CONFERENCE PROCEEDINGS Australian Bushfire Conference, Albury, July 1999

copyright 1999

Abstract

Introduction

Approaches to recording and mapping fire histories in Australia and internationally are many and varied. Dendrochronological interpretation (Banks 1989) and the analysis of soil cores for pollen and charcoal (Clark 1983; Larsen and Macdonald 1998) have been popular in constructing long fire histories. Verbal accounts of fires have been compiled to supplement existing fire history data (Reid et al. 1996). However, all of these techniques are localised to the area of study, so spatial interpolation to larger areas is either not possible or inaccurate, given the highly variable nature of fire.

Fire histories that contain spatial and temporal information are limited, but available for some areas. Examples include the Switzerland wildfire database with records from 1920 (Tinner et al. 1998) and Wood Buffalow National Park in Canada with fire records since 1950 (Larson 1997). In Australia, Nieuwenhuis (1987) used records from the early 1960s for Ku-Ring-Gai Chase National Park to investigate vegetation response and these plus other records from Brisbane and Royal National Park have been used in a number of fire frequency studies (e.g. Benson 1985; Morrison et al. 1995). These databases are rare, and generally fire history information is spatially inaccuracy or contains gaps in the time sequence. Remotely sensed satellite imagery can be used to address the need for spatial and temporal data on fires over long periods.

Digital imagery is derived from a sensing device carried on a satellite, recording the electromagnetic energy emitted or reflected from the earths surface. These data have been collected regularly for many years and are particularly suited to retrospective multi-temporal studies. The images cover a large area and can provide information in remote areas. There are a number of different sensor types each with differing pixel sizes, overpass rates and sensing capabilities (for more information see Harrison and Jupp (1989)) providing an array of options for ecological applications. The most popular satellite sensors for fire mapping have been the Advanced Very High Resolution Radiometer (AVHRR), Landsat Multispectral Scanner (MSS) and Landsat Thematic Mapper (TM). These three data sources were considered for this study. Other remote sensing platforms have been used to map fires, such as satellite radar (Bourgeau-Chavez et al. 1997), but they lacked the retrospective data archive needed to construct a long fire history so were not considered for this study.

Satellite data has been used for many fire applications. Matson (1981) demonstrated the utility of the thermal channels in AVHRR data for detecting high temperature targets such as forest fires over large remote areas. The rapid repeat pass and large pixel size of AVHRR data make it useful for mapping fires in large areas (e.g. Pereira and Setzer 1993) and globally (Robinson 1991; Chuvieco and Martin 1994). The Normalised Difference Vegetation Index (NDVI) derived by Tucker (1979) has been used to monitor vegetation condition and change for multi-temporal land cover studies (Townshend et al. 1985; Chilar et al. 1991) and to map fires globally (Chuvieco and Martin 1994). In Australia, a large-scale application of the AVHRR data maps fires throughout northern Australia (Smith et al. 1998). The northern Australian landscape and the nature of the fires are particularly suited to AVHRR fire mapping as the fires are generally large, more than 50 km² (Allan 1984), so the large pixel size of 1.1 km² is suitable for fire detection.

The Landsat satellite series that began in 1972 with the Landsat MSS that has been used widely for land cover change assessment (e.g. Byrne et al. 1980; Chavez and MacKinnon 1994). In Australia, Landsat MSS was used to map fires and vegetation change associated with bushfires (Richards et al. 1984; Graetz 1990). Similarly, Landsat TM has been used for fire detection and mapping (Chavez et al. 1989), forest fire hazard and risk mapping (Chuvieco et al. 1989; Maselli et al. 1996), fire severity (White et al. 1996) and analysing the post-fire recovery of ecosystems (Viedma et al. 1997). In Kakadu National Park, Landsat MSS satellite imagery was used to derive a fire history from 1980 to 1994 (Russell-Smith et al. 1997). However, it was not possible to apply these techniques to Guy Fawkes River National Park (GFRNP) because of complexity in the landscape, steep slope and frequent small low intensity fires. Evaluation was required on the utility of satellite data to map fires in this terrain.

