Biodiversity publications archive

Refugia for biological diversity in arid and semi-arid Australia

Biodiversity Series, Paper No. 4
S.R. Morton, J. Short and R.D. Barker, with an Appendix by G.F. Griffin and G. Pearce
Biodiversity Unit
Department of the Environment, Sport and Territories, 1995

18. Appendix: Identifying zones of sustained herbage greenness as refugia in arid and semi-arid areas

By Graham F. Griffin and Graham Pearce

18.1. Introduction

The redistribution of rainfall and nutrients in the arid regions of Australia results in dramatic differences in floristic composition and productivity in different parts of the landscape. Areas that receive regular and reliable supplies of water and seed are highly productive (relative to adjacent drier areas) and support larger numbers of species than surrounding drier zones. If left undisturbed, many of these areas become dominated by large perennial plants, and species richness declines (even though some species may be restricted to these areas). Disturbance can reduce the dominance of the long-lived plants, however, and maintain highly productive and biologically rich areas. Disturbance may come in the form of fire, grazing, or periods of water shortage, among others. The nature of these disturbances is frequently difficult to identify and quantify for large areas. Nevertheless, high levels of greenness remain a constant and characterising feature of resource-rich areas.

Morton (1990) argued that zones of high and regular herbage production in the arid zone may have been “ecological refuges” for some mammal species. These small to medium-sized mammals have been most seriously affected by grazing, by introduced animals, and by changes in disturbance regimes in arid Australia. It seems likely that other taxa such as birds have been similarly affected, if more subtly (Reid and Fleming 1992). Unfortunately, the identification of potential “refuges” for animals would be time-consuming and expensive if they had to be located individually by ground observations. If persistence of greenness is the key feature of these zones then it may be possible to use high-resolution satellite imagery to identify them. This Appendix investigates such a possibility.

18.2. Methods

18.2.1. Selection of images

To test the notion that potential ecological refuges could be identified using satellite imagery, we selected a set of images over central Australia and examined them for locations where greenness persisted over long periods. The selection was based on available imagery, the time of collection of the imagery in relation to rainfall, and our familiarity with the area (in order to enhance interpretation).

The selected area covers from Alice Springs south to about 25°S and from about 133°E to 134°30’E. Figure 18.1a shows very dark areas representing outcropping rocks and mountain ranges. The MacDonnell Ranges occur just to the north of the image, and their associated southern floodouts occur on the top of the image. The long cigar-shaped range in the top right of the image is the Waterhouse Range. The range that cuts across the middle of the image from left to right is the James and Krichauff Range, which includes the Ooraminna Range on the right-hand side. The Hugh River bisects the Waterhouse Range and the James Range, then heads east and south-east to join the Finke River in the far bottom right of the image. The Finke River comes in below the James Range on the right centre of the image and heads to the bottom right of the image. It is joined by the Palmer River coming in from the lower right-hand side of the image.

Landsat MSS imagery first became available over central Australia in late 1972 and 1973. There was a long gap in availability until 1979, when images became regularly available. Central Australia had a long and severe drought through the 1950s and 1960, ending in the late 1960s. It was not until late 1973, however, that very substantial rains fell throughout the region (Fig. 18.2). These wet conditions persisted for several years, initiating massive shrub recruitment and resulting in substantial herbage biomass levels in most areas (Griffin and Friedel 1985). Fires were extensive throughout this period (Griffin et al. 1984). Thus, the image acquired in 1973 postdates a severe drought and predates a period of massive shrub and biomass increase. Throughout the early 1980s rainfall decreased, but not to drought levels. In mid 1983 there were unusually high winter rains that stimulated a massive herbage response; an image acquired in May 1983 captured this response. Rainfall continued to decline through the latter part of the 1980s (Fig. 18.2), so by early 1988 the landscape was becoming dry. However, scrub cover initiated in the 1970s remained high where the country had not been recently burned. In order to include in the analysis these dry conditions with persistent high shrub cover, an image from early 1988 was acquired. If our ideas about the importance of certain areas of the landscape as “ecological refuges” are correct, the persistence of greenness through a variety of rainfall levels should be evident by comparing the three images at the selected dates. It may seem simpler to examine only the driest image, but focussed grazing at any single time may disguise the greenness effect in some areas. Thus, we have sought to highlight green areas when grazing might be concentrated (1973, 1988) and dispersed (1983) and when rainfall was high (1983) and low (1973, 1988). Table 18.1 summarises characteristics of the dates selected.

