is located at the same distance from a road in flat or undulating terrain. The improving quality of digital elevation modelling and the increasing availability of elevation data sets means that it is now practicable to introduce terrain factors into accessibility modelling at region and global scale. A global digital elevation data set (DEM) modelled from satellite imagery is, now available at a grid resolution of approximately 1 km2 (USGS EROS Data Center 1996). The accessibility of forest environments has recently been modelled at regional scale using slope measures derived from elevation data at that scale (Lorenzini 1998).
Enhancements of this kind should, however, be treated with caution. There is no doubt that terrain-related factors have significant impact on accessibility in the real world, but terrain attributes are highly sensitive to DEM accuracy and precision, and estimations should be treated very cautiously. Typically, a DEM grid resolution in the order of 20 m is required if the tilt components of topographic variation are to be faithfully reflected in model output (Moore et al.,. 1993). It is therefore doubtful that terrain data at the scale presently available have the capacity to add meaningfully to accessibility analyses of the kind under discussion. However, the prospects for this type of analysis will certainly improve over time.
The second area of model elaboration involves calibration of the intensity of human activity. One simple way this may be developed is by factoring in statistical data on population and spatially relating this to settlement and infrastructure patterns. Indeed, patterns of population density have already been derived in this way on a global basis and show promise in this regard (Tobler et al.,. 1995).
The calibration of the intensity of human activity could potentially be further refined by combining population data with information about resource use. The intensity of demand pressure on forest ecosystems may, for example, be calibrated by inclusion of demand estimates for forest resources. A demand surface for fuelwood could be developed by using statistical data on fuelwood consumption, and distributing this spatially around urban and rural settlements on the basis of population density, forest cover, accessibility and known patterns of fuel wood consumption. A similar surface representing the pressures caused by industrial demand for timber resources could be developed and spatially distributed through settlement, infrastructure and land use components.
Discussion to this point reveals some key principles that guide the development of spatial indicators for assessing the naturalness of forests.
Indicators based on ecological response to human perturbation can only provide limited, and possibly contradictory, answers to questions concerning ecological change at the ecosystem level.
Indicators of naturalness that focus on human activity - the driver of human-induced ecosystem change - are highly promising in generic, landscape-scale applications. Such indicators rely on the assumption that the greater the exposure to human activity, the greater the probability of human interaction and intervention in ecosystems.
Spatially explicit measures of relative environmental isolation from human activity can provide a generic foundation for describing and measuring the potential for human intervention in ecosystems.
Methods for measuring relative environmental isolation should be quantitative, repeatable, transparent, appropriate to the input data, simple to interpret and amenable to elaboration.
Human activity in the landscape can be described in terms of spatial patterns of land use and land occupation. Categories of landscape modification can be represented unambiguously in terms of (1) settlement, (2) infrastructure, and (3) land use.
1. Settlement can be defined as permanent human occupation. It is the focal point for human activity in the landscape where resources are transformed and used. Settlements range in scale from a single point of permanent occupation, such as a house, through to conurbations which may extend over thousands of square kilometres.
2. Infrastructure is the built fabric around which human activity concentrates. Infrastructure provides the physical means for accessing, distributing and transforming resources. Infrastructure includes all built structures, including those associated access and settlement.
3. Land use includes any resource procurement or transformation activity that can be spatially delimited on the land surface.
The representation of human activity using primary data sets comprising settlement and infrastructure features and land use has a number of advantages.
These features represent fundamental elements through which the pattern of human activity in the landscape can be measured and described at both small and large scale, encapsulating complex resources procurement and transformation processes.
These features are unambiguous landscape phenomena, which are amenable to classification and measurement. This provides for some control over accuracy and precision in spatial (and temporal) representation. It also facilitates analyses that use spatial information technologies.
These features provide flexibility in relating human activity in the landscape to ecological effects, allowing for either aggregated or disaggregated analyses.
Using settlement, infrastructure and land use features for representing human activity does, however, have the important limitation of excluding resource manipulation techniques that rely on naturally occurring physical or biotic phenomena, such as fire or specific plants and animals, commonly associated with indigenous societies. In modern societies at least, there may be some association between the distribution of these factors and discernible patterns of settlement, infrastructure and land use.
Spatial analytical technologies such as GIS provide powerful tools for modelling patterns of human activity in the landscape. Established theory connected with the spatial structure of settlement and urban structure provides also some useful principles to guide development of a modelling framework for representing spatial isolation from technological activity in the landscape.
A set of spatial indicators that represent a gradient of exposure to technological activity can be developed using distance-decay functions to rate locations across the landscape on the basis of their relative exposure to settlement, exposure to infrastructure, and land use intensity. Relatively high levels of exposure to technological activity are encountered in places that are close to urban areas. Relatively low exposure is associated with places that are isolated from the settlement, infrastructure and land use practices associated with modern technological society (Lesslie 1997).
