Resource Sciences and Knowledge
Queensland Department of Natural Resources
Natural Sciences Precinct
80 Meiers Road, Indooroopilly, 4068
http://www.dnr.qld.gov.au/slats

The Statewide Landcover and Trees Study (SLATS) is a major vegetation monitoring initiative of the Queensland Department of Natural Resources (QDNR). SLATS is gathering accurate vegetation cover and cover change information for vegetation management planning and for greenhouse gas inventory purposes.

Landsat Thematic Mapper (TM) satellite imagery is being used to compare the vegetation cover between 1988, 1991, 1995 and 1997, and to provide baseline landcover mapping over the entire State of Queensland. The Landsat TM imagery has a resolution of 30 metres, enabling most areas of vegetation change (one hectare or greater) to be detected. Typically, it is used to produce maps at a scale of 1:100 000 or smaller. This study provides a consistent data set covering the entire State at a medium resolution, but it is not intended to be a substitute for high resolution remnant bushland studies using aerial photography. Landsat TM satellite imagery should be used with caution when mapping narrow vegetation corridors, such as riparian vegetation, as the resolution of the data may be less than the size of the vegetation to be mapped.

A project advisory committee was established to monitor SLATS progress. The committee provides input with regard to overall direction and methods and provides advice on the release of project results. The committee consists of representatives of DNR, Department of Primary Industries (DPI), Environment Protection Agency (EPA), Queensland Conservation Council, Cattlemen,s Union, United Graziers Association, Queensland Canegrowers, Local Government Association, academia, Land and Water Resources Research and Development Corporation, and Queensland Grain Growers Association.

An analysis of 1995-97 woody vegetation change has been completed for the entire State and is the subject of this report. Imagery for the 1991-95 period has previously been analysed for vegetation change and reported (QDNR, 1999B). A sample of scenes has been processed for 1988-91 vegetation change with approximately 12 per cent of the State completed. Detailed baseline landcover mapping which discriminates areas of trees from pasture, crop, water, settlement areas etc. is being done using 1991 imagery. This has been completed for approximately 70 per cent of the State, which includes most areas affected by clearing in the 1995-97 period.

Landsat scenes containing approximately 87 per cent of the 1995-97 clearing have been field checked. There may be minor updates to the 1995-97 vegetation change data when the next change period (1997-99) is analysed and consistency checks are completed.

A separate report concerning land cover change in south east Queensland (QDNR, 1999A) over the period 1988-97 has been published. This also includes information on 1995-97 vegetation change and provides a more detailed breakdown of clearing by catchment within south east Queensland.

< to top of this page >< to reports page >< to SLATS homepage >

There are many definitions of what constitutes a forest or woody vegetation. A common definition used by foresters is 20 per cent crown cover which equates to approximately 12 per cent foliage projective cover (FPC). It is clear in pages 5.14 and 5.25 of the Intergovernmental Panel on Climate Change guidelines (IPCC, 1996) that where possible, all changes in woody biomass stocks resulting from human interventions should be measured and included in the national inventory. Hence, we have mapped vegetation change for all perennial woody plants of all sizes which can be distinguished with Landsat TM imagery.

The statistics for vegetation change and woody vegetation cover quoted in this report include all woody vegetation. This includes remaining areas of native vegetation, disturbed areas of native vegetation, regrowth, plantations of native and exotic species and domestic woody vegetation. In higher rainfall coastal areas, woody vegetation of 12 per cent FPC or greater can be detected with a high level of confidence but more sparse vegetation is difficult to distinguish from pastures. However, there is very little vegetation less than 12 per cent FPC in coastal areas due to fast vegetation growth rates. In the lower rainfall areas, it is possible to detect clearing of vegetation with an FPC of 5 per cent or greater.

Landsat TM satellite imagery was purchased from the Australian Centre for Remote Sensing (ACRES). In most cases 1995 and 1997 imagery was used but when cloud-free 1995 and 1997 imagery was not available, 1994 or 1998 imagery was purchased. Approximately 15 percent of the State is covered with 1994 imagery, and 7 percent of the State with 1998 imagery. The remainder is 1995 and 1997 imagery. The 1994 and 1998 data are referred to throughout the report as 1995 and 1997 imagery. Winter imagery was chosen to maximise discrimination between grasses and the woody component of the vegetation.