Considering these issues, this study aimed to assess the potential of satellite imagery to derive fire information in GFRNP. Specifically to (1) assess the quality of existing hardcopy fire data, (2) assess satellite derived data sources to establish their extent and limitations for mapping fire in dissected gorge terrain, (3) determine the best method for mapping the fires and (4) to integrate satellite and hardcopy fire data to compile a comprehensive fire history for GFRNP.

Method

GFRNP is situated in northern New South Wales (NSW) on the eastern edge of the New England Tablelands. The vegetation of the park is Eucalyptus dominated open forest with a grassy understorey. The dominant physiological feature is the north-south flowing Guy Fawkes River with gorges forming a steeply dissected landscape. Fire records from the Dorrigo District of the NSW National Parks and Wildlife Service (NPWS), from previous projects, or as verbal accounts from field officers who had worked on fires, were identified and collated. These included bushfire reports, registers of radio and telephone calls, media releases, incident action plans and summary faxes of fire situations to Head Office. As part of a previous project, the hardcopy fire records for the whole of Dorrigo District had been digitised by NPWS staff (Rollings and van der Lee 1995). For reasons of accuracy and consistency, these data were checked and the data converted to layers in a geographical information system (GIS).

Interviews were conducted with officers who worked on fires in the park. One Park Ranger provided work diaries for dates of fires that were attended. As digitising errors can appear in spatial data when converted from digitising to map units, the spatial location of each record was checked and the records scanned for errors. Each fire, including the satellite-mapped fires described later, was tagged with an attribute indicating the level of accuracy of the original data source.

Data from the AVHRR and Landsat TM satellites were obtained for before and after a major fire in GFRNP in October 1994. The AVHRR data was obtained for the whole park area from AMG Zone 56: 400 000 to 450 000 (easting) to 6 663 000 to 6 710 000 (northing). This area covered the National Park boundary, a larger area identified as wilderness and land contiguous with the park. Images were acquired for January 1994 and December 1994. These image dates were selected as they were soon after the fire when minimal regrowth would have occurred and matched in season to minimise shadow and sun-angle between scenes. The images were from a continental AVHRR database held by the Environmental Resources Information Network (ERIN).

Landsat TM images were acquired for September 1994, December 1994, and May 1995 for a smaller section of the park from 421 500 - 446 650 to 6 655 750 – 6 680 100. The boundary was acquired of the October 1994 fire that had been mapped by the NPWS using the Automated Real-time Mapping System (ARMS) with global positioning system (GPS) accuracy. This boundary was used for analysing the AVHRR data, but only the northern section of the boundary was used for the smaller Landsat TM area of coverage. This fire boundary was used to assess the accuracy with which the fire could be detected as it contained a range of fire intensities from full crown scorch to grass consumption.

For the AVHRR data, the NDVI was calculated for both time periods. The ARMS fire boundary was overlaid and the AVHRR bands including the NDVI were assessed for their ability to duplicate the fire boundary. As this process identified limitations with the AVHRR data that are discussed in the results, no further analysis was undertaken.

The raw Landsat TM data bands were displayed and evaluated for the ability to delineate the fire burn. Burnt and unburnt categories were taken from the images. All bands were analysed in relation to the differences in the spectral response of the burnt and unburnt areas. This analysis was used to assess the significant spectral bands for detecting the fires. The techniques of image differencing, band ratioing, principal component analysis and classification were evaluated. Using the ARMS fire boundary each change detection method was assessed for its accuracy in defining the boundary. To assess the longevity of the fire scar, the May 1995 image, seven months after the fire, was processed using a greenness index to assess regeneration of the area. Using techniques to determine differences in severity developed in a previous study (Kitchin et al. 1998), it was determined how much area remained detectable as burnt within the three burn severity categories. No suitable 1994 Landsat MSS images were available for testing the technique with the same fire boundary, so Landsat TM data were resampled to 80m pixels and the four corresponding bands (1-4) assessed.

