Vegetation condition mapping at a landscape-scale across Victoria. Graeme R. Newell1, Matt D. White1, Peter Griffioen2 and Michael Conroy3 (1Arthur Rylah Institute for Environmental Research, Department of Sustainability and Environment, 123 Brown Street, Heidelberg, Vic. 3084, Australia. Email:; 2Acromap, Pty Ltd, 37 Gloucester Drive, Heidelberg, Vic. 3084, Australia; 3Spatial Information and Infrastructure, Department of Sustainability and Environment, Marland House, 570 Bourke Street, Melbourne, Vic 3000, Australia).

Key words: ecological modelling, mapping, native vegetation, neural networks, vegetation condition.

Introduction.  One of the strongest drivers in conservation planning across southeastern Australia has been information on the level of depletion of remnant native vegetation (i.e. both ‘extent’ and ‘type’, e.g. Regional Forest Agreement process). However, more recently natural resource management (NRM) agencies have also been asked to consider the ‘condition’ or ‘quality’ of native vegetation, and many government agencies at regional, state and federal levels have now instituted policies and requirements to monitor native vegetation condition over time (e.g. DNRE 2002).

Concepts, methods of assessment and monitoring of condition in rangelands, arid and semiarid ecosystems have been around for many decades (e.g. Dyksterhuis 1949), and mapping of condition in these environments has become even become routine (e.g. Reeves et al. 2001; Wallace et al. 2004). However, the concepts of ecological condition or quality of native vegetation (or ‘habitat’) in more mesic and temperate systems are still relatively vague and poorly defined (Gibbons & Freudenberger 2006), and maybe considered idiosyncratic when compared to similar concepts applied to agricultural systems, medicine, engineering, etc. (Tongway & Ludwig 1997).

Useful attempts to clarify these concepts (McIntyre & Hobbs 1999; Gibbons & Freudenberger 2006), and to develop general and widely applicable methods for scoring vegetation condition or habitat value have been made recently. Methods range from relatively generic scores of habitat structural complexity (e.g. Newsome & Catling 1979), to intricate scoring and modelling approaches applicable to single species (e.g. Habitat Suitability Indices; USFWS 1981). Recent approaches such as ‘Habitat Hectares’ (Parkes et al. 2003) provide rapidly obtained indices of native vegetation condition by using comparisons to reference condition states (i.e. ‘benchmarks’, viz. Hopkins 1999), and are intended to provide NRM practitioners with simple measures of vegetation condition. ‘Habitat Hectares’ values are attributed to a specific site or location on a map. However, the highly variable nature of condition across landscapes means that this information does not necessarily infer the condition of other adjacent sites. The need for a complete map rather than individual points of vegetation condition became obvious to land managers, and resulted in this project to attempt to ‘map’ native vegetation condition at a landscape scale. Maps of native vegetation condition in temperate Australia are not novel (e.g. Coops et al. 1998), but have not been formerly produced at the current spatial scale or across a broad range of vegetation types. This paper highlights the need for maps of native vegetation condition for supporting NRM activities, introduces some recent research to produce a landscape scale vegetation condition map, and considers the current usefulness of these maps for stated NRM aims.

Methods.  A full explanation of the development and application of the statistical and spatial modelling routines to develop the map is beyond scope of this short paper. The study area comprised an area of more than 9 million ha (∼ 40% Victorian landmass) within the jurisdictions of the Mallee, North-Central, Goulburn-Broken and Northeast Catchment Management Authorities. ‘Habitat Hectare’ condition assessments were available from 1267 pre-existing sites, and these data were supplemented with a further 1641 sites stratified by vegetation type, tenure and patch size across Mallee and Goulburn Broken CMAs. These data were then used as the dependent variable in a ‘neural network’ modelling procedure (viz. Lek & Guégan 1999) that identified relationships between site condition scores and 13 independent variables assembled from a state-wide spatial data library. These variables included vegetation type, climate and lithology, novel indices of tree density derived from an existing statewide tree cover dataset, and a land-use map. The output of this modelling identified relationships between the 13 independent variables and site condition assessed in the field, with a strong positive correlation evident between the predicted and observed scores explaining 51% of the variance. The modelled ‘neural network’ relationships were then applied to ‘unknown’ sites or cells (30 m2) with recognized extant native terrestrial vegetation to form a condition map that was coincident with extant native vegetation. Data from a further 541 sites were then collected to validate the final map, thus creating a total pool of 3449 site assessments used to develop the map (Fig. 1a).