Imagery must undergo several pre-processing steps prior to use in change detection and mapping procedures. The imagery is first corrected for variation in solar zenith angle and then standardised for atmospheric variations between dates by radiometrically registering all dates to a reference year. Corrections are applied to improve scene-to-scene matching along each path, and an across-scene correction is also applied (Collett et al., 1998).

Geometric correction involves registration of all scenes to the 1991 reference year. Semi-automated image correlation methods are being used to register imagery to better than 0.4 pixels root mean square error. The majority of subsequent processing uses these registered images. To enable rectification of change detection classifications, vegetation mapping and other products, scene to map transformations are calculated. Preliminary transformations are generated using ground control points identified on 1:100 000 and 1:250 000 scale topographic maps. The accuracy of these points is upgraded in the field using differential global positioning systems (Fugro and Garmin type, with sub 10m accuracy) to maintain a constant spatial accuracy of at least 1:100 000 scale (Kuhnell et al., 1998).

The vegetation change has been mapped using classification procedures described in Paudyal et al. (1997). A semi-automated computer classification was used to identify areas of change and a sample of these areas was then field checked.

Each scene was ground truthed;

• to verify vegetation change classification accuracy;
• to measure differential GPS control points;
• to collect vegetation site data for calibration and validation purposes;
• by liaising with local DNR offices and landholders.

A detailed baseline landcover survey was done using 1991 imagery and vegetation site data. This involves the development of woody/non-woody masks, foliage projective cover (overstorey and shrub) and tree basal area layers. A full description of field methods and landcover mapping methods is given by Kuhnell et al. (1998).

To determine vegetation change, the woody vegetation mask was used in conjunction with the change classification to isolate changes in woody vegetation and produce a final change classification. Maps of vegetation change were then edited using visual interpretation techniques aided by data gathered in the field. Finally, the change classification for each scene was checked by an independent operator.

Both clearing and regrowth of woody vegetation were classified. Regrowth is more difficult to measure due to slow rates of change (relative to clearing) and the low initial density of some regrowth stands. The "new woody regrowth" figures included in the tables are regrowth generation, i.e. areas which have changed from non-woody to woody within the 1995-97 period.

The proportion of clearing which is re-clearing of previously cleared areas (regrowth clearing) is unknown. Regrowth clearing is best mapped using a sequence of imagery and identifying cleared-woody-cleared transitions. In the 1995-97 analysis it was only possible to identify areas which were re-cleared since the 1988 or 1991 imagery. Historical Landsat Multi Spectral Scanner (MSS) imagery (1972, 1980, 1984) has been acquired for the entire State. This imagery and the more recent Landsat TM imagery will be analysed to determine the proportion of clearing which is regrowth control, and to provide estimates of historical clearing rates.

While fires can remove a significant proportion of the woody vegetation foliage it is usually a temporary affect, and in most cases the foliage recovers quickly. SLATS site data shows that on average a fire removes less than 2m sq. / ha basal area. Hence it is not common for fire to change the land cover from woody to non-woody in a single event. There were a few cases of very hot fire resulting in land cover change and these areas were assigned a replacement landcover class of pasture.

A significant amount of natural tree death occurred in the area surrounding Charters Towers due to a prolonged drought in this area. The area affected by natural tree death is not included in the "clearing" figures and is covered in a separate section in this report.

Every patch identified in the change analysis as clearing was assigned one of the replacement landcover classes in Table 1. The assignment of these classes is primarily based on visual interpretation. State forest boundaries are used to re-code some of the native forest clearing which would otherwise appear as clearing to pasture. In areas where there are many different forms of land use, it is sometimes difficult to interpret the correct replacement class. Sometimes land is cleared to pasture then later converted to urban development. These factors will cause the interpretation accuracy for replacement class to be lower than the accuracy for identification of woody vegetation change.

Large, seamless mosaics of vegetation change, landcover and vegetation cover were created by joining together the 75 scenes that contained some vegetation change. Scenes were overlapped with preference for the more recent final date in the overlap (e.g 1995-98 used in preference to 1995-97). The size of the digital file for each mosaic was approximately 8 Gigabytes. In order to calculate annual tree clearing rates, a vector geographic information system (GIS) layer containing the extent and dates (Julian date) of mosaiced scenes was created. The mosaic raster of cleared areas was intersected with GIS overlays of one kilometre resolution depicting of date, tenure type, 30' x 30' grid cell, soil type and structural vegetation classifications in order to generate tabular statistics. A spatial resolution of one kilometre was used as it was appropriate given the scale of most input data and was more efficient to compute than full 30m resolution. The tabular statistics show slightly different State clearing totals which is due to the different scales of the GIS overlays used.