Once the method had been established, suitable images for as many years as possible were identified and purchased for the full fire history. Suitable images were those that occurred later in the season, to capture all the fires occurring within that season, with as little cloud as possible. A total of 30 images were obtained with support from ERIN and CSIRO. This spanned the period of 1972 to 1998 with images available for 16 of these fire seasons and four seasons mapped using ARMS.

All images were registered and atmospherically normalised so that a meaningful comparison could be made between images recorded on different dates by different scanners. For initial trials with the Landsat TM data, a pseudo-invariant target normalisation was applied (Caselles and Lopez Garcia 1989) due to the availability of suitable targets. However, when compiling the full fire history it was found the Landsat MSS data, due to its larger cell size, had fewer suitable invariant dark targets and no suitable light targets in the study area. When processing the yearly satellite data an alternative normalisation was undertaken using the haze reduction technique and the digital numbers converted to physical values of radiance (Hill and Sturm 1991).

Due to the lack of a suitable pre-season image for all the years, an alternative difference method was developed for processing all the yearly data. A pre-season running median image was calculated for the four seasons prior to the fire season of interest. Rollings (1998) used a technique of deriving a long-term average from a number of images that spanned the whole study period. It had the benefit of providing an accurate pre-fire image without the additional cost of purchasing more data but contained longer-term changes, such as land clearing. For this project, a method was developed of calculating the median value from four years prior to the fire. By using a small number of years, the longer-term land cover changes were removed. By calculating the median, the analysis was not as strongly influenced by sudden changes in the spectral value due to change, such as fire. The moving average is a technique used in statistics to smooth variable data. This same technique was applied to derive an image that captured the ‘normal’ state of the park. The image for the year of interest was then subtracted from this median image. This technique, along with the raw data bands, was used to detect the fires. However, in some situations, especially for low intensity fires, the ratio image was also referenced and used for verification. The boundary was then mapped on screen. The individual years were combined into one fire history database. Where there was duplication between the hardcopy and satellite mapped fires, the boundary from the satellite image was used.

Results

The hardcopy fire reports varied widely in quality. The fire records began in July 1968 and continue to the present. The older written reports contain less useful information, for example a report from July 1969 refers to the locality of an ignition source as "began on trunk road 76". No road named ‘trunk road 76’ now exists in GFRNP so this and other reports of this nature were of little value for mapping fire boundaries. The more recent records contained increased spatial and non-spatial fire details. These were incident action plans that included maps used at the time of the fire. The maps were of varying quality from hand drawn boundaries on plain paper, to sketches on 1:25 000 topographic maps. Often information on the weather, particularly temperature, humidity, wind speed and wind direction was included. The recording quality improved through time from broad descriptions to more detailed reports by 1984. There was an increase in the number of mapped fire boundaries from 1986 and from 1994 all fires had been mapped digitally using ARMS. The interviews with NPWS staff did not provide any new information but confirmed fire records on file. Eight fires had been missed and these were added to the database. The spatial data derived from the hardcopy files mapped fires from 1975 to 1994. The data had a large area of the park unburnt and there were six years where no fires had been recorded. The method adopted by NPWS to digitally map fires and record details in fire incident reports from 1994 had produced the highest quality information.

The analysis of the AVHRR data showed the most significant response to the fire in the near-infrared band. Contrary to other studies, the NDVI detected little change due to the fire. The near infrared band mapped the fire very broadly with detail being lost due to the large pixel size of 1.1 km ². The loss of detail in detecting the burn and mapping the boundary was noticeable. It was possible the AVHRR data may have enabled the mapping of lower intensity fires if imagery were available directly after the fire however, it was likely that smaller fires could have gone undetected due to the loss of detail from the large pixels. This was similar to findings in Kakadu National Park, where the large pixel size of AVHRR yielded a major loss of mapping detail (Russell-Smith et al. 1997). For this study, a satellite sensor with a smaller pixel size for greater spatial accuracy was pursued.