The procedures devised for measuring the Australian wilderness index (Lesslie et al.,. 1988, Lesslie & Maslen 1995) offer a simple and coherent starting-point for measuring exposure to technological activity at the global scale (Figure 18). In the first instance, data on settlement, access and infrastructure features have been drawn from Digital Chart of the World, and the analysis conducted using a grid resolution of 2.,500 m. No land cover or landuse information was available for use in this analysis, so land use intensity is not included in this pilot analysis. However, in future an appropriate surface could be derived on the basis of elementary land use and land cover information.
This representation of naturalness can then be applied to forest area to provide a characterisation of forest naturalness, which can be displayed in both mapped and statistical forms.
Any attempt to link ecological integrity or naturalness with measures of exposure to human activity raises the issue of validation. It has already been stated that an emphasis on human activity alone means that specific ecological impacts cannot be inferred. However, some form of validation is critical in order to establish the extent to which there is agreement between the model and the real world. The validation process consists of showing that a model accords with facts as known, with what is accepted as true or reasonable, or is justifiable and appropriate for a stated set of purposes (Caswell 1976).
One attempt at validating the AWI approach, which was carried out at WCMC (Kapos 1997), focused on forest reserves in Uganda and Sri Lanka. The study compared a number of measures, including average wilderness scores and surrounding human population density, with expert on-the-ground evaluations of the relative condition or naturalness of each reserve. The AWI averages were the measure most closely correlated with the expert evaluations, and were more effective than, for example, human population density in predicting forest condition. This suggests that the approach may well be appropriate as an indicator of forest naturalness at broader scales. Additional validation exercises would be useful.
Primary attribute data availability is a key issue in determining appropriate procedures for modelling global environmental phenomena. The data used in producing this global analysis were extracted from the Digital Chart of the World (DCW) database, which is drawn from the Defense Mapping Agencys Operational Navigation Charts at 1:1,000,000. Although it remains the principle source of data on access and settlement at global and regional scales, DCW is geographically inconsistent in the level of detail it provides and is rather outdated as some of the ONC charts on which it is based date to the mid-1970s. Therefore, improvement of the spatial data on settlement and infrastructure would be a necessary part of a global assessment of forest naturalness using this approach. Some progress can be made by drawing on the available higher quality national and regional data sets.
The accuracy and spatial precision of any value obtained from a spatial model can be no better than the accuracy and precision of the primary attribute data from which it is derived. This means that the scale of the primary attribute data that are available to an analysis of ecological integrity or naturalness, in effect, represents a limit to the confidence that may be placed on results and their interpretation.
The impact that the accuracy and spatial precision of primary attribute data may have on derived index measures is well illustrated by comparing the results drawn from two analyses of the Australian island state of Tasmania using AWI methodology (Fig. 19). In one case the data are drawn from the global analysis shown in Figure 18, for which the primary attribute data come from DCW. In the other, the data are derived from the Australian assessment, shown in Figure 17, which was derived from a range of very high quality primary attribute data. Most access, settlement and infrastructure data were extracted from local 1:100,000 and 1:25,000 scale topographic mapping. High quality land use and land cover data were utilised, drawn directly from relevant land management and mapping agencies. The Australian assessment was conducted at a grid resolution of 500m.
A comparison of the results of these two analyses shows that both successfully discriminate places with the highest relative wilderness quality. The south-west of Tasmania is notable in this regard as a stand-out feature of both studies. This region is regarded as one of the three key cool temperate wilderness areas of the Southern Hemisphere. Both studies also pick out most other places that are regarded as significant as wilderness within the Australian context.
However, scale difference between these two analyses is evident in the ability of each to discriminate smaller areas of significance as wilderness, and in the accuracy and precision of these assessments. Only the Australian study is capable of identifying areas that may have significance as wilderness in a local or Tasmanian context. Moreover, only the Australian study has sufficient accuracy and precision to be useful for operational evaluation and planning purposes. The global analysis is far too imprecise and incomplete in this respect.
Bearing in mind the constraints on analysis and interpretation discussed above, this approach remains a promising one for providing insight into forest naturalness at global and regional scales. Once a wilderness surface has been generated, it can be intersected or overlaid with mapped forest cover distribution to assign all forest cover to a wilderness quality or naturalness category (Fig. 20). This can is turn be displayed as Forest Naturalness maps that are analogous to the Forest Spatial Integrity maps shown in Figures 9, 11 and 13. Alternatively, statistical summaries of forest area by naturalness class (cf. Figs 10, 12 and 14) can be generated to provide a baseline for monitoring. The maintenance of forest naturalness within some target range would provide a useful basis for evaluating policy and management relating to forests in the context of forest biodiversity preservation.