Note that all statistics were generated based on an Albers equal-area projection, so clearing rates for different regions could be compared; and all the vegetation change statistics in this report except for natural tree death, have been converted to annual rates to account for the variation in scene dates.

The traditional form of accuracy assessment is an assessment against an independent data source of higher resolution, but this is not always possible. While some change areas on each scene are checked in the field, access and cost limit the extent of this checking to a representative sample only. The aerial photography coverage available does not usually align with the same dates of the satellite imagery, so in most cases it is not a viable option and other alternatives need to be considered. A formal accuracy assessment is proposed as part of the national Remote Sensing of Agricultural Land Cover Change project (Barson, 1996) and will use independent methods rather than independent data. DNR was a partner in this project so this will be done with the 1991-95 change data that DNR supplied to this project. It is expected to be completed by December 1999.

A considerable amount of work has gone into ensuring the quality of the change analysis. The satellite imagery was selected at dates which maximise discrimination between the grass and woody layers. Fully documented procedures have been developed to analyse the data and these are available to operators on the project Intranet. Many of the procedures have been scripted with error traps and log files to avoid errors and allow errors to be traced. The change detection method used offers the advantage of automated and visual methods combined with independent checks. Initially, the analysis is biased so that too much change is mapped. Following fieldwork, the incorrect areas are rejected. This ensures that very little change is missed and our fieldwork confirms this is the case.

Hence, it seems the largest source of variation is not misclassification of change but determining the extent (area) of change at a clearing location. The woody vegetation mask is very important for determining the area of vegetation change, as it delineates how much woody vegetation existed before clearing. Considerable effort has gone into ensuring that the woody vegetation mask and FPC layer is well calibrated to ground vegetation measurements. These ground vegetation measurements and the radiometric correction methods used ensure the FPC layer is well matched across scene boundaries. This process ensures a high level of accuracy and consistency in the change area measurements. In the future, laser profiling techniques may provide a further means of calibrating and validating the woody FPC layer. There has been an initial investigation of these laser altimetry methods by Tickle et al. (1998) and a detailed two year testing program is now under way.

The average annual clearing rate over the period 1995 to 1997 was calculated to be 340 000 hectares per year or 0.19 per cent of the land area of Queensland per year. A previous study (Danaher et al., 1992) has shown that the area of wooded vegetation in Queensland is 76 million hectares. Of this, approximately 60 million hectares of vegetation resources occur in agro-ecological zones where some land development could occur. Hence, the annual clearing rate is 0.56 percent of this area per year. In total, the annual clearing figure represents a clearing area of 58km x 58km. If a woody vegetation definition of 12 per cent FPC or approximately 20 per cent crown cover were used, the annual average clearing rate from 1995 to 1997 would be 299 000 hectares per year. A spatial view of where the clearing is occurring within Queensland is shown in Figure 1. It shows the rate of clearing (km²/year) per 30' x 30' (latitude /longitude) grid cell. These 30' x 30' cells are the same size as a 1:100 000 map sheet.

The area of change can also be shown as a proportion of the 1991 woody vegetation cleared each year over the 1995-97 period, as in Figure 2. This map was created by using the 1991 baseline landcover mapping (Kuhnell et al., 1998), produced by SLATS, and combining this with the 1995-97 clearing data. A map of 1991 woody vegetation by 30' x 30' grid cell is shown in Figure 3. This map is not yet complete for the entire State but is complete for all the major clearing areas. Please take care when interpreting the results in Figures 2 and 3 as some of the differences in percentage of 1991 woody vegetation are related to the original type of vegetation cover, not clearing. For example, there are significant areas of natural grassland which were never woody.

Figure 1: Average annual clearing rate (1995-1997)

The 1995-97 woody vegetation change rates have been grouped by tenure in Table 2. Table 3 shows the trend in clearing between the 1991-95 and 1995-97 periods. The information in these tables was derived by combining the woody vegetation change data with the Digital Cadastral Data Base and Tenures Administration System data. Figure 4 is a map of the four broad tenure classes used in Table 2 and Table 3. The area of predominantly freehold tenure is also shown in Figures 1 to 3 using hatching.