The Landsat TM analysis between the pre- and post-fire images found an increase in band 7 due to the burn, a moderate increase in band 5 and little change in bands 4 – 1. Band 7 and band 5 were the most accurate in separating the spectral response of the burnt and non-burnt vegetation and hence for mapping the fire boundary. The degree of separation between the spectral response of burnt and unburnt areas varied between the three main physiographic areas of the park (tablelands, slopes and gorge-base). Both the gorge-base and the tablelands showed a greater degree of spectral separation than on the slopes. The separation between the burnt and non-burnt areas was more distinct in the flatter areas of the tablelands and gorge base as shown in figure 1a. In the sloped areas, the spectral response was mixed (figure 1b). The within scene variation in the spectral response was less marked and complicated by the fact that the results were from different locations in the scene, despite being matched as closely as possible in sun angle, slope and aspect. The within scene variation was similar but with less separation between spectral responses.

The analysis to determine the best change detection method showed that no automated technique could uniquely map the fire boundary. All techniques included changes not associated with fire and required interpretation to distinguish between other causes of change such as forestry or agriculture. Band ratioing and difference to a pre-fire image proved optimum. The best discrimination of the fire boundary was the ratio of band 7 over band 5. After using the ratio and difference to highlight change, the actual fire boundary was determined by visual interpretation, as the eye can detect the fire boundary based on position in the landscape, shape, colour and contrast. As fire forms a relatively ‘predictable’ pattern, either with rounded fronts or ‘tongues’ of fire up ridges, this visual interpretation was used to validate the enhanced imagery.

The analysis of the May 1995 image to determined the longevity of the fire scar showed that by seven months after the burn there was little in the low or moderate intensity categories still detectable as burnt. The high fire intensity area was detectable. This indicated that images needed to be purchased within six months of the burn. Ideally monthly images would have been analysed to map the diminishing fire scar however, due to lack of resources this was not possible.

The results from the Landsat MSS data showed that the fires could be detected and mapped but with a loss of detail. The two bands, band 7 and band 5, found most effective in Landsat TM are not available in Landsat MSS data. The bands 1 and 4 produced the best result in distinguishing burnt and non-burnt areas, but with more spectral mixing especially on the slopes. There was a loss of detail in detecting the fire due to the larger pixel size. This was especially evident in the gorge where topographic changes are sharp and over 80 m of horizontal distance a number of differing vegetation types and conditions can occur, this results in a mixed spectral signal. Landsat MSS was deemed suitable for mapping the fires but a loss of spectral and spatial resolution was evident.

Figure 1. Radiance comparison between burnt and unburnt areas in the park for (a) the tablelands and gorge-base and (b) the sloped areas.

To summarise, the results demonstrated that for mapping fires in GFRNP the optimum satellite data was Landsat TM with its increased spectral and spatial resolution. It had the advantage of more bands in a significant portion of the electromagnetic spectrum for mapping fires and a smaller pixel size, resulting in more spatial detail. Landsat MSS provided a longer multi-temporal study and was cheaper but its four bands covered a less significant portion of the electromagnetic spectrum. The AVHRR data had the advantage of a rapid repeat time but the large pixels were unsuitable for detecting fires in a park of this nature.

Both Landsat TM and MSS imagery were used to map fires for GFRNP between 1972 and 1994. A section of the fire history is shown in figure 2. The fire history showed a majority of the park having been burnt in the last 25 years with an area being burnt eight times. This area is called the ‘Fattening Paddock’, an area that has been used in the past for grazing and known to have experienced frequent fires (Peter Evans pers. comm.). From local knowledge, this area is said to have burnt every 1 – 3 years, so the actual number of fires could be as low as six or as high as eighteen fires in this time.

Figure 2. An example of the fire history data where the fire frequency was highest, for the Fattening Paddock of Guy Fawkes River National Park.