Figure 1. (a) Distribution of Survey Sites with northern Victorian CMAs (Blue, pre-existing data; Orange, new data. Total = 3449 sites). (b) Output model of native vegetation condition. (c) Example output of model at proposed operational scale (1:100 000).

Download figure to PowerPoint

Results. Figure 1(b) provides an example of the coverage of the native vegetation condition map across northern Victoria, and clearly shows a gradient from relative good condition predicted in the large Mallee blocks in the north-west of the State, to poorer native vegetation condition predicted for more fragmented and more intensively farmed landscapes on the inland slopes and plains of central and north-eastern Victoria. Although this ‘regional scale’ view of ecological condition is somewhat interesting, it is not particularly informative for planning local scale NRM activities. Given the constraints of the data and final map we considered that the map is probably most useful at scales of 1:100 000 and above. Figure 1(c) shows an area around Wangaratta at this scale, and clearly shows varying levels of predicted condition across the Killawarra Range in the north, Warby Ranges in the central region, the riverine systems along the Ovens River, and the fragmented remnant vegetation. This diagram demonstrates the ability of the map to represent condition across various landforms and vegetation configurations, but also highlights the inability of the map to clearly distinguish relatively subtle differences in predicted condition at the ‘paddock’ scale.

Discussion.  Our primary aim was to produce a map that would assist NRM staff in making decisions for conservation investments across northern Victoria. Arguably, this aim has been met, but how good is the model and resulting map? Considering the geographical scale of the project, the range of bioregions and biological systems involved, and the range of condition states, we believe that the map aligns well with existing broad scale appraisals of vegetation condition (e.g. National Land & Water Resources Audit 2001; Department of Natural Resources and Environment 2002).

However, there are always strengths and weaknesses. In constructing this broad-scale map, we became aware of the difficulties and challenges to developing accurate models of vegetation condition. First, the sampling of the full range of condition states is important, and the scarcity of high quality sites in lowland southeastern Australia can impact on the model and subsequent map. Second, the rarity of ‘good’ sites means that the current model has a ‘flat’ response. The model tends to overestimate the scores for sites in poorer condition and conversely underestimate scores for sites in very good condition, while providing the most statistically robust predictions within the mid-range of condition states. Third, the quality and consistency of field data are critical (including GPS location), and we had to exclude a large proportion of pre-existing field data from modelling because of obvious errors. Fourth, GIS surfaces used as independent variables often have coarse resolution and can result in maps with similar properties. This coupled with the inappropriate use of zoom function on GIS and image analysis software can often imply false precision when viewing maps.

For these and a number of other reasons (e.g. ‘Habitat Hectare’ condition assessments are intentionally coarse, and were not intended for the fine scale temporal or spatial monitoring of condition; see Parkes et al. 2003, 2004), we believe that the current map is unlikely to provide the accuracy and precision necessary to (i) detect differences in condition between small parcels of land suitable for property or statutory planning purposes (i.e. ‘paddock’ level); and (ii) monitor acute and subtle changes in condition at fine spatial scales and over short time frames (e.g. political and funding cycles).