The replacement landcover after clearing for the periods 1991-95 and 1995-97 is summarised in Table 4. It shows that the majority of clearing is for conversion to pasture for grazing but there has been a large increase in clearing for cropping and infrastructure between the 1991-95 and 1995-97 periods.

The 1991-95 and 1995-97 woody vegetation change data was cross-tabulated with the 1991 woody vegetation cover (FPC) layer which was generated from the Landsat TM imagery, to produce a histogram showing clearing in relation to foliage cover (Figure 5). The highest frequency of clearing occurs at a FPC value of 24 per cent for 1991-95 and at 28 per cent for 1995-97. There is a spike in the 1995-97 curve at zero FPC. These are areas of young regrowth which were cleared in the 1995-97 period but were not mapped as woody vegetation in 1991. Development of the FPC layer is described by Kuhnell et al. (1998).

A woody basal area layer was derived from the foliage cover (FPC) layer using relationships between FPC and basal area based on site measurements. This relationship only holds for mature vegetation, so areas which are known to be young regrowth were set to a nominal basal area of 1m²/ha (the spike in figure 6). These young regrowth areas were identified as non-woody using both the 1988 and 1991 imagery. In the 1995-97 analysis, they account for approximately 18 per cent of all clearing.

The basal area layer will continue to be improved using historical Landsat TM and MSS analyses to identify previously cleared vegetation. Kuhnell et al. (1998) describe the development of the basal area data. Figure 6 shows the frequency distribution of clearing rate in relation to basal area. The highest frequency of clearing occurs at the values of 9m²/ha for the 1991-95 change and of 10m²/ha for the 1995-97 change. An example of a typical woodland of basal area 9m²/ha is shown in Figure 7.

The 1991-95 and 1995-97 clearing rasters were intersected with the foliage cover classes from the 1:5 000 000 Carnahan Present Vegetation Map (Carnahan, 1988) to produce Table 6. All vegetation classes which included some clearing have been included in this table. In the National and State inventories, clearing is grouped into three vegetation types which are combinations of the Carnahan classification.

The three groups are tropical and temperate closed forests (T4,M4,L4,T3), dense woodlands and open forests (M3,L3), and open woodland (M2,L2). In Queensland, the only additional classes which have significant clearing are added to the open woodland class. The Carnahan classes when mapped are geographically large and represent the dominant vegetation formation but may include significant areas of other vegetation types. Apparent anomalies, such as no clearing in T4 in Queensland, are due to this problem.

A table of clearing by soil type (Table 5) was created by intersecting the 1991-95 and 1995-97 clearing areas with data from the 1:2 000 000 Atlas of Australian Soils. This table shows the trend in clearing from the 1991-95 period to the 1995-97 period.

It is quite common for trees to lose foliage during dry spells, then recover. The area where this happens is far greater than where tree death is occurring and it is quite difficult to distinguish tree death from normal variations in the foliage cover using remote sensing alone. Considerable field work and additional imagery is required to map areas affected by natural tree death.

In most parts of the State, natural tree death occurred in small, isolated areas. It was not feasible to check all places where it potentially occurred as they were scattered and mostly in inaccessible locations. Hence, it was mostly included with the areas of seasonal change and thus recorded as no change in land cover. A small proportion may have been included in the clearing to pasture category.

There was a significant amount of natural tree death in the area covered by Dalrymple shire (west of Townsville) due to a prolonged drought. A detailed analysis of this phenomenon was carried out as field data was available from EPA vegetation mapping and DNR bare ground mapping projects. Also additional 1997 and 1998 imagery was available. This information was not included with the other "clearing" data in the tables as it was not mapped to the same standard across the whole State.

In total, an area (not annual rate) of 69 000 ha was affected over the period 1991-97, with most death believed to have occurred in the 1994-97 period. The tree deaths seem to be selective with mainly the ironbark species and larger trees affected. Although trees are killed, it does not always result in change from woody to non-woody at the Landsat TM spatial resolution.

The methods used to map these areas relied heavily on field observations and then extrapolation using satellite imagery. Additional imagery is also required to confirm that the changes are long term and not just seasonal, e.g. 1997 and 1998 imagery is used to confirm possible areas of tree death shown by the 1991-95 change analysis.