Discussion

It is the differing nature of band 7 and band 5 in the burn area that when combined as a ratio, is sensitive to changes associated with fire. Band 5 is highly reflective for bedrock materials. In the areas of greatest fire intensity, the majority of green vegetation was lost, exposing bedrock. The bedrock response was detected in band 5, but more markedly in band 7. This was due to the bedrock containing iron oxide materials that have a high response in band 7. Due to the varying intensity of the fire, some or all of the vegetation was removed. Partial decrease of vegetation correlates with an increased response from exposed soil and rock. This was tested by investigating two soil band ratios. These were band 5 / band 4, sensitive to iron oxide and band 5 / band 7, sensitive to clay. An increase in iron oxide in the burn area was detected. Field validation showed iron oxide mixed with the dominant granite and silica found in the soil. This is being investigated further. To summarise, following the fire a loss of green vegetation and an increase in soil exposure was detected, this change is mappable and derived the basis for detecting changes that were then digitised as fire boundaries. The increased spectral mixing of the classes in the sloped areas was attributed to the variation in slope and aspect.

The band ratio enhanced within scene contrasts and improved the response in the gorge by normalising some of the topographic effects. The difference method enhanced the between scene contrasts. They, along with histogram enhancement of the raw bands, detected the fire but no classification technique was able to map the fire boundary without including areas of shadow or other effects. These techniques are effective in determining change in vegetation but are not always specific to fire, they also highlight change due to other factors.

The satellite mapped fire boundaries produced a fire history database enabling the calculation of fire frequency values for all areas. Field validation revealed evidence that some fires had missed detection. These were either low intensity burns with no change in the overstorey or fires that had occurred more than six months prior to the image causing the fire scar to be lost. One of the confounding factors in GFRNP was that summer rainfall often made images cloudy, hence in some years, suitable images were not available until March or later. There were also the complications of shadow. The majority of the park is steep so south and west-facing ridges are often in shadow. While there are shadow reduction techniques for satellite imagery they require complex algorithms and intensive processing, some topographic normalisation was addressed by utilising the ratio technique. Due to these issues, it was assumed fires had missed detection, so rather than take the absolute number of fires, the fire frequency database was interpreted as having probabilities of fire occurrence. For example, the area shown in figure 2 was interpreted as an area of highest fire probability with an occurrence of eight or more fires within the study period. Using the data as a relative probability of fire was a more accurate interpretation of the true nature of the fire occurrence in the area. In rugged terrain, it is likely some fires will go undetected whether using direct fire mapping (i.e. fire officers are not always aware of all fires) or satellite imagery, so interpreting the data as an absolute number of fires is questioned. Probabilities of fire remove this ambiguity and provide a relative index of fire occurrence. Until our methods for recording fires improve to the level of being certain of all fire occurrence and the accuracy of its mapping, whether it be digitally with imagery or in the field, it is suggested this concept of relative fire frequency values, reflects the accuracy of the data. Work is continuing on a map of fire detection accuracy to indicate areas where non-detection of the fires is highest.

Conclusions

This study demonstrated how fire history information can be derived from satellite imagery when previous records are either not available, or where records are known to be deficient. It has shown there are alternative data sources when field observations of fires are not available. The satellite-based and hardcopy-based data on the fire history of GFRNP has generated a fire frequency database from 1972 to 1998. This information enables ecological data to be interpreted in the context of the fire frequency regime and has the potential for deriving other attributes such as the shortest fire interval and time since last fire, that hold ecological significance. A change detection method has been developed to map fires and minimise the confusing effects of sun illumination and shadow in rugged dissected terrain. The technique is useable and repeatable. The processing method minimises the acquisition of imagery and provides an accurate fire boundary for moderate to high intensity fires, with the detection of lower intensity fires linked to the time of purchase of the imagery. The mapped fire history has proved to be a valuable data source and is being used to interpret the influence of fire frequency on vegetation in the GFRNP.