Nevertheless, the stated objectives, policies and requirements of various Government agencies to monitor native vegetation condition over time remain. Whereas this may be possible using modelling approaches applied solely at fine scales, this will limit the widespread use of the approach. There are currently no substitutes for site inspections and targeted data collection when monitoring site condition, and making planning and conservation management decisions at fine scales. Modelling the condition of native vegetation across the Australian temperate zone is in its infancy, and we can anticipate that emerging technologies, more ecologically relevant GIS data, and remotely sensed datasets may allow maps at finer scales to be produced in the future.

Despite this list of limitations, the model does have a number of strengths. We believe that the map does provide important insights into patterns of native vegetation condition at landscape scales, and is novel in its representation of the condition of native vegetation across a large area of temperate Australia. We believe that the most appropriate use of the map currently is to augment other spatial expressions of landscape attributes (e.g. relative rarity of vegetation types, etc.) and to assist in the identification of locations where NRM activities are most likely to deliver effective conservation outcomes (Higgins 2006). The use of vegetation condition mapping for other purposes will also evolve and become apparent with future improvements.


  1. Top of page
  2. References
  • Coops N., Culvenor D., Preston R. and Catling P. (1998) Procedures for predicting habitat and structural attributes in eucalypt forests using high spatial resolution remotely sensed imagery. Australian Forestry 61 (4), 244252.
  • Department of Natural Resources and Environment (2002) Victoria's Native Vegetation Management. A Framework for Action. The State of Victoria.
  • Dyksterhuis E. J. (1949) Condition and management of range land based on quantitative ecology. Journal of Range Management 2, 104114.
  • Gibbons P. and Freudenberger D. (2006) An overview of methods used to assess vegetation condition at the scale of the site. Ecological Management & Restoration 7 (Suppl. 1), S10S17.
  • Higgins I. (2006) Vegetation condition mapping and catchment management: The North Central Victorian experience. Ecological Management & Restoration 7 (Suppl. 1), S68S71.
  • Hopkins A. (1999) National Land and Water Resources Audit Vegetation Theme. Condition of Vegetation: a Discussion Paper. Department of Conservation and Land Management, Wanneroo.
  • Lek S. and Guégan J. F. (1999) Artifical neural networks as a tool in ecological modelling, an introduction. Ecological Modelling 120, 6573.
  • McIntyre S. and Hobbs R. (1999) A framework for conceptualizing human effects on landscapes and its relevance to management and research models. Conservation Biology 13 (6), 12821292.
  • National Land & Water Resources Audit (2001) Australian Native Vegetation Assessment 2001. Commonwealth of Australia, Canberra.
  • Newsome A. E. and Catling P. C. (1979) Habitat preferences of vertebrates inhabiting heathlands of coastal, montane and alpine regions of south-east Australia. In: Ecosystems of the World. Vol. 9A. Heathlands and Related Shrublands of the World (ed. R. L.Specht), pp. 301316. Elsevier, Amsterdam.
  • Parkes D., Newell G. and Cheal D. (2003) Assessing the quality of native vegetation: The ‘habitat hectares’ approach. Ecological Management & Restoration 4, S29S38.
  • Parkes D., Newell G., Cheal D. (2004) The development and raison d’être of ‘habitat hectares’: a response to McCarthy et al. (2004). Ecological Management & Restoration 5 (1), 2829.
  • Reeves M. C., Winslow J. C. and Running S. W. (2001) Mapping weekly rangeland vegetation productivity using MODIS algorithms. Journal of Range Management 54, A90A105.
  • Tongway D. and Ludwig J. A. (1997) The nature of landscape dysfunction in rangelands. In: Landscape Ecology: Function and Management (eds J. A.Ludwig, D.Tongway, D.Freudenberger, J.Noble and K.Hodgkinson), pp. 4961. CSIRO Publications, Melbourne.
  • USFWS (1981) Standards for the Development of Habitat Suitability Index Models. United States Fish and Wildlife Service, Washington, DC, USA.
  • Wallace J. F., Caccetta P. A. and Kiiveri H. T. (2004) Recent developments in analysis of spatial and temporal data for landscape qualities and monitoring. Austral Ecology 29, 100107.