Table 18.1: Summary of rainfall, herbage and shrub cover recorded in the three images used in this study
Image date Previous rainfall Herb cover Shrub cover
Sep 1973 Low; winter Low Very low
May 1983 High; summer High High
Feb 1988 Low; summer Very low High

18.2.2. Processing of image data

Prior to calculating greenness, a number of corrections to the original data is required. The images must be decalibrated (to remove corrections applied by the Australian Centre for Remote Sensing), corrected for haze (to remove atmospheric effects, sun angle effects, and inter-satellite effects), destriped (to remove a striping artefact in the original data), and geometrically corrected (to match the image to a known map base). All these corrections are necessary to enable images from different satellites at different dates to be compared (see for example Pickup et al. 1994).

Greenness is calculated using the standard normalised difference vegetation index (NDVI). “Healthy green vegetation typically has high reflectance in the near infrared and low reflectance in the red wavelengths ... dividing the near infrared channel by the red channel value for each pixel in the image gives us a ratio result which is high for vegetation pixels only ... indicative of plant vigour” (Harrison and Jupp 1990). NDVI is calculated as:
(near infrared-red)/(near infrared + red).

The NDVI values for each pixel in each image (1973, 1983, 1988) were calculated, and a new image was created where the high values of the NDVI were displayed over a backdrop image (Fig.18.1b). By using the scatter plots of NDVI for the three dates against one another (Fig. 18.3), we selected pixels exhibiting the highest values for each of the three dates (Channel 1 = 1973, Channel 2= 1983, Channel 3= 1988). We divided the high values into 3 classes (extremely high, very high and high). The division was done arbitrarily on the basis of a visual inspection of the plots, and using our knowledge of where areas of persistent greenness might be within the image. We then assigned a colour (red) to a pixel that was extremely high in dates 1973, 1983 and 1988. We assigned yellow to very high value pixels, and green to high values. All other pixels remained a shade of grey. We plotted the image with the assigned pixels displayed over the grey scale image (Fig. 18.1b). On that image, red indicates areas that consistently maintained extremely high greenness over all dates, yellow indicates pixels that maintained very high greenness over the three dates, and green indicates pixels that maintained a high greenness over the three dates.

18.3. Results

The selection of the cutoff values for each of the three classes was arbitrary (Table 18.2, Fig. 18.3), though based on our experience with that landscape they do reflect the highly productive areas. The highlighted areas are known to support unusual fauna and represent known refuge zones in very dry times.

Table 18.2: The NDVI pixel value ranges for the categories used to highlight persistent greenness in Fig 1b
Category NDVI range of pixels Number Percent
of pixels in image
Extremely high greenness >172, >177, >163 1182 0.5
Very high greenness >166, >175, >161 158 0.1
High greenness >160, >173, >159 470 0.2
Unclassified pixels <=159 216,759 99.2
Total pixels   218,569 100

Areas of high and persistent greenness are particularly associated with river margins, floodouts adjacent to hill areas, and drainage depressions. Some areas have a higher density of red areas than others. Prominent areas of persistent greenness were in the far north-west of the area on a floodout from the MacDonnell Ranges. Other floodouts along the southern foothills of the MacDonnells were also persistently green. Many small areas (but quite dense) occur in and about the Waterhouse and James-Krichauff Ranges. Tertiary river systems that have been buried by aeolian sands to the north and south of the Ooraminna Ranges show persistent areas of greenness. The major river systems all support high levels of green vegetation in the form of large river gums Eucalyptus camaldulensis that tap into perennial water supplies. We suggest that many of these red areas are likely to be highly productive and support refuge populations of certain animals in dry times, and that they were likely refuge areas for some mammal species prior to their recent extinction.

18.4. Conclusions

On the basis of what we know about ecology in arid Australia, it seems reasonable to predict that many plants and animals remain highly dependent on areas of persistent cover and greenness during dry times. Our investigation shows that it is possible to identify many of these areas using readily available satellite imagery and technology. The simple analysis conducted here is but one of several types that would yield similar answers. Identification of areas with persistent greenness could be achieved using classification or ordination of the NDVI data, or of the infrared data, and a variety of other methods (Harrison and Jupp 1990). The example worked through here gives an indication of the potential for use of remotely sensed data to identify key refuge areas in arid landscapes.

We suggest that a simple extension of this method would be to take all the known localities of biological records in dry times and examine the greenness values for those areas. If they are significantly higher than surrounding country then it may be possible to determine a threshold level for these values to use to search for other sites of possible populations. If that proves to be a suitable measure of the likelihood of habitat for certain organisms then it would be rational to examine the spatial structure of patches in relation to climate, substrate and vegetation as an added measure of the likelihood of population persistence.