Acknowledgments

This work was undertaken as part of an Australian Postgraduate Award Industry (1997-1999) project, funded by the Australian Research Council and Dorrigo District of the NSW NPWS. Some of the imagery was provided and owned by Environment Australia and CSIRO and acquired by the ACRES. The processing would not have been possible without the support of ERIN and AGSO. Thanks are due to John Wilford and NPWS staff for assisting in the field and Nick Rollings for use of the imagery and his initial work in GFRNP.

Literature Cited

Allan, G.E. (1984).

Criteria for the remote sensing of bushfire history, current fuel load and fire behaviour in Central Australia. 3rd Australasian Remote Sensing Conference, Queensland. pp. 275-280

Banks, J.C.G. (1989).

A history of forest fires in the Australian Alps. In 'The Scientific Significance of the Australian Alps'.(ed Good, R. B.) pp 157-80.(Australian Alps National Parks Liasion Committee: Australia).

Benson, D.H. (1985).

Maturation periods for fire-sensitive shrub species in Hawkesbury sandstone vegetation. Cunninghamia 1, 339-349.

Bourgeau-Chavez, L.L., Harrell, P.A., Kasischke, E.S. and French, N.H.F. (1997).

The detection and mapping of Alaskan wildfires using a spaceborne imaging radar system. International Journal of Remote Sensing 18, 355-373.

Byrne, G.F., Crapper, P.F., and Mayo, K.K. (1980).

Monitoring land-cover change by principal component analysis of multitemporal landsat data. Remote Sensing of the Environment 10, 175-184.

Caselles V. and Lopez Garcia M. J. (1989).

An alternative simple approach to estimate atmospheric correction in multitemporal studies. International Journal of Remote Sensing 10(6), 1127-1134.

Chavez, P.S. and Yaw Kwarteng, A. (1989).

Extracting spectral contrast in Landsat Thematic Mapper image data using selective principal component analysis. Photogrammetric Engineering and Remote Sensing 55, 339-348.

Chavez, P.S. and MacKinnon, D.J. (1994).

Automatic detection of vegetation changes in the southwestern United States using remotely sensed images. Photogrammetric Engineering and Remote Sensing 60, 571-583.

Chilar, J., St-Laurent, L., and Dyer, J.A. (1991).

Relation between the normalised difference vegetation index and ecological variables. Remote Sensing of the Environment 35, 279-289.

Chuvieco, E. and Martin, M.P. (1994).

A simple method for fire growth mapping using AVHRR channel-3 data. International Journal of Remote Sensing 15, 3141-3146.

Chuvieco, E. and Congalton, R.G. (1989).

Application of remote sensing and geographic information systems to forest fire hazard mapping. Remote Sensing of Environment 29, 147-159.

Clark, R.L. (1983).

Pollen and charcoal evidence for the effects of Aboriginal burning on the vegetation of Australia. Archaeology in Oceania 18, 32-37.

Graetz, R. D. (1990).

Analysis of land degradation in Kakadu National Park (Stage III) PHASE II: Landscape dynamics, 1972 - Present. (CSIRO Division of Wildlife and Ecology: Canberra).

Harrison, B.A. and Jupp, D.L.B. (1989).

'Introduction to remotely sensed data', (CSIRO, Division of Water Resources): Canberra, Australia.

Hill, J. and Sturm, B. (1991).

Radiometetric correction of multitemporal Thematic Mapper data for use in agricultural land cover classification and vegetation monitoring. International Journal of Remote Sensing 12, 1471-1491.

Kitchin, M.B., Rollings, N., and Bradstock, R.A., (1998).

Mapping forest fire severity using Landsat TM. 9th Australasian Remote Sensing and Photogrammetry Conference. University of New South Wales, Sydney. Paper No. 62.

Larsen, C.P.S. (1997).

Spatial and temporal variations in boreal forest fire frequency in northern Alberta. Journal of Biogeography 24, 663-673.

Larsen, C.P.S. and MacDonald, G.M. (1998).

An 840-year record of fire and vegetation in Boreal White Spruce forest. Ecology 79, 106-118.

Maselli, F., Rodolfi, A., Bottai, L., and Conese, C. (1996).

Evaluation of forest fire risk by the analysis of environmental data and TM images. International Journal of Remote Sensing 17, 1417-1423.

Matson, M. (1981).

Identification of subresolution high temperature souces using a themal IR sensor. Photogrammetric Engineering and Remote Sensing 47, 1311-1318.

Morrison, D.A., Cary, G.J., Pengelly, S.M., Ross, D.G., Mullins, B.J., Thomas, C.R., and Anderson, T.S. (1995).

Effects of fire frequency on plant species composition of sandstone communities in the Sydney region: Inter-fire interval and time-since fire. Australian Journal of Ecology 20, 239-247.

Nieuvenhuis, A. (1987).

The effect of fire frequency on the sclerophyll vegetation of the West Head, NSW. Australian Journal of Ecology 12, 373-385.

Pereira, M.C. and Setzer, A.W. (1993).

Spectral characteristics of deforestation fires in NOAA/AVHRR images. International Journal of Remote Sensing 14, 583-597.

Reid, N., Gross, C., Jarman, P., and van der Lee, G. (1996).

Fire management criteria and guidelines for conservation of the natural and cultural heritage of Guy Fawkes River National Park. (University of New England: Armidale, NSW).

Richards, J.A. and Milne, A.K., (1984).

Mapping fire burns & vegetation regeneration by classification of multitemporal Landsat MSS image data. 3rd Australasian Remote Sensing Conference. Qld. pp 304-314.

Robinson, J.M. (1991).

Fire from space: Global fire evaluation using infrared remote sensing. International Journal of Remote Sensing 12, 3-24.

Rollings, N., (1998).

Mapping occurrences of fire burns is easier using a temporally-averaged image as a reference. 9th Australasian Remote Sensing and Photogrammetry Conference. University of New South Wales, Sydney. Paper No. 110.

Rollings, N. and van der Lee, G. (1995).

Compilation of a fire burn history and an investigation into mapping fire burn distribution and intensity using satellite imagery. (Ecosystem Management, University of New England: Armidale, NSW).

Russell-Smith, J., Ryan, P.G., and Durieu, R. (1997).

A LANDSAT MSS-derived fire history of Kakadu National Park, monsoonal northern Australia, 1980-94: seasonal extent, frequency and patchiness. Journal of Applied Ecology 34, 748-766.

Smith, R.C.G., Craig, R., Steber, M., Marsden, J., McMillan C. and Adams, J. (1998)

Use of NOAA-AVHRR for fire management in Australia’s tropical savannas. 9th Australasian Remote Sensing and Photogrammetry Conference. University of New South Wales, Sydney. Paper No. 123.

Tinner, W., Conedera, M., Ammann, B., Gaggeler, H.W., Gedye, S., Jones, R. and Sagesser, B. (1998).

Pollen & charcoal in lake sediment compared with historically documented forest fires in southern Switzerland since AD 1920. Holocene 8, 31-42.

Townshend, J.R.G., Goff, T.E., and Tucker, C.J. (1985).

Multitemporal dimensionality of images of normalised difference vegetation index at continental scales. IEEE Transactions on Geoscience and Remote Sensing 23, 222-295.

Tucker, C.J. (1979).

Red and photographic infrared linear combinations for monitoring vegetation. Remote Sensing of Environment 8, 127-150.

Viedma, O., Melia, J., Segarra, D., and Garcia-Haro, J. (1997).

Modelling rates of ecosystem recovery after fires by using Landsat TM data. Remote Sensing of the Environment 61, 383-398.

White, J.D., Ryan, K.C., Key, C.C., and Running, S.W. (1996).

Remote Sensing of forest fire severity and vegetation recovery. International Journal of Wildland Fire 6,125-136.

Published by School of Environmental & Information Sciences     Charles Sturt University